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 */ 21d6694327SYuri Pankov 227c478bd9Sstevel@tonic-gate /* 230db3240dSStephen Hanson * Copyright (c) 1993, 2010, Oracle and/or its affiliates. All rights reserved. 249e3e4df2SGarrett D'Amore * Copyright 2012 DEY Storage Systems, Inc. All rights reserved. 25d6694327SYuri Pankov * Copyright 2017 Nexenta Systems, Inc. 2674ecdb51SJohn Levon * Copyright (c) 2018 Joyent, Inc. 27af868f46SMatthew Ahrens * Copyright (c) 2015 by Delphix. All rights reserved. 287c478bd9Sstevel@tonic-gate */ 29a3114836SGerry Liu /* 30a3114836SGerry Liu * Copyright (c) 2010, Intel Corporation. 31a3114836SGerry Liu * All rights reserved. 32a3114836SGerry Liu */ 337c478bd9Sstevel@tonic-gate 347c478bd9Sstevel@tonic-gate #include <sys/types.h> 357c478bd9Sstevel@tonic-gate #include <sys/t_lock.h> 367c478bd9Sstevel@tonic-gate #include <sys/param.h> 377c478bd9Sstevel@tonic-gate #include <sys/sysmacros.h> 387c478bd9Sstevel@tonic-gate #include <sys/signal.h> 397c478bd9Sstevel@tonic-gate #include <sys/systm.h> 407c478bd9Sstevel@tonic-gate #include <sys/user.h> 417c478bd9Sstevel@tonic-gate #include <sys/mman.h> 427c478bd9Sstevel@tonic-gate #include <sys/vm.h> 437c478bd9Sstevel@tonic-gate #include <sys/conf.h> 447c478bd9Sstevel@tonic-gate #include <sys/avintr.h> 457c478bd9Sstevel@tonic-gate #include <sys/autoconf.h> 467c478bd9Sstevel@tonic-gate #include <sys/disp.h> 477c478bd9Sstevel@tonic-gate #include <sys/class.h> 487c478bd9Sstevel@tonic-gate #include <sys/bitmap.h> 497c478bd9Sstevel@tonic-gate 507c478bd9Sstevel@tonic-gate #include <sys/privregs.h> 517c478bd9Sstevel@tonic-gate 527c478bd9Sstevel@tonic-gate #include <sys/proc.h> 537c478bd9Sstevel@tonic-gate #include <sys/buf.h> 547c478bd9Sstevel@tonic-gate #include <sys/kmem.h> 55ae115bc7Smrj #include <sys/mem.h> 567c478bd9Sstevel@tonic-gate #include <sys/kstat.h> 577c478bd9Sstevel@tonic-gate 587c478bd9Sstevel@tonic-gate #include <sys/reboot.h> 597c478bd9Sstevel@tonic-gate 607c478bd9Sstevel@tonic-gate #include <sys/cred.h> 617c478bd9Sstevel@tonic-gate #include <sys/vnode.h> 627c478bd9Sstevel@tonic-gate #include <sys/file.h> 637c478bd9Sstevel@tonic-gate 647c478bd9Sstevel@tonic-gate #include <sys/procfs.h> 657c478bd9Sstevel@tonic-gate 667c478bd9Sstevel@tonic-gate #include <sys/vfs.h> 677c478bd9Sstevel@tonic-gate #include <sys/cmn_err.h> 687c478bd9Sstevel@tonic-gate #include <sys/utsname.h> 697c478bd9Sstevel@tonic-gate #include <sys/debug.h> 707c478bd9Sstevel@tonic-gate #include <sys/kdi.h> 717c478bd9Sstevel@tonic-gate 727c478bd9Sstevel@tonic-gate #include <sys/dumphdr.h> 737c478bd9Sstevel@tonic-gate #include <sys/bootconf.h> 74a3114836SGerry Liu #include <sys/memlist_plat.h> 757c478bd9Sstevel@tonic-gate #include <sys/varargs.h> 767c478bd9Sstevel@tonic-gate #include <sys/promif.h> 77843e1988Sjohnlev #include <sys/modctl.h> 787c478bd9Sstevel@tonic-gate 797c478bd9Sstevel@tonic-gate #include <sys/sunddi.h> 807c478bd9Sstevel@tonic-gate #include <sys/sunndi.h> 817c478bd9Sstevel@tonic-gate #include <sys/ndi_impldefs.h> 827c478bd9Sstevel@tonic-gate #include <sys/ddidmareq.h> 837c478bd9Sstevel@tonic-gate #include <sys/psw.h> 847c478bd9Sstevel@tonic-gate #include <sys/regset.h> 857c478bd9Sstevel@tonic-gate #include <sys/clock.h> 867c478bd9Sstevel@tonic-gate #include <sys/pte.h> 877c478bd9Sstevel@tonic-gate #include <sys/tss.h> 887c478bd9Sstevel@tonic-gate #include <sys/stack.h> 897c478bd9Sstevel@tonic-gate #include <sys/trap.h> 907c478bd9Sstevel@tonic-gate #include <sys/fp.h> 911d03c31eSjohnlev #include <vm/kboot_mmu.h> 927c478bd9Sstevel@tonic-gate #include <vm/anon.h> 937c478bd9Sstevel@tonic-gate #include <vm/as.h> 947c478bd9Sstevel@tonic-gate #include <vm/page.h> 957c478bd9Sstevel@tonic-gate #include <vm/seg.h> 967c478bd9Sstevel@tonic-gate #include <vm/seg_dev.h> 977c478bd9Sstevel@tonic-gate #include <vm/seg_kmem.h> 987c478bd9Sstevel@tonic-gate #include <vm/seg_kpm.h> 997c478bd9Sstevel@tonic-gate #include <vm/seg_map.h> 1007c478bd9Sstevel@tonic-gate #include <vm/seg_vn.h> 1017c478bd9Sstevel@tonic-gate #include <vm/seg_kp.h> 1027c478bd9Sstevel@tonic-gate #include <sys/memnode.h> 1037c478bd9Sstevel@tonic-gate #include <vm/vm_dep.h> 1047c478bd9Sstevel@tonic-gate #include <sys/thread.h> 1057c478bd9Sstevel@tonic-gate #include <sys/sysconf.h> 1067c478bd9Sstevel@tonic-gate #include <sys/vm_machparam.h> 1077c478bd9Sstevel@tonic-gate #include <sys/archsystm.h> 1087c478bd9Sstevel@tonic-gate #include <sys/machsystm.h> 1097c478bd9Sstevel@tonic-gate #include <vm/hat.h> 1107c478bd9Sstevel@tonic-gate #include <vm/hat_i86.h> 1117c478bd9Sstevel@tonic-gate #include <sys/pmem.h> 1127c478bd9Sstevel@tonic-gate #include <sys/smp_impldefs.h> 1137c478bd9Sstevel@tonic-gate #include <sys/x86_archext.h> 114a563a037Sbholler #include <sys/cpuvar.h> 1157c478bd9Sstevel@tonic-gate #include <sys/segments.h> 1167c478bd9Sstevel@tonic-gate #include <sys/clconf.h> 1177c478bd9Sstevel@tonic-gate #include <sys/kobj.h> 1187c478bd9Sstevel@tonic-gate #include <sys/kobj_lex.h> 1197c478bd9Sstevel@tonic-gate #include <sys/cpc_impl.h> 1207aec1d6eScindi #include <sys/cpu_module.h> 12184ab085aSmws #include <sys/smbios.h> 122ae115bc7Smrj #include <sys/debug_info.h> 1231d03c31eSjohnlev #include <sys/bootinfo.h> 124a288e5a9SJoshua M. Clulow #include <sys/ddi_periodic.h> 1255679c89fSjv #include <sys/systeminfo.h> 12619397407SSherry Moore #include <sys/multiboot.h> 1270181461bSKeith M Wesolowski #include <sys/ramdisk.h> 128ae115bc7Smrj 129a3114836SGerry Liu #ifdef __xpv 1301d03c31eSjohnlev 131843e1988Sjohnlev #include <sys/hypervisor.h> 132843e1988Sjohnlev #include <sys/xen_mmu.h> 133843e1988Sjohnlev #include <sys/evtchn_impl.h> 134843e1988Sjohnlev #include <sys/gnttab.h> 135843e1988Sjohnlev #include <sys/xpv_panic.h> 136843e1988Sjohnlev #include <xen/sys/xenbus_comms.h> 137843e1988Sjohnlev #include <xen/public/physdev.h> 1381d03c31eSjohnlev 139843e1988Sjohnlev extern void xen_late_startup(void); 140ae115bc7Smrj 1411d03c31eSjohnlev struct xen_evt_data cpu0_evt_data; 1421d03c31eSjohnlev 143a3114836SGerry Liu #else /* __xpv */ 144a3114836SGerry Liu #include <sys/memlist_impl.h> 145a3114836SGerry Liu 146a3114836SGerry Liu extern void mem_config_init(void); 1471d03c31eSjohnlev #endif /* __xpv */ 1487c478bd9Sstevel@tonic-gate 1497c478bd9Sstevel@tonic-gate extern void progressbar_init(void); 1509acbbeafSnn extern void brand_init(void); 15106fb6a36Sdv extern void pcf_init(void); 1520e751525SEric Saxe extern void pg_init(void); 1536a59053bSAlex Wilson extern void ssp_init(void); 1547c478bd9Sstevel@tonic-gate 155d7d93655Sblakej extern int size_pse_array(pgcnt_t, int); 156d7d93655Sblakej 157ed5289f9SKen Erickson #if defined(_SOFT_HOSTID) 158ed5289f9SKen Erickson 159ed5289f9SKen Erickson #include <sys/rtc.h> 160ed5289f9SKen Erickson 161ed5289f9SKen Erickson static int32_t set_soft_hostid(void); 162ed5289f9SKen Erickson static char hostid_file[] = "/etc/hostid"; 163ed5289f9SKen Erickson 164ed5289f9SKen Erickson #endif 165ed5289f9SKen Erickson 16694f1124eSVikram Hegde void *gfx_devinfo_list; 167ed5289f9SKen Erickson 1683a634bfcSVikram Hegde #if defined(__amd64) && !defined(__xpv) 1693a634bfcSVikram Hegde extern void immu_startup(void); 1703a634bfcSVikram Hegde #endif 1713a634bfcSVikram Hegde 1727c478bd9Sstevel@tonic-gate /* 1737c478bd9Sstevel@tonic-gate * XXX make declaration below "static" when drivers no longer use this 1747c478bd9Sstevel@tonic-gate * interface. 1757c478bd9Sstevel@tonic-gate */ 1767c478bd9Sstevel@tonic-gate extern caddr_t p0_va; /* Virtual address for accessing physical page 0 */ 1777c478bd9Sstevel@tonic-gate 1787c478bd9Sstevel@tonic-gate /* 1797c478bd9Sstevel@tonic-gate * segkp 1807c478bd9Sstevel@tonic-gate */ 1817c478bd9Sstevel@tonic-gate extern int segkp_fromheap; 1827c478bd9Sstevel@tonic-gate 1837c478bd9Sstevel@tonic-gate static void kvm_init(void); 1847c478bd9Sstevel@tonic-gate static void startup_init(void); 1857c478bd9Sstevel@tonic-gate static void startup_memlist(void); 186ae115bc7Smrj static void startup_kmem(void); 1877c478bd9Sstevel@tonic-gate static void startup_modules(void); 1887c478bd9Sstevel@tonic-gate static void startup_vm(void); 1897c478bd9Sstevel@tonic-gate static void startup_end(void); 19035b1ab99Sjosephb static void layout_kernel_va(void); 1917c478bd9Sstevel@tonic-gate 1927c478bd9Sstevel@tonic-gate /* 1937c478bd9Sstevel@tonic-gate * Declare these as initialized data so we can patch them. 1947c478bd9Sstevel@tonic-gate */ 195f53ad214Skchow #ifdef __i386 19635b1ab99Sjosephb 197f53ad214Skchow /* 198f53ad214Skchow * Due to virtual address space limitations running in 32 bit mode, restrict 19935b1ab99Sjosephb * the amount of physical memory configured to a max of PHYSMEM pages (16g). 200f53ad214Skchow * 201f53ad214Skchow * If the physical max memory size of 64g were allowed to be configured, the 202f53ad214Skchow * size of user virtual address space will be less than 1g. A limited user 203f53ad214Skchow * address space greatly reduces the range of applications that can run. 204f53ad214Skchow * 20535b1ab99Sjosephb * If more physical memory than PHYSMEM is required, users should preferably 20635b1ab99Sjosephb * run in 64 bit mode which has far looser virtual address space limitations. 207f53ad214Skchow * 208f53ad214Skchow * If 64 bit mode is not available (as in IA32) and/or more physical memory 20935b1ab99Sjosephb * than PHYSMEM is required in 32 bit mode, physmem can be set to the desired 210f53ad214Skchow * value or to 0 (to configure all available memory) via eeprom(1M). kernelbase 211f53ad214Skchow * should also be carefully tuned to balance out the need of the user 212f53ad214Skchow * application while minimizing the risk of kernel heap exhaustion due to 213f53ad214Skchow * kernelbase being set too high. 214f53ad214Skchow */ 21535b1ab99Sjosephb #define PHYSMEM 0x400000 216f53ad214Skchow 21735b1ab99Sjosephb #else /* __amd64 */ 21835b1ab99Sjosephb 21935b1ab99Sjosephb /* 22035b1ab99Sjosephb * For now we can handle memory with physical addresses up to about 22135b1ab99Sjosephb * 64 Terabytes. This keeps the kernel above the VA hole, leaving roughly 22235b1ab99Sjosephb * half the VA space for seg_kpm. When systems get bigger than 64TB this 22335b1ab99Sjosephb * code will need revisiting. There is an implicit assumption that there 22435b1ab99Sjosephb * are no *huge* holes in the physical address space too. 22535b1ab99Sjosephb */ 22635b1ab99Sjosephb #define TERABYTE (1ul << 40) 22735b1ab99Sjosephb #define PHYSMEM_MAX64 mmu_btop(64 * TERABYTE) 22835b1ab99Sjosephb #define PHYSMEM PHYSMEM_MAX64 22935b1ab99Sjosephb #define AMD64_VA_HOLE_END 0xFFFF800000000000ul 23035b1ab99Sjosephb 23135b1ab99Sjosephb #endif /* __amd64 */ 23235b1ab99Sjosephb 23335b1ab99Sjosephb pgcnt_t physmem = PHYSMEM; 2347c478bd9Sstevel@tonic-gate pgcnt_t obp_pages; /* Memory used by PROM for its text and data */ 2357c478bd9Sstevel@tonic-gate 2367c478bd9Sstevel@tonic-gate char *kobj_file_buf; 2377c478bd9Sstevel@tonic-gate int kobj_file_bufsize; /* set in /etc/system */ 2387c478bd9Sstevel@tonic-gate 2397c478bd9Sstevel@tonic-gate /* Global variables for MP support. Used in mp_startup */ 24019397407SSherry Moore caddr_t rm_platter_va = 0; 2417c478bd9Sstevel@tonic-gate uint32_t rm_platter_pa; 2427c478bd9Sstevel@tonic-gate 24383f9b804Skchow int auto_lpg_disable = 1; 24483f9b804Skchow 2457c478bd9Sstevel@tonic-gate /* 2467c478bd9Sstevel@tonic-gate * Some CPUs have holes in the middle of the 64-bit virtual address range. 2477c478bd9Sstevel@tonic-gate */ 2487c478bd9Sstevel@tonic-gate uintptr_t hole_start, hole_end; 2497c478bd9Sstevel@tonic-gate 2507c478bd9Sstevel@tonic-gate /* 2517c478bd9Sstevel@tonic-gate * kpm mapping window 2527c478bd9Sstevel@tonic-gate */ 2537c478bd9Sstevel@tonic-gate caddr_t kpm_vbase; 2547c478bd9Sstevel@tonic-gate size_t kpm_size; 25535b1ab99Sjosephb static int kpm_desired; 25635b1ab99Sjosephb #ifdef __amd64 25735b1ab99Sjosephb static uintptr_t segkpm_base = (uintptr_t)SEGKPM_BASE; 25835b1ab99Sjosephb #endif 2597c478bd9Sstevel@tonic-gate 2607c478bd9Sstevel@tonic-gate /* 2617c478bd9Sstevel@tonic-gate * Configuration parameters set at boot time. 2627c478bd9Sstevel@tonic-gate */ 2637c478bd9Sstevel@tonic-gate 2647c478bd9Sstevel@tonic-gate caddr_t econtig; /* end of first block of contiguous kernel */ 2657c478bd9Sstevel@tonic-gate 2667c478bd9Sstevel@tonic-gate struct bootops *bootops = 0; /* passed in from boot */ 2677c478bd9Sstevel@tonic-gate struct bootops **bootopsp; 2687c478bd9Sstevel@tonic-gate struct boot_syscalls *sysp; /* passed in from boot */ 2697c478bd9Sstevel@tonic-gate 2707c478bd9Sstevel@tonic-gate char bootblock_fstype[16]; 2717c478bd9Sstevel@tonic-gate 2727c478bd9Sstevel@tonic-gate char kern_bootargs[OBP_MAXPATHLEN]; 27319397407SSherry Moore char kern_bootfile[OBP_MAXPATHLEN]; 2747c478bd9Sstevel@tonic-gate 275ad23a2dbSjohansen /* 276ad23a2dbSjohansen * ZFS zio segment. This allows us to exclude large portions of ZFS data that 277ad23a2dbSjohansen * gets cached in kmem caches on the heap. If this is set to zero, we allocate 278ad23a2dbSjohansen * zio buffers from their own segment, otherwise they are allocated from the 279ad23a2dbSjohansen * heap. The optimization of allocating zio buffers from their own segment is 280ad23a2dbSjohansen * only valid on 64-bit kernels. 281ad23a2dbSjohansen */ 282ad23a2dbSjohansen #if defined(__amd64) 283ad23a2dbSjohansen int segzio_fromheap = 0; 284ad23a2dbSjohansen #else 285ad23a2dbSjohansen int segzio_fromheap = 1; 286ad23a2dbSjohansen #endif 287ad23a2dbSjohansen 2883ce2fcdcSRobert Mustacchi /* 2893ce2fcdcSRobert Mustacchi * Give folks an escape hatch for disabling SMAP via kmdb. Doesn't work 2903ce2fcdcSRobert Mustacchi * post-boot. 2913ce2fcdcSRobert Mustacchi */ 2923ce2fcdcSRobert Mustacchi int disable_smap = 0; 2933ce2fcdcSRobert Mustacchi 2947c478bd9Sstevel@tonic-gate /* 2957c478bd9Sstevel@tonic-gate * new memory fragmentations are possible in startup() due to BOP_ALLOCs. this 2967c478bd9Sstevel@tonic-gate * depends on number of BOP_ALLOC calls made and requested size, memory size 2977c478bd9Sstevel@tonic-gate * combination and whether boot.bin memory needs to be freed. 2987c478bd9Sstevel@tonic-gate */ 2997c478bd9Sstevel@tonic-gate #define POSS_NEW_FRAGMENTS 12 3007c478bd9Sstevel@tonic-gate 3017c478bd9Sstevel@tonic-gate /* 3027c478bd9Sstevel@tonic-gate * VM data structures 3037c478bd9Sstevel@tonic-gate */ 3047c478bd9Sstevel@tonic-gate long page_hashsz; /* Size of page hash table (power of two) */ 305cb15d5d9SPeter Rival unsigned int page_hashsz_shift; /* log2(page_hashsz) */ 3067c478bd9Sstevel@tonic-gate struct page *pp_base; /* Base of initial system page struct array */ 3077c478bd9Sstevel@tonic-gate struct page **page_hash; /* Page hash table */ 308d7d93655Sblakej pad_mutex_t *pse_mutex; /* Locks protecting pp->p_selock */ 309d7d93655Sblakej size_t pse_table_size; /* Number of mutexes in pse_mutex[] */ 310d7d93655Sblakej int pse_shift; /* log2(pse_table_size) */ 3117c478bd9Sstevel@tonic-gate struct seg ktextseg; /* Segment used for kernel executable image */ 3127c478bd9Sstevel@tonic-gate struct seg kvalloc; /* Segment used for "valloc" mapping */ 3137c478bd9Sstevel@tonic-gate struct seg kpseg; /* Segment used for pageable kernel virt mem */ 3147c478bd9Sstevel@tonic-gate struct seg kmapseg; /* Segment used for generic kernel mappings */ 3157c478bd9Sstevel@tonic-gate struct seg kdebugseg; /* Segment used for the kernel debugger */ 3167c478bd9Sstevel@tonic-gate 3177c478bd9Sstevel@tonic-gate struct seg *segkmap = &kmapseg; /* Kernel generic mapping segment */ 318ae115bc7Smrj static struct seg *segmap = &kmapseg; /* easier to use name for in here */ 319ae115bc7Smrj 3207c478bd9Sstevel@tonic-gate struct seg *segkp = &kpseg; /* Pageable kernel virtual memory segment */ 3217c478bd9Sstevel@tonic-gate 3227c478bd9Sstevel@tonic-gate struct seg kvseg_core; /* Segment used for the core heap */ 3237c478bd9Sstevel@tonic-gate struct seg kpmseg; /* Segment used for physical mapping */ 3247c478bd9Sstevel@tonic-gate struct seg *segkpm = &kpmseg; /* 64bit kernel physical mapping segment */ 3257c478bd9Sstevel@tonic-gate 3267c478bd9Sstevel@tonic-gate caddr_t segkp_base; /* Base address of segkp */ 327ad23a2dbSjohansen caddr_t segzio_base; /* Base address of segzio */ 3287c478bd9Sstevel@tonic-gate pgcnt_t segkpsize = btop(SEGKPDEFSIZE); /* size of segkp segment in pages */ 329*04909c8cSJohn Levon caddr_t segkvmm_base; 330*04909c8cSJohn Levon pgcnt_t segkvmmsize; 331*04909c8cSJohn Levon pgcnt_t segziosize; 3327c478bd9Sstevel@tonic-gate 333a3114836SGerry Liu /* 334a3114836SGerry Liu * A static DR page_t VA map is reserved that can map the page structures 335a3114836SGerry Liu * for a domain's entire RA space. The pages that back this space are 336a3114836SGerry Liu * dynamically allocated and need not be physically contiguous. The DR 337a3114836SGerry Liu * map size is derived from KPM size. 338a3114836SGerry Liu * This mechanism isn't used by x86 yet, so just stubs here. 339a3114836SGerry Liu */ 340a3114836SGerry Liu int ppvm_enable = 0; /* Static virtual map for page structs */ 341a3114836SGerry Liu page_t *ppvm_base = NULL; /* Base of page struct map */ 342a3114836SGerry Liu pgcnt_t ppvm_size = 0; /* Size of page struct map */ 343a3114836SGerry Liu 344ae115bc7Smrj /* 345ae115bc7Smrj * VA range available to the debugger 346ae115bc7Smrj */ 347ae115bc7Smrj const caddr_t kdi_segdebugbase = (const caddr_t)SEGDEBUGBASE; 348ae115bc7Smrj const size_t kdi_segdebugsize = SEGDEBUGSIZE; 349ae115bc7Smrj 3507c478bd9Sstevel@tonic-gate struct memseg *memseg_base; 3517c478bd9Sstevel@tonic-gate struct vnode unused_pages_vp; 3527c478bd9Sstevel@tonic-gate 3537c478bd9Sstevel@tonic-gate #define FOURGB 0x100000000LL 3547c478bd9Sstevel@tonic-gate 3557c478bd9Sstevel@tonic-gate struct memlist *memlist; 3567c478bd9Sstevel@tonic-gate 3577c478bd9Sstevel@tonic-gate caddr_t s_text; /* start of kernel text segment */ 3587c478bd9Sstevel@tonic-gate caddr_t e_text; /* end of kernel text segment */ 3597c478bd9Sstevel@tonic-gate caddr_t s_data; /* start of kernel data segment */ 3607c478bd9Sstevel@tonic-gate caddr_t e_data; /* end of kernel data segment */ 3617c478bd9Sstevel@tonic-gate caddr_t modtext; /* start of loadable module text reserved */ 3627c478bd9Sstevel@tonic-gate caddr_t e_modtext; /* end of loadable module text reserved */ 3637c478bd9Sstevel@tonic-gate caddr_t moddata; /* start of loadable module data reserved */ 3647c478bd9Sstevel@tonic-gate caddr_t e_moddata; /* end of loadable module data reserved */ 3657c478bd9Sstevel@tonic-gate 3667c478bd9Sstevel@tonic-gate struct memlist *phys_install; /* Total installed physical memory */ 3677c478bd9Sstevel@tonic-gate struct memlist *phys_avail; /* Total available physical memory */ 3681de082f7SVikram Hegde struct memlist *bios_rsvd; /* Bios reserved memory */ 3697c478bd9Sstevel@tonic-gate 3707c478bd9Sstevel@tonic-gate /* 3717c478bd9Sstevel@tonic-gate * kphysm_init returns the number of pages that were processed 3727c478bd9Sstevel@tonic-gate */ 373ae115bc7Smrj static pgcnt_t kphysm_init(page_t *, pgcnt_t); 3747c478bd9Sstevel@tonic-gate 3757c478bd9Sstevel@tonic-gate #define IO_PROP_SIZE 64 /* device property size */ 3767c478bd9Sstevel@tonic-gate 3777c478bd9Sstevel@tonic-gate /* 3787c478bd9Sstevel@tonic-gate * a couple useful roundup macros 3797c478bd9Sstevel@tonic-gate */ 3807c478bd9Sstevel@tonic-gate #define ROUND_UP_PAGE(x) \ 3817c478bd9Sstevel@tonic-gate ((uintptr_t)P2ROUNDUP((uintptr_t)(x), (uintptr_t)MMU_PAGESIZE)) 3827c478bd9Sstevel@tonic-gate #define ROUND_UP_LPAGE(x) \ 3837c478bd9Sstevel@tonic-gate ((uintptr_t)P2ROUNDUP((uintptr_t)(x), mmu.level_size[1])) 3847c478bd9Sstevel@tonic-gate #define ROUND_UP_4MEG(x) \ 385ae115bc7Smrj ((uintptr_t)P2ROUNDUP((uintptr_t)(x), (uintptr_t)FOUR_MEG)) 3867c478bd9Sstevel@tonic-gate #define ROUND_UP_TOPLEVEL(x) \ 3877c478bd9Sstevel@tonic-gate ((uintptr_t)P2ROUNDUP((uintptr_t)(x), mmu.level_size[mmu.max_level])) 3887c478bd9Sstevel@tonic-gate 3897c478bd9Sstevel@tonic-gate /* 3907c478bd9Sstevel@tonic-gate * 32-bit Kernel's Virtual memory layout. 