17c478bd9Sstevel@tonic-gate /* 27c478bd9Sstevel@tonic-gate * CDDL HEADER START 37c478bd9Sstevel@tonic-gate * 47c478bd9Sstevel@tonic-gate * The contents of this file are subject to the terms of the 5100b72f4Sandrei * Common Development and Distribution License (the "License"). 6100b72f4Sandrei * You may not use this file except in compliance with the License. 77c478bd9Sstevel@tonic-gate * 87c478bd9Sstevel@tonic-gate * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 97c478bd9Sstevel@tonic-gate * or http://www.opensolaris.org/os/licensing. 107c478bd9Sstevel@tonic-gate * See the License for the specific language governing permissions 117c478bd9Sstevel@tonic-gate * and limitations under the License. 127c478bd9Sstevel@tonic-gate * 137c478bd9Sstevel@tonic-gate * When distributing Covered Code, include this CDDL HEADER in each 147c478bd9Sstevel@tonic-gate * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 157c478bd9Sstevel@tonic-gate * If applicable, add the following below this CDDL HEADER, with the 167c478bd9Sstevel@tonic-gate * fields enclosed by brackets "[]" replaced with your own identifying 177c478bd9Sstevel@tonic-gate * information: Portions Copyright [yyyy] [name of copyright owner] 187c478bd9Sstevel@tonic-gate * 197c478bd9Sstevel@tonic-gate * CDDL HEADER END 207c478bd9Sstevel@tonic-gate */ 217c478bd9Sstevel@tonic-gate /* 22*0db3240dSStephen Hanson * Copyright (c) 1993, 2010, Oracle and/or its affiliates. All rights reserved. 237c478bd9Sstevel@tonic-gate */ 24a3114836SGerry Liu /* 25a3114836SGerry Liu * Copyright (c) 2010, Intel Corporation. 26a3114836SGerry Liu * All rights reserved. 27a3114836SGerry Liu */ 287c478bd9Sstevel@tonic-gate 297c478bd9Sstevel@tonic-gate #include <sys/types.h> 307c478bd9Sstevel@tonic-gate #include <sys/t_lock.h> 317c478bd9Sstevel@tonic-gate #include <sys/param.h> 327c478bd9Sstevel@tonic-gate #include <sys/sysmacros.h> 337c478bd9Sstevel@tonic-gate #include <sys/signal.h> 347c478bd9Sstevel@tonic-gate #include <sys/systm.h> 357c478bd9Sstevel@tonic-gate #include <sys/user.h> 367c478bd9Sstevel@tonic-gate #include <sys/mman.h> 377c478bd9Sstevel@tonic-gate #include <sys/vm.h> 387c478bd9Sstevel@tonic-gate #include <sys/conf.h> 397c478bd9Sstevel@tonic-gate #include <sys/avintr.h> 407c478bd9Sstevel@tonic-gate #include <sys/autoconf.h> 417c478bd9Sstevel@tonic-gate #include <sys/disp.h> 427c478bd9Sstevel@tonic-gate #include <sys/class.h> 437c478bd9Sstevel@tonic-gate #include <sys/bitmap.h> 447c478bd9Sstevel@tonic-gate 457c478bd9Sstevel@tonic-gate #include <sys/privregs.h> 467c478bd9Sstevel@tonic-gate 477c478bd9Sstevel@tonic-gate #include <sys/proc.h> 487c478bd9Sstevel@tonic-gate #include <sys/buf.h> 497c478bd9Sstevel@tonic-gate #include <sys/kmem.h> 50ae115bc7Smrj #include <sys/mem.h> 517c478bd9Sstevel@tonic-gate #include <sys/kstat.h> 527c478bd9Sstevel@tonic-gate 537c478bd9Sstevel@tonic-gate #include <sys/reboot.h> 547c478bd9Sstevel@tonic-gate 557c478bd9Sstevel@tonic-gate #include <sys/cred.h> 567c478bd9Sstevel@tonic-gate #include <sys/vnode.h> 577c478bd9Sstevel@tonic-gate #include <sys/file.h> 587c478bd9Sstevel@tonic-gate 597c478bd9Sstevel@tonic-gate #include <sys/procfs.h> 607c478bd9Sstevel@tonic-gate 617c478bd9Sstevel@tonic-gate #include <sys/vfs.h> 627c478bd9Sstevel@tonic-gate #include <sys/cmn_err.h> 637c478bd9Sstevel@tonic-gate #include <sys/utsname.h> 647c478bd9Sstevel@tonic-gate #include <sys/debug.h> 657c478bd9Sstevel@tonic-gate #include <sys/kdi.h> 667c478bd9Sstevel@tonic-gate 677c478bd9Sstevel@tonic-gate #include <sys/dumphdr.h> 687c478bd9Sstevel@tonic-gate #include <sys/bootconf.h> 69a3114836SGerry Liu #include <sys/memlist_plat.h> 707c478bd9Sstevel@tonic-gate #include <sys/varargs.h> 717c478bd9Sstevel@tonic-gate #include <sys/promif.h> 72843e1988Sjohnlev #include <sys/modctl.h> 737c478bd9Sstevel@tonic-gate 747c478bd9Sstevel@tonic-gate #include <sys/sunddi.h> 757c478bd9Sstevel@tonic-gate #include <sys/sunndi.h> 767c478bd9Sstevel@tonic-gate #include <sys/ndi_impldefs.h> 777c478bd9Sstevel@tonic-gate #include <sys/ddidmareq.h> 787c478bd9Sstevel@tonic-gate #include <sys/psw.h> 797c478bd9Sstevel@tonic-gate #include <sys/regset.h> 807c478bd9Sstevel@tonic-gate #include <sys/clock.h> 817c478bd9Sstevel@tonic-gate #include <sys/pte.h> 827c478bd9Sstevel@tonic-gate #include <sys/tss.h> 837c478bd9Sstevel@tonic-gate #include <sys/stack.h> 847c478bd9Sstevel@tonic-gate #include <sys/trap.h> 857c478bd9Sstevel@tonic-gate #include <sys/fp.h> 861d03c31eSjohnlev #include <vm/kboot_mmu.h> 877c478bd9Sstevel@tonic-gate #include <vm/anon.h> 887c478bd9Sstevel@tonic-gate #include <vm/as.h> 897c478bd9Sstevel@tonic-gate #include <vm/page.h> 907c478bd9Sstevel@tonic-gate #include <vm/seg.h> 917c478bd9Sstevel@tonic-gate #include <vm/seg_dev.h> 927c478bd9Sstevel@tonic-gate #include <vm/seg_kmem.h> 937c478bd9Sstevel@tonic-gate #include <vm/seg_kpm.h> 947c478bd9Sstevel@tonic-gate #include <vm/seg_map.h> 957c478bd9Sstevel@tonic-gate #include <vm/seg_vn.h> 967c478bd9Sstevel@tonic-gate #include <vm/seg_kp.h> 977c478bd9Sstevel@tonic-gate #include <sys/memnode.h> 987c478bd9Sstevel@tonic-gate #include <vm/vm_dep.h> 997c478bd9Sstevel@tonic-gate #include <sys/thread.h> 1007c478bd9Sstevel@tonic-gate #include <sys/sysconf.h> 1017c478bd9Sstevel@tonic-gate #include <sys/vm_machparam.h> 1027c478bd9Sstevel@tonic-gate #include <sys/archsystm.h> 1037c478bd9Sstevel@tonic-gate #include <sys/machsystm.h> 1047c478bd9Sstevel@tonic-gate #include <vm/hat.h> 1057c478bd9Sstevel@tonic-gate #include <vm/hat_i86.h> 1067c478bd9Sstevel@tonic-gate #include <sys/pmem.h> 1077c478bd9Sstevel@tonic-gate #include <sys/smp_impldefs.h> 1087c478bd9Sstevel@tonic-gate #include <sys/x86_archext.h> 109a563a037Sbholler #include <sys/cpuvar.h> 1107c478bd9Sstevel@tonic-gate #include <sys/segments.h> 1117c478bd9Sstevel@tonic-gate #include <sys/clconf.h> 1127c478bd9Sstevel@tonic-gate #include <sys/kobj.h> 1137c478bd9Sstevel@tonic-gate #include <sys/kobj_lex.h> 1147c478bd9Sstevel@tonic-gate #include <sys/cpc_impl.h> 1157aec1d6eScindi #include <sys/cpu_module.h> 11684ab085aSmws #include <sys/smbios.h> 117ae115bc7Smrj #include <sys/debug_info.h> 1181d03c31eSjohnlev #include <sys/bootinfo.h> 119dd4eeefdSeota #include <sys/ddi_timer.h> 1205679c89fSjv #include <sys/systeminfo.h> 12119397407SSherry Moore #include <sys/multiboot.h> 122ae115bc7Smrj 123a3114836SGerry Liu #ifdef __xpv 1241d03c31eSjohnlev 125843e1988Sjohnlev #include <sys/hypervisor.h> 126843e1988Sjohnlev #include <sys/xen_mmu.h> 127843e1988Sjohnlev #include <sys/evtchn_impl.h> 128843e1988Sjohnlev #include <sys/gnttab.h> 129843e1988Sjohnlev #include <sys/xpv_panic.h> 130843e1988Sjohnlev #include <xen/sys/xenbus_comms.h> 131843e1988Sjohnlev #include <xen/public/physdev.h> 1321d03c31eSjohnlev 133843e1988Sjohnlev extern void xen_late_startup(void); 134ae115bc7Smrj 