3917c478bd9Sstevel@tonic-gate * +-----------------------+ 392ae115bc7Smrj * | | 3937c478bd9Sstevel@tonic-gate * 0xFFC00000 -|-----------------------|- ARGSBASE 3947c478bd9Sstevel@tonic-gate * | debugger | 3957c478bd9Sstevel@tonic-gate * 0xFF800000 -|-----------------------|- SEGDEBUGBASE 3967c478bd9Sstevel@tonic-gate * | Kernel Data | 3977c478bd9Sstevel@tonic-gate * 0xFEC00000 -|-----------------------| 3987c478bd9Sstevel@tonic-gate * | Kernel Text | 399843e1988Sjohnlev * 0xFE800000 -|-----------------------|- KERNEL_TEXT (0xFB400000 on Xen) 400ae115bc7Smrj * |--- GDT ---|- GDT page (GDT_VA) 401ae115bc7Smrj * |--- debug info ---|- debug info (DEBUG_INFO_VA) 402ae115bc7Smrj * | | 403af868f46SMatthew Ahrens * | page_t structures | 404af868f46SMatthew Ahrens * | memsegs, memlists, | 405af868f46SMatthew Ahrens * | page hash, etc. | 406ae115bc7Smrj * --- -|-----------------------|- ekernelheap, valloc_base (floating) 407ae115bc7Smrj * | | (segkp is just an arena in the heap) 4087c478bd9Sstevel@tonic-gate * | | 4097c478bd9Sstevel@tonic-gate * | kvseg | 4107c478bd9Sstevel@tonic-gate * | | 4117c478bd9Sstevel@tonic-gate * | | 4127c478bd9Sstevel@tonic-gate * --- -|-----------------------|- kernelheap (floating) 413af868f46SMatthew Ahrens * | Segkmap | 414ae115bc7Smrj * 0xC3002000 -|-----------------------|- segmap_start (floating) 4157c478bd9Sstevel@tonic-gate * | Red Zone | 4167c478bd9Sstevel@tonic-gate * 0xC3000000 -|-----------------------|- kernelbase / userlimit (floating) 4177c478bd9Sstevel@tonic-gate * | | || 4187c478bd9Sstevel@tonic-gate * | Shared objects | \/ 4197c478bd9Sstevel@tonic-gate * | | 4207c478bd9Sstevel@tonic-gate * : : 4217c478bd9Sstevel@tonic-gate * | user data | 4227c478bd9Sstevel@tonic-gate * |-----------------------| 4237c478bd9Sstevel@tonic-gate * | user text | 4247c478bd9Sstevel@tonic-gate * 0x08048000 -|-----------------------| 4257c478bd9Sstevel@tonic-gate * | user stack | 4267c478bd9Sstevel@tonic-gate * : : 4277c478bd9Sstevel@tonic-gate * | invalid | 4287c478bd9Sstevel@tonic-gate * 0x00000000 +-----------------------+ 4297c478bd9Sstevel@tonic-gate * 4307c478bd9Sstevel@tonic-gate * 4317c478bd9Sstevel@tonic-gate * 64-bit Kernel's Virtual memory layout. (assuming 64 bit app) 4327c478bd9Sstevel@tonic-gate * +-----------------------+ 433ae115bc7Smrj * | | 4347c478bd9Sstevel@tonic-gate * 0xFFFFFFFF.FFC00000 |-----------------------|- ARGSBASE 4357c478bd9Sstevel@tonic-gate * | debugger (?) | 4367c478bd9Sstevel@tonic-gate * 0xFFFFFFFF.FF800000 |-----------------------|- SEGDEBUGBASE 437af868f46SMatthew Ahrens * | unused | 4387c478bd9Sstevel@tonic-gate * +-----------------------+ 4397c478bd9Sstevel@tonic-gate * | Kernel Data | 4407c478bd9Sstevel@tonic-gate * 0xFFFFFFFF.FBC00000 |-----------------------| 4417c478bd9Sstevel@tonic-gate * | Kernel Text | 4427c478bd9Sstevel@tonic-gate * 0xFFFFFFFF.FB800000 |-----------------------|- KERNEL_TEXT 443ae115bc7Smrj * |--- debug info ---|- debug info (DEBUG_INFO_VA) 44474ecdb51SJohn Levon * |--- GDT ---|- GDT page (GDT_VA) 44574ecdb51SJohn Levon * |--- IDT ---|- IDT page (IDT_VA) 44674ecdb51SJohn Levon * |--- LDT ---|- LDT pages (LDT_VA) 447ae115bc7Smrj * | | 448af868f46SMatthew Ahrens * | Core heap | (used for loadable modules) 4497c478bd9Sstevel@tonic-gate * 0xFFFFFFFF.C0000000 |-----------------------|- core_base / ekernelheap 4507c478bd9Sstevel@tonic-gate * | Kernel | 4517c478bd9Sstevel@tonic-gate * | heap | 452*04909c8cSJohn Levon * | | 453*04909c8cSJohn Levon * | | 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) 459*04909c8cSJohn Levon * | segzio | 460ad23a2dbSjohansen * 0xFFFFFXXX.XXX00000 |-----------------------|- segzio_base (floating) 461*04909c8cSJohn Levon * | segkvmm | 462*04909c8cSJohn Levon * | | 463*04909c8cSJohn Levon * | | 464*04909c8cSJohn Levon * | | 465*04909c8cSJohn Levon * 0xFFFFFXXX.XXX00000 |-----------------------|- segkvmm_base (floating) 466*04909c8cSJohn Levon * | segkp | 467*04909c8cSJohn Levon * |-----------------------|- segkp_base (floating) 468af868f46SMatthew Ahrens * | page_t structures | valloc_base + valloc_sz 469af868f46SMatthew Ahrens * | memsegs, memlists, | 470af868f46SMatthew Ahrens * | page hash, etc. | 471*04909c8cSJohn Levon * 0xFFFFFE00.00000000 |-----------------------|- valloc_base (lower if >256GB) 4727c478bd9Sstevel@tonic-gate * | segkpm | 473*04909c8cSJohn Levon * | | 474*04909c8cSJohn Levon * 0xFFFFFD00.00000000 |-----------------------|- SEGKPM_BASE (lower if >256GB) 4757c478bd9Sstevel@tonic-gate * | Red Zone | 476*04909c8cSJohn Levon * 0xFFFFFC80.00000000 |-----------------------|- KERNELBASE (lower if >256GB) 477*04909c8cSJohn Levon * 0xFFFFFC7F.FFE00000 |-----------------------|- USERLIMIT (lower if >256GB) 4787c478bd9Sstevel@tonic-gate * | User stack |- User space memory 479af868f46SMatthew Ahrens * | | 480af868f46SMatthew Ahrens * | shared objects, etc | (grows downwards) 4817c478bd9Sstevel@tonic-gate * : : 482af868f46SMatthew Ahrens * | | 4837c478bd9Sstevel@tonic-gate * 0xFFFF8000.00000000 |-----------------------| 484af868f46SMatthew Ahrens * | | 485af868f46SMatthew Ahrens * | VA Hole / unused | 486af868f46SMatthew Ahrens * | | 4877c478bd9Sstevel@tonic-gate * 0x00008000.00000000 |-----------------------| 4887c478bd9Sstevel@tonic-gate * | | 4897c478bd9Sstevel@tonic-gate * | | 4907c478bd9Sstevel@tonic-gate * : : 4917c478bd9Sstevel@tonic-gate * | user heap | (grows upwards) 4927c478bd9Sstevel@tonic-gate * | | 4937c478bd9Sstevel@tonic-gate * | user data | 4947c478bd9Sstevel@tonic-gate * |-----------------------| 4957c478bd9Sstevel@tonic-gate * | user text | 4967c478bd9Sstevel@tonic-gate * 0x00000000.04000000 |-----------------------| 4977c478bd9Sstevel@tonic-gate * | invalid | 4987c478bd9Sstevel@tonic-gate * 0x00000000.00000000 +-----------------------+ 4997c478bd9Sstevel@tonic-gate * 5007c478bd9Sstevel@tonic-gate * A 32 bit app on the 64 bit kernel sees the same layout as on the 32 bit 5017c478bd9Sstevel@tonic-gate * kernel, except that userlimit is raised to 0xfe000000 5027c478bd9Sstevel@tonic-gate * 5037c478bd9Sstevel@tonic-gate * Floating values: 5047c478bd9Sstevel@tonic-gate * 5057c478bd9Sstevel@tonic-gate * valloc_base: start of the kernel's memory management/tracking data 506ae115bc7Smrj * structures. This region contains page_t structures for 507ae115bc7Smrj * physical memory, memsegs, memlists, and the page hash. 5087c478bd9Sstevel@tonic-gate * 5097c478bd9Sstevel@tonic-gate * core_base: start of the kernel's "core" heap area on 64-bit systems. 5107c478bd9Sstevel@tonic-gate * This area is intended to be used for global data as well as for module 5117c478bd9Sstevel@tonic-gate * text/data that does not fit into the nucleus pages. The core heap is 5127c478bd9Sstevel@tonic-gate * restricted to a 2GB range, allowing every address within it to be 5137c478bd9Sstevel@tonic-gate * accessed using rip-relative addressing 5147c478bd9Sstevel@tonic-gate * 5157c478bd9Sstevel@tonic-gate * ekernelheap: end of kernelheap and start of segmap. 5167c478bd9Sstevel@tonic-gate * 5177c478bd9Sstevel@tonic-gate * kernelheap: start of kernel heap. On 32-bit systems, this starts right 5187c478bd9Sstevel@tonic-gate * above a red zone that separates the user's address space from the 5197c478bd9Sstevel@tonic-gate * kernel's. On 64-bit systems, it sits above segkp and segkpm. 5207c478bd9Sstevel@tonic-gate * 521ae115bc7Smrj * segmap_start: start of segmap. The length of segmap can be modified 52252eef812SAmrita Sadhukhan * through eeprom. The default length is 16MB on 32-bit systems and 64MB 52352eef812SAmrita Sadhukhan * on 64-bit systems. 5247c478bd9Sstevel@tonic-gate * 5257c478bd9Sstevel@tonic-gate * kernelbase: On a 32-bit kernel the default value of 0xd4000000 will be 5267c478bd9Sstevel@tonic-gate * decreased by 2X the size required for page_t. This allows the kernel 5277c478bd9Sstevel@tonic-gate * heap to grow in size with physical memory. With sizeof(page_t) == 80 5287c478bd9Sstevel@tonic-gate * bytes, the following shows the values of kernelbase and kernel heap 5297c478bd9Sstevel@tonic-gate * sizes for different memory configurations (assuming default segmap and 5307c478bd9Sstevel@tonic-gate * segkp sizes). 5317c478bd9Sstevel@tonic-gate * 5327c478bd9Sstevel@tonic-gate * mem size for kernelbase kernel heap 5337c478bd9Sstevel@tonic-gate * size page_t's size 5347c478bd9Sstevel@tonic-gate * ---- --------- ---------- ----------- 5357c478bd9Sstevel@tonic-gate * 1gb 0x01400000 0xd1800000 684MB 5367c478bd9Sstevel@tonic-gate * 2gb 0x02800000 0xcf000000 704MB 5377c478bd9Sstevel@tonic-gate * 4gb 0x05000000 0xca000000 744MB 5387c478bd9Sstevel@tonic-gate * 6gb 0x07800000 0xc5000000 784MB 5397c478bd9Sstevel@tonic-gate * 8gb 0x0a000000 0xc0000000 824MB 5407c478bd9Sstevel@tonic-gate * 16gb 0x14000000 0xac000000 984MB 5417c478bd9Sstevel@tonic-gate * 32gb 0x28000000 0x84000000 1304MB 5427c478bd9Sstevel@tonic-gate * 64gb 0x50000000 0x34000000 1944MB (*) 5437c478bd9Sstevel@tonic-gate * 5447c478bd9Sstevel@tonic-gate * kernelbase is less than the abi minimum of 0xc0000000 for memory 5457c478bd9Sstevel@tonic-gate * configurations above 8gb. 5467c478bd9Sstevel@tonic-gate * 5477c478bd9Sstevel@tonic-gate * (*) support for memory configurations above 32gb will require manual tuning 5487c478bd9Sstevel@tonic-gate * of kernelbase to balance out the need of user applications. 5497c478bd9Sstevel@tonic-gate */ 5507c478bd9Sstevel@tonic-gate 5517c478bd9Sstevel@tonic-gate /* real-time-clock initialization parameters */ 552ae115bc7Smrj extern time_t process_rtc_config_file(void); 5537c478bd9Sstevel@tonic-gate 5547c478bd9Sstevel@tonic-gate uintptr_t kernelbase; 555ae115bc7Smrj uintptr_t postbootkernelbase; /* not set till boot loader is gone */ 5567c478bd9Sstevel@tonic-gate uintptr_t eprom_kernelbase; 5577c478bd9Sstevel@tonic-gate size_t segmapsize; 558ae115bc7Smrj uintptr_t segmap_start; 5597c478bd9Sstevel@tonic-gate int segmapfreelists; 5607c478bd9Sstevel@tonic-gate pgcnt_t npages; 561ae115bc7Smrj pgcnt_t orig_npages; 5627c478bd9Sstevel@tonic-gate size_t core_size; /* size of "core" heap */ 5637c478bd9Sstevel@tonic-gate uintptr_t core_base; /* base address of "core" heap */ 5647c478bd9Sstevel@tonic-gate 5657c478bd9Sstevel@tonic-gate /* 5667c478bd9Sstevel@tonic-gate * List of bootstrap pages. We mark these as allocated in startup. 5677c478bd9Sstevel@tonic-gate * release_bootstrap() will free them when we're completely done with 5687c478bd9Sstevel@tonic-gate * the bootstrap. 5697c478bd9Sstevel@tonic-gate */ 570ae115bc7Smrj static page_t *bootpages; 571ae115bc7Smrj 572ae115bc7Smrj /* 573ae115bc7Smrj * boot time pages that have a vnode from the ramdisk will keep that forever. 574ae115bc7Smrj */ 575ae115bc7Smrj static page_t *rd_pages; 5767c478bd9Sstevel@tonic-gate 57719397407SSherry Moore /* 57819397407SSherry Moore * Lower 64K 57919397407SSherry Moore */ 58019397407SSherry Moore static page_t *lower_pages = NULL; 58119397407SSherry Moore static int lower_pages_count = 0; 58219397407SSherry Moore 5837c478bd9Sstevel@tonic-gate struct system_hardware system_hardware; 5847c478bd9Sstevel@tonic-gate 5857c478bd9Sstevel@tonic-gate /* 5867c478bd9Sstevel@tonic-gate * Enable some debugging messages concerning memory usage... 5877c478bd9Sstevel@tonic-gate */ 5887c478bd9Sstevel@tonic-gate static void 589ae115bc7Smrj print_memlist(char *title, struct memlist *mp) 5907c478bd9Sstevel@tonic-gate { 5917c478bd9Sstevel@tonic-gate prom_printf("MEMLIST: %s:\n", title); 5927c478bd9Sstevel@tonic-gate while (mp != NULL) { 5937c478bd9Sstevel@tonic-gate prom_printf("\tAddress 0x%" PRIx64 ", size 0x%" PRIx64 "\n", 59456f33205SJonathan Adams mp->ml_address, mp->ml_size); 59556f33205SJonathan Adams mp = mp->ml_next; 5967c478bd9Sstevel@tonic-gate } 5977c478bd9Sstevel@tonic-gate } 5987c478bd9Sstevel@tonic-gate 5997c478bd9Sstevel@tonic-gate /* 6007c478bd9Sstevel@tonic-gate * XX64 need a comment here.. are these just default values, surely 6017c478bd9Sstevel@tonic-gate * we read the "cpuid" type information to figure this out. 6027c478bd9Sstevel@tonic-gate */ 6037c478bd9Sstevel@tonic-gate int l2cache_sz = 0x80000; 6047c478bd9Sstevel@tonic-gate int l2cache_linesz = 0x40; 6057c478bd9Sstevel@tonic-gate int l2cache_assoc = 1; 6067c478bd9Sstevel@tonic-gate 6072cb27123Saguzovsk static size_t textrepl_min_gb = 10; 6082cb27123Saguzovsk 6097c478bd9Sstevel@tonic-gate /* 6107c478bd9Sstevel@tonic-gate * on 64 bit we use a predifined VA range for mapping devices in the kernel 6117c478bd9Sstevel@tonic-gate * on 32 bit the mappings are intermixed in the heap, so we use a bit map 6127c478bd9Sstevel@tonic-gate */ 6137c478bd9Sstevel@tonic-gate #ifdef __amd64 6147c478bd9Sstevel@tonic-gate 6157c478bd9Sstevel@tonic-gate vmem_t *device_arena; 6167c478bd9Sstevel@tonic-gate uintptr_t toxic_addr = (uintptr_t)NULL; 617ae115bc7Smrj size_t toxic_size = 1024 * 1024 * 1024; /* Sparc uses 1 gig too */ 6187c478bd9Sstevel@tonic-gate 6197c478bd9Sstevel@tonic-gate #else /* __i386 */ 6207c478bd9Sstevel@tonic-gate 6217c478bd9Sstevel@tonic-gate ulong_t *toxic_bit_map; /* one bit for each 4k of VA in heap_arena */ 6227c478bd9Sstevel@tonic-gate size_t toxic_bit_map_len = 0; /* in bits */ 6237c478bd9Sstevel@tonic-gate 6247c478bd9Sstevel@tonic-gate #endif /* __i386 */ 6257c478bd9Sstevel@tonic-gate 6267c478bd9Sstevel@tonic-gate /* 6277c478bd9Sstevel@tonic-gate * Simple boot time debug facilities 6287c478bd9Sstevel@tonic-gate */ 6297c478bd9Sstevel@tonic-gate static char *prm_dbg_str[] = { 6307c478bd9Sstevel@tonic-gate "%s:%d: '%s' is 0x%x\n", 6317c478bd9Sstevel@tonic-gate "%s:%d: '%s' is 0x%llx\n" 6327c478bd9Sstevel@tonic-gate }; 6337c478bd9Sstevel@tonic-gate 6347c478bd9Sstevel@tonic-gate int prom_debug; 6357c478bd9Sstevel@tonic-gate 6364da99751SToomas Soome #define PRM_DEBUG(q) if (prom_debug) \ 6377c478bd9Sstevel@tonic-gate prom_printf(prm_dbg_str[sizeof (q) >> 3], "startup.c", __LINE__, #q, q); 6384da99751SToomas Soome #define PRM_POINT(q) if (prom_debug) \ 6397c478bd9Sstevel@tonic-gate prom_printf("%s:%d: %s\n", "startup.c", __LINE__, q); 6407c478bd9Sstevel@tonic-gate 6417c478bd9Sstevel@tonic-gate /* 6427c478bd9Sstevel@tonic-gate * This structure is used to keep track of the intial allocations 6437c478bd9Sstevel@tonic-gate * done in startup_memlist(). The value of NUM_ALLOCATIONS needs to 6447c478bd9Sstevel@tonic-gate * be >= the number of ADD_TO_ALLOCATIONS() executed in the code. 6457c478bd9Sstevel@tonic-gate */ 6461de082f7SVikram Hegde #define NUM_ALLOCATIONS 8 6477c478bd9Sstevel@tonic-gate int num_allocations = 0; 6487c478bd9Sstevel@tonic-gate struct { 6497c478bd9Sstevel@tonic-gate void **al_ptr; 6507c478bd9Sstevel@tonic-gate size_t al_size; 6517c478bd9Sstevel@tonic-gate } allocations[NUM_ALLOCATIONS]; 6527c478bd9Sstevel@tonic-gate size_t valloc_sz = 0; 6537c478bd9Sstevel@tonic-gate uintptr_t valloc_base; 6547c478bd9Sstevel@tonic-gate 6557c478bd9Sstevel@tonic-gate #define ADD_TO_ALLOCATIONS(ptr, size) { \ 6564da99751SToomas Soome size = ROUND_UP_PAGE(size); \ 6577c478bd9Sstevel@tonic-gate if (num_allocations == NUM_ALLOCATIONS) \ 6587c478bd9Sstevel@tonic-gate panic("too many ADD_TO_ALLOCATIONS()"); \ 6597c478bd9Sstevel@tonic-gate allocations[num_allocations].al_ptr = (void**)&ptr; \ 6607c478bd9Sstevel@tonic-gate allocations[num_allocations].al_size = size; \ 6617c478bd9Sstevel@tonic-gate valloc_sz += size; \ 6624da99751SToomas Soome ++num_allocations; \ 6637c478bd9Sstevel@tonic-gate } 6647c478bd9Sstevel@tonic-gate 665ae115bc7Smrj /* 666ae115bc7Smrj * Allocate all the initial memory needed by the page allocator. 667ae115bc7Smrj */ 6687c478bd9Sstevel@tonic-gate static void 6697c478bd9Sstevel@tonic-gate perform_allocations(void) 6707c478bd9Sstevel@tonic-gate { 6717c478bd9Sstevel@tonic-gate caddr_t mem; 6727c478bd9Sstevel@tonic-gate int i; 673ae115bc7Smrj int valloc_align; 6747c478bd9Sstevel@tonic-gate 675ae115bc7Smrj PRM_DEBUG(valloc_base); 676ae115bc7Smrj PRM_DEBUG(valloc_sz); 677ae115bc7Smrj valloc_align = mmu.level_size[mmu.max_page_level > 0]; 678ae115bc7Smrj mem = BOP_ALLOC(bootops, (caddr_t)valloc_base, valloc_sz, valloc_align); 6797c478bd9Sstevel@tonic-gate if (mem != (caddr_t)valloc_base) 6807c478bd9Sstevel@tonic-gate panic("BOP_ALLOC() failed"); 6817c478bd9Sstevel@tonic-gate bzero(mem, valloc_sz); 6827c478bd9Sstevel@tonic-gate for (i = 0; i < num_allocations; ++i) { 6837c478bd9Sstevel@tonic-gate *allocations[i].al_ptr = (void *)mem; 6847c478bd9Sstevel@tonic-gate mem += allocations[i].al_size; 6857c478bd9Sstevel@tonic-gate } 6867c478bd9Sstevel@tonic-gate } 6877c478bd9Sstevel@tonic-gate 6883ce2fcdcSRobert Mustacchi /* 6893ce2fcdcSRobert Mustacchi * Set up and enable SMAP now before we start other CPUs, but after the kernel's 6903ce2fcdcSRobert Mustacchi * VM has been set up so we can use hot_patch_kernel_text(). 6913ce2fcdcSRobert Mustacchi * 6923ce2fcdcSRobert Mustacchi * We can only patch 1, 2, or 4 bytes, but not three bytes. So instead, we 6933ce2fcdcSRobert Mustacchi * replace the four byte word at the patch point. See uts/intel/ia32/ml/copy.s 6943ce2fcdcSRobert Mustacchi * for more information on what's going on here. 6953ce2fcdcSRobert Mustacchi */ 6963ce2fcdcSRobert Mustacchi static void 6973ce2fcdcSRobert Mustacchi startup_smap(void) 6983ce2fcdcSRobert Mustacchi { 6993ce2fcdcSRobert Mustacchi int i; 7003ce2fcdcSRobert Mustacchi uint32_t inst; 7013ce2fcdcSRobert Mustacchi uint8_t *instp; 7023ce2fcdcSRobert Mustacchi char sym[128]; 703425251fdSSam Gwydir struct modctl *modp; 7043ce2fcdcSRobert Mustacchi 7053ce2fcdcSRobert Mustacchi extern int _smap_enable_patch_count; 7063ce2fcdcSRobert Mustacchi extern int _smap_disable_patch_count; 7073ce2fcdcSRobert Mustacchi 7083ce2fcdcSRobert Mustacchi if (disable_smap != 0) 7093ce2fcdcSRobert Mustacchi remove_x86_feature(x86_featureset, X86FSET_SMAP); 7103ce2fcdcSRobert Mustacchi 7113ce2fcdcSRobert Mustacchi if (is_x86_feature(x86_featureset, X86FSET_SMAP) == B_FALSE) 7123ce2fcdcSRobert Mustacchi return; 7133ce2fcdcSRobert Mustacchi 7143ce2fcdcSRobert Mustacchi for (i = 0; i < _smap_enable_patch_count; i++) { 7153ce2fcdcSRobert Mustacchi int sizep; 7163ce2fcdcSRobert Mustacchi 7173ce2fcdcSRobert Mustacchi VERIFY3U(i, <, _smap_enable_patch_count); 7183ce2fcdcSRobert Mustacchi VERIFY(snprintf(sym, sizeof (sym), "_smap_enable_patch_%d", i) < 7193ce2fcdcSRobert Mustacchi sizeof (sym)); 7203ce2fcdcSRobert Mustacchi instp = (uint8_t *)(void *)kobj_getelfsym(sym, NULL, &sizep); 7213ce2fcdcSRobert Mustacchi VERIFY(instp != 0); 7223ce2fcdcSRobert Mustacchi inst = (instp[3] << 24) | (SMAP_CLAC_INSTR & 0x00ffffff); 7233ce2fcdcSRobert Mustacchi hot_patch_kernel_text((caddr_t)instp, inst, 4); 7243ce2fcdcSRobert Mustacchi } 7253ce2fcdcSRobert Mustacchi 7263ce2fcdcSRobert Mustacchi for (i = 0; i < _smap_disable_patch_count; i++) { 7273ce2fcdcSRobert Mustacchi int sizep; 7283ce2fcdcSRobert Mustacchi 7293ce2fcdcSRobert Mustacchi VERIFY(snprintf(sym, sizeof (sym), "_smap_disable_patch_%d", 7303ce2fcdcSRobert Mustacchi i) < sizeof (sym)); 7313ce2fcdcSRobert Mustacchi instp = (uint8_t *)(void *)kobj_getelfsym(sym, NULL, &sizep); 7323ce2fcdcSRobert Mustacchi VERIFY(instp != 0); 7333ce2fcdcSRobert Mustacchi inst = (instp[3] << 24) | (SMAP_STAC_INSTR & 0x00ffffff); 7343ce2fcdcSRobert Mustacchi hot_patch_kernel_text((caddr_t)instp, inst, 4); 7353ce2fcdcSRobert Mustacchi } 7363ce2fcdcSRobert Mustacchi 737425251fdSSam Gwydir /* 738425251fdSSam Gwydir * Hotinline calls to smap_enable and smap_disable within 739425251fdSSam Gwydir * unix module. Hotinlines in other modules are done on 740425251fdSSam Gwydir * mod_load(). 741425251fdSSam Gwydir */ 742425251fdSSam Gwydir modp = mod_hold_by_name("unix"); 743425251fdSSam Gwydir do_hotinlines(modp->mod_mp); 744425251fdSSam Gwydir mod_release_mod(modp); 745425251fdSSam Gwydir 7463ce2fcdcSRobert Mustacchi setcr4(getcr4() | CR4_SMAP); 7473ce2fcdcSRobert Mustacchi smap_enable(); 7483ce2fcdcSRobert Mustacchi } 7493ce2fcdcSRobert Mustacchi 7507c478bd9Sstevel@tonic-gate /* 7517c478bd9Sstevel@tonic-gate * Our world looks like this at startup time. 7527c478bd9Sstevel@tonic-gate * 7537c478bd9Sstevel@tonic-gate * In a 32-bit OS, boot loads the kernel text at 0xfe800000 and kernel data 7547c478bd9Sstevel@tonic-gate * at 0xfec00000. On a 64-bit OS, kernel text and data are loaded at 7557c478bd9Sstevel@tonic-gate * 0xffffffff.fe800000 and 0xffffffff.fec00000 respectively. Those 7567c478bd9Sstevel@tonic-gate * addresses are fixed in the binary at link time. 7577c478bd9Sstevel@tonic-gate * 7587c478bd9Sstevel@tonic-gate * On the text page: 7597c478bd9Sstevel@tonic-gate * unix/genunix/krtld/module text loads. 7607c478bd9Sstevel@tonic-gate * 7617c478bd9Sstevel@tonic-gate * On the data page: 762ae115bc7Smrj * unix/genunix/krtld/module data loads. 763ae115bc7Smrj * 7647c478bd9Sstevel@tonic-gate * Machine-dependent startup code 7657c478bd9Sstevel@tonic-gate */ 7667c478bd9Sstevel@tonic-gate void 7677c478bd9Sstevel@tonic-gate startup(void) 7687c478bd9Sstevel@tonic-gate { 769843e1988Sjohnlev #if !defined(__xpv) 77075bcd456Sjg extern void startup_pci_bios(void); 771843e1988Sjohnlev #endif 772a563a037Sbholler extern cpuset_t cpu_ready_set; 773a563a037Sbholler 774843e1988Sjohnlev /* 775843e1988Sjohnlev * Make sure that nobody tries to use sekpm until we have 776843e1988Sjohnlev * initialized it properly. 777843e1988Sjohnlev */ 7787c478bd9Sstevel@tonic-gate #if defined(__amd64) 77935b1ab99Sjosephb kpm_desired = 1; 7807c478bd9Sstevel@tonic-gate #endif 7817c478bd9Sstevel@tonic-gate kpm_enable = 0; 782a563a037Sbholler CPUSET_ONLY(cpu_ready_set, 0); /* cpu 0 is boot cpu */ 7837c478bd9Sstevel@tonic-gate 784843e1988Sjohnlev #if defined(__xpv) /* XXPV fix me! */ 785843e1988Sjohnlev { 786843e1988Sjohnlev extern int segvn_use_regions; 787843e1988Sjohnlev segvn_use_regions = 0; 788843e1988Sjohnlev } 789843e1988Sjohnlev #endif 7906a59053bSAlex Wilson ssp_init(); 7917c478bd9Sstevel@tonic-gate progressbar_init(); 7927c478bd9Sstevel@tonic-gate startup_init(); 793ab4a9bebSjohnlev #if defined(__xpv) 794ab4a9bebSjohnlev startup_xen_version(); 795ab4a9bebSjohnlev #endif 7967c478bd9Sstevel@tonic-gate startup_memlist(); 797ae115bc7Smrj startup_kmem(); 79835b1ab99Sjosephb startup_vm(); 799843e1988Sjohnlev #if !defined(__xpv) 8000db3240dSStephen Hanson /* 8010db3240dSStephen Hanson * Note we need to do this even on fast reboot in order to access 8020db3240dSStephen Hanson * the irq routing table (used for pci labels). 8030db3240dSStephen Hanson */ 8040db3240dSStephen Hanson startup_pci_bios(); 8053ce2fcdcSRobert Mustacchi startup_smap(); 806e4b86885SCheng Sean Ye #endif 807e4b86885SCheng Sean Ye #if defined(__xpv) 808e4b86885SCheng Sean Ye startup_xen_mca(); 809843e1988Sjohnlev #endif 8107c478bd9Sstevel@tonic-gate startup_modules(); 8118770118eSlipeng sang - Sun Microsystems - Beijing China 8127c478bd9Sstevel@tonic-gate startup_end(); 8137c478bd9Sstevel@tonic-gate } 8147c478bd9Sstevel@tonic-gate 8157c478bd9Sstevel@tonic-gate static void 8167c478bd9Sstevel@tonic-gate startup_init() 8177c478bd9Sstevel@tonic-gate { 8187c478bd9Sstevel@tonic-gate PRM_POINT("startup_init() starting..."); 8197c478bd9Sstevel@tonic-gate 8207c478bd9Sstevel@tonic-gate /* 8217c478bd9Sstevel@tonic-gate * Complete the extraction of cpuid data 8227c478bd9Sstevel@tonic-gate */ 8237c478bd9Sstevel@tonic-gate cpuid_pass2(CPU); 8247c478bd9Sstevel@tonic-gate 8257c478bd9Sstevel@tonic-gate (void) check_boot_version(BOP_GETVERSION(bootops)); 8267c478bd9Sstevel@tonic-gate 8277c478bd9Sstevel@tonic-gate /* 8287c478bd9Sstevel@tonic-gate * Check for prom_debug in boot environment 8297c478bd9Sstevel@tonic-gate */ 8307c478bd9Sstevel@tonic-gate if (BOP_GETPROPLEN(bootops, "prom_debug") >= 0) { 8317c478bd9Sstevel@tonic-gate ++prom_debug; 8327c478bd9Sstevel@tonic-gate PRM_POINT("prom_debug found in boot enviroment"); 8337c478bd9Sstevel@tonic-gate } 8347c478bd9Sstevel@tonic-gate 8357c478bd9Sstevel@tonic-gate /* 8367c478bd9Sstevel@tonic-gate * Collect node, cpu and memory configuration information. 8377c478bd9Sstevel@tonic-gate */ 8387c478bd9Sstevel@tonic-gate get_system_configuration(); 8397c478bd9Sstevel@tonic-gate 8407c478bd9Sstevel@tonic-gate /* 8417c478bd9Sstevel@tonic-gate * Halt if this is an unsupported processor. 8427c478bd9Sstevel@tonic-gate */ 8437c478bd9Sstevel@tonic-gate if (x86_type == X86_TYPE_486 || x86_type == X86_TYPE_CYRIX_486) { 8447c478bd9Sstevel@tonic-gate printf("\n486 processor (\"%s\") detected.\n", 8457c478bd9Sstevel@tonic-gate CPU->cpu_brandstr); 8467c478bd9Sstevel@tonic-gate halt("This processor is not supported by this release " 8477c478bd9Sstevel@tonic-gate "of Solaris."); 8487c478bd9Sstevel@tonic-gate } 8497c478bd9Sstevel@tonic-gate 8507c478bd9Sstevel@tonic-gate PRM_POINT("startup_init() done"); 8517c478bd9Sstevel@tonic-gate } 8527c478bd9Sstevel@tonic-gate 8537c478bd9Sstevel@tonic-gate /* 8547c478bd9Sstevel@tonic-gate * Callback for copy_memlist_filter() to filter nucleus, kadb/kmdb, (ie. 8557c478bd9Sstevel@tonic-gate * everything mapped above KERNEL_TEXT) pages from phys_avail. Note it 8567c478bd9Sstevel@tonic-gate * also filters out physical page zero. There is some reliance on the 8577c478bd9Sstevel@tonic-gate * boot loader allocating only a few contiguous physical memory chunks. 8587c478bd9Sstevel@tonic-gate */ 8597c478bd9Sstevel@tonic-gate static void 8607c478bd9Sstevel@tonic-gate avail_filter(uint64_t *addr, uint64_t *size) 8617c478bd9Sstevel@tonic-gate { 8627c478bd9Sstevel@tonic-gate uintptr_t va; 8637c478bd9Sstevel@tonic-gate uintptr_t next_va; 8647c478bd9Sstevel@tonic-gate pfn_t pfn; 8657c478bd9Sstevel@tonic-gate uint64_t pfn_addr; 8667c478bd9Sstevel@tonic-gate uint64_t pfn_eaddr; 8677c478bd9Sstevel@tonic-gate uint_t prot; 8687c478bd9Sstevel@tonic-gate size_t len; 8697c478bd9Sstevel@tonic-gate uint_t change; 8707c478bd9Sstevel@tonic-gate 8717c478bd9Sstevel@tonic-gate if (prom_debug) 8727c478bd9Sstevel@tonic-gate prom_printf("\tFilter: in: a=%" PRIx64 ", s=%" PRIx64 "\n", 8737c478bd9Sstevel@tonic-gate *addr, *size); 8747c478bd9Sstevel@tonic-gate 8757c478bd9Sstevel@tonic-gate /* 8767c478bd9Sstevel@tonic-gate * page zero is required for BIOS.. never make it available 8777c478bd9Sstevel@tonic-gate */ 8787c478bd9Sstevel@tonic-gate if (*addr == 0) { 8797c478bd9Sstevel@tonic-gate *addr += MMU_PAGESIZE; 8807c478bd9Sstevel@tonic-gate *size -= MMU_PAGESIZE; 8817c478bd9Sstevel@tonic-gate } 8827c478bd9Sstevel@tonic-gate 8837c478bd9Sstevel@tonic-gate /* 884ae115bc7Smrj * First we trim from the front of the range. Since kbm_probe() 8857c478bd9Sstevel@tonic-gate * walks ranges in virtual order, but addr/size are physical, we need 8867c478bd9Sstevel@tonic-gate * to the list until no changes are seen. This deals with the case 8877c478bd9Sstevel@tonic-gate * where page "p" is mapped at v, page "p + PAGESIZE" is mapped at w 8887c478bd9Sstevel@tonic-gate * but w < v. 8897c478bd9Sstevel@tonic-gate */ 8907c478bd9Sstevel@tonic-gate do { 8917c478bd9Sstevel@tonic-gate change = 0; 8927c478bd9Sstevel@tonic-gate for (va = KERNEL_TEXT; 893ae115bc7Smrj *size > 0 && kbm_probe(&va, &len, &pfn, &prot) != 0; 8947c478bd9Sstevel@tonic-gate va = next_va) { 8957c478bd9Sstevel@tonic-gate 8967c478bd9Sstevel@tonic-gate next_va = va + len; 897ae115bc7Smrj pfn_addr = pfn_to_pa(pfn); 8987c478bd9Sstevel@tonic-gate pfn_eaddr = pfn_addr + len; 8997c478bd9Sstevel@tonic-gate 9007c478bd9Sstevel@tonic-gate if (pfn_addr <= *addr && pfn_eaddr > *addr) { 9017c478bd9Sstevel@tonic-gate change = 1; 9027c478bd9Sstevel@tonic-gate while (*size > 0 && len > 0) { 9037c478bd9Sstevel@tonic-gate *addr += MMU_PAGESIZE; 9047c478bd9Sstevel@tonic-gate *size -= MMU_PAGESIZE; 9057c478bd9Sstevel@tonic-gate len -= MMU_PAGESIZE; 9067c478bd9Sstevel@tonic-gate } 9077c478bd9Sstevel@tonic-gate } 9087c478bd9Sstevel@tonic-gate } 9097c478bd9Sstevel@tonic-gate if (change && prom_debug) 9107c478bd9Sstevel@tonic-gate prom_printf("\t\ttrim: a=%" PRIx64 ", s=%" PRIx64 "\n", 9117c478bd9Sstevel@tonic-gate *addr, *size); 9127c478bd9Sstevel@tonic-gate } while (change); 9137c478bd9Sstevel@tonic-gate 9147c478bd9Sstevel@tonic-gate /* 9157c478bd9Sstevel@tonic-gate * Trim pages from the end of the range. 9167c478bd9Sstevel@tonic-gate */ 9177c478bd9Sstevel@tonic-gate for (va = KERNEL_TEXT; 918ae115bc7Smrj *size > 0 && kbm_probe(&va, &len, &pfn, &prot) != 0; 9197c478bd9Sstevel@tonic-gate va = next_va) { 9207c478bd9Sstevel@tonic-gate 9217c478bd9Sstevel@tonic-gate next_va = va + len; 922ae115bc7Smrj pfn_addr = pfn_to_pa(pfn); 9237c478bd9Sstevel@tonic-gate 9247c478bd9Sstevel@tonic-gate if (pfn_addr >= *addr && pfn_addr < *addr + *size) 9257c478bd9Sstevel@tonic-gate *size = pfn_addr - *addr; 9267c478bd9Sstevel@tonic-gate } 9277c478bd9Sstevel@tonic-gate 9287c478bd9Sstevel@tonic-gate if (prom_debug) 9297c478bd9Sstevel@tonic-gate prom_printf("\tFilter out: a=%" PRIx64 ", s=%" PRIx64 "\n", 9307c478bd9Sstevel@tonic-gate *addr, *size); 9317c478bd9Sstevel@tonic-gate } 9327c478bd9Sstevel@tonic-gate 9337c478bd9Sstevel@tonic-gate static void 9347c478bd9Sstevel@tonic-gate kpm_init() 9357c478bd9Sstevel@tonic-gate { 9367c478bd9Sstevel@tonic-gate struct segkpm_crargs b; 9377c478bd9Sstevel@tonic-gate 9387c478bd9Sstevel@tonic-gate /* 9397c478bd9Sstevel@tonic-gate * These variables were all designed for sfmmu in which segkpm is 9407c478bd9Sstevel@tonic-gate * mapped using a single pagesize - either 8KB or 4MB. On x86, we 9417c478bd9Sstevel@tonic-gate * might use 2+ page sizes on a single machine, so none of these 9427c478bd9Sstevel@tonic-gate * variables have a single correct value. They are set up as if we 9437c478bd9Sstevel@tonic-gate * always use a 4KB pagesize, which should do no harm. In the long 9447c478bd9Sstevel@tonic-gate * run, we should get rid of KPM's assumption that only a single 9457c478bd9Sstevel@tonic-gate * pagesize is used. 9467c478bd9Sstevel@tonic-gate */ 9477c478bd9Sstevel@tonic-gate kpm_pgshft = MMU_PAGESHIFT; 9487c478bd9Sstevel@tonic-gate kpm_pgsz = MMU_PAGESIZE; 9497c478bd9Sstevel@tonic-gate kpm_pgoff = MMU_PAGEOFFSET; 9507c478bd9Sstevel@tonic-gate kpmp2pshft = 0; 9517c478bd9Sstevel@tonic-gate kpmpnpgs = 1; 9527c478bd9Sstevel@tonic-gate ASSERT(((uintptr_t)kpm_vbase & (kpm_pgsz - 1)) == 0); 9537c478bd9Sstevel@tonic-gate 9547c478bd9Sstevel@tonic-gate PRM_POINT("about to create segkpm"); 9557c478bd9Sstevel@tonic-gate rw_enter(&kas.a_lock, RW_WRITER); 9567c478bd9Sstevel@tonic-gate 9577c478bd9Sstevel@tonic-gate if (seg_attach(&kas, kpm_vbase, kpm_size, segkpm) < 0) 9587c478bd9Sstevel@tonic-gate panic("cannot attach segkpm"); 9597c478bd9Sstevel@tonic-gate 9607c478bd9Sstevel@tonic-gate b.prot = PROT_READ | PROT_WRITE; 9617c478bd9Sstevel@tonic-gate b.nvcolors = 1; 9627c478bd9Sstevel@tonic-gate 9637c478bd9Sstevel@tonic-gate if (segkpm_create(segkpm, (caddr_t)&b) != 0) 9647c478bd9Sstevel@tonic-gate panic("segkpm_create segkpm"); 9657c478bd9Sstevel@tonic-gate 9667c478bd9Sstevel@tonic-gate rw_exit(&kas.a_lock); 96774ecdb51SJohn Levon 96874ecdb51SJohn Levon kpm_enable = 1; 96974ecdb51SJohn Levon 97074ecdb51SJohn Levon /* 97174ecdb51SJohn Levon * As the KPM was disabled while setting up the system, go back and fix 97274ecdb51SJohn Levon * CPU zero's access to its user page table. This is a bit gross, but 97374ecdb51SJohn Levon * we have a chicken and egg problem otherwise. 97474ecdb51SJohn Levon */ 97574ecdb51SJohn Levon ASSERT(CPU->cpu_hat_info->hci_user_l3ptes == NULL); 97674ecdb51SJohn Levon CPU->cpu_hat_info->hci_user_l3ptes = 97774ecdb51SJohn Levon (x86pte_t *)hat_kpm_mapin_pfn(CPU->cpu_hat_info->hci_user_l3pfn); 978ae115bc7Smrj } 9797c478bd9Sstevel@tonic-gate 980ae115bc7Smrj /* 981ae115bc7Smrj * The debug info page provides enough information to allow external 982ae115bc7Smrj * inspectors (e.g. when running under a hypervisor) to bootstrap 983ae115bc7Smrj * themselves into allowing full-blown kernel debugging. 984ae115bc7Smrj */ 985ae115bc7Smrj static void 986ae115bc7Smrj init_debug_info(void) 987ae115bc7Smrj { 988ae115bc7Smrj caddr_t mem; 989ae115bc7Smrj debug_info_t *di; 990ae115bc7Smrj 991ae115bc7Smrj #ifndef __lint 992ae115bc7Smrj ASSERT(sizeof (debug_info_t) < MMU_PAGESIZE); 993ae115bc7Smrj #endif 994ae115bc7Smrj 995ae115bc7Smrj mem = BOP_ALLOC(bootops, (caddr_t)DEBUG_INFO_VA, MMU_PAGESIZE, 996ae115bc7Smrj MMU_PAGESIZE); 997ae115bc7Smrj 998ae115bc7Smrj if (mem != (caddr_t)DEBUG_INFO_VA) 999ae115bc7Smrj panic("BOP_ALLOC() failed"); 1000ae115bc7Smrj bzero(mem, MMU_PAGESIZE); 1001ae115bc7Smrj 1002ae115bc7Smrj di = (debug_info_t *)mem; 1003ae115bc7Smrj 1004ae115bc7Smrj di->di_magic = DEBUG_INFO_MAGIC; 1005ae115bc7Smrj di->di_version = DEBUG_INFO_VERSION; 1006843e1988Sjohnlev di->di_modules = (uintptr_t)&modules; 1007843e1988Sjohnlev di->di_s_text = (uintptr_t)s_text; 1008843e1988Sjohnlev di->di_e_text = (uintptr_t)e_text; 1009843e1988Sjohnlev di->di_s_data = (uintptr_t)s_data; 1010843e1988Sjohnlev di->di_e_data = (uintptr_t)e_data; 1011843e1988Sjohnlev di->di_hat_htable_off = offsetof(hat_t, hat_htable); 1012843e1988Sjohnlev di->di_ht_pfn_off = offsetof(htable_t, ht_pfn); 10137c478bd9Sstevel@tonic-gate } 10147c478bd9Sstevel@tonic-gate 10157c478bd9Sstevel@tonic-gate /* 1016ae115bc7Smrj * Build the memlists and other kernel essential memory system data structures. 1017ae115bc7Smrj * This is everything at valloc_base. 10187c478bd9Sstevel@tonic-gate */ 10197c478bd9Sstevel@tonic-gate static void 10207c478bd9Sstevel@tonic-gate startup_memlist(void) 10217c478bd9Sstevel@tonic-gate { 10227c478bd9Sstevel@tonic-gate size_t memlist_sz; 10237c478bd9Sstevel@tonic-gate size_t memseg_sz; 10247c478bd9Sstevel@tonic-gate size_t pagehash_sz; 10257c478bd9Sstevel@tonic-gate size_t pp_sz; 10267c478bd9Sstevel@tonic-gate uintptr_t va; 10277c478bd9Sstevel@tonic-gate size_t len; 10287c478bd9Sstevel@tonic-gate uint_t prot; 10297c478bd9Sstevel@tonic-gate pfn_t pfn; 10307c478bd9Sstevel@tonic-gate int memblocks; 10311de082f7SVikram Hegde pfn_t rsvd_high_pfn; 10321de082f7SVikram Hegde pgcnt_t rsvd_pgcnt; 10331de082f7SVikram Hegde size_t rsvdmemlist_sz; 10341de082f7SVikram Hegde int rsvdmemblocks; 10357c478bd9Sstevel@tonic-gate caddr_t pagecolor_mem; 10367c478bd9Sstevel@tonic-gate size_t pagecolor_memsz; 10377c478bd9Sstevel@tonic-gate caddr_t page_ctrs_mem; 10387c478bd9Sstevel@tonic-gate size_t page_ctrs_size; 1039d7d93655Sblakej size_t pse_table_alloc_size; 10407c478bd9Sstevel@tonic-gate struct memlist *current; 10417c478bd9Sstevel@tonic-gate extern void startup_build_mem_nodes(struct memlist *); 10427c478bd9Sstevel@tonic-gate 10437c478bd9Sstevel@tonic-gate /* XX64 fix these - they should be in include files */ 10447c478bd9Sstevel@tonic-gate extern size_t page_coloring_init(uint_t, int, int); 10457c478bd9Sstevel@tonic-gate extern void page_coloring_setup(caddr_t); 10467c478bd9Sstevel@tonic-gate 10477c478bd9Sstevel@tonic-gate PRM_POINT("startup_memlist() starting..."); 10487c478bd9Sstevel@tonic-gate 10497c478bd9Sstevel@tonic-gate /* 10507c478bd9Sstevel@tonic-gate * Use leftover large page nucleus text/data space for loadable modules. 10517c478bd9Sstevel@tonic-gate * Use at most MODTEXT/MODDATA. 10527c478bd9Sstevel@tonic-gate */ 1053ae115bc7Smrj len = kbm_nucleus_size; 1054ae115bc7Smrj ASSERT(len > MMU_PAGESIZE); 10557c478bd9Sstevel@tonic-gate 1056ae115bc7Smrj moddata = (caddr_t)ROUND_UP_PAGE(e_data); 1057ae115bc7Smrj e_moddata = (caddr_t)P2ROUNDUP((uintptr_t)e_data, (uintptr_t)len); 1058ae115bc7Smrj if (e_moddata - moddata > MODDATA) 1059ae115bc7Smrj e_moddata = moddata + MODDATA; 10607c478bd9Sstevel@tonic-gate 1061ae115bc7Smrj modtext = (caddr_t)ROUND_UP_PAGE(e_text); 1062ae115bc7Smrj e_modtext = (caddr_t)P2ROUNDUP((uintptr_t)e_text, (uintptr_t)len); 1063ae115bc7Smrj if (e_modtext - modtext > MODTEXT) 1064ae115bc7Smrj e_modtext = modtext + MODTEXT; 10657c478bd9Sstevel@tonic-gate 10667c478bd9Sstevel@tonic-gate econtig = e_moddata; 10677c478bd9Sstevel@tonic-gate 10687c478bd9Sstevel@tonic-gate PRM_DEBUG(modtext); 10697c478bd9Sstevel@tonic-gate PRM_DEBUG(e_modtext); 10707c478bd9Sstevel@tonic-gate PRM_DEBUG(moddata); 10717c478bd9Sstevel@tonic-gate PRM_DEBUG(e_moddata); 10727c478bd9Sstevel@tonic-gate PRM_DEBUG(econtig); 10737c478bd9Sstevel@tonic-gate 10747c478bd9Sstevel@tonic-gate /* 1075ae115bc7Smrj * Examine the boot loader physical memory map to find out: 10767c478bd9Sstevel@tonic-gate * - total memory in system - physinstalled 10777c478bd9Sstevel@tonic-gate * - the max physical address - physmax 1078ae115bc7Smrj * - the number of discontiguous segments of memory. 10797c478bd9Sstevel@tonic-gate */ 10807c478bd9Sstevel@tonic-gate if (prom_debug) 1081ae115bc7Smrj print_memlist("boot physinstalled", 10827c478bd9Sstevel@tonic-gate bootops->boot_mem->physinstalled); 1083a3114836SGerry Liu installed_top_size_ex(bootops->boot_mem->physinstalled, &physmax, 10847c478bd9Sstevel@tonic-gate &physinstalled, &memblocks); 10857c478bd9Sstevel@tonic-gate PRM_DEBUG(physmax); 10867c478bd9Sstevel@tonic-gate PRM_DEBUG(physinstalled); 10877c478bd9Sstevel@tonic-gate PRM_DEBUG(memblocks); 10887c478bd9Sstevel@tonic-gate 1089a3114836SGerry Liu /* 1090*04909c8cSJohn Levon * We no longer support any form of memory DR. 1091a3114836SGerry Liu */ 1092*04909c8cSJohn Levon plat_dr_physmax = 0; 1093a3114836SGerry Liu 10941de082f7SVikram Hegde /* 10951de082f7SVikram Hegde * Examine the bios reserved memory to find out: 10961de082f7SVikram Hegde * - the number of discontiguous segments of memory. 10971de082f7SVikram Hegde */ 10981de082f7SVikram Hegde if (prom_debug) 10991de082f7SVikram Hegde print_memlist("boot reserved mem", 11001de082f7SVikram Hegde bootops->boot_mem->rsvdmem); 1101a3114836SGerry Liu installed_top_size_ex(bootops->boot_mem->rsvdmem, &rsvd_high_pfn, 11021de082f7SVikram Hegde &rsvd_pgcnt, &rsvdmemblocks); 11031de082f7SVikram Hegde PRM_DEBUG(rsvd_high_pfn); 11041de082f7SVikram Hegde PRM_DEBUG(rsvd_pgcnt); 11051de082f7SVikram Hegde PRM_DEBUG(rsvdmemblocks); 11061de082f7SVikram Hegde 11077c478bd9Sstevel@tonic-gate /* 11087c478bd9Sstevel@tonic-gate * Initialize hat's mmu parameters. 11097c478bd9Sstevel@tonic-gate * Check for enforce-prot-exec in boot environment. It's used to 11107c478bd9Sstevel@tonic-gate * enable/disable support for the page table entry NX bit. 11117c478bd9Sstevel@tonic-gate * The default is to enforce PROT_EXEC on processors that support NX. 11127c478bd9Sstevel@tonic-gate * Boot seems to round up the "len", but 8 seems to be big enough. 11137c478bd9Sstevel@tonic-gate */ 11147c478bd9Sstevel@tonic-gate mmu_init(); 11157c478bd9Sstevel@tonic-gate 11167c478bd9Sstevel@tonic-gate #ifdef __i386 11177c478bd9Sstevel@tonic-gate /* 11187c478bd9Sstevel@tonic-gate * physmax is lowered if there is more memory than can be 11197c478bd9Sstevel@tonic-gate * physically addressed in 32 bit (PAE/non-PAE) modes. 11207c478bd9Sstevel@tonic-gate */ 11217c478bd9Sstevel@tonic-gate if (mmu.pae_hat) { 11227c478bd9Sstevel@tonic-gate if (PFN_ABOVE64G(physmax)) { 11237c478bd9Sstevel@tonic-gate physinstalled -= (physmax - (PFN_64G - 1)); 11247c478bd9Sstevel@tonic-gate physmax = PFN_64G - 1; 11257c478bd9Sstevel@tonic-gate } 11267c478bd9Sstevel@tonic-gate } else { 11277c478bd9Sstevel@tonic-gate if (PFN_ABOVE4G(physmax)) { 11287c478bd9Sstevel@tonic-gate physinstalled -= (physmax - (PFN_4G - 1)); 11297c478bd9Sstevel@tonic-gate physmax = PFN_4G - 1; 11307c478bd9Sstevel@tonic-gate } 11317c478bd9Sstevel@tonic-gate } 11327c478bd9Sstevel@tonic-gate #endif 11337c478bd9Sstevel@tonic-gate 11347c478bd9Sstevel@tonic-gate startup_build_mem_nodes(bootops->boot_mem->physinstalled); 11357c478bd9Sstevel@tonic-gate 11367c478bd9Sstevel@tonic-gate if (BOP_GETPROPLEN(bootops, "enforce-prot-exec") >= 0) { 11377c478bd9Sstevel@tonic-gate int len = BOP_GETPROPLEN(bootops, "enforce-prot-exec"); 11387c478bd9Sstevel@tonic-gate char value[8]; 11397c478bd9Sstevel@tonic-gate 11407c478bd9Sstevel@tonic-gate if (len < 8) 11417c478bd9Sstevel@tonic-gate (void) BOP_GETPROP(bootops, "enforce-prot-exec", value); 11427c478bd9Sstevel@tonic-gate else 11437c478bd9Sstevel@tonic-gate (void) strcpy(value, ""); 11447c478bd9Sstevel@tonic-gate if (strcmp(value, "off") == 0) 11457c478bd9Sstevel@tonic-gate mmu.pt_nx = 0; 11467c478bd9Sstevel@tonic-gate } 11477c478bd9Sstevel@tonic-gate PRM_DEBUG(mmu.pt_nx); 11487c478bd9Sstevel@tonic-gate 11497c478bd9Sstevel@tonic-gate /* 11507c478bd9Sstevel@tonic-gate * We will need page_t's for every page in the system, except for 11517c478bd9Sstevel@tonic-gate * memory mapped at or above above the start of the kernel text segment. 11527c478bd9Sstevel@tonic-gate * 11537c478bd9Sstevel@tonic-gate * pages above e_modtext are attributed to kernel debugger (obp_pages) 11547c478bd9Sstevel@tonic-gate */ 11557c478bd9Sstevel@tonic-gate npages = physinstalled - 1; /* avail_filter() skips page 0, so "- 1" */ 11567c478bd9Sstevel@tonic-gate obp_pages = 0; 11577c478bd9Sstevel@tonic-gate va = KERNEL_TEXT; 1158ae115bc7Smrj while (kbm_probe(&va, &len, &pfn, &prot) != 0) { 11597c478bd9Sstevel@tonic-gate npages -= len >> MMU_PAGESHIFT; 11607c478bd9Sstevel@tonic-gate if (va >= (uintptr_t)e_moddata) 11617c478bd9Sstevel@tonic-gate obp_pages += len >> MMU_PAGESHIFT; 11627c478bd9Sstevel@tonic-gate va += len; 11637c478bd9Sstevel@tonic-gate } 11647c478bd9Sstevel@tonic-gate PRM_DEBUG(npages); 11657c478bd9Sstevel@tonic-gate PRM_DEBUG(obp_pages); 11667c478bd9Sstevel@tonic-gate 11677c478bd9Sstevel@tonic-gate /* 116835b1ab99Sjosephb * If physmem is patched to be non-zero, use it instead of the computed 116935b1ab99Sjosephb * value unless it is larger than the actual amount of memory on hand. 11707c478bd9Sstevel@tonic-gate */ 11714944b02eSkchow if (physmem == 0 || physmem > npages) { 11727c478bd9Sstevel@tonic-gate physmem = npages; 11734944b02eSkchow } else if (physmem < npages) { 1174f53ad214Skchow orig_npages = npages; 11757c478bd9Sstevel@tonic-gate npages = physmem; 11764944b02eSkchow } 11777c478bd9Sstevel@tonic-gate PRM_DEBUG(physmem); 11787c478bd9Sstevel@tonic-gate 11797c478bd9Sstevel@tonic-gate /* 11807c478bd9Sstevel@tonic-gate * We now compute the sizes of all the initial allocations for 11817c478bd9Sstevel@tonic-gate * structures the kernel needs in order do kmem_alloc(). These 11827c478bd9Sstevel@tonic-gate * include: 11837c478bd9Sstevel@tonic-gate * memsegs 11847c478bd9Sstevel@tonic-gate * memlists 11857c478bd9Sstevel@tonic-gate * page hash table 11867c478bd9Sstevel@tonic-gate * page_t's 11877c478bd9Sstevel@tonic-gate * page coloring data structs 11887c478bd9Sstevel@tonic-gate */ 11897c478bd9Sstevel@tonic-gate memseg_sz = sizeof (struct memseg) * (memblocks + POSS_NEW_FRAGMENTS); 11907c478bd9Sstevel@tonic-gate ADD_TO_ALLOCATIONS(memseg_base, memseg_sz); 11917c478bd9Sstevel@tonic-gate PRM_DEBUG(memseg_sz); 11927c478bd9Sstevel@tonic-gate 11937c478bd9Sstevel@tonic-gate /* 1194ae115bc7Smrj * Reserve space for memlists. There's no real good way to know exactly 1195ae115bc7Smrj * how much room we'll need, but this should be a good upper bound. 11967c478bd9Sstevel@tonic-gate */ 11977c478bd9Sstevel@tonic-gate memlist_sz = ROUND_UP_PAGE(2 * sizeof (struct memlist) * 11987c478bd9Sstevel@tonic-gate (memblocks + POSS_NEW_FRAGMENTS)); 11997c478bd9Sstevel@tonic-gate ADD_TO_ALLOCATIONS(memlist, memlist_sz); 12007c478bd9Sstevel@tonic-gate PRM_DEBUG(memlist_sz); 12017c478bd9Sstevel@tonic-gate 12021de082f7SVikram Hegde /* 12031de082f7SVikram Hegde * Reserve space for bios reserved memlists. 12041de082f7SVikram Hegde */ 12051de082f7SVikram Hegde rsvdmemlist_sz = ROUND_UP_PAGE(2 * sizeof (struct memlist) * 12061de082f7SVikram Hegde (rsvdmemblocks + POSS_NEW_FRAGMENTS)); 12071de082f7SVikram Hegde ADD_TO_ALLOCATIONS(bios_rsvd, rsvdmemlist_sz); 12081de082f7SVikram Hegde PRM_DEBUG(rsvdmemlist_sz); 12091de082f7SVikram Hegde 1210cb15d5d9SPeter Rival /* LINTED */ 1211cb15d5d9SPeter Rival ASSERT(P2SAMEHIGHBIT((1 << PP_SHIFT), sizeof (struct page))); 12127c478bd9Sstevel@tonic-gate /* 12137c478bd9Sstevel@tonic-gate * The page structure hash table size is a power of 2 12147c478bd9Sstevel@tonic-gate * such that the average hash chain length is PAGE_HASHAVELEN. 