1351d03c31eSjohnlev struct xen_evt_data cpu0_evt_data; 1361d03c31eSjohnlev 137a3114836SGerry Liu #else /* __xpv */ 138a3114836SGerry Liu #include <sys/memlist_impl.h> 139a3114836SGerry Liu 140a3114836SGerry Liu extern void mem_config_init(void); 1411d03c31eSjohnlev #endif /* __xpv */ 1427c478bd9Sstevel@tonic-gate 1437c478bd9Sstevel@tonic-gate extern void progressbar_init(void); 1449acbbeafSnn extern void brand_init(void); 14506fb6a36Sdv extern void pcf_init(void); 1460e751525SEric Saxe extern void pg_init(void); 1477c478bd9Sstevel@tonic-gate 148d7d93655Sblakej extern int size_pse_array(pgcnt_t, int); 149d7d93655Sblakej 150ed5289f9SKen Erickson #if defined(_SOFT_HOSTID) 151ed5289f9SKen Erickson 152ed5289f9SKen Erickson #include <sys/rtc.h> 153ed5289f9SKen Erickson 154ed5289f9SKen Erickson static int32_t set_soft_hostid(void); 155ed5289f9SKen Erickson static char hostid_file[] = "/etc/hostid"; 156ed5289f9SKen Erickson 157ed5289f9SKen Erickson #endif 158ed5289f9SKen Erickson 15994f1124eSVikram Hegde void *gfx_devinfo_list; 160ed5289f9SKen Erickson 1613a634bfcSVikram Hegde #if defined(__amd64) && !defined(__xpv) 1623a634bfcSVikram Hegde extern void immu_startup(void); 1633a634bfcSVikram Hegde #endif 1643a634bfcSVikram Hegde 1657c478bd9Sstevel@tonic-gate /* 1667c478bd9Sstevel@tonic-gate * XXX make declaration below "static" when drivers no longer use this 1677c478bd9Sstevel@tonic-gate * interface. 1687c478bd9Sstevel@tonic-gate */ 1697c478bd9Sstevel@tonic-gate extern caddr_t p0_va; /* Virtual address for accessing physical page 0 */ 1707c478bd9Sstevel@tonic-gate 1717c478bd9Sstevel@tonic-gate /* 1727c478bd9Sstevel@tonic-gate * segkp 1737c478bd9Sstevel@tonic-gate */ 1747c478bd9Sstevel@tonic-gate extern int segkp_fromheap; 1757c478bd9Sstevel@tonic-gate 1767c478bd9Sstevel@tonic-gate static void kvm_init(void); 1777c478bd9Sstevel@tonic-gate static void startup_init(void); 1787c478bd9Sstevel@tonic-gate static void startup_memlist(void); 179ae115bc7Smrj static void startup_kmem(void); 1807c478bd9Sstevel@tonic-gate static void startup_modules(void); 1817c478bd9Sstevel@tonic-gate static void startup_vm(void); 1827c478bd9Sstevel@tonic-gate static void startup_end(void); 18335b1ab99Sjosephb static void layout_kernel_va(void); 1847c478bd9Sstevel@tonic-gate 1857c478bd9Sstevel@tonic-gate /* 1867c478bd9Sstevel@tonic-gate * Declare these as initialized data so we can patch them. 1877c478bd9Sstevel@tonic-gate */ 188f53ad214Skchow #ifdef __i386 18935b1ab99Sjosephb 190f53ad214Skchow /* 191f53ad214Skchow * Due to virtual address space limitations running in 32 bit mode, restrict 19235b1ab99Sjosephb * the amount of physical memory configured to a max of PHYSMEM pages (16g). 193f53ad214Skchow * 194f53ad214Skchow * If the physical max memory size of 64g were allowed to be configured, the 195f53ad214Skchow * size of user virtual address space will be less than 1g. A limited user 196f53ad214Skchow * address space greatly reduces the range of applications that can run. 197f53ad214Skchow * 19835b1ab99Sjosephb * If more physical memory than PHYSMEM is required, users should preferably 19935b1ab99Sjosephb * run in 64 bit mode which has far looser virtual address space limitations. 200f53ad214Skchow * 201f53ad214Skchow * If 64 bit mode is not available (as in IA32) and/or more physical memory 20235b1ab99Sjosephb * than PHYSMEM is required in 32 bit mode, physmem can be set to the desired 203f53ad214Skchow * value or to 0 (to configure all available memory) via eeprom(1M). kernelbase 204f53ad214Skchow * should also be carefully tuned to balance out the need of the user 205f53ad214Skchow * application while minimizing the risk of kernel heap exhaustion due to 206f53ad214Skchow * kernelbase being set too high. 207f53ad214Skchow */ 20835b1ab99Sjosephb #define PHYSMEM 0x400000 209f53ad214Skchow 21035b1ab99Sjosephb #else /* __amd64 */ 21135b1ab99Sjosephb 21235b1ab99Sjosephb /* 21335b1ab99Sjosephb * For now we can handle memory with physical addresses up to about 21435b1ab99Sjosephb * 64 Terabytes. This keeps the kernel above the VA hole, leaving roughly 21535b1ab99Sjosephb * half the VA space for seg_kpm. When systems get bigger than 64TB this 21635b1ab99Sjosephb * code will need revisiting. There is an implicit assumption that there 21735b1ab99Sjosephb * are no *huge* holes in the physical address space too. 21835b1ab99Sjosephb */ 21935b1ab99Sjosephb #define TERABYTE (1ul << 40) 22035b1ab99Sjosephb #define PHYSMEM_MAX64 mmu_btop(64 * TERABYTE) 22135b1ab99Sjosephb #define PHYSMEM PHYSMEM_MAX64 22235b1ab99Sjosephb #define AMD64_VA_HOLE_END 0xFFFF800000000000ul 22335b1ab99Sjosephb 22435b1ab99Sjosephb #endif /* __amd64 */ 22535b1ab99Sjosephb 22635b1ab99Sjosephb pgcnt_t physmem = PHYSMEM; 2277c478bd9Sstevel@tonic-gate pgcnt_t obp_pages; /* Memory used by PROM for its text and data */ 2287c478bd9Sstevel@tonic-gate 2297c478bd9Sstevel@tonic-gate char *kobj_file_buf; 2307c478bd9Sstevel@tonic-gate int kobj_file_bufsize; /* set in /etc/system */ 2317c478bd9Sstevel@tonic-gate 2327c478bd9Sstevel@tonic-gate /* Global variables for MP support. Used in mp_startup */ 23319397407SSherry Moore caddr_t rm_platter_va = 0; 2347c478bd9Sstevel@tonic-gate uint32_t rm_platter_pa; 2357c478bd9Sstevel@tonic-gate 23683f9b804Skchow int auto_lpg_disable = 1; 23783f9b804Skchow 2387c478bd9Sstevel@tonic-gate /* 2397c478bd9Sstevel@tonic-gate * Some CPUs have holes in the middle of the 64-bit virtual address range. 2407c478bd9Sstevel@tonic-gate */ 2417c478bd9Sstevel@tonic-gate uintptr_t hole_start, hole_end; 2427c478bd9Sstevel@tonic-gate 2437c478bd9Sstevel@tonic-gate /* 2447c478bd9Sstevel@tonic-gate * kpm mapping window 2457c478bd9Sstevel@tonic-gate */ 2467c478bd9Sstevel@tonic-gate caddr_t kpm_vbase; 2477c478bd9Sstevel@tonic-gate size_t kpm_size; 24835b1ab99Sjosephb static int kpm_desired; 24935b1ab99Sjosephb #ifdef __amd64 25035b1ab99Sjosephb static uintptr_t segkpm_base = (uintptr_t)SEGKPM_BASE; 25135b1ab99Sjosephb #endif 2527c478bd9Sstevel@tonic-gate 2537c478bd9Sstevel@tonic-gate /* 2547c478bd9Sstevel@tonic-gate * Configuration parameters set at boot time. 2557c478bd9Sstevel@tonic-gate */ 2567c478bd9Sstevel@tonic-gate 2577c478bd9Sstevel@tonic-gate caddr_t econtig; /* end of first block of contiguous kernel */ 2587c478bd9Sstevel@tonic-gate 2597c478bd9Sstevel@tonic-gate struct bootops *bootops = 0; /* passed in from boot */ 2607c478bd9Sstevel@tonic-gate struct bootops **bootopsp; 2617c478bd9Sstevel@tonic-gate struct boot_syscalls *sysp; /* passed in from boot */ 2627c478bd9Sstevel@tonic-gate 2637c478bd9Sstevel@tonic-gate char bootblock_fstype[16]; 2647c478bd9Sstevel@tonic-gate 2657c478bd9Sstevel@tonic-gate char kern_bootargs[OBP_MAXPATHLEN]; 26619397407SSherry Moore char kern_bootfile[OBP_MAXPATHLEN]; 2677c478bd9Sstevel@tonic-gate 268ad23a2dbSjohansen /* 269ad23a2dbSjohansen * ZFS zio segment. This allows us to exclude large portions of ZFS data that 270ad23a2dbSjohansen * gets cached in kmem caches on the heap. If this is set to zero, we allocate 271ad23a2dbSjohansen * zio buffers from their own segment, otherwise they are allocated from the 272ad23a2dbSjohansen * heap. The optimization of allocating zio buffers from their own segment is 273ad23a2dbSjohansen * only valid on 64-bit kernels. 