12157c478bd9Sstevel@tonic-gate */ 12167c478bd9Sstevel@tonic-gate page_hashsz = npages / PAGE_HASHAVELEN; 1217cb15d5d9SPeter Rival page_hashsz_shift = highbit(page_hashsz); 1218cb15d5d9SPeter Rival page_hashsz = 1 << page_hashsz_shift; 12197c478bd9Sstevel@tonic-gate pagehash_sz = sizeof (struct page *) * page_hashsz; 12207c478bd9Sstevel@tonic-gate ADD_TO_ALLOCATIONS(page_hash, pagehash_sz); 12217c478bd9Sstevel@tonic-gate PRM_DEBUG(pagehash_sz); 12227c478bd9Sstevel@tonic-gate 12237c478bd9Sstevel@tonic-gate /* 1224ae115bc7Smrj * Set aside room for the page structures themselves. 12257c478bd9Sstevel@tonic-gate */ 1226ae115bc7Smrj PRM_DEBUG(npages); 1227ae115bc7Smrj pp_sz = sizeof (struct page) * npages; 12287c478bd9Sstevel@tonic-gate ADD_TO_ALLOCATIONS(pp_base, pp_sz); 12297c478bd9Sstevel@tonic-gate PRM_DEBUG(pp_sz); 12307c478bd9Sstevel@tonic-gate 12317c478bd9Sstevel@tonic-gate /* 12327c478bd9Sstevel@tonic-gate * determine l2 cache info and memory size for page coloring 12337c478bd9Sstevel@tonic-gate */ 12347c478bd9Sstevel@tonic-gate (void) getl2cacheinfo(CPU, 12357c478bd9Sstevel@tonic-gate &l2cache_sz, &l2cache_linesz, &l2cache_assoc); 12367c478bd9Sstevel@tonic-gate pagecolor_memsz = 12377c478bd9Sstevel@tonic-gate page_coloring_init(l2cache_sz, l2cache_linesz, l2cache_assoc); 12387c478bd9Sstevel@tonic-gate ADD_TO_ALLOCATIONS(pagecolor_mem, pagecolor_memsz); 12397c478bd9Sstevel@tonic-gate PRM_DEBUG(pagecolor_memsz); 12407c478bd9Sstevel@tonic-gate 12417c478bd9Sstevel@tonic-gate page_ctrs_size = page_ctrs_sz(); 12427c478bd9Sstevel@tonic-gate ADD_TO_ALLOCATIONS(page_ctrs_mem, page_ctrs_size); 12437c478bd9Sstevel@tonic-gate PRM_DEBUG(page_ctrs_size); 12447c478bd9Sstevel@tonic-gate 1245d7d93655Sblakej /* 1246d7d93655Sblakej * Allocate the array that protects pp->p_selock. 1247d7d93655Sblakej */ 1248d7d93655Sblakej pse_shift = size_pse_array(physmem, max_ncpus); 1249d7d93655Sblakej pse_table_size = 1 << pse_shift; 1250d7d93655Sblakej pse_table_alloc_size = pse_table_size * sizeof (pad_mutex_t); 1251d7d93655Sblakej ADD_TO_ALLOCATIONS(pse_mutex, pse_table_alloc_size); 1252d7d93655Sblakej 12537c478bd9Sstevel@tonic-gate valloc_sz = ROUND_UP_LPAGE(valloc_sz); 1254ae115bc7Smrj valloc_base = VALLOC_BASE; 125535b1ab99Sjosephb 125635b1ab99Sjosephb /* 1257*04909c8cSJohn Levon * The signicant memory-sized regions are roughly sized as follows in 1258*04909c8cSJohn Levon * the default layout with max physmem: 1259*04909c8cSJohn Levon * segkpm: 1x physmem allocated (but 1Tb room, below VALLOC_BASE) 1260*04909c8cSJohn Levon * segzio: 1.5x physmem 1261*04909c8cSJohn Levon * segkvmm: 4x physmem 1262*04909c8cSJohn Levon * heap: whatever's left up to COREHEAP_BASE, at least 1.5x physmem 1263*04909c8cSJohn Levon * 1264*04909c8cSJohn Levon * The idea is that we leave enough room to avoid fragmentation issues, 1265*04909c8cSJohn Levon * so we would like the VA arenas to have some extra. 1266af868f46SMatthew Ahrens * 1267*04909c8cSJohn Levon * Ignoring the loose change of segkp, valloc, and such, this means that 1268*04909c8cSJohn Levon * as COREHEAP_BASE-VALLOC_BASE=2Tb, we can accommodate a physmem up to 1269*04909c8cSJohn Levon * about (2Tb / 7.0), rounded down to 256Gb in the check below. 1270af868f46SMatthew Ahrens * 1271*04909c8cSJohn Levon * Note that KPM lives below VALLOC_BASE, but we want to include it in 1272*04909c8cSJohn Levon * adjustments, hence the 8 below. 1273*04909c8cSJohn Levon * 1274*04909c8cSJohn Levon * Beyond 256Gb, we push segkpm_base (and hence kernelbase and 1275*04909c8cSJohn Levon * _userlimit) down to accommodate the VA requirements above. 127635b1ab99Sjosephb */ 1277*04909c8cSJohn Levon if (physmax + 1 > mmu_btop(TERABYTE / 4)) { 1278*04909c8cSJohn Levon uint64_t physmem_bytes = mmu_ptob(physmax + 1); 1279*04909c8cSJohn Levon uint64_t adjustment = 8 * (physmem_bytes - (TERABYTE / 4)); 128035b1ab99Sjosephb 1281*04909c8cSJohn Levon PRM_DEBUG(adjustment); 1282a3114836SGerry Liu 1283af868f46SMatthew Ahrens /* 1284*04909c8cSJohn Levon * segkpm_base is always aligned on a L3 PTE boundary. 1285af868f46SMatthew Ahrens */ 1286*04909c8cSJohn Levon segkpm_base -= P2ROUNDUP(adjustment, KERNEL_REDZONE_SIZE); 128735b1ab99Sjosephb 1288*04909c8cSJohn Levon /* 1289*04909c8cSJohn Levon * But make sure we leave some space for user apps above hole. 1290*04909c8cSJohn Levon */ 129135b1ab99Sjosephb segkpm_base = MAX(segkpm_base, AMD64_VA_HOLE_END + TERABYTE); 129235b1ab99Sjosephb 1293*04909c8cSJohn Levon ASSERT(segkpm_base <= SEGKPM_BASE); 1294*04909c8cSJohn Levon 1295*04909c8cSJohn Levon valloc_base = segkpm_base + P2ROUNDUP(physmem_bytes, ONE_GIG); 12966c8c89eaSKit Chow if (valloc_base < segkpm_base) 12976c8c89eaSKit Chow panic("not enough kernel VA to support memory size"); 129835b1ab99Sjosephb } 1299*04909c8cSJohn Levon 1300*04909c8cSJohn Levon PRM_DEBUG(segkpm_base); 13017c478bd9Sstevel@tonic-gate PRM_DEBUG(valloc_base); 1302ae115bc7Smrj 1303ae115bc7Smrj /* 1304ae115bc7Smrj * do all the initial allocations 1305ae115bc7Smrj */ 1306ae115bc7Smrj perform_allocations(); 1307ae115bc7Smrj 1308ae115bc7Smrj /* 1309ae115bc7Smrj * Build phys_install and phys_avail in kernel memspace. 1310ae115bc7Smrj * - phys_install should be all memory in the system. 1311ae115bc7Smrj * - phys_avail is phys_install minus any memory mapped before this 1312ae115bc7Smrj * point above KERNEL_TEXT. 1313ae115bc7Smrj */ 1314ae115bc7Smrj current = phys_install = memlist; 1315ae115bc7Smrj copy_memlist_filter(bootops->boot_mem->physinstalled, ¤t, NULL); 1316ae115bc7Smrj if ((caddr_t)current > (caddr_t)memlist + memlist_sz) 1317ae115bc7Smrj panic("physinstalled was too big!"); 1318ae115bc7Smrj if (prom_debug) 1319ae115bc7Smrj print_memlist("phys_install", phys_install); 1320ae115bc7Smrj 1321ae115bc7Smrj phys_avail = current; 1322ae115bc7Smrj PRM_POINT("Building phys_avail:\n"); 1323ae115bc7Smrj copy_memlist_filter(bootops->boot_mem->physinstalled, ¤t, 1324ae115bc7Smrj avail_filter); 1325ae115bc7Smrj if ((caddr_t)current > (caddr_t)memlist + memlist_sz) 1326ae115bc7Smrj panic("physavail was too big!"); 1327ae115bc7Smrj if (prom_debug) 1328ae115bc7Smrj print_memlist("phys_avail", phys_avail); 1329a3114836SGerry Liu #ifndef __xpv 1330a3114836SGerry Liu /* 1331a3114836SGerry Liu * Free unused memlist items, which may be used by memory DR driver 1332a3114836SGerry Liu * at runtime. 1333a3114836SGerry Liu */ 1334a3114836SGerry Liu if ((caddr_t)current < (caddr_t)memlist + memlist_sz) { 1335a3114836SGerry Liu memlist_free_block((caddr_t)current, 1336a3114836SGerry Liu (caddr_t)memlist + memlist_sz - (caddr_t)current); 1337a3114836SGerry Liu } 1338a3114836SGerry Liu #endif 1339ae115bc7Smrj 13401de082f7SVikram Hegde /* 13411de082f7SVikram Hegde * Build bios reserved memspace 13421de082f7SVikram Hegde */ 13431de082f7SVikram Hegde current = bios_rsvd; 13441de082f7SVikram Hegde copy_memlist_filter(bootops->boot_mem->rsvdmem, ¤t, NULL); 13451de082f7SVikram Hegde if ((caddr_t)current > (caddr_t)bios_rsvd + rsvdmemlist_sz) 13461de082f7SVikram Hegde panic("bios_rsvd was too big!"); 13471de082f7SVikram Hegde if (prom_debug) 13481de082f7SVikram Hegde print_memlist("bios_rsvd", bios_rsvd); 1349a3114836SGerry Liu #ifndef __xpv 1350a3114836SGerry Liu /* 1351a3114836SGerry Liu * Free unused memlist items, which may be used by memory DR driver 1352a3114836SGerry Liu * at runtime. 1353a3114836SGerry Liu */ 1354a3114836SGerry Liu if ((caddr_t)current < (caddr_t)bios_rsvd + rsvdmemlist_sz) { 1355a3114836SGerry Liu memlist_free_block((caddr_t)current, 1356a3114836SGerry Liu (caddr_t)bios_rsvd + rsvdmemlist_sz - (caddr_t)current); 1357a3114836SGerry Liu } 1358a3114836SGerry Liu #endif 13591de082f7SVikram Hegde 1360ae115bc7Smrj /* 1361ae115bc7Smrj * setup page coloring 1362ae115bc7Smrj */ 1363ae115bc7Smrj page_coloring_setup(pagecolor_mem); 1364ae115bc7Smrj page_lock_init(); /* currently a no-op */ 1365ae115bc7Smrj 1366ae115bc7Smrj /* 1367ae115bc7Smrj * free page list counters 1368ae115bc7Smrj */ 1369ae115bc7Smrj (void) page_ctrs_alloc(page_ctrs_mem); 1370ae115bc7Smrj 137106fb6a36Sdv /* 137206fb6a36Sdv * Size the pcf array based on the number of cpus in the box at 137306fb6a36Sdv * boot time. 137406fb6a36Sdv */ 137506fb6a36Sdv 137606fb6a36Sdv pcf_init(); 137706fb6a36Sdv 1378ae115bc7Smrj /* 1379ae115bc7Smrj * Initialize the page structures from the memory lists. 1380ae115bc7Smrj */ 1381ae115bc7Smrj availrmem_initial = availrmem = freemem = 0; 1382ae115bc7Smrj PRM_POINT("Calling kphysm_init()..."); 1383ae115bc7Smrj npages = kphysm_init(pp_base, npages); 1384ae115bc7Smrj PRM_POINT("kphysm_init() done"); 1385ae115bc7Smrj PRM_DEBUG(npages); 1386ae115bc7Smrj 1387ae115bc7Smrj init_debug_info(); 1388ae115bc7Smrj 1389ae115bc7Smrj /* 1390ae115bc7Smrj * Now that page_t's have been initialized, remove all the 1391ae115bc7Smrj * initial allocation pages from the kernel free page lists. 1392ae115bc7Smrj */ 1393ae115bc7Smrj boot_mapin((caddr_t)valloc_base, valloc_sz); 13940cfdb603Sjosephb boot_mapin((caddr_t)MISC_VA_BASE, MISC_VA_SIZE); 1395ae115bc7Smrj PRM_POINT("startup_memlist() done"); 1396ae115bc7Smrj 1397ae115bc7Smrj PRM_DEBUG(valloc_sz); 13982cb27123Saguzovsk 1399567d55e1Saguzovsk #if defined(__amd64) 1400567d55e1Saguzovsk if ((availrmem >> (30 - MMU_PAGESHIFT)) >= 1401567d55e1Saguzovsk textrepl_min_gb && l2cache_sz <= 2 << 20) { 1402567d55e1Saguzovsk extern size_t textrepl_size_thresh; 14032cb27123Saguzovsk textrepl_size_thresh = (16 << 20) - 1; 14042cb27123Saguzovsk } 1405567d55e1Saguzovsk #endif 1406ae115bc7Smrj } 1407ae115bc7Smrj 1408ae115bc7Smrj /* 1409ae115bc7Smrj * Layout the kernel's part of address space and initialize kmem allocator. 1410ae115bc7Smrj */ 1411ae115bc7Smrj static void 1412ae115bc7Smrj startup_kmem(void) 1413ae115bc7Smrj { 1414932dc8e5Sdp extern void page_set_colorequiv_arr(void); 141574ecdb51SJohn Levon #if !defined(__xpv) 141674ecdb51SJohn Levon extern uint64_t kpti_kbase; 141774ecdb51SJohn Levon #endif 1418932dc8e5Sdp 1419ae115bc7Smrj PRM_POINT("startup_kmem() starting..."); 14207c478bd9Sstevel@tonic-gate 14217c478bd9Sstevel@tonic-gate #if defined(__amd64) 14227c478bd9Sstevel@tonic-gate if (eprom_kernelbase && eprom_kernelbase != KERNELBASE) 14237c478bd9Sstevel@tonic-gate cmn_err(CE_NOTE, "!kernelbase cannot be changed on 64-bit " 14247c478bd9Sstevel@tonic-gate "systems."); 142535b1ab99Sjosephb kernelbase = segkpm_base - KERNEL_REDZONE_SIZE; 14267c478bd9Sstevel@tonic-gate core_base = (uintptr_t)COREHEAP_BASE; 1427ae115bc7Smrj core_size = (size_t)MISC_VA_BASE - COREHEAP_BASE; 14287c478bd9Sstevel@tonic-gate #else /* __i386 */ 14297c478bd9Sstevel@tonic-gate /* 14307c478bd9Sstevel@tonic-gate * We configure kernelbase based on: 14317c478bd9Sstevel@tonic-gate * 14327c478bd9Sstevel@tonic-gate * 1. user specified kernelbase via eeprom command. Value cannot exceed 14337c478bd9Sstevel@tonic-gate * KERNELBASE_MAX. we large page align eprom_kernelbase 14347c478bd9Sstevel@tonic-gate * 14357c478bd9Sstevel@tonic-gate * 2. Default to KERNELBASE and adjust to 2X less the size for page_t. 14367c478bd9Sstevel@tonic-gate * On large memory systems we must lower kernelbase to allow 14377c478bd9Sstevel@tonic-gate * enough room for page_t's for all of memory. 14387c478bd9Sstevel@tonic-gate * 14397c478bd9Sstevel@tonic-gate * The value set here, might be changed a little later. 14407c478bd9Sstevel@tonic-gate */ 14417c478bd9Sstevel@tonic-gate if (eprom_kernelbase) { 14427c478bd9Sstevel@tonic-gate kernelbase = eprom_kernelbase & mmu.level_mask[1]; 14437c478bd9Sstevel@tonic-gate if (kernelbase > KERNELBASE_MAX) 14447c478bd9Sstevel@tonic-gate kernelbase = KERNELBASE_MAX; 14457c478bd9Sstevel@tonic-gate } else { 14467c478bd9Sstevel@tonic-gate kernelbase = (uintptr_t)KERNELBASE; 14477c478bd9Sstevel@tonic-gate kernelbase -= ROUND_UP_4MEG(2 * valloc_sz); 14487c478bd9Sstevel@tonic-gate } 14497c478bd9Sstevel@tonic-gate ASSERT((kernelbase & mmu.level_offset[1]) == 0); 1450ae115bc7Smrj core_base = valloc_base; 14517c478bd9Sstevel@tonic-gate core_size = 0; 1452ae115bc7Smrj #endif /* __i386 */ 14537c478bd9Sstevel@tonic-gate 14547c478bd9Sstevel@tonic-gate PRM_DEBUG(core_base); 14557c478bd9Sstevel@tonic-gate PRM_DEBUG(core_size); 1456ae115bc7Smrj PRM_DEBUG(kernelbase); 14577c478bd9Sstevel@tonic-gate 145835b1ab99Sjosephb #if defined(__i386) 14597c478bd9Sstevel@tonic-gate segkp_fromheap = 1; 1460ae115bc7Smrj #endif /* __i386 */ 146135b1ab99Sjosephb 14627c478bd9Sstevel@tonic-gate ekernelheap = (char *)core_base; 1463ae115bc7Smrj PRM_DEBUG(ekernelheap); 14647c478bd9Sstevel@tonic-gate 14657c478bd9Sstevel@tonic-gate /* 14667c478bd9Sstevel@tonic-gate * Now that we know the real value of kernelbase, 14677c478bd9Sstevel@tonic-gate * update variables that were initialized with a value of 14687c478bd9Sstevel@tonic-gate * KERNELBASE (in common/conf/param.c). 14697c478bd9Sstevel@tonic-gate * 14707c478bd9Sstevel@tonic-gate * XXX The problem with this sort of hackery is that the 14717c478bd9Sstevel@tonic-gate * compiler just may feel like putting the const declarations 14727c478bd9Sstevel@tonic-gate * (in param.c) into the .text section. Perhaps they should 14737c478bd9Sstevel@tonic-gate * just be declared as variables there? 14747c478bd9Sstevel@tonic-gate */ 14757c478bd9Sstevel@tonic-gate 14767c478bd9Sstevel@tonic-gate *(uintptr_t *)&_kernelbase = kernelbase; 14777c478bd9Sstevel@tonic-gate *(uintptr_t *)&_userlimit = kernelbase; 1478ccbaea4fSjosephb #if defined(__amd64) 1479ccbaea4fSjosephb *(uintptr_t *)&_userlimit -= KERNELBASE - USERLIMIT; 148074ecdb51SJohn Levon #if !defined(__xpv) 148174ecdb51SJohn Levon kpti_kbase = kernelbase; 148274ecdb51SJohn Levon #endif 1483ccbaea4fSjosephb #else 14847c478bd9Sstevel@tonic-gate *(uintptr_t *)&_userlimit32 = _userlimit; 14857c478bd9Sstevel@tonic-gate #endif 14867c478bd9Sstevel@tonic-gate PRM_DEBUG(_kernelbase); 14877c478bd9Sstevel@tonic-gate PRM_DEBUG(_userlimit); 14887c478bd9Sstevel@tonic-gate PRM_DEBUG(_userlimit32); 14897c478bd9Sstevel@tonic-gate 149074ecdb51SJohn Levon /* We have to re-do this now that we've modified _userlimit. */ 149174ecdb51SJohn Levon mmu_calc_user_slots(); 149274ecdb51SJohn Levon 149335b1ab99Sjosephb layout_kernel_va(); 149435b1ab99Sjosephb 149535b1ab99Sjosephb #if defined(__i386) 149635b1ab99Sjosephb /* 149735b1ab99Sjosephb * If segmap is too large we can push the bottom of the kernel heap 149835b1ab99Sjosephb * higher than the base. Or worse, it could exceed the top of the 149935b1ab99Sjosephb * VA space entirely, causing it to wrap around. 150035b1ab99Sjosephb */ 150135b1ab99Sjosephb if (kernelheap >= ekernelheap || (uintptr_t)kernelheap < kernelbase) 150235b1ab99Sjosephb panic("too little address space available for kernelheap," 150335b1ab99Sjosephb " use eeprom for lower kernelbase or smaller segmapsize"); 150435b1ab99Sjosephb #endif /* __i386 */ 150535b1ab99Sjosephb 15067c478bd9Sstevel@tonic-gate /* 15077c478bd9Sstevel@tonic-gate * Initialize the kernel heap. Note 3rd argument must be > 1st. 15087c478bd9Sstevel@tonic-gate */ 150935b1ab99Sjosephb kernelheap_init(kernelheap, ekernelheap, 151035b1ab99Sjosephb kernelheap + MMU_PAGESIZE, 1511ae115bc7Smrj (void *)core_base, (void *)(core_base + core_size)); 15127c478bd9Sstevel@tonic-gate 1513843e1988Sjohnlev #if defined(__xpv) 1514843e1988Sjohnlev /* 1515843e1988Sjohnlev * Link pending events struct into cpu struct 1516843e1988Sjohnlev */ 1517843e1988Sjohnlev CPU->cpu_m.mcpu_evt_pend = &cpu0_evt_data; 1518843e1988Sjohnlev #endif 15197c478bd9Sstevel@tonic-gate /* 15207c478bd9Sstevel@tonic-gate * Initialize kernel memory allocator. 15217c478bd9Sstevel@tonic-gate */ 15227c478bd9Sstevel@tonic-gate kmem_init(); 15237c478bd9Sstevel@tonic-gate 1524932dc8e5Sdp /* 1525932dc8e5Sdp * Factor in colorequiv to check additional 'equivalent' bins 1526932dc8e5Sdp */ 1527932dc8e5Sdp page_set_colorequiv_arr(); 1528932dc8e5Sdp 15297c478bd9Sstevel@tonic-gate /* 15307c478bd9Sstevel@tonic-gate * print this out early so that we know what's going on 15317c478bd9Sstevel@tonic-gate */ 15327417cfdeSKuriakose Kuruvilla print_x86_featureset(x86_featureset); 15337c478bd9Sstevel@tonic-gate 15347c478bd9Sstevel@tonic-gate /* 15357c478bd9Sstevel@tonic-gate * Initialize bp_mapin(). 15367c478bd9Sstevel@tonic-gate */ 15377c478bd9Sstevel@tonic-gate bp_init(MMU_PAGESIZE, HAT_STORECACHING_OK); 15387c478bd9Sstevel@tonic-gate 1539f53ad214Skchow /* 1540f53ad214Skchow * orig_npages is non-zero if physmem has been configured for less 1541f53ad214Skchow * than the available memory. 1542f53ad214Skchow */ 1543f53ad214Skchow if (orig_npages) { 154435b1ab99Sjosephb cmn_err(CE_WARN, "!%slimiting physmem to 0x%lx of 0x%lx pages", 154535b1ab99Sjosephb (npages == PHYSMEM ? "Due to virtual address space " : ""), 154635b1ab99Sjosephb npages, orig_npages); 1547f53ad214Skchow } 15487c478bd9Sstevel@tonic-gate #if defined(__i386) 15497c478bd9Sstevel@tonic-gate if (eprom_kernelbase && (eprom_kernelbase != kernelbase)) 15507c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "kernelbase value, User specified 0x%lx, " 15517c478bd9Sstevel@tonic-gate "System using 0x%lx", 15527c478bd9Sstevel@tonic-gate (uintptr_t)eprom_kernelbase, (uintptr_t)kernelbase); 15537c478bd9Sstevel@tonic-gate #endif 15547c478bd9Sstevel@tonic-gate 15557c478bd9Sstevel@tonic-gate #ifdef KERNELBASE_ABI_MIN 15567c478bd9Sstevel@tonic-gate if (kernelbase < (uintptr_t)KERNELBASE_ABI_MIN) { 15577c478bd9Sstevel@tonic-gate cmn_err(CE_NOTE, "!kernelbase set to 0x%lx, system is not " 15587c478bd9Sstevel@tonic-gate "i386 ABI compliant.", (uintptr_t)kernelbase); 15597c478bd9Sstevel@tonic-gate } 15607c478bd9Sstevel@tonic-gate #endif 15617c478bd9Sstevel@tonic-gate 1562a3114836SGerry Liu #ifndef __xpv 1563a3114836SGerry Liu if (plat_dr_support_memory()) { 1564a3114836SGerry Liu mem_config_init(); 1565a3114836SGerry Liu } 1566a3114836SGerry Liu #else /* __xpv */ 1567843e1988Sjohnlev /* 1568843e1988Sjohnlev * Some of the xen start information has to be relocated up 1569843e1988Sjohnlev * into the kernel's permanent address space. 1570843e1988Sjohnlev */ 1571843e1988Sjohnlev PRM_POINT("calling xen_relocate_start_info()"); 1572843e1988Sjohnlev xen_relocate_start_info(); 1573843e1988Sjohnlev PRM_POINT("xen_relocate_start_info() done"); 1574843e1988Sjohnlev 1575843e1988Sjohnlev /* 1576843e1988Sjohnlev * (Update the vcpu pointer in our cpu structure to point into 1577843e1988Sjohnlev * the relocated shared info.) 1578843e1988Sjohnlev */ 1579843e1988Sjohnlev CPU->cpu_m.mcpu_vcpu_info = 1580843e1988Sjohnlev &HYPERVISOR_shared_info->vcpu_info[CPU->cpu_id]; 1581a3114836SGerry Liu #endif /* __xpv */ 1582843e1988Sjohnlev 1583ae115bc7Smrj PRM_POINT("startup_kmem() done"); 15847c478bd9Sstevel@tonic-gate } 15857c478bd9Sstevel@tonic-gate 1586551bc2a6Smrj #ifndef __xpv 1587551bc2a6Smrj /* 1588551bc2a6Smrj * If we have detected that we are running in an HVM environment, we need 1589551bc2a6Smrj * to prepend the PV driver directory to the module search path. 1590551bc2a6Smrj */ 1591551bc2a6Smrj #define HVM_MOD_DIR "/platform/i86hvm/kernel" 1592551bc2a6Smrj static void 1593551bc2a6Smrj update_default_path() 1594551bc2a6Smrj { 1595551bc2a6Smrj char *current, *newpath; 1596551bc2a6Smrj int newlen; 1597551bc2a6Smrj 1598551bc2a6Smrj /* 1599551bc2a6Smrj * We are about to resync with krtld. krtld will reset its 1600551bc2a6Smrj * internal module search path iff Solaris has set default_path. 1601551bc2a6Smrj * We want to be sure we're prepending this new directory to the 1602551bc2a6Smrj * right search path. 1603551bc2a6Smrj */ 1604551bc2a6Smrj current = (default_path == NULL) ? kobj_module_path : default_path; 1605551bc2a6Smrj 1606b6611e3bSStuart Maybee newlen = strlen(HVM_MOD_DIR) + strlen(current) + 2; 1607551bc2a6Smrj newpath = kmem_alloc(newlen, KM_SLEEP); 1608551bc2a6Smrj (void) strcpy(newpath, HVM_MOD_DIR); 1609551bc2a6Smrj (void) strcat(newpath, " "); 1610551bc2a6Smrj (void) strcat(newpath, current); 1611551bc2a6Smrj 1612551bc2a6Smrj default_path = newpath; 1613551bc2a6Smrj } 1614551bc2a6Smrj #endif 1615551bc2a6Smrj 16167c478bd9Sstevel@tonic-gate static void 16177c478bd9Sstevel@tonic-gate startup_modules(void) 16187c478bd9Sstevel@tonic-gate { 1619ed5289f9SKen Erickson int cnt; 16207c478bd9Sstevel@tonic-gate extern void prom_setup(void); 1621ed5289f9SKen Erickson int32_t v, h; 1622ed5289f9SKen Erickson char d[11]; 1623ed5289f9SKen Erickson char *cp; 1624e4b86885SCheng Sean Ye cmi_hdl_t hdl; 16257c478bd9Sstevel@tonic-gate 16267c478bd9Sstevel@tonic-gate PRM_POINT("startup_modules() starting..."); 1627843e1988Sjohnlev 1628843e1988Sjohnlev #ifndef __xpv 16297c478bd9Sstevel@tonic-gate /* 16307c478bd9Sstevel@tonic-gate * Initialize ten-micro second timer so that drivers will 16317c478bd9Sstevel@tonic-gate * not get short changed in their init phase. This was 16327c478bd9Sstevel@tonic-gate * not getting called until clkinit which, on fast cpu's 16337c478bd9Sstevel@tonic-gate * caused the drv_usecwait to be way too short. 16347c478bd9Sstevel@tonic-gate */ 16357c478bd9Sstevel@tonic-gate microfind(); 1636551bc2a6Smrj 163779ec9da8SYuri Pankov if ((get_hwenv() & HW_XEN_HVM) != 0) 1638551bc2a6Smrj update_default_path(); 1639843e1988Sjohnlev #endif 16407c478bd9Sstevel@tonic-gate 16417c478bd9Sstevel@tonic-gate /* 16427c478bd9Sstevel@tonic-gate * Read the GMT lag from /etc/rtc_config. 16437c478bd9Sstevel@tonic-gate */ 1644ae115bc7Smrj sgmtl(process_rtc_config_file()); 16457c478bd9Sstevel@tonic-gate 16467c478bd9Sstevel@tonic-gate /* 16477c478bd9Sstevel@tonic-gate * Calculate default settings of system parameters based upon 16487c478bd9Sstevel@tonic-gate * maxusers, yet allow to be overridden via the /etc/system file. 16497c478bd9Sstevel@tonic-gate */ 16507c478bd9Sstevel@tonic-gate param_calc(0); 16517c478bd9Sstevel@tonic-gate 16527c478bd9Sstevel@tonic-gate mod_setup(); 16537c478bd9Sstevel@tonic-gate 16547c478bd9Sstevel@tonic-gate /* 16557c478bd9Sstevel@tonic-gate * Initialize system parameters. 16567c478bd9Sstevel@tonic-gate */ 16577c478bd9Sstevel@tonic-gate param_init(); 16587c478bd9Sstevel@tonic-gate 16599acbbeafSnn /* 16609acbbeafSnn * Initialize the default brands 16619acbbeafSnn */ 16629acbbeafSnn brand_init(); 16639acbbeafSnn 16647c478bd9Sstevel@tonic-gate /* 16657c478bd9Sstevel@tonic-gate * maxmem is the amount of physical memory we're playing with. 16667c478bd9Sstevel@tonic-gate */ 16677c478bd9Sstevel@tonic-gate maxmem = physmem; 16687c478bd9Sstevel@tonic-gate 16697c478bd9Sstevel@tonic-gate /* 16707c478bd9Sstevel@tonic-gate * Initialize segment management stuff. 16717c478bd9Sstevel@tonic-gate */ 16727c478bd9Sstevel@tonic-gate seg_init(); 16737c478bd9Sstevel@tonic-gate 16747c478bd9Sstevel@tonic-gate if (modload("fs", "specfs") == -1) 16757c478bd9Sstevel@tonic-gate halt("Can't load specfs"); 16767c478bd9Sstevel@tonic-gate 16777c478bd9Sstevel@tonic-gate if (modload("fs", "devfs") == -1) 16787c478bd9Sstevel@tonic-gate halt("Can't load devfs"); 16797c478bd9Sstevel@tonic-gate 1680facf4a8dSllai if (modload("fs", "dev") == -1) 1681facf4a8dSllai halt("Can't load dev"); 1682facf4a8dSllai 168335a5a358SJonathan Adams if (modload("fs", "procfs") == -1) 168435a5a358SJonathan Adams halt("Can't load procfs"); 168535a5a358SJonathan Adams 168645916cd2Sjpk (void) modloadonly("sys", "lbl_edition"); 168745916cd2Sjpk 16887c478bd9Sstevel@tonic-gate dispinit(); 16897c478bd9Sstevel@tonic-gate 16907c478bd9Sstevel@tonic-gate /* Read cluster configuration data. */ 16917c478bd9Sstevel@tonic-gate clconf_init(); 16927c478bd9Sstevel@tonic-gate 1693843e1988Sjohnlev #if defined(__xpv) 1694349b53ddSStuart Maybee (void) ec_init(); 1695843e1988Sjohnlev gnttab_init(); 1696843e1988Sjohnlev (void) xs_early_init(); 1697843e1988Sjohnlev #endif /* __xpv */ 1698843e1988Sjohnlev 16997c478bd9Sstevel@tonic-gate /* 17007c478bd9Sstevel@tonic-gate * Create a kernel device tree. First, create rootnex and 17017c478bd9Sstevel@tonic-gate * then invoke bus specific code to probe devices. 17027c478bd9Sstevel@tonic-gate */ 17037c478bd9Sstevel@tonic-gate setup_ddi(); 17047aec1d6eScindi 1705074bb90dSTom Pothier #ifdef __xpv 1706074bb90dSTom Pothier if (DOMAIN_IS_INITDOMAIN(xen_info)) 1707074bb90dSTom Pothier #endif 1708074bb90dSTom Pothier { 1709e41d648bSToomas Soome id_t smid; 1710e41d648bSToomas Soome smbios_system_t smsys; 1711e41d648bSToomas Soome smbios_info_t sminfo; 1712e41d648bSToomas Soome char *mfg; 1713074bb90dSTom Pothier /* 1714074bb90dSTom Pothier * Load the System Management BIOS into the global ksmbios 1715074bb90dSTom Pothier * handle, if an SMBIOS is present on this system. 1716e41d648bSToomas Soome * Also set "si-hw-provider" property, if not already set. 1717074bb90dSTom Pothier */ 1718074bb90dSTom Pothier ksmbios = smbios_open(NULL, SMB_VERSION, ksmbios_flags, NULL); 1719e41d648bSToomas Soome if (ksmbios != NULL && 1720e41d648bSToomas Soome ((smid = smbios_info_system(ksmbios, &smsys)) != SMB_ERR) && 1721e41d648bSToomas Soome (smbios_info_common(ksmbios, smid, &sminfo)) != SMB_ERR) { 1722e41d648bSToomas Soome mfg = (char *)sminfo.smbi_manufacturer; 1723e41d648bSToomas Soome if (BOP_GETPROPLEN(bootops, "si-hw-provider") < 0) { 1724e41d648bSToomas Soome extern char hw_provider[]; 1725e41d648bSToomas Soome int i; 1726e41d648bSToomas Soome for (i = 0; i < SYS_NMLN; i++) { 1727e41d648bSToomas Soome if (isprint(mfg[i])) 1728e41d648bSToomas Soome hw_provider[i] = mfg[i]; 1729e41d648bSToomas Soome else { 1730e41d648bSToomas Soome hw_provider[i] = '\0'; 1731e41d648bSToomas Soome break; 1732e41d648bSToomas Soome } 1733e41d648bSToomas Soome } 1734e41d648bSToomas Soome hw_provider[SYS_NMLN - 1] = '\0'; 1735e41d648bSToomas Soome } 1736e41d648bSToomas Soome } 1737074bb90dSTom Pothier } 1738074bb90dSTom Pothier 1739074bb90dSTom Pothier 17409e3e4df2SGarrett D'Amore /* 17419e3e4df2SGarrett D'Amore * Originally clconf_init() apparently needed the hostid. But 17429e3e4df2SGarrett D'Amore * this no longer appears to be true - it uses its own nodeid. 17439e3e4df2SGarrett D'Amore * By placing the hostid logic here, we are able to make use of 17449e3e4df2SGarrett D'Amore * the SMBIOS UUID. 17459e3e4df2SGarrett D'Amore */ 17469e3e4df2SGarrett D'Amore if ((h = set_soft_hostid()) == HW_INVALID_HOSTID) { 17479e3e4df2SGarrett D'Amore cmn_err(CE_WARN, "Unable to set hostid"); 17489e3e4df2SGarrett D'Amore } else { 17499e3e4df2SGarrett D'Amore for (v = h, cnt = 0; cnt < 10; cnt++) { 17509e3e4df2SGarrett D'Amore d[cnt] = (char)(v % 10); 17519e3e4df2SGarrett D'Amore v /= 10; 17529e3e4df2SGarrett D'Amore if (v == 0) 17539e3e4df2SGarrett D'Amore break; 17549e3e4df2SGarrett D'Amore } 17559e3e4df2SGarrett D'Amore for (cp = hw_serial; cnt >= 0; cnt--) 17569e3e4df2SGarrett D'Amore *cp++ = d[cnt] + '0'; 17579e3e4df2SGarrett D'Amore *cp = 0; 17589e3e4df2SGarrett D'Amore } 17599e3e4df2SGarrett D'Amore 1760e4b86885SCheng Sean Ye /* 1761e4b86885SCheng Sean Ye * Set up the CPU module subsystem for the boot cpu in the native 1762e4b86885SCheng Sean Ye * case, and all physical cpu resource in the xpv dom0 case. 1763e4b86885SCheng Sean Ye * Modifies the device tree, so this must be done after 1764e4b86885SCheng Sean Ye * setup_ddi(). 1765e4b86885SCheng Sean Ye */ 1766e4b86885SCheng Sean Ye #ifdef __xpv 1767e4b86885SCheng Sean Ye /* 1768e4b86885SCheng Sean Ye * If paravirtualized and on dom0 then we initialize all physical 1769e4b86885SCheng Sean Ye * cpu handles now; if paravirtualized on a domU then do not 1770e4b86885SCheng Sean Ye * initialize. 1771e4b86885SCheng Sean Ye */ 1772e4b86885SCheng Sean Ye if (DOMAIN_IS_INITDOMAIN(xen_info)) { 1773e4b86885SCheng Sean Ye xen_mc_lcpu_cookie_t cpi; 1774e4b86885SCheng Sean Ye 1775e4b86885SCheng Sean Ye for (cpi = xen_physcpu_next(NULL); cpi != NULL; 1776e4b86885SCheng Sean Ye cpi = xen_physcpu_next(cpi)) { 1777e4b86885SCheng Sean Ye if ((hdl = cmi_init(CMI_HDL_SOLARIS_xVM_MCA, 1778e4b86885SCheng Sean Ye xen_physcpu_chipid(cpi), xen_physcpu_coreid(cpi), 1779e4b86885SCheng Sean Ye xen_physcpu_strandid(cpi))) != NULL && 17807417cfdeSKuriakose Kuruvilla is_x86_feature(x86_featureset, X86FSET_MCA)) 178120c794b3Sgavinm cmi_mca_init(hdl); 178220c794b3Sgavinm } 178320c794b3Sgavinm } 1784e4b86885SCheng Sean Ye #else 1785e4b86885SCheng Sean Ye /* 1786e4b86885SCheng Sean Ye * Initialize a handle for the boot cpu - others will initialize 1787d6694327SYuri Pankov * as they startup. 1788e4b86885SCheng Sean Ye */ 1789d6694327SYuri Pankov if ((hdl = cmi_init(CMI_HDL_NATIVE, cmi_ntv_hwchipid(CPU), 1790d6694327SYuri Pankov cmi_ntv_hwcoreid(CPU), cmi_ntv_hwstrandid(CPU))) != NULL) { 1791d6694327SYuri Pankov if (is_x86_feature(x86_featureset, X86FSET_MCA)) 1792e4b86885SCheng Sean Ye cmi_mca_init(hdl); 1793d6694327SYuri Pankov CPU->cpu_m.mcpu_cmi_hdl = hdl; 1794a3114836SGerry Liu } 179520c794b3Sgavinm #endif /* __xpv */ 17967aec1d6eScindi 17977c478bd9Sstevel@tonic-gate /* 17987c478bd9Sstevel@tonic-gate * Fake a prom tree such that /dev/openprom continues to work 17997c478bd9Sstevel@tonic-gate */ 1800ae115bc7Smrj PRM_POINT("startup_modules: calling prom_setup..."); 18017c478bd9Sstevel@tonic-gate prom_setup(); 1802ae115bc7Smrj PRM_POINT("startup_modules: done"); 18037c478bd9Sstevel@tonic-gate 18047c478bd9Sstevel@tonic-gate /* 18057c478bd9Sstevel@tonic-gate * Load all platform specific modules 18067c478bd9Sstevel@tonic-gate */ 1807ae115bc7Smrj PRM_POINT("startup_modules: calling psm_modload..."); 18087c478bd9Sstevel@tonic-gate psm_modload(); 18097c478bd9Sstevel@tonic-gate 18107c478bd9Sstevel@tonic-gate PRM_POINT("startup_modules() done"); 18117c478bd9Sstevel@tonic-gate } 18127c478bd9Sstevel@tonic-gate 1813ae115bc7Smrj /* 1814ae115bc7Smrj * claim a "setaside" boot page for use in the kernel 1815ae115bc7Smrj */ 1816ae115bc7Smrj page_t * 1817ae115bc7Smrj boot_claim_page(pfn_t pfn) 18187c478bd9Sstevel@tonic-gate { 1819ae115bc7Smrj page_t *pp; 18207c478bd9Sstevel@tonic-gate 1821ae115bc7Smrj pp = page_numtopp_nolock(pfn); 1822ae115bc7Smrj ASSERT(pp != NULL); 18237c478bd9Sstevel@tonic-gate 1824ae115bc7Smrj if (PP_ISBOOTPAGES(pp)) { 1825ae115bc7Smrj if (pp->p_next != NULL) 1826ae115bc7Smrj pp->p_next->p_prev = pp->p_prev; 1827ae115bc7Smrj if (pp->p_prev == NULL) 1828ae115bc7Smrj bootpages = pp->p_next; 1829ae115bc7Smrj else 1830ae115bc7Smrj pp->p_prev->p_next = pp->p_next; 1831ae115bc7Smrj } else { 1832ae115bc7Smrj /* 1833ae115bc7Smrj * htable_attach() expects a base pagesize page 1834ae115bc7Smrj */ 1835ae115bc7Smrj if (pp->p_szc != 0) 1836ae115bc7Smrj page_boot_demote(pp); 1837ae115bc7Smrj pp = page_numtopp(pfn, SE_EXCL); 1838ae115bc7Smrj } 1839ae115bc7Smrj return (pp); 18407c478bd9Sstevel@tonic-gate } 18417c478bd9Sstevel@tonic-gate 18427c478bd9Sstevel@tonic-gate /* 18437c478bd9Sstevel@tonic-gate * Walk through the pagetables looking for pages mapped in by boot. If the 18447c478bd9Sstevel@tonic-gate * setaside flag is set the pages are expected to be returned to the 18457c478bd9Sstevel@tonic-gate * kernel later in boot, so we add them to the bootpages list. 18467c478bd9Sstevel@tonic-gate */ 18477c478bd9Sstevel@tonic-gate static void 18487c478bd9Sstevel@tonic-gate protect_boot_range(uintptr_t low, uintptr_t high, int setaside) 18497c478bd9Sstevel@tonic-gate { 18507c478bd9Sstevel@tonic-gate uintptr_t va = low; 18517c478bd9Sstevel@tonic-gate size_t len; 18527c478bd9Sstevel@tonic-gate uint_t prot; 18537c478bd9Sstevel@tonic-gate pfn_t pfn; 18547c478bd9Sstevel@tonic-gate page_t *pp; 18557c478bd9Sstevel@tonic-gate pgcnt_t boot_protect_cnt = 0; 18567c478bd9Sstevel@tonic-gate 1857ae115bc7Smrj while (kbm_probe(&va, &len, &pfn, &prot) != 0 && va < high) { 18587c478bd9Sstevel@tonic-gate if (va + len >= high) 18597c478bd9Sstevel@tonic-gate panic("0x%lx byte mapping at 0x%p exceeds boot's " 18607c478bd9Sstevel@tonic-gate "legal range.", len, (void *)va); 18617c478bd9Sstevel@tonic-gate 18627c478bd9Sstevel@tonic-gate while (len > 0) { 18637c478bd9Sstevel@tonic-gate pp = page_numtopp_alloc(pfn); 18647c478bd9Sstevel@tonic-gate if (pp != NULL) { 18657c478bd9Sstevel@tonic-gate if (setaside == 0) 18667c478bd9Sstevel@tonic-gate panic("Unexpected mapping by boot. " 18677c478bd9Sstevel@tonic-gate "addr=%p pfn=%lx\n", 18687c478bd9Sstevel@tonic-gate (void *)va, pfn); 18697c478bd9Sstevel@tonic-gate 18707c478bd9Sstevel@tonic-gate pp->p_next = bootpages; 1871ae115bc7Smrj pp->p_prev = NULL; 1872ae115bc7Smrj PP_SETBOOTPAGES(pp); 1873ae115bc7Smrj if (bootpages != NULL) { 1874ae115bc7Smrj bootpages->p_prev = pp; 1875ae115bc7Smrj } 18767c478bd9Sstevel@tonic-gate bootpages = pp; 18777c478bd9Sstevel@tonic-gate ++boot_protect_cnt; 18787c478bd9Sstevel@tonic-gate } 18797c478bd9Sstevel@tonic-gate 18807c478bd9Sstevel@tonic-gate ++pfn; 18817c478bd9Sstevel@tonic-gate len -= MMU_PAGESIZE; 18827c478bd9Sstevel@tonic-gate va += MMU_PAGESIZE; 18837c478bd9Sstevel@tonic-gate } 18847c478bd9Sstevel@tonic-gate } 18857c478bd9Sstevel@tonic-gate PRM_DEBUG(boot_protect_cnt); 18867c478bd9Sstevel@tonic-gate } 18877c478bd9Sstevel@tonic-gate 1888ae115bc7Smrj /* 1889*04909c8cSJohn Levon * Establish the final size of the kernel's heap, size of segmap, segkp, etc. 1890ae115bc7Smrj */ 18917c478bd9Sstevel@tonic-gate static void 189235b1ab99Sjosephb layout_kernel_va(void) 18937c478bd9Sstevel@tonic-gate { 1894*04909c8cSJohn Levon const size_t physmem_size = mmu_ptob(physmem); 1895*04909c8cSJohn Levon size_t size; 18967c478bd9Sstevel@tonic-gate 1897*04909c8cSJohn Levon PRM_POINT("layout_kernel_va() starting..."); 18987c478bd9Sstevel@tonic-gate 189935b1ab99Sjosephb kpm_vbase = (caddr_t)segkpm_base; 1900*04909c8cSJohn Levon kpm_size = ROUND_UP_LPAGE(mmu_ptob(physmax + 1)); 190135b1ab99Sjosephb if ((uintptr_t)kpm_vbase + kpm_size > (uintptr_t)valloc_base) 190235b1ab99Sjosephb panic("not enough room for kpm!"); 1903ae115bc7Smrj PRM_DEBUG(kpm_size); 1904ae115bc7Smrj PRM_DEBUG(kpm_vbase); 19057c478bd9Sstevel@tonic-gate 1906ae115bc7Smrj segkp_base = (caddr_t)valloc_base + valloc_sz; 19077c478bd9Sstevel@tonic-gate if (!segkp_fromheap) { 1908*04909c8cSJohn Levon size = mmu_ptob(segkpsize); 19097c478bd9Sstevel@tonic-gate 19107c478bd9Sstevel@tonic-gate /* 1911ae115bc7Smrj * determine size of segkp 19127c478bd9Sstevel@tonic-gate */ 1913*04909c8cSJohn Levon if (size < SEGKPMINSIZE || size > SEGKPMAXSIZE) { 1914*04909c8cSJohn Levon size = SEGKPDEFSIZE; 19157c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "!Illegal value for segkpsize. " 19167c478bd9Sstevel@tonic-gate "segkpsize has been reset to %ld pages", 1917*04909c8cSJohn Levon mmu_btop(size)); 19187c478bd9Sstevel@tonic-gate } 1919*04909c8cSJohn Levon size = MIN(size, MAX(SEGKPMINSIZE, physmem_size)); 19207c478bd9Sstevel@tonic-gate 1921*04909c8cSJohn Levon segkpsize = mmu_btop(ROUND_UP_LPAGE(size)); 19227c478bd9Sstevel@tonic-gate } 1923ae115bc7Smrj PRM_DEBUG(segkp_base); 1924ae115bc7Smrj PRM_DEBUG(segkpsize); 19257c478bd9Sstevel@tonic-gate 192635b1ab99Sjosephb /* 1927*04909c8cSJohn Levon * segkvmm: backing for vmm guest memory. Like segzio, we have a 1928*04909c8cSJohn Levon * separate segment for two reasons: it makes it easy to skip our pages 1929*04909c8cSJohn Levon * on kernel crash dumps, and it helps avoid fragmentation. With this 1930*04909c8cSJohn Levon * segment, we're expecting significantly-sized allocations only; we'll 1931*04909c8cSJohn Levon * default to 4x the size of physmem. 193235b1ab99Sjosephb */ 1933*04909c8cSJohn Levon segkvmm_base = segkp_base + mmu_ptob(segkpsize); 1934*04909c8cSJohn Levon size = segkvmmsize != 0 ? mmu_ptob(segkvmmsize) : (physmem_size * 4); 1935*04909c8cSJohn Levon 1936*04909c8cSJohn Levon size = MAX(size, SEGVMMMINSIZE); 1937*04909c8cSJohn Levon segkvmmsize = mmu_btop(ROUND_UP_LPAGE(size)); 1938*04909c8cSJohn Levon 1939*04909c8cSJohn Levon PRM_DEBUG(segkvmmsize); 1940*04909c8cSJohn Levon PRM_DEBUG(segkvmm_base); 1941*04909c8cSJohn Levon 1942*04909c8cSJohn Levon /* 1943*04909c8cSJohn Levon * segzio is used for ZFS cached data. For segzio, we use 1.5x physmem. 1944*04909c8cSJohn Levon */ 1945*04909c8cSJohn Levon segzio_base = segkvmm_base + mmu_ptob(segkvmmsize); 1946ae115bc7Smrj if (segzio_fromheap) { 1947ae115bc7Smrj segziosize = 0; 1948ae115bc7Smrj } else { 1949*04909c8cSJohn Levon size = (segziosize != 0) ? mmu_ptob(segziosize) : 1950*04909c8cSJohn Levon (physmem_size * 3) / 2; 1951ad23a2dbSjohansen 1952*04909c8cSJohn Levon size = MAX(size, SEGZIOMINSIZE); 1953ad23a2dbSjohansen segziosize = mmu_btop(ROUND_UP_LPAGE(size)); 1954ad23a2dbSjohansen } 1955ae115bc7Smrj PRM_DEBUG(segziosize); 1956ae115bc7Smrj PRM_DEBUG(segzio_base); 1957ad23a2dbSjohansen 19587c478bd9Sstevel@tonic-gate /* 1959ae115bc7Smrj * Put the range of VA for device mappings next, kmdb knows to not 1960ae115bc7Smrj * grep in this range of addresses. 19617c478bd9Sstevel@tonic-gate */ 1962ae115bc7Smrj toxic_addr = 1963ae115bc7Smrj ROUND_UP_LPAGE((uintptr_t)segzio_base + mmu_ptob(segziosize)); 19647c478bd9Sstevel@tonic-gate PRM_DEBUG(toxic_addr); 1965ae115bc7Smrj segmap_start = ROUND_UP_LPAGE(toxic_addr + toxic_size); 19667c478bd9Sstevel@tonic-gate 19677c478bd9Sstevel@tonic-gate /* 196852eef812SAmrita Sadhukhan * Users can change segmapsize through eeprom. If the variable 196952eef812SAmrita Sadhukhan * is tuned through eeprom, there is no upper bound on the 1970ed5289f9SKen Erickson * size of segmap. 19717c478bd9Sstevel@tonic-gate */ 19727c478bd9Sstevel@tonic-gate segmapsize = MAX(ROUND_UP_LPAGE(segmapsize), SEGMAPDEFAULT); 19737c478bd9Sstevel@tonic-gate 1974ae115bc7Smrj PRM_DEBUG(segmap_start); 19757c478bd9Sstevel@tonic-gate PRM_DEBUG(segmapsize); 197635b1ab99Sjosephb kernelheap = (caddr_t)ROUND_UP_LPAGE(segmap_start + segmapsize); 197735b1ab99Sjosephb PRM_DEBUG(kernelheap); 197835b1ab99Sjosephb PRM_POINT("layout_kernel_va() done..."); 197935b1ab99Sjosephb } 198035b1ab99Sjosephb 198135b1ab99Sjosephb /* 198235b1ab99Sjosephb * Finish initializing the VM system, now that we are no longer 198335b1ab99Sjosephb * relying on the boot time memory allocators. 198435b1ab99Sjosephb */ 198535b1ab99Sjosephb static void 198635b1ab99Sjosephb startup_vm(void) 198735b1ab99Sjosephb { 198835b1ab99Sjosephb struct segmap_crargs a; 198935b1ab99Sjosephb 199035b1ab99Sjosephb extern int use_brk_lpg, use_stk_lpg; 199135b1ab99Sjosephb 199235b1ab99Sjosephb PRM_POINT("startup_vm() starting..."); 199335b1ab99Sjosephb 199435b1ab99Sjosephb /* 199535b1ab99Sjosephb * Initialize the hat layer. 199635b1ab99Sjosephb */ 199735b1ab99Sjosephb hat_init(); 19987c478bd9Sstevel@tonic-gate 1999ae115bc7Smrj /* 2000ae115bc7Smrj * Do final allocations of HAT data structures that need to 2001ae115bc7Smrj * be allocated before quiescing the boot loader. 2002ae115bc7Smrj */ 2003ae115bc7Smrj PRM_POINT("Calling hat_kern_alloc()..."); 2004ae115bc7Smrj hat_kern_alloc((caddr_t)segmap_start, segmapsize, ekernelheap); 2005ae115bc7Smrj PRM_POINT("hat_kern_alloc() done"); 2006ae115bc7Smrj 2007843e1988Sjohnlev #ifndef __xpv 2008ae115bc7Smrj /* 20091d03c31eSjohnlev * Setup Page Attribute Table 2010ae115bc7Smrj */ 20111d03c31eSjohnlev pat_sync(); 2012843e1988Sjohnlev #endif 2013ae115bc7Smrj 2014ae115bc7Smrj /* 2015ae115bc7Smrj * The next two loops are done in distinct steps in order 2016ae115bc7Smrj * to be sure that any page that is doubly mapped (both above 2017ae115bc7Smrj * KERNEL_TEXT and below kernelbase) is dealt with correctly. 2018ae115bc7Smrj * Note this may never happen, but it might someday. 2019ae115bc7Smrj */ 2020ae115bc7Smrj bootpages = NULL; 2021ae115bc7Smrj PRM_POINT("Protecting boot pages"); 2022ae115bc7Smrj 2023ae115bc7Smrj /* 2024ae115bc7Smrj * Protect any pages mapped above KERNEL_TEXT that somehow have 2025ae115bc7Smrj * page_t's. This can only happen if something weird allocated 2026ae115bc7Smrj * in this range (like kadb/kmdb). 2027ae115bc7Smrj */ 2028ae115bc7Smrj protect_boot_range(KERNEL_TEXT, (uintptr_t)-1, 0); 2029ae115bc7Smrj 2030ae115bc7Smrj /* 2031ae115bc7Smrj * Before we can take over memory allocation/mapping from the boot 2032ae115bc7Smrj * loader we must remove from our free page lists any boot allocated 2033ae115bc7Smrj * pages that stay mapped until release_bootstrap(). 2034ae115bc7Smrj */ 2035ae115bc7Smrj protect_boot_range(0, kernelbase, 1); 2036ae115bc7Smrj 2037843e1988Sjohnlev 2038ae115bc7Smrj /* 2039ae115bc7Smrj * Switch to running on regular HAT (not boot_mmu) 2040ae115bc7Smrj */ 2041ae115bc7Smrj PRM_POINT("Calling hat_kern_setup()..."); 2042ae115bc7Smrj hat_kern_setup(); 2043ae115bc7Smrj 2044ae115bc7Smrj /* 2045ae115bc7Smrj * It is no longer safe to call BOP_ALLOC(), so make sure we don't. 2046ae115bc7Smrj */ 2047ae115bc7Smrj bop_no_more_mem(); 2048ae115bc7Smrj 2049ae115bc7Smrj PRM_POINT("hat_kern_setup() done"); 2050ae115bc7Smrj 2051ae115bc7Smrj hat_cpu_online(CPU); 2052ae115bc7Smrj 20537c478bd9Sstevel@tonic-gate /* 20547c478bd9Sstevel@tonic-gate * Initialize VM system 20557c478bd9Sstevel@tonic-gate */ 20567c478bd9Sstevel@tonic-gate PRM_POINT("Calling kvm_init()..."); 20577c478bd9Sstevel@tonic-gate kvm_init(); 20587c478bd9Sstevel@tonic-gate PRM_POINT("kvm_init() done"); 20597c478bd9Sstevel@tonic-gate 20607c478bd9Sstevel@tonic-gate /* 20617c478bd9Sstevel@tonic-gate * Tell kmdb that the VM system is now working 20627c478bd9Sstevel@tonic-gate */ 20637c478bd9Sstevel@tonic-gate if (boothowto & RB_DEBUG) 20647c478bd9Sstevel@tonic-gate kdi_dvec_vmready(); 20657c478bd9Sstevel@tonic-gate 2066843e1988Sjohnlev #if defined(__xpv) 2067843e1988Sjohnlev /* 2068843e1988Sjohnlev * Populate the I/O pool on domain 0 2069843e1988Sjohnlev */ 2070843e1988Sjohnlev if (DOMAIN_IS_INITDOMAIN(xen_info)) { 2071843e1988Sjohnlev extern long populate_io_pool(void); 2072843e1988Sjohnlev long init_io_pool_cnt; 2073843e1988Sjohnlev 2074843e1988Sjohnlev PRM_POINT("Populating reserve I/O page pool"); 2075843e1988Sjohnlev init_io_pool_cnt = populate_io_pool(); 2076843e1988Sjohnlev PRM_DEBUG(init_io_pool_cnt); 2077843e1988Sjohnlev } 2078843e1988Sjohnlev #endif 20797c478bd9Sstevel@tonic-gate /* 20807c478bd9Sstevel@tonic-gate * Mangle the brand string etc. 20817c478bd9Sstevel@tonic-gate */ 20827c478bd9Sstevel@tonic-gate cpuid_pass3(CPU); 20837c478bd9Sstevel@tonic-gate 20847c478bd9Sstevel@tonic-gate #if defined(__amd64) 20857c478bd9Sstevel@tonic-gate 20867c478bd9Sstevel@tonic-gate /* 20877c478bd9Sstevel@tonic-gate * Create the device arena for toxic (to dtrace/kmdb) mappings. 