274ad23a2dbSjohansen */ 275ad23a2dbSjohansen #if defined(__amd64) 276ad23a2dbSjohansen int segzio_fromheap = 0; 277ad23a2dbSjohansen #else 278ad23a2dbSjohansen int segzio_fromheap = 1; 279ad23a2dbSjohansen #endif 280ad23a2dbSjohansen 2817c478bd9Sstevel@tonic-gate /* 2827c478bd9Sstevel@tonic-gate * new memory fragmentations are possible in startup() due to BOP_ALLOCs. this 2837c478bd9Sstevel@tonic-gate * depends on number of BOP_ALLOC calls made and requested size, memory size 2847c478bd9Sstevel@tonic-gate * combination and whether boot.bin memory needs to be freed. 2857c478bd9Sstevel@tonic-gate */ 2867c478bd9Sstevel@tonic-gate #define POSS_NEW_FRAGMENTS 12 2877c478bd9Sstevel@tonic-gate 2887c478bd9Sstevel@tonic-gate /* 2897c478bd9Sstevel@tonic-gate * VM data structures 2907c478bd9Sstevel@tonic-gate */ 2917c478bd9Sstevel@tonic-gate long page_hashsz; /* Size of page hash table (power of two) */ 2927c478bd9Sstevel@tonic-gate struct page *pp_base; /* Base of initial system page struct array */ 2937c478bd9Sstevel@tonic-gate struct page **page_hash; /* Page hash table */ 294d7d93655Sblakej pad_mutex_t *pse_mutex; /* Locks protecting pp->p_selock */ 295d7d93655Sblakej size_t pse_table_size; /* Number of mutexes in pse_mutex[] */ 296d7d93655Sblakej int pse_shift; /* log2(pse_table_size) */ 2977c478bd9Sstevel@tonic-gate struct seg ktextseg; /* Segment used for kernel executable image */ 2987c478bd9Sstevel@tonic-gate struct seg kvalloc; /* Segment used for "valloc" mapping */ 2997c478bd9Sstevel@tonic-gate struct seg kpseg; /* Segment used for pageable kernel virt mem */ 3007c478bd9Sstevel@tonic-gate struct seg kmapseg; /* Segment used for generic kernel mappings */ 3017c478bd9Sstevel@tonic-gate struct seg kdebugseg; /* Segment used for the kernel debugger */ 3027c478bd9Sstevel@tonic-gate 3037c478bd9Sstevel@tonic-gate struct seg *segkmap = &kmapseg; /* Kernel generic mapping segment */ 304ae115bc7Smrj static struct seg *segmap = &kmapseg; /* easier to use name for in here */ 305ae115bc7Smrj 3067c478bd9Sstevel@tonic-gate struct seg *segkp = &kpseg; /* Pageable kernel virtual memory segment */ 3077c478bd9Sstevel@tonic-gate 3087c478bd9Sstevel@tonic-gate #if defined(__amd64) 3097c478bd9Sstevel@tonic-gate struct seg kvseg_core; /* Segment used for the core heap */ 3107c478bd9Sstevel@tonic-gate struct seg kpmseg; /* Segment used for physical mapping */ 3117c478bd9Sstevel@tonic-gate struct seg *segkpm = &kpmseg; /* 64bit kernel physical mapping segment */ 3127c478bd9Sstevel@tonic-gate #else 3137c478bd9Sstevel@tonic-gate struct seg *segkpm = NULL; /* Unused on IA32 */ 3147c478bd9Sstevel@tonic-gate #endif 3157c478bd9Sstevel@tonic-gate 3167c478bd9Sstevel@tonic-gate caddr_t segkp_base; /* Base address of segkp */ 317ad23a2dbSjohansen caddr_t segzio_base; /* Base address of segzio */ 3187c478bd9Sstevel@tonic-gate #if defined(__amd64) 3197c478bd9Sstevel@tonic-gate pgcnt_t segkpsize = btop(SEGKPDEFSIZE); /* size of segkp segment in pages */ 3207c478bd9Sstevel@tonic-gate #else 3217c478bd9Sstevel@tonic-gate pgcnt_t segkpsize = 0; 3227c478bd9Sstevel@tonic-gate #endif 323ad23a2dbSjohansen pgcnt_t segziosize = 0; /* size of zio segment in pages */ 3247c478bd9Sstevel@tonic-gate 325a3114836SGerry Liu /* 326a3114836SGerry Liu * A static DR page_t VA map is reserved that can map the page structures 327a3114836SGerry Liu * for a domain's entire RA space. The pages that back this space are 328a3114836SGerry Liu * dynamically allocated and need not be physically contiguous. The DR 329a3114836SGerry Liu * map size is derived from KPM size. 330a3114836SGerry Liu * This mechanism isn't used by x86 yet, so just stubs here. 331a3114836SGerry Liu */ 332a3114836SGerry Liu int ppvm_enable = 0; /* Static virtual map for page structs */ 333a3114836SGerry Liu page_t *ppvm_base = NULL; /* Base of page struct map */ 334a3114836SGerry Liu pgcnt_t ppvm_size = 0; /* Size of page struct map */ 335a3114836SGerry Liu 336ae115bc7Smrj /* 337ae115bc7Smrj * VA range available to the debugger 338ae115bc7Smrj */ 339ae115bc7Smrj const caddr_t kdi_segdebugbase = (const caddr_t)SEGDEBUGBASE; 340ae115bc7Smrj const size_t kdi_segdebugsize = SEGDEBUGSIZE; 341ae115bc7Smrj 3427c478bd9Sstevel@tonic-gate struct memseg *memseg_base; 3437c478bd9Sstevel@tonic-gate struct vnode unused_pages_vp; 3447c478bd9Sstevel@tonic-gate 3457c478bd9Sstevel@tonic-gate #define FOURGB 0x100000000LL 3467c478bd9Sstevel@tonic-gate 3477c478bd9Sstevel@tonic-gate struct memlist *memlist; 3487c478bd9Sstevel@tonic-gate 3497c478bd9Sstevel@tonic-gate caddr_t s_text; /* start of kernel text segment */ 3507c478bd9Sstevel@tonic-gate caddr_t e_text; /* end of kernel text segment */ 3517c478bd9Sstevel@tonic-gate caddr_t s_data; /* start of kernel data segment */ 3527c478bd9Sstevel@tonic-gate caddr_t e_data; /* end of kernel data segment */ 3537c478bd9Sstevel@tonic-gate caddr_t modtext; /* start of loadable module text reserved */ 3547c478bd9Sstevel@tonic-gate caddr_t e_modtext; /* end of loadable module text reserved */ 3557c478bd9Sstevel@tonic-gate caddr_t moddata; /* start of loadable module data reserved */ 3567c478bd9Sstevel@tonic-gate caddr_t e_moddata; /* end of loadable module data reserved */ 3577c478bd9Sstevel@tonic-gate 3587c478bd9Sstevel@tonic-gate struct memlist *phys_install; /* Total installed physical memory */ 3597c478bd9Sstevel@tonic-gate struct memlist *phys_avail; /* Total available physical memory */ 3601de082f7SVikram Hegde struct memlist *bios_rsvd; /* Bios reserved memory */ 3617c478bd9Sstevel@tonic-gate 3627c478bd9Sstevel@tonic-gate /* 3637c478bd9Sstevel@tonic-gate * kphysm_init returns the number of pages that were processed 3647c478bd9Sstevel@tonic-gate */ 365ae115bc7Smrj static pgcnt_t kphysm_init(page_t *, pgcnt_t); 3667c478bd9Sstevel@tonic-gate 3677c478bd9Sstevel@tonic-gate #define IO_PROP_SIZE 64 /* device property size */ 3687c478bd9Sstevel@tonic-gate 3697c478bd9Sstevel@tonic-gate /* 3707c478bd9Sstevel@tonic-gate * a couple useful roundup macros 3717c478bd9Sstevel@tonic-gate */ 3727c478bd9Sstevel@tonic-gate #define ROUND_UP_PAGE(x) \ 3737c478bd9Sstevel@tonic-gate ((uintptr_t)P2ROUNDUP((uintptr_t)(x), (uintptr_t)MMU_PAGESIZE)) 3747c478bd9Sstevel@tonic-gate #define ROUND_UP_LPAGE(x) \ 3757c478bd9Sstevel@tonic-gate ((uintptr_t)P2ROUNDUP((uintptr_t)(x), mmu.level_size[1])) 3767c478bd9Sstevel@tonic-gate #define ROUND_UP_4MEG(x) \ 377ae115bc7Smrj ((uintptr_t)P2ROUNDUP((uintptr_t)(x), (uintptr_t)FOUR_MEG)) 3787c478bd9Sstevel@tonic-gate #define ROUND_UP_TOPLEVEL(x) \ 3797c478bd9Sstevel@tonic-gate ((uintptr_t)P2ROUNDUP((uintptr_t)(x), mmu.level_size[mmu.max_level])) 3807c478bd9Sstevel@tonic-gate 3817c478bd9Sstevel@tonic-gate /* 3827c478bd9Sstevel@tonic-gate * 32-bit Kernel's Virtual memory layout. 