20887c478bd9Sstevel@tonic-gate */ 20897c478bd9Sstevel@tonic-gate device_arena = vmem_create("device", (void *)toxic_addr, 20907c478bd9Sstevel@tonic-gate toxic_size, MMU_PAGESIZE, NULL, NULL, NULL, 0, VM_SLEEP); 20917c478bd9Sstevel@tonic-gate 20927c478bd9Sstevel@tonic-gate #else /* __i386 */ 20937c478bd9Sstevel@tonic-gate 20947c478bd9Sstevel@tonic-gate /* 20957c478bd9Sstevel@tonic-gate * allocate the bit map that tracks toxic pages 20967c478bd9Sstevel@tonic-gate */ 2097ae115bc7Smrj toxic_bit_map_len = btop((ulong_t)(valloc_base - kernelbase)); 20987c478bd9Sstevel@tonic-gate PRM_DEBUG(toxic_bit_map_len); 20997c478bd9Sstevel@tonic-gate toxic_bit_map = 21007c478bd9Sstevel@tonic-gate kmem_zalloc(BT_SIZEOFMAP(toxic_bit_map_len), KM_NOSLEEP); 21017c478bd9Sstevel@tonic-gate ASSERT(toxic_bit_map != NULL); 21027c478bd9Sstevel@tonic-gate PRM_DEBUG(toxic_bit_map); 21037c478bd9Sstevel@tonic-gate 21047c478bd9Sstevel@tonic-gate #endif /* __i386 */ 21057c478bd9Sstevel@tonic-gate 21067c478bd9Sstevel@tonic-gate 21077c478bd9Sstevel@tonic-gate /* 21087c478bd9Sstevel@tonic-gate * Now that we've got more VA, as well as the ability to allocate from 21097c478bd9Sstevel@tonic-gate * it, tell the debugger. 21107c478bd9Sstevel@tonic-gate */ 21117c478bd9Sstevel@tonic-gate if (boothowto & RB_DEBUG) 21127c478bd9Sstevel@tonic-gate kdi_dvec_memavail(); 21137c478bd9Sstevel@tonic-gate 2114843e1988Sjohnlev #if !defined(__xpv) 21157c478bd9Sstevel@tonic-gate /* 21167c478bd9Sstevel@tonic-gate * Map page pfn=0 for drivers, such as kd, that need to pick up 21177c478bd9Sstevel@tonic-gate * parameters left there by controllers/BIOS. 21187c478bd9Sstevel@tonic-gate */ 21197c478bd9Sstevel@tonic-gate PRM_POINT("setup up p0_va"); 21207c478bd9Sstevel@tonic-gate p0_va = i86devmap(0, 1, PROT_READ); 21217c478bd9Sstevel@tonic-gate PRM_DEBUG(p0_va); 2122843e1988Sjohnlev #endif 21237c478bd9Sstevel@tonic-gate 21247c478bd9Sstevel@tonic-gate cmn_err(CE_CONT, "?mem = %luK (0x%lx)\n", 21257c478bd9Sstevel@tonic-gate physinstalled << (MMU_PAGESHIFT - 10), ptob(physinstalled)); 21267c478bd9Sstevel@tonic-gate 212783f9b804Skchow /* 212883f9b804Skchow * disable automatic large pages for small memory systems or 212983f9b804Skchow * when the disable flag is set. 213002bc52beSkchow * 213102bc52beSkchow * Do not yet consider page sizes larger than 2m/4m. 213283f9b804Skchow */ 2133ec25b48fSsusans if (!auto_lpg_disable && mmu.max_page_level > 0) { 2134ec25b48fSsusans max_uheap_lpsize = LEVEL_SIZE(1); 2135ec25b48fSsusans max_ustack_lpsize = LEVEL_SIZE(1); 2136ec25b48fSsusans max_privmap_lpsize = LEVEL_SIZE(1); 2137ec25b48fSsusans max_uidata_lpsize = LEVEL_SIZE(1); 2138ec25b48fSsusans max_utext_lpsize = LEVEL_SIZE(1); 2139ec25b48fSsusans max_shm_lpsize = LEVEL_SIZE(1); 2140ec25b48fSsusans } 2141ec25b48fSsusans if (physmem < privm_lpg_min_physmem || mmu.max_page_level == 0 || 2142ec25b48fSsusans auto_lpg_disable) { 2143beb1bda0Sdavemq use_brk_lpg = 0; 2144beb1bda0Sdavemq use_stk_lpg = 0; 2145ec25b48fSsusans } 214602bc52beSkchow mcntl0_lpsize = LEVEL_SIZE(mmu.umax_page_level); 2147beb1bda0Sdavemq 21487c478bd9Sstevel@tonic-gate PRM_POINT("Calling hat_init_finish()..."); 21497c478bd9Sstevel@tonic-gate hat_init_finish(); 21507c478bd9Sstevel@tonic-gate PRM_POINT("hat_init_finish() done"); 21517c478bd9Sstevel@tonic-gate 21527c478bd9Sstevel@tonic-gate /* 21537c478bd9Sstevel@tonic-gate * Initialize the segkp segment type. 21547c478bd9Sstevel@tonic-gate */ 21557c478bd9Sstevel@tonic-gate rw_enter(&kas.a_lock, RW_WRITER); 2156ae115bc7Smrj PRM_POINT("Attaching segkp"); 2157ae115bc7Smrj if (segkp_fromheap) { 21587c478bd9Sstevel@tonic-gate segkp->s_as = &kas; 2159ae115bc7Smrj } else if (seg_attach(&kas, (caddr_t)segkp_base, mmu_ptob(segkpsize), 2160ae115bc7Smrj segkp) < 0) { 2161ae115bc7Smrj panic("startup: cannot attach segkp"); 2162ae115bc7Smrj /*NOTREACHED*/ 21637c478bd9Sstevel@tonic-gate } 2164ae115bc7Smrj PRM_POINT("Doing segkp_create()"); 21657c478bd9Sstevel@tonic-gate if (segkp_create(segkp) != 0) { 21667c478bd9Sstevel@tonic-gate panic("startup: segkp_create failed"); 21677c478bd9Sstevel@tonic-gate /*NOTREACHED*/ 21687c478bd9Sstevel@tonic-gate } 21697c478bd9Sstevel@tonic-gate PRM_DEBUG(segkp); 21707c478bd9Sstevel@tonic-gate rw_exit(&kas.a_lock); 21717c478bd9Sstevel@tonic-gate 21727c478bd9Sstevel@tonic-gate /* 21737c478bd9Sstevel@tonic-gate * kpm segment 21747c478bd9Sstevel@tonic-gate */ 21757c478bd9Sstevel@tonic-gate segmap_kpm = 0; 217674ecdb51SJohn Levon if (kpm_desired) 21777c478bd9Sstevel@tonic-gate kpm_init(); 21787c478bd9Sstevel@tonic-gate 21797c478bd9Sstevel@tonic-gate /* 21807c478bd9Sstevel@tonic-gate * Now create segmap segment. 21817c478bd9Sstevel@tonic-gate */ 21827c478bd9Sstevel@tonic-gate rw_enter(&kas.a_lock, RW_WRITER); 2183ae115bc7Smrj if (seg_attach(&kas, (caddr_t)segmap_start, segmapsize, segmap) < 0) { 2184ae115bc7Smrj panic("cannot attach segmap"); 21857c478bd9Sstevel@tonic-gate /*NOTREACHED*/ 21867c478bd9Sstevel@tonic-gate } 2187ae115bc7Smrj PRM_DEBUG(segmap); 21887c478bd9Sstevel@tonic-gate 21897c478bd9Sstevel@tonic-gate a.prot = PROT_READ | PROT_WRITE; 21907c478bd9Sstevel@tonic-gate a.shmsize = 0; 21917c478bd9Sstevel@tonic-gate a.nfreelist = segmapfreelists; 21927c478bd9Sstevel@tonic-gate 2193ae115bc7Smrj if (segmap_create(segmap, (caddr_t)&a) != 0) 2194ae115bc7Smrj panic("segmap_create segmap"); 21957c478bd9Sstevel@tonic-gate rw_exit(&kas.a_lock); 21967c478bd9Sstevel@tonic-gate 21977c478bd9Sstevel@tonic-gate setup_vaddr_for_ppcopy(CPU); 21987c478bd9Sstevel@tonic-gate 21997c478bd9Sstevel@tonic-gate segdev_init(); 2200843e1988Sjohnlev #if defined(__xpv) 2201843e1988Sjohnlev if (DOMAIN_IS_INITDOMAIN(xen_info)) 2202843e1988Sjohnlev #endif 2203843e1988Sjohnlev pmem_init(); 2204ae115bc7Smrj 22057c478bd9Sstevel@tonic-gate PRM_POINT("startup_vm() done"); 22067c478bd9Sstevel@tonic-gate } 22077c478bd9Sstevel@tonic-gate 2208ae115bc7Smrj /* 2209ae115bc7Smrj * Load a tod module for the non-standard tod part found on this system. 2210ae115bc7Smrj */ 2211ae115bc7Smrj static void 2212ae115bc7Smrj load_tod_module(char *todmod) 2213ae115bc7Smrj { 2214ae115bc7Smrj if (modload("tod", todmod) == -1) 2215ae115bc7Smrj halt("Can't load TOD module"); 2216ae115bc7Smrj } 2217ae115bc7Smrj 22187c478bd9Sstevel@tonic-gate static void 22197c478bd9Sstevel@tonic-gate startup_end(void) 22207c478bd9Sstevel@tonic-gate { 2221dd4eeefdSeota int i; 22227c478bd9Sstevel@tonic-gate extern void setx86isalist(void); 2223fb2caebeSRandy Fishel extern void cpu_event_init(void); 22247c478bd9Sstevel@tonic-gate 22257c478bd9Sstevel@tonic-gate PRM_POINT("startup_end() starting..."); 22267c478bd9Sstevel@tonic-gate 22277c478bd9Sstevel@tonic-gate /* 22287c478bd9Sstevel@tonic-gate * Perform tasks that get done after most of the VM 22297c478bd9Sstevel@tonic-gate * initialization has been done but before the clock 22307c478bd9Sstevel@tonic-gate * and other devices get started. 22317c478bd9Sstevel@tonic-gate */ 22327c478bd9Sstevel@tonic-gate kern_setup1(); 22337c478bd9Sstevel@tonic-gate 22347c478bd9Sstevel@tonic-gate /* 22357c478bd9Sstevel@tonic-gate * Perform CPC initialization for this CPU. 22367c478bd9Sstevel@tonic-gate */ 22377c478bd9Sstevel@tonic-gate kcpc_hw_init(CPU); 22387c478bd9Sstevel@tonic-gate 2239fb2caebeSRandy Fishel /* 2240fb2caebeSRandy Fishel * Initialize cpu event framework. 2241fb2caebeSRandy Fishel */ 2242fb2caebeSRandy Fishel cpu_event_init(); 2243fb2caebeSRandy Fishel 2244ee88d2b9Skchow #if defined(OPTERON_WORKAROUND_6323525) 2245ee88d2b9Skchow if (opteron_workaround_6323525) 2246ee88d2b9Skchow patch_workaround_6323525(); 2247ee88d2b9Skchow #endif 2248ae115bc7Smrj /* 2249ae115bc7Smrj * If needed, load TOD module now so that ddi_get_time(9F) etc. work 2250ae115bc7Smrj * (For now, "needed" is defined as set tod_module_name in /etc/system) 2251ae115bc7Smrj */ 2252ae115bc7Smrj if (tod_module_name != NULL) { 2253ae115bc7Smrj PRM_POINT("load_tod_module()"); 2254ae115bc7Smrj load_tod_module(tod_module_name); 2255ae115bc7Smrj } 2256ae115bc7Smrj 2257843e1988Sjohnlev #if defined(__xpv) 2258843e1988Sjohnlev /* 2259843e1988Sjohnlev * Forceload interposing TOD module for the hypervisor. 2260843e1988Sjohnlev */ 2261843e1988Sjohnlev PRM_POINT("load_tod_module()"); 2262843e1988Sjohnlev load_tod_module("xpvtod"); 2263843e1988Sjohnlev #endif 2264843e1988Sjohnlev 22657c478bd9Sstevel@tonic-gate /* 22667c478bd9Sstevel@tonic-gate * Configure the system. 22677c478bd9Sstevel@tonic-gate */ 22687c478bd9Sstevel@tonic-gate PRM_POINT("Calling configure()..."); 22697c478bd9Sstevel@tonic-gate configure(); /* set up devices */ 22707c478bd9Sstevel@tonic-gate PRM_POINT("configure() done"); 22717c478bd9Sstevel@tonic-gate 22727af88ac7SKuriakose Kuruvilla /* 22737af88ac7SKuriakose Kuruvilla * We can now setup for XSAVE because fpu_probe is done in configure(). 22747af88ac7SKuriakose Kuruvilla */ 22757af88ac7SKuriakose Kuruvilla if (fp_save_mech == FP_XSAVE) { 22767af88ac7SKuriakose Kuruvilla xsave_setup_msr(CPU); 22777af88ac7SKuriakose Kuruvilla } 22787af88ac7SKuriakose Kuruvilla 22797c478bd9Sstevel@tonic-gate /* 22807c478bd9Sstevel@tonic-gate * Set the isa_list string to the defined instruction sets we 22817c478bd9Sstevel@tonic-gate * support. 22827c478bd9Sstevel@tonic-gate */ 22837c478bd9Sstevel@tonic-gate setx86isalist(); 2284100b72f4Sandrei cpu_intr_alloc(CPU, NINTR_THREADS); 22857c478bd9Sstevel@tonic-gate psm_install(); 22867c478bd9Sstevel@tonic-gate 22877c478bd9Sstevel@tonic-gate /* 22887c478bd9Sstevel@tonic-gate * We're done with bootops. We don't unmap the bootstrap yet because 22897c478bd9Sstevel@tonic-gate * we're still using bootsvcs. 22907c478bd9Sstevel@tonic-gate */ 2291ae115bc7Smrj PRM_POINT("NULLing out bootops"); 2292ae115bc7Smrj *bootopsp = (struct bootops *)NULL; 22937c478bd9Sstevel@tonic-gate bootops = (struct bootops *)NULL; 22947c478bd9Sstevel@tonic-gate 2295843e1988Sjohnlev #if defined(__xpv) 2296843e1988Sjohnlev ec_init_debug_irq(); 2297843e1988Sjohnlev xs_domu_init(); 2298843e1988Sjohnlev #endif 22991de082f7SVikram Hegde 230074ecdb51SJohn Levon #if !defined(__xpv) 23013a634bfcSVikram Hegde /* 23023a634bfcSVikram Hegde * Intel IOMMU has been setup/initialized in ddi_impl.c 23033a634bfcSVikram Hegde * Start it up now. 23043a634bfcSVikram Hegde */ 23053a634bfcSVikram Hegde immu_startup(); 230674ecdb51SJohn Levon 230774ecdb51SJohn Levon /* 230874ecdb51SJohn Levon * Now that we're no longer going to drop into real mode for a BIOS call 230974ecdb51SJohn Levon * via bootops, we can enable PCID (which requires CR0.PG). 231074ecdb51SJohn Levon */ 231174ecdb51SJohn Levon enable_pcid(); 23121de082f7SVikram Hegde #endif 23133a634bfcSVikram Hegde 23147c478bd9Sstevel@tonic-gate PRM_POINT("Enabling interrupts"); 23157c478bd9Sstevel@tonic-gate (*picinitf)(); 23167c478bd9Sstevel@tonic-gate sti(); 2317843e1988Sjohnlev #if defined(__xpv) 2318843e1988Sjohnlev ASSERT(CPU->cpu_m.mcpu_vcpu_info->evtchn_upcall_mask == 0); 2319843e1988Sjohnlev xen_late_startup(); 2320843e1988Sjohnlev #endif 23217c478bd9Sstevel@tonic-gate 23227c478bd9Sstevel@tonic-gate (void) add_avsoftintr((void *)&softlevel1_hdl, 1, softlevel1, 232335b1ab99Sjosephb "softlevel1", NULL, NULL); /* XXX to be moved later */ 23247c478bd9Sstevel@tonic-gate 2325dd4eeefdSeota /* 2326a288e5a9SJoshua M. Clulow * Register software interrupt handlers for ddi_periodic_add(9F). 2327dd4eeefdSeota * Software interrupts up to the level 10 are supported. 2328dd4eeefdSeota */ 2329dd4eeefdSeota for (i = DDI_IPL_1; i <= DDI_IPL_10; i++) { 2330dd4eeefdSeota (void) add_avsoftintr((void *)&softlevel_hdl[i-1], i, 2331027bcc9fSToomas Soome (avfunc)(uintptr_t)ddi_periodic_softintr, "ddi_periodic", 2332a288e5a9SJoshua M. Clulow (caddr_t)(uintptr_t)i, NULL); 2333dd4eeefdSeota } 2334dd4eeefdSeota 23357125fcbdSVikram Hegde #if !defined(__xpv) 23367125fcbdSVikram Hegde if (modload("drv", "amd_iommu") < 0) { 23377125fcbdSVikram Hegde PRM_POINT("No AMD IOMMU present\n"); 233894f1124eSVikram Hegde } else if (ddi_hold_installed_driver(ddi_name_to_major( 233994f1124eSVikram Hegde "amd_iommu")) == NULL) { 23407125fcbdSVikram Hegde prom_printf("ERROR: failed to attach AMD IOMMU\n"); 23417125fcbdSVikram Hegde } 23427125fcbdSVikram Hegde #endif 2343e774b42bSBill Holler post_startup_cpu_fixups(); 23447125fcbdSVikram Hegde 23457c478bd9Sstevel@tonic-gate PRM_POINT("startup_end() done"); 23467c478bd9Sstevel@tonic-gate } 23477c478bd9Sstevel@tonic-gate 234896d008c2SKen Erickson /* 234996d008c2SKen Erickson * Don't remove the following 2 variables. They are necessary 235096d008c2SKen Erickson * for reading the hostid from the legacy file (/kernel/misc/sysinit). 235196d008c2SKen Erickson */ 23527c478bd9Sstevel@tonic-gate char *_hs1107 = hw_serial; 23537c478bd9Sstevel@tonic-gate ulong_t _bdhs34; 23547c478bd9Sstevel@tonic-gate 23557c478bd9Sstevel@tonic-gate void 23567c478bd9Sstevel@tonic-gate post_startup(void) 23577c478bd9Sstevel@tonic-gate { 23580e751525SEric Saxe extern void cpupm_init(cpu_t *); 2359fb2caebeSRandy Fishel extern void cpu_event_init_cpu(cpu_t *); 23600e751525SEric Saxe 23617c478bd9Sstevel@tonic-gate /* 23627c478bd9Sstevel@tonic-gate * Set the system wide, processor-specific flags to be passed 23637c478bd9Sstevel@tonic-gate * to userland via the aux vector for performance hints and 23647c478bd9Sstevel@tonic-gate * instruction set extensions. 23657c478bd9Sstevel@tonic-gate */ 23667c478bd9Sstevel@tonic-gate bind_hwcap(); 23677c478bd9Sstevel@tonic-gate 2368843e1988Sjohnlev #ifdef __xpv 2369843e1988Sjohnlev if (DOMAIN_IS_INITDOMAIN(xen_info)) 2370843e1988Sjohnlev #endif 2371843e1988Sjohnlev { 2372843e1988Sjohnlev #if defined(__xpv) 2373843e1988Sjohnlev xpv_panic_init(); 2374843e1988Sjohnlev #else 2375843e1988Sjohnlev /* 2376843e1988Sjohnlev * Startup the memory scrubber. 2377843e1988Sjohnlev * XXPV This should be running somewhere .. 2378843e1988Sjohnlev */ 237979ec9da8SYuri Pankov if ((get_hwenv() & HW_VIRTUAL) == 0) 2380e4b86885SCheng Sean Ye memscrub_init(); 2381843e1988Sjohnlev #endif 2382843e1988Sjohnlev } 23837c478bd9Sstevel@tonic-gate 23847aec1d6eScindi /* 23857aec1d6eScindi * Complete CPU module initialization 23867aec1d6eScindi */ 238720c794b3Sgavinm cmi_post_startup(); 23887aec1d6eScindi 23897c478bd9Sstevel@tonic-gate /* 239084ab085aSmws * Perform forceloading tasks for /etc/system. 23917c478bd9Sstevel@tonic-gate */ 239284ab085aSmws (void) mod_sysctl(SYS_FORCELOAD, NULL); 23937c478bd9Sstevel@tonic-gate 23947c478bd9Sstevel@tonic-gate /* 23957c478bd9Sstevel@tonic-gate * ON4.0: Force /proc module in until clock interrupt handle fixed 23967c478bd9Sstevel@tonic-gate * ON4.0: This must be fixed or restated in /etc/systems. 23977c478bd9Sstevel@tonic-gate */ 23987c478bd9Sstevel@tonic-gate (void) modload("fs", "procfs"); 23997c478bd9Sstevel@tonic-gate 2400c35aa225Smarx (void) i_ddi_attach_hw_nodes("pit_beep"); 2401c35aa225Smarx 24027c478bd9Sstevel@tonic-gate #if defined(__i386) 24037c478bd9Sstevel@tonic-gate /* 24047c478bd9Sstevel@tonic-gate * Check for required functional Floating Point hardware, 24057c478bd9Sstevel@tonic-gate * unless FP hardware explicitly disabled. 24067c478bd9Sstevel@tonic-gate */ 24077c478bd9Sstevel@tonic-gate if (fpu_exists && (fpu_pentium_fdivbug || fp_kind == FP_NO)) 24087c478bd9Sstevel@tonic-gate halt("No working FP hardware found"); 24097c478bd9Sstevel@tonic-gate #endif 24107c478bd9Sstevel@tonic-gate 24117c478bd9Sstevel@tonic-gate maxmem = freemem; 24127c478bd9Sstevel@tonic-gate 2413fb2caebeSRandy Fishel cpu_event_init_cpu(CPU); 24140e751525SEric Saxe cpupm_init(CPU); 2415fa96bd91SMichael Corcoran (void) mach_cpu_create_device_node(CPU, NULL); 24160e751525SEric Saxe 24170e751525SEric Saxe pg_init(); 24187c478bd9Sstevel@tonic-gate } 24197c478bd9Sstevel@tonic-gate 24207c478bd9Sstevel@tonic-gate static int 242119397407SSherry Moore pp_in_range(page_t *pp, uint64_t low_addr, uint64_t high_addr) 24227c478bd9Sstevel@tonic-gate { 242319397407SSherry Moore return ((pp->p_pagenum >= btop(low_addr)) && 242419397407SSherry Moore (pp->p_pagenum < btopr(high_addr))); 24257c478bd9Sstevel@tonic-gate } 24267c478bd9Sstevel@tonic-gate 24270181461bSKeith M Wesolowski static int 24280181461bSKeith M Wesolowski pp_in_module(page_t *pp, const rd_existing_t *modranges) 24290181461bSKeith M Wesolowski { 24300181461bSKeith M Wesolowski uint_t i; 24310181461bSKeith M Wesolowski 24320181461bSKeith M Wesolowski for (i = 0; modranges[i].phys != 0; i++) { 24330181461bSKeith M Wesolowski if (pp_in_range(pp, modranges[i].phys, 24340181461bSKeith M Wesolowski modranges[i].phys + modranges[i].size)) 24350181461bSKeith M Wesolowski return (1); 24360181461bSKeith M Wesolowski } 24370181461bSKeith M Wesolowski 24380181461bSKeith M Wesolowski return (0); 24390181461bSKeith M Wesolowski } 24400181461bSKeith M Wesolowski 24417c478bd9Sstevel@tonic-gate void 24427c478bd9Sstevel@tonic-gate release_bootstrap(void) 24437c478bd9Sstevel@tonic-gate { 24447c478bd9Sstevel@tonic-gate int root_is_ramdisk; 24457c478bd9Sstevel@tonic-gate page_t *pp; 24467c478bd9Sstevel@tonic-gate extern void kobj_boot_unmountroot(void); 24477c478bd9Sstevel@tonic-gate extern dev_t rootdev; 24480181461bSKeith M Wesolowski uint_t i; 24490181461bSKeith M Wesolowski char propname[32]; 24500181461bSKeith M Wesolowski rd_existing_t *modranges; 245155d507a9SSeth Goldberg #if !defined(__xpv) 245255d507a9SSeth Goldberg pfn_t pfn; 245355d507a9SSeth Goldberg #endif 24547c478bd9Sstevel@tonic-gate 24550181461bSKeith M Wesolowski /* 24560181461bSKeith M Wesolowski * Save the bootfs module ranges so that we can reserve them below 24570181461bSKeith M Wesolowski * for the real bootfs. 24580181461bSKeith M Wesolowski */ 24590181461bSKeith M Wesolowski modranges = kmem_alloc(sizeof (rd_existing_t) * MAX_BOOT_MODULES, 24600181461bSKeith M Wesolowski KM_SLEEP); 24610181461bSKeith M Wesolowski for (i = 0; ; i++) { 24620181461bSKeith M Wesolowski uint64_t start, size; 24630181461bSKeith M Wesolowski 24640181461bSKeith M Wesolowski modranges[i].phys = 0; 24650181461bSKeith M Wesolowski 24660181461bSKeith M Wesolowski (void) snprintf(propname, sizeof (propname), 24670181461bSKeith M Wesolowski "module-addr-%u", i); 24680181461bSKeith M Wesolowski if (do_bsys_getproplen(NULL, propname) <= 0) 24690181461bSKeith M Wesolowski break; 24700181461bSKeith M Wesolowski (void) do_bsys_getprop(NULL, propname, &start); 24710181461bSKeith M Wesolowski 24720181461bSKeith M Wesolowski (void) snprintf(propname, sizeof (propname), 24730181461bSKeith M Wesolowski "module-size-%u", i); 24740181461bSKeith M Wesolowski if (do_bsys_getproplen(NULL, propname) <= 0) 24750181461bSKeith M Wesolowski break; 24760181461bSKeith M Wesolowski (void) do_bsys_getprop(NULL, propname, &size); 24770181461bSKeith M Wesolowski 24780181461bSKeith M Wesolowski modranges[i].phys = start; 24790181461bSKeith M Wesolowski modranges[i].size = size; 24800181461bSKeith M Wesolowski } 24810181461bSKeith M Wesolowski 24827c478bd9Sstevel@tonic-gate /* unmount boot ramdisk and release kmem usage */ 24837c478bd9Sstevel@tonic-gate kobj_boot_unmountroot(); 24847c478bd9Sstevel@tonic-gate 24857c478bd9Sstevel@tonic-gate /* 24867c478bd9Sstevel@tonic-gate * We're finished using the boot loader so free its pages. 24877c478bd9Sstevel@tonic-gate */ 24887c478bd9Sstevel@tonic-gate PRM_POINT("Unmapping lower boot pages"); 248919397407SSherry Moore 2490ae115bc7Smrj clear_boot_mappings(0, _userlimit); 249119397407SSherry Moore 2492ae115bc7Smrj postbootkernelbase = kernelbase; 24937c478bd9Sstevel@tonic-gate 24947c478bd9Sstevel@tonic-gate /* 24957c478bd9Sstevel@tonic-gate * If root isn't on ramdisk, destroy the hardcoded 24967c478bd9Sstevel@tonic-gate * ramdisk node now and release the memory. Else, 24977c478bd9Sstevel@tonic-gate * ramdisk memory is kept in rd_pages. 24987c478bd9Sstevel@tonic-gate */ 24997c478bd9Sstevel@tonic-gate root_is_ramdisk = (getmajor(rootdev) == ddi_name_to_major("ramdisk")); 25007c478bd9Sstevel@tonic-gate if (!root_is_ramdisk) { 25017c478bd9Sstevel@tonic-gate dev_info_t *dip = ddi_find_devinfo("ramdisk", -1, 0); 25027c478bd9Sstevel@tonic-gate ASSERT(dip && ddi_get_parent(dip) == ddi_root_node()); 25037c478bd9Sstevel@tonic-gate ndi_rele_devi(dip); /* held from ddi_find_devinfo */ 25047c478bd9Sstevel@tonic-gate (void) ddi_remove_child(dip, 0); 25057c478bd9Sstevel@tonic-gate } 25067c478bd9Sstevel@tonic-gate 25077c478bd9Sstevel@tonic-gate PRM_POINT("Releasing boot pages"); 25087c478bd9Sstevel@tonic-gate while (bootpages) { 250919397407SSherry Moore extern uint64_t ramdisk_start, ramdisk_end; 25107c478bd9Sstevel@tonic-gate pp = bootpages; 25117c478bd9Sstevel@tonic-gate bootpages = pp->p_next; 251219397407SSherry Moore 251319397407SSherry Moore 251419397407SSherry Moore /* Keep pages for the lower 64K */ 251519397407SSherry Moore if (pp_in_range(pp, 0, 0x40000)) { 251619397407SSherry Moore pp->p_next = lower_pages; 251719397407SSherry Moore lower_pages = pp; 251819397407SSherry Moore lower_pages_count++; 251919397407SSherry Moore continue; 252019397407SSherry Moore } 252119397407SSherry Moore 252219397407SSherry Moore if (root_is_ramdisk && pp_in_range(pp, ramdisk_start, 25230181461bSKeith M Wesolowski ramdisk_end) || pp_in_module(pp, modranges)) { 25247c478bd9Sstevel@tonic-gate pp->p_next = rd_pages; 25257c478bd9Sstevel@tonic-gate rd_pages = pp; 25267c478bd9Sstevel@tonic-gate continue; 25277c478bd9Sstevel@tonic-gate } 25287c478bd9Sstevel@tonic-gate pp->p_next = (struct page *)0; 2529ae115bc7Smrj pp->p_prev = (struct page *)0; 2530ae115bc7Smrj PP_CLRBOOTPAGES(pp); 25317c478bd9Sstevel@tonic-gate page_free(pp, 1); 25327c478bd9Sstevel@tonic-gate } 2533ae115bc7Smrj PRM_POINT("Boot pages released"); 25347c478bd9Sstevel@tonic-gate 25350181461bSKeith M Wesolowski kmem_free(modranges, sizeof (rd_existing_t) * 99); 25360181461bSKeith M Wesolowski 2537843e1988Sjohnlev #if !defined(__xpv) 2538843e1988Sjohnlev /* XXPV -- note this following bunch of code needs to be revisited in Xen 3.0 */ 25397c478bd9Sstevel@tonic-gate /* 254055d507a9SSeth Goldberg * Find 1 page below 1 MB so that other processors can boot up or 254155d507a9SSeth Goldberg * so that any processor can resume. 25427c478bd9Sstevel@tonic-gate * Make sure it has a kernel VA as well as a 1:1 mapping. 25437c478bd9Sstevel@tonic-gate * We should have just free'd one up. 25447c478bd9Sstevel@tonic-gate */ 254519397407SSherry Moore 254619397407SSherry Moore /* 254719397407SSherry Moore * 0x10 pages is 64K. Leave the bottom 64K alone 254819397407SSherry Moore * for BIOS. 254919397407SSherry Moore */ 255019397407SSherry Moore for (pfn = 0x10; pfn < btop(1*1024*1024); pfn++) { 255155d507a9SSeth Goldberg if (page_numtopp_alloc(pfn) == NULL) 255255d507a9SSeth Goldberg continue; 255355d507a9SSeth Goldberg rm_platter_va = i86devmap(pfn, 1, 255455d507a9SSeth Goldberg PROT_READ | PROT_WRITE | PROT_EXEC); 255555d507a9SSeth Goldberg rm_platter_pa = ptob(pfn); 255655d507a9SSeth Goldberg break; 25577c478bd9Sstevel@tonic-gate } 255855d507a9SSeth Goldberg if (pfn == btop(1*1024*1024) && use_mp) 255955d507a9SSeth Goldberg panic("No page below 1M available for starting " 256055d507a9SSeth Goldberg "other processors or for resuming from system-suspend"); 2561843e1988Sjohnlev #endif /* !__xpv */ 25627c478bd9Sstevel@tonic-gate } 25637c478bd9Sstevel@tonic-gate 25647c478bd9Sstevel@tonic-gate /* 25657c478bd9Sstevel@tonic-gate * Initialize the platform-specific parts of a page_t. 25667c478bd9Sstevel@tonic-gate */ 25677c478bd9Sstevel@tonic-gate void 25687c478bd9Sstevel@tonic-gate add_physmem_cb(page_t *pp, pfn_t pnum) 25697c478bd9Sstevel@tonic-gate { 25707c478bd9Sstevel@tonic-gate pp->p_pagenum = pnum; 25717c478bd9Sstevel@tonic-gate pp->p_mapping = NULL; 25727c478bd9Sstevel@tonic-gate pp->p_embed = 0; 25737c478bd9Sstevel@tonic-gate pp->p_share = 0; 25747c478bd9Sstevel@tonic-gate pp->p_mlentry = 0; 25757c478bd9Sstevel@tonic-gate } 25767c478bd9Sstevel@tonic-gate 25777c478bd9Sstevel@tonic-gate /* 25787c478bd9Sstevel@tonic-gate * kphysm_init() initializes physical memory. 25797c478bd9Sstevel@tonic-gate */ 25807c478bd9Sstevel@tonic-gate static pgcnt_t 2581584b574aSToomas Soome kphysm_init(page_t *pp, pgcnt_t npages) 25827c478bd9Sstevel@tonic-gate { 25837c478bd9Sstevel@tonic-gate struct memlist *pmem; 25847c478bd9Sstevel@tonic-gate struct memseg *cur_memseg; 25857c478bd9Sstevel@tonic-gate pfn_t base_pfn; 258618968004SKit Chow pfn_t end_pfn; 25877c478bd9Sstevel@tonic-gate pgcnt_t num; 25887c478bd9Sstevel@tonic-gate pgcnt_t pages_done = 0; 25897c478bd9Sstevel@tonic-gate uint64_t addr; 25907c478bd9Sstevel@tonic-gate uint64_t size; 25917c478bd9Sstevel@tonic-gate extern pfn_t ddiphysmin; 259218968004SKit Chow extern int mnode_xwa; 259318968004SKit Chow int ms = 0, me = 0; 25947c478bd9Sstevel@tonic-gate 25957c478bd9Sstevel@tonic-gate ASSERT(page_hash != NULL && page_hashsz != 0); 25967c478bd9Sstevel@tonic-gate 2597ae115bc7Smrj cur_memseg = memseg_base; 259856f33205SJonathan Adams for (pmem = phys_avail; pmem && npages; pmem = pmem->ml_next) { 25997c478bd9Sstevel@tonic-gate /* 26007c478bd9Sstevel@tonic-gate * In a 32 bit kernel can't use higher memory if we're 26017c478bd9Sstevel@tonic-gate * not booting in PAE mode. This check takes care of that. 26027c478bd9Sstevel@tonic-gate */ 260356f33205SJonathan Adams addr = pmem->ml_address; 260456f33205SJonathan Adams size = pmem->ml_size; 26057c478bd9Sstevel@tonic-gate if (btop(addr) > physmax) 26067c478bd9Sstevel@tonic-gate continue; 26077c478bd9Sstevel@tonic-gate 26087c478bd9Sstevel@tonic-gate /* 26097c478bd9Sstevel@tonic-gate * align addr and size - they may not be at page boundaries 26107c478bd9Sstevel@tonic-gate */ 26117c478bd9Sstevel@tonic-gate if ((addr & MMU_PAGEOFFSET) != 0) { 26127c478bd9Sstevel@tonic-gate addr += MMU_PAGEOFFSET; 26137c478bd9Sstevel@tonic-gate addr &= ~(uint64_t)MMU_PAGEOFFSET; 261456f33205SJonathan Adams size -= addr - pmem->ml_address; 26157c478bd9Sstevel@tonic-gate } 26167c478bd9Sstevel@tonic-gate 26176bb54764Skchow /* only process pages below or equal to physmax */ 26186bb54764Skchow if ((btop(addr + size) - 1) > physmax) 26196bb54764Skchow size = ptob(physmax - btop(addr) + 1); 26207c478bd9Sstevel@tonic-gate 26217c478bd9Sstevel@tonic-gate num = btop(size); 26227c478bd9Sstevel@tonic-gate if (num == 0) 26237c478bd9Sstevel@tonic-gate continue; 26247c478bd9Sstevel@tonic-gate 26257c478bd9Sstevel@tonic-gate if (num > npages) 26267c478bd9Sstevel@tonic-gate num = npages; 26277c478bd9Sstevel@tonic-gate 26287c478bd9Sstevel@tonic-gate npages -= num; 26297c478bd9Sstevel@tonic-gate pages_done += num; 26307c478bd9Sstevel@tonic-gate base_pfn = btop(addr); 26317c478bd9Sstevel@tonic-gate 26327c478bd9Sstevel@tonic-gate if (prom_debug) 26337c478bd9Sstevel@tonic-gate prom_printf("MEMSEG addr=0x%" PRIx64 26347c478bd9Sstevel@tonic-gate " pgs=0x%lx pfn 0x%lx-0x%lx\n", 26357c478bd9Sstevel@tonic-gate addr, num, base_pfn, base_pfn + num); 26367c478bd9Sstevel@tonic-gate 26377c478bd9Sstevel@tonic-gate /* 2638ae115bc7Smrj * Ignore pages below ddiphysmin to simplify ddi memory 26397c478bd9Sstevel@tonic-gate * allocation with non-zero addr_lo requests. 26407c478bd9Sstevel@tonic-gate */ 26417c478bd9Sstevel@tonic-gate if (base_pfn < ddiphysmin) { 2642ae115bc7Smrj if (base_pfn + num <= ddiphysmin) 26437c478bd9Sstevel@tonic-gate continue; 26447c478bd9Sstevel@tonic-gate pp += (ddiphysmin - base_pfn); 26457c478bd9Sstevel@tonic-gate num -= (ddiphysmin - base_pfn); 26467c478bd9Sstevel@tonic-gate base_pfn = ddiphysmin; 26477c478bd9Sstevel@tonic-gate } 2648ae115bc7Smrj 26497c478bd9Sstevel@tonic-gate /* 265018968004SKit Chow * mnode_xwa is greater than 1 when large pages regions can 265118968004SKit Chow * cross memory node boundaries. To prevent the formation 265218968004SKit Chow * of these large pages, configure the memsegs based on the 265318968004SKit Chow * memory node ranges which had been made non-contiguous. 26547c478bd9Sstevel@tonic-gate */ 2655584b574aSToomas Soome end_pfn = base_pfn + num - 1; 265618968004SKit Chow if (mnode_xwa > 1) { 265718968004SKit Chow ms = PFN_2_MEM_NODE(base_pfn); 265818968004SKit Chow me = PFN_2_MEM_NODE(end_pfn); 265918968004SKit Chow 266018968004SKit Chow if (ms != me) { 266118968004SKit Chow /* 266218968004SKit Chow * current range spans more than 1 memory node. 266318968004SKit Chow * Set num to only the pfn range in the start 266418968004SKit Chow * memory node. 266518968004SKit Chow */ 266618968004SKit Chow num = mem_node_config[ms].physmax - base_pfn 266718968004SKit Chow + 1; 266818968004SKit Chow ASSERT(end_pfn > mem_node_config[ms].physmax); 266918968004SKit Chow } 26707c478bd9Sstevel@tonic-gate } 26717c478bd9Sstevel@tonic-gate 267218968004SKit Chow for (;;) { 267318968004SKit Chow /* 267418968004SKit Chow * Build the memsegs entry 267518968004SKit Chow */ 267618968004SKit Chow cur_memseg->pages = pp; 267718968004SKit Chow cur_memseg->epages = pp + num; 267818968004SKit Chow cur_memseg->pages_base = base_pfn; 267918968004SKit Chow cur_memseg->pages_end = base_pfn + num; 268018968004SKit Chow 268118968004SKit Chow /* 268218968004SKit Chow * Insert into memseg list in decreasing pfn range 268318968004SKit Chow * order. Low memory is typically more fragmented such 268418968004SKit Chow * that this ordering keeps the larger ranges at the 268518968004SKit Chow * front of the list for code that searches memseg. 268618968004SKit Chow * This ASSERTS that the memsegs coming in from boot 268718968004SKit Chow * are in increasing physical address order and not 268818968004SKit Chow * contiguous. 268918968004SKit Chow */ 269018968004SKit Chow if (memsegs != NULL) { 269118968004SKit Chow ASSERT(cur_memseg->pages_base >= 269218968004SKit Chow memsegs->pages_end); 269318968004SKit Chow cur_memseg->next = memsegs; 269418968004SKit Chow } 269518968004SKit Chow memsegs = cur_memseg; 269618968004SKit Chow 269718968004SKit Chow /* 269818968004SKit Chow * add_physmem() initializes the PSM part of the page 269918968004SKit Chow * struct by calling the PSM back with add_physmem_cb(). 270018968004SKit Chow * In addition it coalesces pages into larger pages as 270118968004SKit Chow * it initializes them. 270218968004SKit Chow */ 270318968004SKit Chow add_physmem(pp, num, base_pfn); 270418968004SKit Chow cur_memseg++; 270518968004SKit Chow availrmem_initial += num; 270618968004SKit Chow availrmem += num; 270718968004SKit Chow 270818968004SKit Chow pp += num; 270918968004SKit Chow if (ms >= me) 271018968004SKit Chow break; 27117c478bd9Sstevel@tonic-gate 271218968004SKit Chow /* process next memory node range */ 271318968004SKit Chow ms++; 271418968004SKit Chow base_pfn = mem_node_config[ms].physbase; 2715584b574aSToomas Soome 2716584b574aSToomas Soome if (mnode_xwa > 1) { 2717584b574aSToomas Soome num = MIN(mem_node_config[ms].physmax, 2718584b574aSToomas Soome end_pfn) - base_pfn + 1; 2719584b574aSToomas Soome } else { 2720584b574aSToomas Soome num = mem_node_config[ms].physmax - 2721584b574aSToomas Soome base_pfn + 1; 2722584b574aSToomas Soome } 272318968004SKit Chow } 27247c478bd9Sstevel@tonic-gate } 27257c478bd9Sstevel@tonic-gate 27267c478bd9Sstevel@tonic-gate PRM_DEBUG(availrmem_initial); 27277c478bd9Sstevel@tonic-gate PRM_DEBUG(availrmem); 27287c478bd9Sstevel@tonic-gate PRM_DEBUG(freemem); 27297c478bd9Sstevel@tonic-gate build_pfn_hash(); 27307c478bd9Sstevel@tonic-gate return (pages_done); 27317c478bd9Sstevel@tonic-gate } 27327c478bd9Sstevel@tonic-gate 27337c478bd9Sstevel@tonic-gate /* 27347c478bd9Sstevel@tonic-gate * Kernel VM initialization. 27357c478bd9Sstevel@tonic-gate */ 27367c478bd9Sstevel@tonic-gate static void 27377c478bd9Sstevel@tonic-gate kvm_init(void) 27387c478bd9Sstevel@tonic-gate { 27397c478bd9Sstevel@tonic-gate ASSERT((((uintptr_t)s_text) & MMU_PAGEOFFSET) == 0); 27407c478bd9Sstevel@tonic-gate 27417c478bd9Sstevel@tonic-gate /* 27427c478bd9Sstevel@tonic-gate * Put the kernel segments in kernel address space. 27437c478bd9Sstevel@tonic-gate */ 27447c478bd9Sstevel@tonic-gate rw_enter(&kas.a_lock, RW_WRITER); 27457c478bd9Sstevel@tonic-gate as_avlinit(&kas); 27467c478bd9Sstevel@tonic-gate 27477c478bd9Sstevel@tonic-gate (void) seg_attach(&kas, s_text, e_moddata - s_text, &ktextseg); 27487c478bd9Sstevel@tonic-gate (void) segkmem_create(&ktextseg); 27497c478bd9Sstevel@tonic-gate 27507c478bd9Sstevel@tonic-gate (void) seg_attach(&kas, (caddr_t)valloc_base, valloc_sz, &kvalloc); 27517c478bd9Sstevel@tonic-gate (void) segkmem_create(&kvalloc); 27527c478bd9Sstevel@tonic-gate 275335b1ab99Sjosephb (void) seg_attach(&kas, kernelheap, 275435b1ab99Sjosephb ekernelheap - kernelheap, &kvseg); 27557c478bd9Sstevel@tonic-gate (void) segkmem_create(&kvseg); 27567c478bd9Sstevel@tonic-gate 2757ae115bc7Smrj if (core_size > 0) { 2758ae115bc7Smrj PRM_POINT("attaching kvseg_core"); 2759ae115bc7Smrj (void) seg_attach(&kas, (caddr_t)core_base, core_size, 2760ae115bc7Smrj &kvseg_core); 2761ae115bc7Smrj (void) segkmem_create(&kvseg_core); 2762ae115bc7Smrj } 2763ad23a2dbSjohansen 2764*04909c8cSJohn Levon PRM_POINT("attaching segkvmm"); 2765*04909c8cSJohn Levon (void) seg_attach(&kas, segkvmm_base, mmu_ptob(segkvmmsize), &kvmmseg); 2766*04909c8cSJohn Levon (void) segkmem_create(&kvmmseg); 2767*04909c8cSJohn Levon segkmem_kvmm_init(segkvmm_base, mmu_ptob(segkvmmsize)); 2768*04909c8cSJohn Levon 2769ae115bc7Smrj if (segziosize > 0) { 2770ae115bc7Smrj PRM_POINT("attaching segzio"); 2771ad23a2dbSjohansen (void) seg_attach(&kas, segzio_base, mmu_ptob(segziosize), 2772ad23a2dbSjohansen &kzioseg); 2773*04909c8cSJohn Levon (void) segkmem_create(&kzioseg); 2774ad23a2dbSjohansen 2775ad23a2dbSjohansen /* create zio area covering new segment */ 2776ad23a2dbSjohansen segkmem_zio_init(segzio_base, mmu_ptob(segziosize)); 2777ad23a2dbSjohansen } 27787c478bd9Sstevel@tonic-gate 2779ae115bc7Smrj (void) seg_attach(&kas, kdi_segdebugbase, kdi_segdebugsize, &kdebugseg); 27807c478bd9Sstevel@tonic-gate (void) segkmem_create(&kdebugseg); 27817c478bd9Sstevel@tonic-gate 27827c478bd9Sstevel@tonic-gate rw_exit(&kas.a_lock); 27837c478bd9Sstevel@tonic-gate 27847c478bd9Sstevel@tonic-gate /* 27857c478bd9Sstevel@tonic-gate * Ensure that the red zone at kernelbase is never accessible. 27867c478bd9Sstevel@tonic-gate */ 2787ae115bc7Smrj PRM_POINT("protecting redzone"); 27887c478bd9Sstevel@tonic-gate (void) as_setprot(&kas, (caddr_t)kernelbase, KERNEL_REDZONE_SIZE, 0); 27897c478bd9Sstevel@tonic-gate 27907c478bd9Sstevel@tonic-gate /* 27917c478bd9Sstevel@tonic-gate * Make the text writable so that it can be hot patched by DTrace. 27927c478bd9Sstevel@tonic-gate */ 27937c478bd9Sstevel@tonic-gate (void) as_setprot(&kas, s_text, e_modtext - s_text, 27947c478bd9Sstevel@tonic-gate PROT_READ | PROT_WRITE | PROT_EXEC); 27957c478bd9Sstevel@tonic-gate 27967c478bd9Sstevel@tonic-gate /* 27977c478bd9Sstevel@tonic-gate * Make data writable until end. 27987c478bd9Sstevel@tonic-gate */ 27997c478bd9Sstevel@tonic-gate (void) as_setprot(&kas, s_data, e_moddata - s_data, 28007c478bd9Sstevel@tonic-gate PROT_READ | PROT_WRITE | PROT_EXEC); 28017c478bd9Sstevel@tonic-gate } 28027c478bd9Sstevel@tonic-gate 2803843e1988Sjohnlev #ifndef __xpv 28047c478bd9Sstevel@tonic-gate /* 28051d03c31eSjohnlev * Solaris adds an entry for Write Combining caching to the PAT 28067c478bd9Sstevel@tonic-gate */ 28071d03c31eSjohnlev static uint64_t pat_attr_reg = PAT_DEFAULT_ATTRIBUTE; 28087c478bd9Sstevel@tonic-gate 28097c478bd9Sstevel@tonic-gate void 28101d03c31eSjohnlev pat_sync(void) 28117c478bd9Sstevel@tonic-gate { 28121d03c31eSjohnlev ulong_t cr0, cr0_orig, cr4; 28137c478bd9Sstevel@tonic-gate 28147417cfdeSKuriakose Kuruvilla if (!is_x86_feature(x86_featureset, X86FSET_PAT)) 28157c478bd9Sstevel@tonic-gate return; 28161d03c31eSjohnlev cr0_orig = cr0 = getcr0(); 28171d03c31eSjohnlev cr4 = getcr4(); 28187c478bd9Sstevel@tonic-gate 28191d03c31eSjohnlev /* disable caching and flush all caches and TLBs */ 28201d03c31eSjohnlev cr0 |= CR0_CD; 28211d03c31eSjohnlev cr0 &= ~CR0_NW; 28221d03c31eSjohnlev setcr0(cr0); 28231d03c31eSjohnlev invalidate_cache(); 28241d03c31eSjohnlev if (cr4 & CR4_PGE) { 28251d03c31eSjohnlev setcr4(cr4 & ~(ulong_t)CR4_PGE); 28261d03c31eSjohnlev setcr4(cr4); 28271d03c31eSjohnlev } else { 28281d03c31eSjohnlev reload_cr3(); 28297c478bd9Sstevel@tonic-gate } 28307c478bd9Sstevel@tonic-gate 28311d03c31eSjohnlev /* add our entry to the PAT */ 28321d03c31eSjohnlev wrmsr(REG_PAT, pat_attr_reg); 28337c478bd9Sstevel@tonic-gate 28341d03c31eSjohnlev /* flush TLBs and cache again, then reenable cr0 caching */ 28351d03c31eSjohnlev if (cr4 & CR4_PGE) { 28361d03c31eSjohnlev setcr4(cr4 & ~(ulong_t)CR4_PGE); 28371d03c31eSjohnlev setcr4(cr4); 28381d03c31eSjohnlev } else { 28391d03c31eSjohnlev reload_cr3(); 28407c478bd9Sstevel@tonic-gate } 28417c478bd9Sstevel@tonic-gate invalidate_cache(); 28427c478bd9Sstevel@tonic-gate setcr0(cr0_orig); 28437c478bd9Sstevel@tonic-gate } 28447c478bd9Sstevel@tonic-gate 28451d03c31eSjohnlev #endif /* !__xpv */ 28467c478bd9Sstevel@tonic-gate 2847ed5289f9SKen Erickson #if defined(_SOFT_HOSTID) 2848ed5289f9SKen Erickson /* 2849ed5289f9SKen Erickson * On platforms that do not have a hardware serial number, attempt 2850ed5289f9SKen Erickson * to set one based on the contents of /etc/hostid. If this file does 2851ed5289f9SKen Erickson * not exist, assume that we are to generate a new hostid and set 2852ed5289f9SKen Erickson * it in the kernel, for subsequent saving by a userland process 2853ed5289f9SKen Erickson * once the system is up and the root filesystem is mounted r/w. 2854ed5289f9SKen Erickson * 285596d008c2SKen Erickson * In order to gracefully support upgrade on OpenSolaris, if 285696d008c2SKen Erickson * /etc/hostid does not exist, we will attempt to get a serial number 285796d008c2SKen Erickson * using the legacy method (/kernel/misc/sysinit). 285896d008c2SKen Erickson * 28599e3e4df2SGarrett D'Amore * If that isn't present, we attempt to use an SMBIOS UUID, which is 28609e3e4df2SGarrett D'Amore * a hardware serial number. Note that we don't automatically trust 28619e3e4df2SGarrett D'Amore * all SMBIOS UUIDs (some older platforms are defective and ship duplicate 28629e3e4df2SGarrett D'Amore * UUIDs in violation of the standard), we check against a blacklist. 28639e3e4df2SGarrett D'Amore * 2864ed5289f9SKen Erickson * In an attempt to make the hostid less prone to abuse 2865ed5289f9SKen Erickson * (for license circumvention, etc), we store it in /etc/hostid 2866ed5289f9SKen Erickson * in rot47 format. 2867ed5289f9SKen Erickson */ 2868ed5289f9SKen Erickson extern volatile unsigned long tenmicrodata; 286996d008c2SKen Erickson static int atoi(char *); 2870ed5289f9SKen Erickson 28719e3e4df2SGarrett D'Amore /* 28729e3e4df2SGarrett D'Amore * Set this to non-zero in /etc/system if you think your SMBIOS returns a 28739e3e4df2SGarrett D'Amore * UUID that is not unique. (Also report it so that the smbios_uuid_blacklist 28749e3e4df2SGarrett D'Amore * array can be updated.) 28759e3e4df2SGarrett D'Amore */ 28769e3e4df2SGarrett D'Amore int smbios_broken_uuid = 0; 28779e3e4df2SGarrett D'Amore 28789e3e4df2SGarrett D'Amore /* 28799e3e4df2SGarrett D'Amore * List of known bad UUIDs. This is just the lower 32-bit values, since 28809e3e4df2SGarrett D'Amore * that's what we use for the host id. If your hostid falls here, you need 28819e3e4df2SGarrett D'Amore * to contact your hardware OEM for a fix for your BIOS. 28829e3e4df2SGarrett D'Amore */ 28839e3e4df2SGarrett D'Amore static unsigned char 28849e3e4df2SGarrett D'Amore smbios_uuid_blacklist[][16] = { 28859e3e4df2SGarrett D'Amore 28869e3e4df2SGarrett D'Amore { /* Reported bad UUID (Google search) */ 28879e3e4df2SGarrett D'Amore 0x00, 0x02, 0x00, 0x03, 0x00, 0x04, 0x00, 0x05, 28889e3e4df2SGarrett D'Amore 0x00, 0x06, 0x00, 0x07, 0x00, 0x08, 0x00, 0x09, 28899e3e4df2SGarrett D'Amore }, 28909e3e4df2SGarrett D'Amore { /* Known bad DELL UUID */ 28919e3e4df2SGarrett D'Amore 0x4C, 0x4C, 0x45, 0x44, 0x00, 0x00, 0x20, 0x10, 28929e3e4df2SGarrett D'Amore 0x80, 0x20, 0x80, 0xC0, 0x4F, 0x20, 0x20, 0x20, 28939e3e4df2SGarrett D'Amore }, 28949e3e4df2SGarrett D'Amore { /* Uninitialized flash */ 28959e3e4df2SGarrett D'Amore 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 28969e3e4df2SGarrett D'Amore 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff 28979e3e4df2SGarrett D'Amore }, 28989e3e4df2SGarrett D'Amore { /* All zeros */ 28999e3e4df2SGarrett D'Amore 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 29009e3e4df2SGarrett D'Amore 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00 29019e3e4df2SGarrett D'Amore }, 29029e3e4df2SGarrett D'Amore }; 29039e3e4df2SGarrett D'Amore 29049e3e4df2SGarrett D'Amore static int32_t 29059e3e4df2SGarrett D'Amore uuid_to_hostid(const uint8_t *uuid) 29069e3e4df2SGarrett D'Amore { 29079e3e4df2SGarrett D'Amore /* 29089e3e4df2SGarrett D'Amore * Although the UUIDs are 128-bits, they may not distribute entropy 29099e3e4df2SGarrett D'Amore * evenly. We would like to use SHA or MD5, but those are located 29109e3e4df2SGarrett D'Amore * in loadable modules and not available this early in boot. As we 29119e3e4df2SGarrett D'Amore * don't need the values to be cryptographically strong, we just 29129e3e4df2SGarrett D'Amore * generate 32-bit vaue by xor'ing the various sequences together, 29132df37efaSGarrett D'Amore * which ensures that the entire UUID contributes to the hostid. 29149e3e4df2SGarrett D'Amore */ 29152df37efaSGarrett D'Amore uint32_t id = 0; 29169e3e4df2SGarrett D'Amore 29179e3e4df2SGarrett D'Amore /* first check against the blacklist */ 29189e3e4df2SGarrett D'Amore for (int i = 0; i < (sizeof (smbios_uuid_blacklist) / 16); i++) { 29199e3e4df2SGarrett D'Amore if (bcmp(smbios_uuid_blacklist[0], uuid, 16) == 0) { 29209e3e4df2SGarrett D'Amore cmn_err(CE_CONT, "?Broken SMBIOS UUID. " 29219e3e4df2SGarrett D'Amore "Contact BIOS manufacturer for repair.\n"); 29229e3e4df2SGarrett D'Amore return ((int32_t)HW_INVALID_HOSTID); 29239e3e4df2SGarrett D'Amore } 29249e3e4df2SGarrett D'Amore } 29259e3e4df2SGarrett D'Amore 29269e3e4df2SGarrett D'Amore for (int i = 0; i < 16; i++) 29279e3e4df2SGarrett D'Amore id ^= ((uuid[i]) << (8 * (i % sizeof (id)))); 29289e3e4df2SGarrett D'Amore 29292df37efaSGarrett D'Amore /* Make sure return value is positive */ 29302df37efaSGarrett D'Amore return (id & 0x7fffffff); 29319e3e4df2SGarrett D'Amore } 29329e3e4df2SGarrett D'Amore 2933ed5289f9SKen Erickson static int32_t 2934ed5289f9SKen Erickson set_soft_hostid(void) 2935ed5289f9SKen Erickson { 2936ed5289f9SKen Erickson struct _buf *file; 2937ed5289f9SKen Erickson char tokbuf[MAXNAMELEN]; 2938ed5289f9SKen Erickson token_t token; 2939ed5289f9SKen Erickson int done = 0; 2940ed5289f9SKen Erickson u_longlong_t tmp; 294196d008c2SKen Erickson int i; 29425679c89fSjv int32_t hostid = (int32_t)HW_INVALID_HOSTID; 2943ed5289f9SKen Erickson unsigned char *c; 2944ed5289f9SKen Erickson hrtime_t tsc; 29459e3e4df2SGarrett D'Amore smbios_system_t smsys; 2946ed5289f9SKen Erickson 2947ed5289f9SKen Erickson /* 2948ed5289f9SKen Erickson * If /etc/hostid file not found, we'd like to get a pseudo 2949ed5289f9SKen Erickson * random number to use at the hostid. A nice way to do this 2950ed5289f9SKen Erickson * is to read the real time clock. To remain xen-compatible, 2951ed5289f9SKen Erickson * we can't poke the real hardware, so we use tsc_read() to 2952ed5289f9SKen Erickson * read the real time clock. However, there is an ominous 2953ed5289f9SKen Erickson * warning in tsc_read that says it can return zero, so we 2954ed5289f9SKen Erickson * deal with that possibility by falling back to using the 2955ed5289f9SKen Erickson * (hopefully random enough) value in tenmicrodata. 