3837c478bd9Sstevel@tonic-gate * +-----------------------+ 384ae115bc7Smrj * | | 3857c478bd9Sstevel@tonic-gate * 0xFFC00000 -|-----------------------|- ARGSBASE 3867c478bd9Sstevel@tonic-gate * | debugger | 3877c478bd9Sstevel@tonic-gate * 0xFF800000 -|-----------------------|- SEGDEBUGBASE 3887c478bd9Sstevel@tonic-gate * | Kernel Data | 3897c478bd9Sstevel@tonic-gate * 0xFEC00000 -|-----------------------| 3907c478bd9Sstevel@tonic-gate * | Kernel Text | 391843e1988Sjohnlev * 0xFE800000 -|-----------------------|- KERNEL_TEXT (0xFB400000 on Xen) 392ae115bc7Smrj * |--- GDT ---|- GDT page (GDT_VA) 393ae115bc7Smrj * |--- debug info ---|- debug info (DEBUG_INFO_VA) 394ae115bc7Smrj * | | 395ae115bc7Smrj * | page_t structures | 3967c478bd9Sstevel@tonic-gate * | memsegs, memlists, | 3977c478bd9Sstevel@tonic-gate * | page hash, etc. | 398ae115bc7Smrj * --- -|-----------------------|- ekernelheap, valloc_base (floating) 399ae115bc7Smrj * | | (segkp is just an arena in the heap) 4007c478bd9Sstevel@tonic-gate * | | 4017c478bd9Sstevel@tonic-gate * | kvseg | 4027c478bd9Sstevel@tonic-gate * | | 4037c478bd9Sstevel@tonic-gate * | | 4047c478bd9Sstevel@tonic-gate * --- -|-----------------------|- kernelheap (floating) 4057c478bd9Sstevel@tonic-gate * | Segkmap | 406ae115bc7Smrj * 0xC3002000 -|-----------------------|- segmap_start (floating) 4077c478bd9Sstevel@tonic-gate * | Red Zone | 4087c478bd9Sstevel@tonic-gate * 0xC3000000 -|-----------------------|- kernelbase / userlimit (floating) 4097c478bd9Sstevel@tonic-gate * | | || 4107c478bd9Sstevel@tonic-gate * | Shared objects | \/ 4117c478bd9Sstevel@tonic-gate * | | 4127c478bd9Sstevel@tonic-gate * : : 4137c478bd9Sstevel@tonic-gate * | user data | 4147c478bd9Sstevel@tonic-gate * |-----------------------| 4157c478bd9Sstevel@tonic-gate * | user text | 4167c478bd9Sstevel@tonic-gate * 0x08048000 -|-----------------------| 4177c478bd9Sstevel@tonic-gate * | user stack | 4187c478bd9Sstevel@tonic-gate * : : 4197c478bd9Sstevel@tonic-gate * | invalid | 4207c478bd9Sstevel@tonic-gate * 0x00000000 +-----------------------+ 4217c478bd9Sstevel@tonic-gate * 4227c478bd9Sstevel@tonic-gate * 4237c478bd9Sstevel@tonic-gate * 64-bit Kernel's Virtual memory layout. (assuming 64 bit app) 4247c478bd9Sstevel@tonic-gate * +-----------------------+ 425ae115bc7Smrj * | | 4267c478bd9Sstevel@tonic-gate * 0xFFFFFFFF.FFC00000 |-----------------------|- ARGSBASE 4277c478bd9Sstevel@tonic-gate * | debugger (?) | 4287c478bd9Sstevel@tonic-gate * 0xFFFFFFFF.FF800000 |-----------------------|- SEGDEBUGBASE 4297c478bd9Sstevel@tonic-gate * | unused | 4307c478bd9Sstevel@tonic-gate * +-----------------------+ 4317c478bd9Sstevel@tonic-gate * | Kernel Data | 4327c478bd9Sstevel@tonic-gate * 0xFFFFFFFF.FBC00000 |-----------------------| 4337c478bd9Sstevel@tonic-gate * | Kernel Text | 4347c478bd9Sstevel@tonic-gate * 0xFFFFFFFF.FB800000 |-----------------------|- KERNEL_TEXT 435ae115bc7Smrj * |--- GDT ---|- GDT page (GDT_VA) 436ae115bc7Smrj * |--- debug info ---|- debug info (DEBUG_INFO_VA) 437ae115bc7Smrj * | | 4387c478bd9Sstevel@tonic-gate * | Core heap | (used for loadable modules) 4397c478bd9Sstevel@tonic-gate * 0xFFFFFFFF.C0000000 |-----------------------|- core_base / ekernelheap 4407c478bd9Sstevel@tonic-gate * | Kernel | 4417c478bd9Sstevel@tonic-gate * | heap | 4427c478bd9Sstevel@tonic-gate * 0xFFFFFXXX.XXX00000 |-----------------------|- kernelheap (floating) 443ae115bc7Smrj * | segmap | 444ae115bc7Smrj * 0xFFFFFXXX.XXX00000 |-----------------------|- segmap_start (floating) 4457c478bd9Sstevel@tonic-gate * | device mappings | 4467c478bd9Sstevel@tonic-gate * 0xFFFFFXXX.XXX00000 |-----------------------|- toxic_addr (floating) 447ad23a2dbSjohansen * | segzio | 448ad23a2dbSjohansen * 0xFFFFFXXX.XXX00000 |-----------------------|- segzio_base (floating) 4497c478bd9Sstevel@tonic-gate * | segkp | 450ae115bc7Smrj * --- |-----------------------|- segkp_base (floating) 451ae115bc7Smrj * | page_t structures | valloc_base + valloc_sz 452ae115bc7Smrj * | memsegs, memlists, | 453ae115bc7Smrj * | page hash, etc. | 45435b1ab99Sjosephb * 0xFFFFFF00.00000000 |-----------------------|- valloc_base (lower if > 1TB) 4557c478bd9Sstevel@tonic-gate * | segkpm | 4567c478bd9Sstevel@tonic-gate * 0xFFFFFE00.00000000 |-----------------------| 4577c478bd9Sstevel@tonic-gate * | Red Zone | 45835b1ab99Sjosephb * 0xFFFFFD80.00000000 |-----------------------|- KERNELBASE (lower if > 1TB) 4597c478bd9Sstevel@tonic-gate * | User stack |- User space memory 4607c478bd9Sstevel@tonic-gate * | | 4617c478bd9Sstevel@tonic-gate * | shared objects, etc | (grows downwards) 4627c478bd9Sstevel@tonic-gate * : : 4637c478bd9Sstevel@tonic-gate * | | 4647c478bd9Sstevel@tonic-gate * 0xFFFF8000.00000000 |-----------------------| 4657c478bd9Sstevel@tonic-gate * | | 4667c478bd9Sstevel@tonic-gate * | VA Hole / unused | 4677c478bd9Sstevel@tonic-gate * | | 4687c478bd9Sstevel@tonic-gate * 0x00008000.00000000 |-----------------------| 4697c478bd9Sstevel@tonic-gate * | | 4707c478bd9Sstevel@tonic-gate * | | 4717c478bd9Sstevel@tonic-gate * : : 4727c478bd9Sstevel@tonic-gate * | user heap | (grows upwards) 4737c478bd9Sstevel@tonic-gate * | | 4747c478bd9Sstevel@tonic-gate * | user data | 4757c478bd9Sstevel@tonic-gate * |-----------------------| 4767c478bd9Sstevel@tonic-gate * | user text | 4777c478bd9Sstevel@tonic-gate * 0x00000000.04000000 |-----------------------| 4787c478bd9Sstevel@tonic-gate * | invalid | 4797c478bd9Sstevel@tonic-gate * 0x00000000.00000000 +-----------------------+ 4807c478bd9Sstevel@tonic-gate * 4817c478bd9Sstevel@tonic-gate * A 32 bit app on the 64 bit kernel sees the same layout as on the 32 bit 4827c478bd9Sstevel@tonic-gate * kernel, except that userlimit is raised to 0xfe000000 4837c478bd9Sstevel@tonic-gate * 4847c478bd9Sstevel@tonic-gate * Floating values: 4857c478bd9Sstevel@tonic-gate * 4867c478bd9Sstevel@tonic-gate * valloc_base: start of the kernel's memory management/tracking data 487ae115bc7Smrj * structures. This region contains page_t structures for 488ae115bc7Smrj * physical memory, memsegs, memlists, and the page hash. 4897c478bd9Sstevel@tonic-gate * 4907c478bd9Sstevel@tonic-gate * core_base: start of the kernel's "core" heap area on 64-bit systems. 