2956ed5289f9SKen Erickson */ 2957ed5289f9SKen Erickson 2958ed5289f9SKen Erickson if ((file = kobj_open_file(hostid_file)) == (struct _buf *)-1) { 295996d008c2SKen Erickson /* 296096d008c2SKen Erickson * hostid file not found - try to load sysinit module 296196d008c2SKen Erickson * and see if it has a nonzero hostid value...use that 296296d008c2SKen Erickson * instead of generating a new hostid here if so. 296396d008c2SKen Erickson */ 296496d008c2SKen Erickson if ((i = modload("misc", "sysinit")) != -1) { 296596d008c2SKen Erickson if (strlen(hw_serial) > 0) 296696d008c2SKen Erickson hostid = (int32_t)atoi(hw_serial); 296796d008c2SKen Erickson (void) modunload(i); 296896d008c2SKen Erickson } 29699e3e4df2SGarrett D'Amore 29709e3e4df2SGarrett D'Amore /* 29719e3e4df2SGarrett D'Amore * We try to use the SMBIOS UUID. But not if it is blacklisted 29729e3e4df2SGarrett D'Amore * in /etc/system. 29739e3e4df2SGarrett D'Amore */ 29749e3e4df2SGarrett D'Amore if ((hostid == HW_INVALID_HOSTID) && 29759e3e4df2SGarrett D'Amore (smbios_broken_uuid == 0) && 29769e3e4df2SGarrett D'Amore (ksmbios != NULL) && 29779e3e4df2SGarrett D'Amore (smbios_info_system(ksmbios, &smsys) != SMB_ERR) && 29789e3e4df2SGarrett D'Amore (smsys.smbs_uuidlen >= 16)) { 29799e3e4df2SGarrett D'Amore hostid = uuid_to_hostid(smsys.smbs_uuid); 29809e3e4df2SGarrett D'Amore } 29819e3e4df2SGarrett D'Amore 29829e3e4df2SGarrett D'Amore /* 29839e3e4df2SGarrett D'Amore * Generate a "random" hostid using the clock. These 29849e3e4df2SGarrett D'Amore * hostids will change on each boot if the value is not 29859e3e4df2SGarrett D'Amore * saved to a persistent /etc/hostid file. 29869e3e4df2SGarrett D'Amore */ 29875679c89fSjv if (hostid == HW_INVALID_HOSTID) { 298896d008c2SKen Erickson tsc = tsc_read(); 298996d008c2SKen Erickson if (tsc == 0) /* tsc_read can return zero sometimes */ 299096d008c2SKen Erickson hostid = (int32_t)tenmicrodata & 0x0CFFFFF; 299196d008c2SKen Erickson else 299296d008c2SKen Erickson hostid = (int32_t)tsc & 0x0CFFFFF; 299396d008c2SKen Erickson } 2994ed5289f9SKen Erickson } else { 2995ed5289f9SKen Erickson /* hostid file found */ 2996ed5289f9SKen Erickson while (!done) { 2997ed5289f9SKen Erickson token = kobj_lex(file, tokbuf, sizeof (tokbuf)); 2998ed5289f9SKen Erickson 2999ed5289f9SKen Erickson switch (token) { 3000ed5289f9SKen Erickson case POUND: 3001ed5289f9SKen Erickson /* 3002ed5289f9SKen Erickson * skip comments 3003ed5289f9SKen Erickson */ 3004ed5289f9SKen Erickson kobj_find_eol(file); 3005ed5289f9SKen Erickson break; 3006ed5289f9SKen Erickson case STRING: 3007ed5289f9SKen Erickson /* 3008ed5289f9SKen Erickson * un-rot47 - obviously this 3009ed5289f9SKen Erickson * nonsense is ascii-specific 3010ed5289f9SKen Erickson */ 3011ed5289f9SKen Erickson for (c = (unsigned char *)tokbuf; 3012ed5289f9SKen Erickson *c != '\0'; c++) { 3013ed5289f9SKen Erickson *c += 47; 3014ed5289f9SKen Erickson if (*c > '~') 3015ed5289f9SKen Erickson *c -= 94; 3016ed5289f9SKen Erickson else if (*c < '!') 3017ed5289f9SKen Erickson *c += 94; 3018ed5289f9SKen Erickson } 3019ed5289f9SKen Erickson /* 3020ed5289f9SKen Erickson * now we should have a real number 3021ed5289f9SKen Erickson */ 3022ed5289f9SKen Erickson 3023ed5289f9SKen Erickson if (kobj_getvalue(tokbuf, &tmp) != 0) 3024ed5289f9SKen Erickson kobj_file_err(CE_WARN, file, 3025ed5289f9SKen Erickson "Bad value %s for hostid", 3026ed5289f9SKen Erickson tokbuf); 3027ed5289f9SKen Erickson else 3028ed5289f9SKen Erickson hostid = (int32_t)tmp; 3029ed5289f9SKen Erickson 3030ed5289f9SKen Erickson break; 3031ed5289f9SKen Erickson case EOF: 3032ed5289f9SKen Erickson done = 1; 3033ed5289f9SKen Erickson /* FALLTHROUGH */ 3034ed5289f9SKen Erickson case NEWLINE: 3035ed5289f9SKen Erickson kobj_newline(file); 3036ed5289f9SKen Erickson break; 3037ed5289f9SKen Erickson default: 3038ed5289f9SKen Erickson break; 3039ed5289f9SKen Erickson 3040ed5289f9SKen Erickson } 3041ed5289f9SKen Erickson } 30425679c89fSjv if (hostid == HW_INVALID_HOSTID) /* didn't find a hostid */ 3043ed5289f9SKen Erickson kobj_file_err(CE_WARN, file, 3044ed5289f9SKen Erickson "hostid missing or corrupt"); 3045ed5289f9SKen Erickson 3046ed5289f9SKen Erickson kobj_close_file(file); 3047ed5289f9SKen Erickson } 3048ed5289f9SKen Erickson /* 3049ed5289f9SKen Erickson * hostid is now the value read from /etc/hostid, or the 30505679c89fSjv * new hostid we generated in this routine or HW_INVALID_HOSTID if not 30515679c89fSjv * set. 3052ed5289f9SKen Erickson */ 3053ed5289f9SKen Erickson return (hostid); 3054ed5289f9SKen Erickson } 305596d008c2SKen Erickson 305696d008c2SKen Erickson static int 305796d008c2SKen Erickson atoi(char *p) 305896d008c2SKen Erickson { 305996d008c2SKen Erickson int i = 0; 306096d008c2SKen Erickson 306196d008c2SKen Erickson while (*p != '\0') 306296d008c2SKen Erickson i = 10 * i + (*p++ - '0'); 306396d008c2SKen Erickson 306496d008c2SKen Erickson return (i); 306596d008c2SKen Erickson } 306696d008c2SKen Erickson 3067ed5289f9SKen Erickson #endif /* _SOFT_HOSTID */ 3068ed5289f9SKen Erickson 30697c478bd9Sstevel@tonic-gate void 307045916cd2Sjpk get_system_configuration(void) 30717c478bd9Sstevel@tonic-gate { 30727c478bd9Sstevel@tonic-gate char prop[32]; 30737c478bd9Sstevel@tonic-gate u_longlong_t nodes_ll, cpus_pernode_ll, lvalue; 30747c478bd9Sstevel@tonic-gate 307535b1ab99Sjosephb if (BOP_GETPROPLEN(bootops, "nodes") > sizeof (prop) || 307635b1ab99Sjosephb BOP_GETPROP(bootops, "nodes", prop) < 0 || 307735b1ab99Sjosephb kobj_getvalue(prop, &nodes_ll) == -1 || 307835b1ab99Sjosephb nodes_ll > MAXNODES || 307935b1ab99Sjosephb BOP_GETPROPLEN(bootops, "cpus_pernode") > sizeof (prop) || 308035b1ab99Sjosephb BOP_GETPROP(bootops, "cpus_pernode", prop) < 0 || 308135b1ab99Sjosephb kobj_getvalue(prop, &cpus_pernode_ll) == -1) { 30827c478bd9Sstevel@tonic-gate system_hardware.hd_nodes = 1; 30837c478bd9Sstevel@tonic-gate system_hardware.hd_cpus_per_node = 0; 30847c478bd9Sstevel@tonic-gate } else { 30857c478bd9Sstevel@tonic-gate system_hardware.hd_nodes = (int)nodes_ll; 30867c478bd9Sstevel@tonic-gate system_hardware.hd_cpus_per_node = (int)cpus_pernode_ll; 30877c478bd9Sstevel@tonic-gate } 308835b1ab99Sjosephb 308935b1ab99Sjosephb if (BOP_GETPROPLEN(bootops, "kernelbase") > sizeof (prop) || 309035b1ab99Sjosephb BOP_GETPROP(bootops, "kernelbase", prop) < 0 || 309135b1ab99Sjosephb kobj_getvalue(prop, &lvalue) == -1) 30924da99751SToomas Soome eprom_kernelbase = 0; 30937c478bd9Sstevel@tonic-gate else 309435b1ab99Sjosephb eprom_kernelbase = (uintptr_t)lvalue; 30957c478bd9Sstevel@tonic-gate 309635b1ab99Sjosephb if (BOP_GETPROPLEN(bootops, "segmapsize") > sizeof (prop) || 309735b1ab99Sjosephb BOP_GETPROP(bootops, "segmapsize", prop) < 0 || 309835b1ab99Sjosephb kobj_getvalue(prop, &lvalue) == -1) 30997c478bd9Sstevel@tonic-gate segmapsize = SEGMAPDEFAULT; 310035b1ab99Sjosephb else 31017c478bd9Sstevel@tonic-gate segmapsize = (uintptr_t)lvalue; 31027c478bd9Sstevel@tonic-gate 310335b1ab99Sjosephb if (BOP_GETPROPLEN(bootops, "segmapfreelists") > sizeof (prop) || 310435b1ab99Sjosephb BOP_GETPROP(bootops, "segmapfreelists", prop) < 0 || 310535b1ab99Sjosephb kobj_getvalue(prop, &lvalue) == -1) 31067c478bd9Sstevel@tonic-gate segmapfreelists = 0; /* use segmap driver default */ 310735b1ab99Sjosephb else 31087c478bd9Sstevel@tonic-gate segmapfreelists = (int)lvalue; 31094944b02eSkchow 3110aac11643Sjosephb /* physmem used to be here, but moved much earlier to fakebop.c */ 31117c478bd9Sstevel@tonic-gate } 31127c478bd9Sstevel@tonic-gate 31137c478bd9Sstevel@tonic-gate /* 31147c478bd9Sstevel@tonic-gate * Add to a memory list. 31157c478bd9Sstevel@tonic-gate * start = start of new memory segment 31167c478bd9Sstevel@tonic-gate * len = length of new memory segment in bytes 31177c478bd9Sstevel@tonic-gate * new = pointer to a new struct memlist 31187c478bd9Sstevel@tonic-gate * memlistp = memory list to which to add segment. 31197c478bd9Sstevel@tonic-gate */ 3120ae115bc7Smrj void 31217c478bd9Sstevel@tonic-gate memlist_add( 31227c478bd9Sstevel@tonic-gate uint64_t start, 31237c478bd9Sstevel@tonic-gate uint64_t len, 31247c478bd9Sstevel@tonic-gate struct memlist *new, 31257c478bd9Sstevel@tonic-gate struct memlist **memlistp) 31267c478bd9Sstevel@tonic-gate { 31277c478bd9Sstevel@tonic-gate struct memlist *cur; 31287c478bd9Sstevel@tonic-gate uint64_t end = start + len; 31297c478bd9Sstevel@tonic-gate 313056f33205SJonathan Adams new->ml_address = start; 313156f33205SJonathan Adams new->ml_size = len; 31327c478bd9Sstevel@tonic-gate 31337c478bd9Sstevel@tonic-gate cur = *memlistp; 31347c478bd9Sstevel@tonic-gate 31357c478bd9Sstevel@tonic-gate while (cur) { 313656f33205SJonathan Adams if (cur->ml_address >= end) { 313756f33205SJonathan Adams new->ml_next = cur; 31387c478bd9Sstevel@tonic-gate *memlistp = new; 313956f33205SJonathan Adams new->ml_prev = cur->ml_prev; 314056f33205SJonathan Adams cur->ml_prev = new; 31417c478bd9Sstevel@tonic-gate return; 31427c478bd9Sstevel@tonic-gate } 314356f33205SJonathan Adams ASSERT(cur->ml_address + cur->ml_size <= start); 314456f33205SJonathan Adams if (cur->ml_next == NULL) { 314556f33205SJonathan Adams cur->ml_next = new; 314656f33205SJonathan Adams new->ml_prev = cur; 314756f33205SJonathan Adams new->ml_next = NULL; 31487c478bd9Sstevel@tonic-gate return; 31497c478bd9Sstevel@tonic-gate } 315056f33205SJonathan Adams memlistp = &cur->ml_next; 315156f33205SJonathan Adams cur = cur->ml_next; 31527c478bd9Sstevel@tonic-gate } 31537c478bd9Sstevel@tonic-gate } 31547c478bd9Sstevel@tonic-gate 31557c478bd9Sstevel@tonic-gate void 31567c478bd9Sstevel@tonic-gate kobj_vmem_init(vmem_t **text_arena, vmem_t **data_arena) 31577c478bd9Sstevel@tonic-gate { 31587c478bd9Sstevel@tonic-gate size_t tsize = e_modtext - modtext; 31597c478bd9Sstevel@tonic-gate size_t dsize = e_moddata - moddata; 31607c478bd9Sstevel@tonic-gate 31617c478bd9Sstevel@tonic-gate *text_arena = vmem_create("module_text", tsize ? modtext : NULL, tsize, 31627c478bd9Sstevel@tonic-gate 1, segkmem_alloc, segkmem_free, heaptext_arena, 0, VM_SLEEP); 31637c478bd9Sstevel@tonic-gate *data_arena = vmem_create("module_data", dsize ? moddata : NULL, dsize, 31647c478bd9Sstevel@tonic-gate 1, segkmem_alloc, segkmem_free, heap32_arena, 0, VM_SLEEP); 31657c478bd9Sstevel@tonic-gate } 31667c478bd9Sstevel@tonic-gate 31677c478bd9Sstevel@tonic-gate caddr_t 31687c478bd9Sstevel@tonic-gate kobj_text_alloc(vmem_t *arena, size_t size) 31697c478bd9Sstevel@tonic-gate { 31707c478bd9Sstevel@tonic-gate return (vmem_alloc(arena, size, VM_SLEEP | VM_BESTFIT)); 31717c478bd9Sstevel@tonic-gate } 31727c478bd9Sstevel@tonic-gate 31737c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 31747c478bd9Sstevel@tonic-gate caddr_t 31757c478bd9Sstevel@tonic-gate kobj_texthole_alloc(caddr_t addr, size_t size) 31767c478bd9Sstevel@tonic-gate { 31777c478bd9Sstevel@tonic-gate panic("unexpected call to kobj_texthole_alloc()"); 31787c478bd9Sstevel@tonic-gate /*NOTREACHED*/ 31797c478bd9Sstevel@tonic-gate return (0); 31807c478bd9Sstevel@tonic-gate } 31817c478bd9Sstevel@tonic-gate 31827c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 31837c478bd9Sstevel@tonic-gate void 31847c478bd9Sstevel@tonic-gate kobj_texthole_free(caddr_t addr, size_t size) 31857c478bd9Sstevel@tonic-gate { 31867c478bd9Sstevel@tonic-gate panic("unexpected call to kobj_texthole_free()"); 31877c478bd9Sstevel@tonic-gate } 31887c478bd9Sstevel@tonic-gate 31897c478bd9Sstevel@tonic-gate /* 31907c478bd9Sstevel@tonic-gate * This is called just after configure() in startup(). 31917c478bd9Sstevel@tonic-gate * 31927c478bd9Sstevel@tonic-gate * The ISALIST concept is a bit hopeless on Intel, because 31937c478bd9Sstevel@tonic-gate * there's no guarantee of an ever-more-capable processor 31947c478bd9Sstevel@tonic-gate * given that various parts of the instruction set may appear 31957c478bd9Sstevel@tonic-gate * and disappear between different implementations. 31967c478bd9Sstevel@tonic-gate * 31977c478bd9Sstevel@tonic-gate * While it would be possible to correct it and even enhance 31987c478bd9Sstevel@tonic-gate * it somewhat, the explicit hardware capability bitmask allows 31997c478bd9Sstevel@tonic-gate * more flexibility. 32007c478bd9Sstevel@tonic-gate * 32017c478bd9Sstevel@tonic-gate * So, we just leave this alone. 32027c478bd9Sstevel@tonic-gate */ 32037c478bd9Sstevel@tonic-gate void 32047c478bd9Sstevel@tonic-gate setx86isalist(void) 32057c478bd9Sstevel@tonic-gate { 32067c478bd9Sstevel@tonic-gate char *tp; 32077c478bd9Sstevel@tonic-gate size_t len; 32087c478bd9Sstevel@tonic-gate extern char *isa_list; 32097c478bd9Sstevel@tonic-gate 32107c478bd9Sstevel@tonic-gate #define TBUFSIZE 1024 32117c478bd9Sstevel@tonic-gate 32127c478bd9Sstevel@tonic-gate tp = kmem_alloc(TBUFSIZE, KM_SLEEP); 32137c478bd9Sstevel@tonic-gate *tp = '\0'; 32147c478bd9Sstevel@tonic-gate 32157c478bd9Sstevel@tonic-gate #if defined(__amd64) 32167c478bd9Sstevel@tonic-gate (void) strcpy(tp, "amd64 "); 32177c478bd9Sstevel@tonic-gate #endif 32187c478bd9Sstevel@tonic-gate 32197c478bd9Sstevel@tonic-gate switch (x86_vendor) { 32207c478bd9Sstevel@tonic-gate case X86_VENDOR_Intel: 32217c478bd9Sstevel@tonic-gate case X86_VENDOR_AMD: 32227c478bd9Sstevel@tonic-gate case X86_VENDOR_TM: 32237417cfdeSKuriakose Kuruvilla if (is_x86_feature(x86_featureset, X86FSET_CMOV)) { 32247c478bd9Sstevel@tonic-gate /* 32257c478bd9Sstevel@tonic-gate * Pentium Pro or later 32267c478bd9Sstevel@tonic-gate */ 32277c478bd9Sstevel@tonic-gate (void) strcat(tp, "pentium_pro"); 32287417cfdeSKuriakose Kuruvilla (void) strcat(tp, 32297417cfdeSKuriakose Kuruvilla is_x86_feature(x86_featureset, X86FSET_MMX) ? 32307c478bd9Sstevel@tonic-gate "+mmx pentium_pro " : " "); 32317c478bd9Sstevel@tonic-gate } 32327c478bd9Sstevel@tonic-gate /*FALLTHROUGH*/ 32337c478bd9Sstevel@tonic-gate case X86_VENDOR_Cyrix: 32347c478bd9Sstevel@tonic-gate /* 32357c478bd9Sstevel@tonic-gate * The Cyrix 6x86 does not have any Pentium features 32367c478bd9Sstevel@tonic-gate * accessible while not at privilege level 0. 32377c478bd9Sstevel@tonic-gate */ 32387417cfdeSKuriakose Kuruvilla if (is_x86_feature(x86_featureset, X86FSET_CPUID)) { 32397c478bd9Sstevel@tonic-gate (void) strcat(tp, "pentium"); 32407417cfdeSKuriakose Kuruvilla (void) strcat(tp, 32417417cfdeSKuriakose Kuruvilla is_x86_feature(x86_featureset, X86FSET_MMX) ? 32427c478bd9Sstevel@tonic-gate "+mmx pentium " : " "); 32437c478bd9Sstevel@tonic-gate } 32447c478bd9Sstevel@tonic-gate break; 32457c478bd9Sstevel@tonic-gate default: 32467c478bd9Sstevel@tonic-gate break; 32477c478bd9Sstevel@tonic-gate } 32487c478bd9Sstevel@tonic-gate (void) strcat(tp, "i486 i386 i86"); 32497c478bd9Sstevel@tonic-gate len = strlen(tp) + 1; /* account for NULL at end of string */ 32507c478bd9Sstevel@tonic-gate isa_list = strcpy(kmem_alloc(len, KM_SLEEP), tp); 32517c478bd9Sstevel@tonic-gate kmem_free(tp, TBUFSIZE); 32527c478bd9Sstevel@tonic-gate 32537c478bd9Sstevel@tonic-gate #undef TBUFSIZE 32547c478bd9Sstevel@tonic-gate } 32557c478bd9Sstevel@tonic-gate 32567c478bd9Sstevel@tonic-gate 32577c478bd9Sstevel@tonic-gate #ifdef __amd64 32587c478bd9Sstevel@tonic-gate 32597c478bd9Sstevel@tonic-gate void * 32607c478bd9Sstevel@tonic-gate device_arena_alloc(size_t size, int vm_flag) 32617c478bd9Sstevel@tonic-gate { 32627c478bd9Sstevel@tonic-gate return (vmem_alloc(device_arena, size, vm_flag)); 32637c478bd9Sstevel@tonic-gate } 32647c478bd9Sstevel@tonic-gate 32657c478bd9Sstevel@tonic-gate void 32667c478bd9Sstevel@tonic-gate device_arena_free(void *vaddr, size_t size) 32677c478bd9Sstevel@tonic-gate { 32687c478bd9Sstevel@tonic-gate vmem_free(device_arena, vaddr, size); 32697c478bd9Sstevel@tonic-gate } 32707c478bd9Sstevel@tonic-gate 3271ae115bc7Smrj #else /* __i386 */ 32727c478bd9Sstevel@tonic-gate 32737c478bd9Sstevel@tonic-gate void * 32747c478bd9Sstevel@tonic-gate device_arena_alloc(size_t size, int vm_flag) 32757c478bd9Sstevel@tonic-gate { 32767c478bd9Sstevel@tonic-gate caddr_t vaddr; 32777c478bd9Sstevel@tonic-gate uintptr_t v; 32787c478bd9Sstevel@tonic-gate size_t start; 32797c478bd9Sstevel@tonic-gate size_t end; 32807c478bd9Sstevel@tonic-gate 32817c478bd9Sstevel@tonic-gate vaddr = vmem_alloc(heap_arena, size, vm_flag); 32827c478bd9Sstevel@tonic-gate if (vaddr == NULL) 32837c478bd9Sstevel@tonic-gate return (NULL); 32847c478bd9Sstevel@tonic-gate 32857c478bd9Sstevel@tonic-gate v = (uintptr_t)vaddr; 32867c478bd9Sstevel@tonic-gate ASSERT(v >= kernelbase); 3287ae115bc7Smrj ASSERT(v + size <= valloc_base); 32887c478bd9Sstevel@tonic-gate 32897c478bd9Sstevel@tonic-gate start = btop(v - kernelbase); 32907c478bd9Sstevel@tonic-gate end = btop(v + size - 1 - kernelbase); 32917c478bd9Sstevel@tonic-gate ASSERT(start < toxic_bit_map_len); 32927c478bd9Sstevel@tonic-gate ASSERT(end < toxic_bit_map_len); 32937c478bd9Sstevel@tonic-gate 32947c478bd9Sstevel@tonic-gate while (start <= end) { 32957c478bd9Sstevel@tonic-gate BT_ATOMIC_SET(toxic_bit_map, start); 32967c478bd9Sstevel@tonic-gate ++start; 32977c478bd9Sstevel@tonic-gate } 32987c478bd9Sstevel@tonic-gate return (vaddr); 32997c478bd9Sstevel@tonic-gate } 33007c478bd9Sstevel@tonic-gate 33017c478bd9Sstevel@tonic-gate void 33027c478bd9Sstevel@tonic-gate device_arena_free(void *vaddr, size_t size) 33037c478bd9Sstevel@tonic-gate { 33047c478bd9Sstevel@tonic-gate uintptr_t v = (uintptr_t)vaddr; 33057c478bd9Sstevel@tonic-gate size_t start; 33067c478bd9Sstevel@tonic-gate size_t end; 33077c478bd9Sstevel@tonic-gate 33087c478bd9Sstevel@tonic-gate ASSERT(v >= kernelbase); 3309ae115bc7Smrj ASSERT(v + size <= valloc_base); 33107c478bd9Sstevel@tonic-gate 33117c478bd9Sstevel@tonic-gate start = btop(v - kernelbase); 33127c478bd9Sstevel@tonic-gate end = btop(v + size - 1 - kernelbase); 33137c478bd9Sstevel@tonic-gate ASSERT(start < toxic_bit_map_len); 33147c478bd9Sstevel@tonic-gate ASSERT(end < toxic_bit_map_len); 33157c478bd9Sstevel@tonic-gate 33167c478bd9Sstevel@tonic-gate while (start <= end) { 33177c478bd9Sstevel@tonic-gate ASSERT(BT_TEST(toxic_bit_map, start) != 0); 33187c478bd9Sstevel@tonic-gate BT_ATOMIC_CLEAR(toxic_bit_map, start); 33197c478bd9Sstevel@tonic-gate ++start; 33207c478bd9Sstevel@tonic-gate } 33217c478bd9Sstevel@tonic-gate vmem_free(heap_arena, vaddr, size); 33227c478bd9Sstevel@tonic-gate } 33237c478bd9Sstevel@tonic-gate 33247c478bd9Sstevel@tonic-gate /* 33257c478bd9Sstevel@tonic-gate * returns 1st address in range that is in device arena, or NULL 33267c478bd9Sstevel@tonic-gate * if len is not NULL it returns the length of the toxic range 33277c478bd9Sstevel@tonic-gate */ 33287c478bd9Sstevel@tonic-gate void * 33297c478bd9Sstevel@tonic-gate device_arena_contains(void *vaddr, size_t size, size_t *len) 33307c478bd9Sstevel@tonic-gate { 33317c478bd9Sstevel@tonic-gate uintptr_t v = (uintptr_t)vaddr; 33327c478bd9Sstevel@tonic-gate uintptr_t eaddr = v + size; 33337c478bd9Sstevel@tonic-gate size_t start; 33347c478bd9Sstevel@tonic-gate size_t end; 33357c478bd9Sstevel@tonic-gate 33367c478bd9Sstevel@tonic-gate /* 33377c478bd9Sstevel@tonic-gate * if called very early by kmdb, just return NULL 33387c478bd9Sstevel@tonic-gate */ 33397c478bd9Sstevel@tonic-gate if (toxic_bit_map == NULL) 33407c478bd9Sstevel@tonic-gate return (NULL); 33417c478bd9Sstevel@tonic-gate 33427c478bd9Sstevel@tonic-gate /* 33437c478bd9Sstevel@tonic-gate * First check if we're completely outside the bitmap range. 33447c478bd9Sstevel@tonic-gate */ 3345ae115bc7Smrj if (v >= valloc_base || eaddr < kernelbase) 33467c478bd9Sstevel@tonic-gate return (NULL); 33477c478bd9Sstevel@tonic-gate 33487c478bd9Sstevel@tonic-gate /* 33497c478bd9Sstevel@tonic-gate * Trim ends of search to look at only what the bitmap covers. 33507c478bd9Sstevel@tonic-gate */ 33517c478bd9Sstevel@tonic-gate if (v < kernelbase) 33527c478bd9Sstevel@tonic-gate v = kernelbase; 33537c478bd9Sstevel@tonic-gate start = btop(v - kernelbase); 33547c478bd9Sstevel@tonic-gate end = btop(eaddr - kernelbase); 33557c478bd9Sstevel@tonic-gate if (end >= toxic_bit_map_len) 33567c478bd9Sstevel@tonic-gate end = toxic_bit_map_len; 33577c478bd9Sstevel@tonic-gate 33587c478bd9Sstevel@tonic-gate if (bt_range(toxic_bit_map, &start, &end, end) == 0) 33597c478bd9Sstevel@tonic-gate return (NULL); 33607c478bd9Sstevel@tonic-gate 33617c478bd9Sstevel@tonic-gate v = kernelbase + ptob(start); 33627c478bd9Sstevel@tonic-gate if (len != NULL) 33637c478bd9Sstevel@tonic-gate *len = ptob(end - start); 33647c478bd9Sstevel@tonic-gate return ((void *)v); 33657c478bd9Sstevel@tonic-gate } 33667c478bd9Sstevel@tonic-gate 3367ae115bc7Smrj #endif /* __i386 */ 3368