4917c478bd9Sstevel@tonic-gate * This area is intended to be used for global data as well as for module 4927c478bd9Sstevel@tonic-gate * text/data that does not fit into the nucleus pages. The core heap is 4937c478bd9Sstevel@tonic-gate * restricted to a 2GB range, allowing every address within it to be 4947c478bd9Sstevel@tonic-gate * accessed using rip-relative addressing 4957c478bd9Sstevel@tonic-gate * 4967c478bd9Sstevel@tonic-gate * ekernelheap: end of kernelheap and start of segmap. 4977c478bd9Sstevel@tonic-gate * 4987c478bd9Sstevel@tonic-gate * kernelheap: start of kernel heap. On 32-bit systems, this starts right 4997c478bd9Sstevel@tonic-gate * above a red zone that separates the user's address space from the 5007c478bd9Sstevel@tonic-gate * kernel's. On 64-bit systems, it sits above segkp and segkpm. 5017c478bd9Sstevel@tonic-gate * 502ae115bc7Smrj * segmap_start: start of segmap. The length of segmap can be modified 50352eef812SAmrita Sadhukhan * through eeprom. The default length is 16MB on 32-bit systems and 64MB 50452eef812SAmrita Sadhukhan * on 64-bit systems. 5057c478bd9Sstevel@tonic-gate * 5067c478bd9Sstevel@tonic-gate * kernelbase: On a 32-bit kernel the default value of 0xd4000000 will be 5077c478bd9Sstevel@tonic-gate * decreased by 2X the size required for page_t. This allows the kernel 5087c478bd9Sstevel@tonic-gate * heap to grow in size with physical memory. With sizeof(page_t) == 80 5097c478bd9Sstevel@tonic-gate * bytes, the following shows the values of kernelbase and kernel heap 5107c478bd9Sstevel@tonic-gate * sizes for different memory configurations (assuming default segmap and 5117c478bd9Sstevel@tonic-gate * segkp sizes). 5127c478bd9Sstevel@tonic-gate * 5137c478bd9Sstevel@tonic-gate * mem size for kernelbase kernel heap 5147c478bd9Sstevel@tonic-gate * size page_t's size 5157c478bd9Sstevel@tonic-gate * ---- --------- ---------- ----------- 5167c478bd9Sstevel@tonic-gate * 1gb 0x01400000 0xd1800000 684MB 5177c478bd9Sstevel@tonic-gate * 2gb 0x02800000 0xcf000000 704MB 5187c478bd9Sstevel@tonic-gate * 4gb 0x05000000 0xca000000 744MB 5197c478bd9Sstevel@tonic-gate * 6gb 0x07800000 0xc5000000 784MB 5207c478bd9Sstevel@tonic-gate * 8gb 0x0a000000 0xc0000000 824MB 5217c478bd9Sstevel@tonic-gate * 16gb 0x14000000 0xac000000 984MB 5227c478bd9Sstevel@tonic-gate * 32gb 0x28000000 0x84000000 1304MB 5237c478bd9Sstevel@tonic-gate * 64gb 0x50000000 0x34000000 1944MB (*) 5247c478bd9Sstevel@tonic-gate * 5257c478bd9Sstevel@tonic-gate * kernelbase is less than the abi minimum of 0xc0000000 for memory 5267c478bd9Sstevel@tonic-gate * configurations above 8gb. 5277c478bd9Sstevel@tonic-gate * 5287c478bd9Sstevel@tonic-gate * (*) support for memory configurations above 32gb will require manual tuning 5297c478bd9Sstevel@tonic-gate * of kernelbase to balance out the need of user applications. 5307c478bd9Sstevel@tonic-gate */ 5317c478bd9Sstevel@tonic-gate 5327c478bd9Sstevel@tonic-gate /* real-time-clock initialization parameters */ 533ae115bc7Smrj extern time_t process_rtc_config_file(void); 5347c478bd9Sstevel@tonic-gate 5357c478bd9Sstevel@tonic-gate uintptr_t kernelbase; 536ae115bc7Smrj uintptr_t postbootkernelbase; /* not set till boot loader is gone */ 5377c478bd9Sstevel@tonic-gate uintptr_t eprom_kernelbase; 5387c478bd9Sstevel@tonic-gate size_t segmapsize; 539ae115bc7Smrj uintptr_t segmap_start; 5407c478bd9Sstevel@tonic-gate int segmapfreelists; 5417c478bd9Sstevel@tonic-gate pgcnt_t npages; 542ae115bc7Smrj pgcnt_t orig_npages; 5437c478bd9Sstevel@tonic-gate size_t core_size; /* size of "core" heap */ 5447c478bd9Sstevel@tonic-gate uintptr_t core_base; /* base address of "core" heap */ 5457c478bd9Sstevel@tonic-gate 5467c478bd9Sstevel@tonic-gate /* 5477c478bd9Sstevel@tonic-gate * List of bootstrap pages. We mark these as allocated in startup. 5487c478bd9Sstevel@tonic-gate * release_bootstrap() will free them when we're completely done with 5497c478bd9Sstevel@tonic-gate * the bootstrap. 5507c478bd9Sstevel@tonic-gate */ 551ae115bc7Smrj static page_t *bootpages; 552ae115bc7Smrj 553ae115bc7Smrj /* 554ae115bc7Smrj * boot time pages that have a vnode from the ramdisk will keep that forever. 555ae115bc7Smrj */ 556ae115bc7Smrj static page_t *rd_pages; 5577c478bd9Sstevel@tonic-gate 55819397407SSherry Moore /* 55919397407SSherry Moore * Lower 64K 56019397407SSherry Moore */ 56119397407SSherry Moore static page_t *lower_pages = NULL; 56219397407SSherry Moore static int lower_pages_count = 0; 56319397407SSherry Moore 5647c478bd9Sstevel@tonic-gate struct system_hardware system_hardware; 5657c478bd9Sstevel@tonic-gate 5667c478bd9Sstevel@tonic-gate /* 5677c478bd9Sstevel@tonic-gate * Enable some debugging messages concerning memory usage... 5687c478bd9Sstevel@tonic-gate */ 5697c478bd9Sstevel@tonic-gate static void 570ae115bc7Smrj print_memlist(char *title, struct memlist *mp) 5717c478bd9Sstevel@tonic-gate { 5727c478bd9Sstevel@tonic-gate prom_printf("MEMLIST: %s:\n", title); 5737c478bd9Sstevel@tonic-gate while (mp != NULL) { 5747c478bd9Sstevel@tonic-gate prom_printf("\tAddress 0x%" PRIx64 ", size 0x%" PRIx64 "\n", 57556f33205SJonathan Adams mp->ml_address, mp->ml_size); 57656f33205SJonathan Adams mp = mp->ml_next; 5777c478bd9Sstevel@tonic-gate } 5787c478bd9Sstevel@tonic-gate } 5797c478bd9Sstevel@tonic-gate 5807c478bd9Sstevel@tonic-gate /* 5817c478bd9Sstevel@tonic-gate * XX64 need a comment here.. are these just default values, surely 5827c478bd9Sstevel@tonic-gate * we read the "cpuid" type information to figure this out. 5837c478bd9Sstevel@tonic-gate */ 5847c478bd9Sstevel@tonic-gate int l2cache_sz = 0x80000; 5857c478bd9Sstevel@tonic-gate int l2cache_linesz = 0x40; 5867c478bd9Sstevel@tonic-gate int l2cache_assoc = 1; 5877c478bd9Sstevel@tonic-gate 5882cb27123Saguzovsk static size_t textrepl_min_gb = 10; 5892cb27123Saguzovsk 5907c478bd9Sstevel@tonic-gate /* 5917c478bd9Sstevel@tonic-gate * on 64 bit we use a predifined VA range for mapping devices in the kernel 5927c478bd9Sstevel@tonic-gate * on 32 bit the mappings are intermixed in the heap, so we use a bit map 5937c478bd9Sstevel@tonic-gate */ 5947c478bd9Sstevel@tonic-gate #ifdef __amd64 5957c478bd9Sstevel@tonic-gate 5967c478bd9Sstevel@tonic-gate vmem_t *device_arena; 5977c478bd9Sstevel@tonic-gate uintptr_t toxic_addr = (uintptr_t)NULL; 598ae115bc7Smrj size_t toxic_size = 1024 * 1024 * 1024; /* Sparc uses 1 gig too */ 5997c478bd9Sstevel@tonic-gate 6007c478bd9Sstevel@tonic-gate #else /* __i386 */ 6017c478bd9Sstevel@tonic-gate 6027c478bd9Sstevel@tonic-gate ulong_t *toxic_bit_map; /* one bit for each 4k of VA in heap_arena */ 6037c478bd9Sstevel@tonic-gate size_t toxic_bit_map_len = 0; /* in bits */ 6047c478bd9Sstevel@tonic-gate 6057c478bd9Sstevel@tonic-gate #endif /* __i386 */ 6067c478bd9Sstevel@tonic-gate 6077c478bd9Sstevel@tonic-gate /* 6087c478bd9Sstevel@tonic-gate * Simple boot time debug facilities 6097c478bd9Sstevel@tonic-gate */ 6107c478bd9Sstevel@tonic-gate static char *prm_dbg_str[] = { 6117c478bd9Sstevel@tonic-gate "%s:%d: '%s' is 0x%x\n", 6127c478bd9Sstevel@tonic-gate "%s:%d: '%s' is 0x%llx\n" 6137c478bd9Sstevel@tonic-gate }; 6147c478bd9Sstevel@tonic-gate 6157c478bd9Sstevel@tonic-gate int prom_debug; 6167c478bd9Sstevel@tonic-gate 6177c478bd9Sstevel@tonic-gate #define PRM_DEBUG(q) if (prom_debug) \ 6187c478bd9Sstevel@tonic-gate prom_printf(prm_dbg_str[sizeof (q) >> 3], "startup.c", __LINE__, #q, q); 6197c478bd9Sstevel@tonic-gate #define PRM_POINT(q) if (prom_debug) \ 6207c478bd9Sstevel@tonic-gate prom_printf("%s:%d: %s\n", "startup.c", __LINE__, q); 6217c478bd9Sstevel@tonic-gate 6227c478bd9Sstevel@tonic-gate /* 6237c478bd9Sstevel@tonic-gate * This structure is used to keep track of the intial allocations 6247c478bd9Sstevel@tonic-gate * done in startup_memlist(). The value of NUM_ALLOCATIONS needs to 6257c478bd9Sstevel@tonic-gate * be >= the number of ADD_TO_ALLOCATIONS() executed in the code. 6267c478bd9Sstevel@tonic-gate */ 6271de082f7SVikram Hegde #define NUM_ALLOCATIONS 8 6287c478bd9Sstevel@tonic-gate int num_allocations = 0; 6297c478bd9Sstevel@tonic-gate struct { 6307c478bd9Sstevel@tonic-gate void **al_ptr; 6317c478bd9Sstevel@tonic-gate size_t al_size; 6327c478bd9Sstevel@tonic-gate } allocations[NUM_ALLOCATIONS]; 6337c478bd9Sstevel@tonic-gate size_t valloc_sz = 0; 6347c478bd9Sstevel@tonic-gate uintptr_t valloc_base; 6357c478bd9Sstevel@tonic-gate 6367c478bd9Sstevel@tonic-gate #define ADD_TO_ALLOCATIONS(ptr, size) { \ 6377c478bd9Sstevel@tonic-gate size = ROUND_UP_PAGE(size); \ 6387c478bd9Sstevel@tonic-gate if (num_allocations == NUM_ALLOCATIONS) \ 6397c478bd9Sstevel@tonic-gate panic("too many ADD_TO_ALLOCATIONS()"); \ 6407c478bd9Sstevel@tonic-gate allocations[num_allocations].al_ptr = (void**)&ptr; \ 6417c478bd9Sstevel@tonic-gate allocations[num_allocations].al_size = size; \ 6427c478bd9Sstevel@tonic-gate valloc_sz += size; \ 6437c478bd9Sstevel@tonic-gate ++num_allocations; \ 6447c478bd9Sstevel@tonic-gate } 6457c478bd9Sstevel@tonic-gate 646ae115bc7Smrj /* 647ae115bc7Smrj * Allocate all the initial memory needed by the page allocator. 648ae115bc7Smrj */ 6497c478bd9Sstevel@tonic-gate static void 6507c478bd9Sstevel@tonic-gate perform_allocations(void) 6517c478bd9Sstevel@tonic-gate { 6527c478bd9Sstevel@tonic-gate caddr_t mem; 6537c478bd9Sstevel@tonic-gate int i; 654ae115bc7Smrj int valloc_align; 6557c478bd9Sstevel@tonic-gate 656ae115bc7Smrj PRM_DEBUG(valloc_base); 657ae115bc7Smrj PRM_DEBUG(valloc_sz); 658ae115bc7Smrj valloc_align = mmu.level_size[mmu.max_page_level > 0]; 659ae115bc7Smrj mem = BOP_ALLOC(bootops, (caddr_t)valloc_base, valloc_sz, valloc_align); 6607c478bd9Sstevel@tonic-gate if (mem != (caddr_t)valloc_base) 6617c478bd9Sstevel@tonic-gate panic("BOP_ALLOC() failed"); 6627c478bd9Sstevel@tonic-gate bzero(mem, valloc_sz); 6637c478bd9Sstevel@tonic-gate for (i = 0; i < num_allocations; ++i) { 6647c478bd9Sstevel@tonic-gate *allocations[i].al_ptr = (void *)mem; 6657c478bd9Sstevel@tonic-gate mem += allocations[i].al_size; 6667c478bd9Sstevel@tonic-gate } 6677c478bd9Sstevel@tonic-gate } 6687c478bd9Sstevel@tonic-gate 6697c478bd9Sstevel@tonic-gate /* 6707c478bd9Sstevel@tonic-gate * Our world looks like this at startup time. 6717c478bd9Sstevel@tonic-gate * 6727c478bd9Sstevel@tonic-gate * In a 32-bit OS, boot loads the kernel text at 0xfe800000 and kernel data 6737c478bd9Sstevel@tonic-gate * at 0xfec00000. On a 64-bit OS, kernel text and data are loaded at 6747c478bd9Sstevel@tonic-gate * 0xffffffff.fe800000 and 0xffffffff.fec00000 respectively. Those 6757c478bd9Sstevel@tonic-gate * addresses are fixed in the binary at link time. 6767c478bd9Sstevel@tonic-gate * 6777c478bd9Sstevel@tonic-gate * On the text page: 6787c478bd9Sstevel@tonic-gate * unix/genunix/krtld/module text loads. 6797c478bd9Sstevel@tonic-gate * 6807c478bd9Sstevel@tonic-gate * On the data page: 681ae115bc7Smrj * unix/genunix/krtld/module data loads. 682ae115bc7Smrj * 6837c478bd9Sstevel@tonic-gate * Machine-dependent startup code 6847c478bd9Sstevel@tonic-gate */ 6857c478bd9Sstevel@tonic-gate void 6867c478bd9Sstevel@tonic-gate startup(void) 6877c478bd9Sstevel@tonic-gate { 688843e1988Sjohnlev #if !defined(__xpv) 68975bcd456Sjg extern void startup_pci_bios(void); 690843e1988Sjohnlev #endif 691a563a037Sbholler extern cpuset_t cpu_ready_set; 692a563a037Sbholler 693843e1988Sjohnlev /* 694843e1988Sjohnlev * Make sure that nobody tries to use sekpm until we have 695843e1988Sjohnlev * initialized it properly. 696843e1988Sjohnlev */ 6977c478bd9Sstevel@tonic-gate #if defined(__amd64) 69835b1ab99Sjosephb kpm_desired = 1; 6997c478bd9Sstevel@tonic-gate #endif 7007c478bd9Sstevel@tonic-gate kpm_enable = 0; 701a563a037Sbholler CPUSET_ONLY(cpu_ready_set, 0); /* cpu 0 is boot cpu */ 7027c478bd9Sstevel@tonic-gate 703843e1988Sjohnlev #if defined(__xpv) /* XXPV fix me! */ 704843e1988Sjohnlev { 705843e1988Sjohnlev extern int segvn_use_regions; 706843e1988Sjohnlev segvn_use_regions = 0; 707843e1988Sjohnlev } 708843e1988Sjohnlev #endif 7097c478bd9Sstevel@tonic-gate progressbar_init(); 7107c478bd9Sstevel@tonic-gate startup_init(); 711ab4a9bebSjohnlev #if defined(__xpv) 712ab4a9bebSjohnlev startup_xen_version(); 713ab4a9bebSjohnlev #endif 7147c478bd9Sstevel@tonic-gate startup_memlist(); 715ae115bc7Smrj startup_kmem(); 71635b1ab99Sjosephb startup_vm(); 717843e1988Sjohnlev #if !defined(__xpv) 718*0db3240dSStephen Hanson /* 719*0db3240dSStephen Hanson * Note we need to do this even on fast reboot in order to access 720*0db3240dSStephen Hanson * the irq routing table (used for pci labels). 721*0db3240dSStephen Hanson */ 722*0db3240dSStephen Hanson startup_pci_bios(); 723e4b86885SCheng Sean Ye #endif 724e4b86885SCheng Sean Ye #if defined(__xpv) 725e4b86885SCheng Sean Ye startup_xen_mca(); 726843e1988Sjohnlev #endif 7277c478bd9Sstevel@tonic-gate startup_modules(); 7288770118eSlipeng sang - Sun Microsystems - Beijing China 7297c478bd9Sstevel@tonic-gate startup_end(); 7307c478bd9Sstevel@tonic-gate } 7317c478bd9Sstevel@tonic-gate 7327c478bd9Sstevel@tonic-gate static void 7337c478bd9Sstevel@tonic-gate startup_init() 7347c478bd9Sstevel@tonic-gate { 7357c478bd9Sstevel@tonic-gate PRM_POINT("startup_init() starting..."); 7367c478bd9Sstevel@tonic-gate 7377c478bd9Sstevel@tonic-gate /* 7387c478bd9Sstevel@tonic-gate * Complete the extraction of cpuid data 7397c478bd9Sstevel@tonic-gate */ 7407c478bd9Sstevel@tonic-gate cpuid_pass2(CPU); 7417c478bd9Sstevel@tonic-gate 7427c478bd9Sstevel@tonic-gate (void) check_boot_version(BOP_GETVERSION(bootops)); 7437c478bd9Sstevel@tonic-gate 7447c478bd9Sstevel@tonic-gate /* 7457c478bd9Sstevel@tonic-gate * Check for prom_debug in boot environment 7467c478bd9Sstevel@tonic-gate */ 7477c478bd9Sstevel@tonic-gate if (BOP_GETPROPLEN(bootops, "prom_debug") >= 0) { 7487c478bd9Sstevel@tonic-gate ++prom_debug; 7497c478bd9Sstevel@tonic-gate PRM_POINT("prom_debug found in boot enviroment"); 7507c478bd9Sstevel@tonic-gate } 7517c478bd9Sstevel@tonic-gate 7527c478bd9Sstevel@tonic-gate /* 7537c478bd9Sstevel@tonic-gate * Collect node, cpu and memory configuration information. 7547c478bd9Sstevel@tonic-gate */ 7557c478bd9Sstevel@tonic-gate get_system_configuration(); 7567c478bd9Sstevel@tonic-gate 7577c478bd9Sstevel@tonic-gate /* 7587c478bd9Sstevel@tonic-gate * Halt if this is an unsupported processor. 7597c478bd9Sstevel@tonic-gate */ 7607c478bd9Sstevel@tonic-gate if (x86_type == X86_TYPE_486 || x86_type == X86_TYPE_CYRIX_486) { 7617c478bd9Sstevel@tonic-gate printf("\n486 processor (\"%s\") detected.\n", 7627c478bd9Sstevel@tonic-gate CPU->cpu_brandstr); 7637c478bd9Sstevel@tonic-gate halt("This processor is not supported by this release " 7647c478bd9Sstevel@tonic-gate "of Solaris."); 7657c478bd9Sstevel@tonic-gate } 7667c478bd9Sstevel@tonic-gate 7677c478bd9Sstevel@tonic-gate PRM_POINT("startup_init() done"); 7687c478bd9Sstevel@tonic-gate } 7697c478bd9Sstevel@tonic-gate 7707c478bd9Sstevel@tonic-gate /* 7717c478bd9Sstevel@tonic-gate * Callback for copy_memlist_filter() to filter nucleus, kadb/kmdb, (ie. 7727c478bd9Sstevel@tonic-gate * everything mapped above KERNEL_TEXT) pages from phys_avail. Note it 7737c478bd9Sstevel@tonic-gate * also filters out physical page zero. There is some reliance on the 7747c478bd9Sstevel@tonic-gate * boot loader allocating only a few contiguous physical memory chunks. 7757c478bd9Sstevel@tonic-gate */ 7767c478bd9Sstevel@tonic-gate static void 7777c478bd9Sstevel@tonic-gate avail_filter(uint64_t *addr, uint64_t *size) 7787c478bd9Sstevel@tonic-gate { 7797c478bd9Sstevel@tonic-gate uintptr_t va; 7807c478bd9Sstevel@tonic-gate uintptr_t next_va; 7817c478bd9Sstevel@tonic-gate pfn_t pfn; 7827c478bd9Sstevel@tonic-gate uint64_t pfn_addr; 7837c478bd9Sstevel@tonic-gate uint64_t pfn_eaddr; 7847c478bd9Sstevel@tonic-gate uint_t prot; 7857c478bd9Sstevel@tonic-gate size_t len; 7867c478bd9Sstevel@tonic-gate uint_t change; 7877c478bd9Sstevel@tonic-gate 7887c478bd9Sstevel@tonic-gate if (prom_debug) 7897c478bd9Sstevel@tonic-gate prom_printf("\tFilter: in: a=%" PRIx64 ", s=%" PRIx64 "\n", 7907c478bd9Sstevel@tonic-gate *addr, *size); 7917c478bd9Sstevel@tonic-gate 7927c478bd9Sstevel@tonic-gate /* 7937c478bd9Sstevel@tonic-gate * page zero is required for BIOS.. never make it available 7947c478bd9Sstevel@tonic-gate */ 7957c478bd9Sstevel@tonic-gate if (*addr == 0) { 7967c478bd9Sstevel@tonic-gate *addr += MMU_PAGESIZE; 7977c478bd9Sstevel@tonic-gate *size -= MMU_PAGESIZE; 7987c478bd9Sstevel@tonic-gate } 7997c478bd9Sstevel@tonic-gate 8007c478bd9Sstevel@tonic-gate /* 801ae115bc7Smrj * First we trim from the front of the range. Since kbm_probe() 8027c478bd9Sstevel@tonic-gate * walks ranges in virtual order, but addr/size are physical, we need 8037c478bd9Sstevel@tonic-gate * to the list until no changes are seen. This deals with the case 8047c478bd9Sstevel@tonic-gate * where page "p" is mapped at v, page "p + PAGESIZE" is mapped at w 8057c478bd9Sstevel@tonic-gate * but w < v. 8067c478bd9Sstevel@tonic-gate */ 8077c478bd9Sstevel@tonic-gate do { 8087c478bd9Sstevel@tonic-gate change = 0; 8097c478bd9Sstevel@tonic-gate for (va = KERNEL_TEXT; 810ae115bc7Smrj *size > 0 && kbm_probe(&va, &len, &pfn, &prot) != 0; 8117c478bd9Sstevel@tonic-gate va = next_va) { 8127c478bd9Sstevel@tonic-gate 8137c478bd9Sstevel@tonic-gate next_va = va + len; 814ae115bc7Smrj pfn_addr = pfn_to_pa(pfn); 8157c478bd9Sstevel@tonic-gate pfn_eaddr = pfn_addr + len; 8167c478bd9Sstevel@tonic-gate 8177c478bd9Sstevel@tonic-gate if (pfn_addr <= *addr && pfn_eaddr > *addr) { 8187c478bd9Sstevel@tonic-gate change = 1; 8197c478bd9Sstevel@tonic-gate while (*size > 0 && len > 0) { 8207c478bd9Sstevel@tonic-gate *addr += MMU_PAGESIZE; 8217c478bd9Sstevel@tonic-gate *size -= MMU_PAGESIZE; 8227c478bd9Sstevel@tonic-gate len -= MMU_PAGESIZE; 8237c478bd9Sstevel@tonic-gate } 8247c478bd9Sstevel@tonic-gate } 8257c478bd9Sstevel@tonic-gate } 8267c478bd9Sstevel@tonic-gate if (change && prom_debug) 8277c478bd9Sstevel@tonic-gate prom_printf("\t\ttrim: a=%" PRIx64 ", s=%" PRIx64 "\n", 8287c478bd9Sstevel@tonic-gate *addr, *size); 8297c478bd9Sstevel@tonic-gate } while (change); 8307c478bd9Sstevel@tonic-gate 8317c478bd9Sstevel@tonic-gate /* 8327c478bd9Sstevel@tonic-gate * Trim pages from the end of the range. 8337c478bd9Sstevel@tonic-gate */ 8347c478bd9Sstevel@tonic-gate for (va = KERNEL_TEXT; 835ae115bc7Smrj *size > 0 && kbm_probe(&va, &len, &pfn, &prot) != 0; 8367c478bd9Sstevel@tonic-gate va = next_va) { 8377c478bd9Sstevel@tonic-gate 8387c478bd9Sstevel@tonic-gate next_va = va + len; 839ae115bc7Smrj pfn_addr = pfn_to_pa(pfn); 8407c478bd9Sstevel@tonic-gate 8417c478bd9Sstevel@tonic-gate if (pfn_addr >= *addr && pfn_addr < *addr + *size) 8427c478bd9Sstevel@tonic-gate *size = pfn_addr - *addr; 8437c478bd9Sstevel@tonic-gate } 8447c478bd9Sstevel@tonic-gate 8457c478bd9Sstevel@tonic-gate if (prom_debug) 8467c478bd9Sstevel@tonic-gate prom_printf("\tFilter out: a=%" PRIx64 ", s=%" PRIx64 "\n", 8477c478bd9Sstevel@tonic-gate *addr, *size); 8487c478bd9Sstevel@tonic-gate } 8497c478bd9Sstevel@tonic-gate 8507c478bd9Sstevel@tonic-gate static void 8517c478bd9Sstevel@tonic-gate kpm_init() 8527c478bd9Sstevel@tonic-gate { 8537c478bd9Sstevel@tonic-gate struct segkpm_crargs b; 8547c478bd9Sstevel@tonic-gate 8557c478bd9Sstevel@tonic-gate /* 8567c478bd9Sstevel@tonic-gate * These variables were all designed for sfmmu in which segkpm is 8577c478bd9Sstevel@tonic-gate * mapped using a single pagesize - either 8KB or 4MB. On x86, we 8587c478bd9Sstevel@tonic-gate * might use 2+ page sizes on a single machine, so none of these 8597c478bd9Sstevel@tonic-gate * variables have a single correct value. They are set up as if we 8607c478bd9Sstevel@tonic-gate * always use a 4KB pagesize, which should do no harm. In the long 8617c478bd9Sstevel@tonic-gate * run, we should get rid of KPM's assumption that only a single 8627c478bd9Sstevel@tonic-gate * pagesize is used. 8637c478bd9Sstevel@tonic-gate */ 8647c478bd9Sstevel@tonic-gate kpm_pgshft = MMU_PAGESHIFT; 8657c478bd9Sstevel@tonic-gate kpm_pgsz = MMU_PAGESIZE; 8667c478bd9Sstevel@tonic-gate kpm_pgoff = MMU_PAGEOFFSET; 8677c478bd9Sstevel@tonic-gate kpmp2pshft = 0; 8687c478bd9Sstevel@tonic-gate kpmpnpgs = 1; 8697c478bd9Sstevel@tonic-gate ASSERT(((uintptr_t)kpm_vbase & (kpm_pgsz - 1)) == 0); 8707c478bd9Sstevel@tonic-gate 8717c478bd9Sstevel@tonic-gate PRM_POINT("about to create segkpm"); 8727c478bd9Sstevel@tonic-gate rw_enter(&kas.a_lock, RW_WRITER); 8737c478bd9Sstevel@tonic-gate 8747c478bd9Sstevel@tonic-gate if (seg_attach(&kas, kpm_vbase, kpm_size, segkpm) < 0) 8757c478bd9Sstevel@tonic-gate panic("cannot attach segkpm"); 8767c478bd9Sstevel@tonic-gate 8777c478bd9Sstevel@tonic-gate b.prot = PROT_READ | PROT_WRITE; 8787c478bd9Sstevel@tonic-gate b.nvcolors = 1; 8797c478bd9Sstevel@tonic-gate 8807c478bd9Sstevel@tonic-gate if (segkpm_create(segkpm, (caddr_t)&b) != 0) 8817c478bd9Sstevel@tonic-gate panic("segkpm_create segkpm"); 8827c478bd9Sstevel@tonic-gate 8837c478bd9Sstevel@tonic-gate rw_exit(&kas.a_lock); 884ae115bc7Smrj } 8857c478bd9Sstevel@tonic-gate 886ae115bc7Smrj /* 887ae115bc7Smrj * The debug info page provides enough information to allow external 888ae115bc7Smrj * inspectors (e.g. when running under a hypervisor) to bootstrap 889ae115bc7Smrj * themselves into allowing full-blown kernel debugging. 890ae115bc7Smrj */ 891ae115bc7Smrj static void 892ae115bc7Smrj init_debug_info(void) 893ae115bc7Smrj { 894ae115bc7Smrj caddr_t mem; 895ae115bc7Smrj debug_info_t *di; 896ae115bc7Smrj 897ae115bc7Smrj #ifndef __lint 898ae115bc7Smrj ASSERT(sizeof (debug_info_t) < MMU_PAGESIZE); 899ae115bc7Smrj #endif 900ae115bc7Smrj 901ae115bc7Smrj mem = BOP_ALLOC(bootops, (caddr_t)DEBUG_INFO_VA, MMU_PAGESIZE, 902ae115bc7Smrj MMU_PAGESIZE); 903ae115bc7Smrj 904ae115bc7Smrj if (mem != (caddr_t)DEBUG_INFO_VA) 905ae115bc7Smrj panic("BOP_ALLOC() failed"); 906ae115bc7Smrj bzero(mem, MMU_PAGESIZE); 907ae115bc7Smrj 908ae115bc7Smrj di = (debug_info_t *)mem; 909ae115bc7Smrj 910ae115bc7Smrj di->di_magic = DEBUG_INFO_MAGIC; 911ae115bc7Smrj di->di_version = DEBUG_INFO_VERSION; 912843e1988Sjohnlev di->di_modules = (uintptr_t)&modules; 913843e1988Sjohnlev di->di_s_text = (uintptr_t)s_text; 914843e1988Sjohnlev di->di_e_text = (uintptr_t)e_text; 915843e1988Sjohnlev di->di_s_data = (uintptr_t)s_data; 916843e1988Sjohnlev di->di_e_data = (uintptr_t)e_data; 917843e1988Sjohnlev di->di_hat_htable_off = offsetof(hat_t, hat_htable); 918843e1988Sjohnlev di->di_ht_pfn_off = offsetof(htable_t, ht_pfn); 9197c478bd9Sstevel@tonic-gate } 9207c478bd9Sstevel@tonic-gate 9217c478bd9Sstevel@tonic-gate /* 922ae115bc7Smrj * Build the memlists and other kernel essential memory system data structures. 923ae115bc7Smrj * This is everything at valloc_base. 9247c478bd9Sstevel@tonic-gate */ 9257c478bd9Sstevel@tonic-gate static void 9267c478bd9Sstevel@tonic-gate startup_memlist(void) 9277c478bd9Sstevel@tonic-gate { 9287c478bd9Sstevel@tonic-gate size_t memlist_sz; 9297c478bd9Sstevel@tonic-gate size_t memseg_sz; 9307c478bd9Sstevel@tonic-gate size_t pagehash_sz; 9317c478bd9Sstevel@tonic-gate size_t pp_sz; 9327c478bd9Sstevel@tonic-gate uintptr_t va; 9337c478bd9Sstevel@tonic-gate size_t len; 9347c478bd9Sstevel@tonic-gate uint_t prot; 9357c478bd9Sstevel@tonic-gate pfn_t pfn; 9367c478bd9Sstevel@tonic-gate int memblocks; 9371de082f7SVikram Hegde pfn_t rsvd_high_pfn; 9381de082f7SVikram Hegde pgcnt_t rsvd_pgcnt; 9391de082f7SVikram Hegde size_t rsvdmemlist_sz; 9401de082f7SVikram Hegde int rsvdmemblocks; 9417c478bd9Sstevel@tonic-gate caddr_t pagecolor_mem; 9427c478bd9Sstevel@tonic-gate size_t pagecolor_memsz; 9437c478bd9Sstevel@tonic-gate caddr_t page_ctrs_mem; 9447c478bd9Sstevel@tonic-gate size_t page_ctrs_size; 945d7d93655Sblakej size_t pse_table_alloc_size; 9467c478bd9Sstevel@tonic-gate struct memlist *current; 9477c478bd9Sstevel@tonic-gate extern void startup_build_mem_nodes(struct memlist *); 9487c478bd9Sstevel@tonic-gate 9497c478bd9Sstevel@tonic-gate /* XX64 fix these - they should be in include files */ 9507c478bd9Sstevel@tonic-gate extern size_t page_coloring_init(uint_t, int, int); 9517c478bd9Sstevel@tonic-gate extern void page_coloring_setup(caddr_t); 9527c478bd9Sstevel@tonic-gate 9537c478bd9Sstevel@tonic-gate PRM_POINT("startup_memlist() starting..."); 9547c478bd9Sstevel@tonic-gate 9557c478bd9Sstevel@tonic-gate /* 9567c478bd9Sstevel@tonic-gate * Use leftover large page nucleus text/data space for loadable modules. 9577c478bd9Sstevel@tonic-gate * Use at most MODTEXT/MODDATA. 9587c478bd9Sstevel@tonic-gate */ 959ae115bc7Smrj len = kbm_nucleus_size; 960ae115bc7Smrj ASSERT(len > MMU_PAGESIZE); 9617c478bd9Sstevel@tonic-gate 962ae115bc7Smrj moddata = (caddr_t)ROUND_UP_PAGE(e_data); 963ae115bc7Smrj e_moddata = (caddr_t)P2ROUNDUP((uintptr_t)e_data, (uintptr_t)len); 964ae115bc7Smrj if (e_moddata - moddata > MODDATA) 965ae115bc7Smrj e_moddata = moddata + MODDATA; 9667c478bd9Sstevel@tonic-gate 967ae115bc7Smrj modtext = (caddr_t)ROUND_UP_PAGE(e_text); 968ae115bc7Smrj e_modtext = (caddr_t)P2ROUNDUP((uintptr_t)e_text, (uintptr_t)len); 969ae115bc7Smrj if (e_modtext - modtext > MODTEXT) 970ae115bc7Smrj e_modtext = modtext + MODTEXT; 9717c478bd9Sstevel@tonic-gate 9727c478bd9Sstevel@tonic-gate econtig = e_moddata; 9737c478bd9Sstevel@tonic-gate 9747c478bd9Sstevel@tonic-gate PRM_DEBUG(modtext); 9757c478bd9Sstevel@tonic-gate PRM_DEBUG(e_modtext); 9767c478bd9Sstevel@tonic-gate PRM_DEBUG(moddata); 9777c478bd9Sstevel@tonic-gate PRM_DEBUG(e_moddata); 9787c478bd9Sstevel@tonic-gate PRM_DEBUG(econtig); 9797c478bd9Sstevel@tonic-gate 9807c478bd9Sstevel@tonic-gate /* 981ae115bc7Smrj * Examine the boot loader physical memory map to find out: 9827c478bd9Sstevel@tonic-gate * - total memory in system - physinstalled 9837c478bd9Sstevel@tonic-gate * - the max physical address - physmax 984ae115bc7Smrj * - the number of discontiguous segments of memory. 9857c478bd9Sstevel@tonic-gate */ 9867c478bd9Sstevel@tonic-gate if (prom_debug) 987ae115bc7Smrj print_memlist("boot physinstalled", 9887c478bd9Sstevel@tonic-gate bootops->boot_mem->physinstalled); 989a3114836SGerry Liu installed_top_size_ex(bootops->boot_mem->physinstalled, &physmax, 9907c478bd9Sstevel@tonic-gate &physinstalled, &memblocks); 9917c478bd9Sstevel@tonic-gate PRM_DEBUG(physmax); 9927c478bd9Sstevel@tonic-gate PRM_DEBUG(physinstalled); 9937c478bd9Sstevel@tonic-gate PRM_DEBUG(memblocks); 9947c478bd9Sstevel@tonic-gate 995a3114836SGerry Liu /* 996a3114836SGerry Liu * Compute maximum physical address for memory DR operations. 997a3114836SGerry Liu * Memory DR operations are unsupported on xpv or 32bit OSes. 998a3114836SGerry Liu */ 999a3114836SGerry Liu #ifdef __amd64 1000a3114836SGerry Liu if (plat_dr_support_memory()) { 1001a3114836SGerry Liu if (plat_dr_physmax == 0) { 1002a3114836SGerry Liu uint_t pabits = UINT_MAX; 1003a3114836SGerry Liu 1004a3114836SGerry Liu cpuid_get_addrsize(CPU, &pabits, NULL); 1005a3114836SGerry Liu plat_dr_physmax = btop(1ULL << pabits); 1006a3114836SGerry Liu } 1007a3114836SGerry Liu if (plat_dr_physmax > PHYSMEM_MAX64) 1008a3114836SGerry Liu plat_dr_physmax = PHYSMEM_MAX64; 1009a3114836SGerry Liu } else 1010a3114836SGerry Liu #endif 1011a3114836SGerry Liu plat_dr_physmax = 0; 1012a3114836SGerry Liu 10131de082f7SVikram Hegde /* 10141de082f7SVikram Hegde * Examine the bios reserved memory to find out: 10151de082f7SVikram Hegde * - the number of discontiguous segments of memory. 10161de082f7SVikram Hegde */ 10171de082f7SVikram Hegde if (prom_debug) 10181de082f7SVikram Hegde print_memlist("boot reserved mem", 10191de082f7SVikram Hegde bootops->boot_mem->rsvdmem); 1020a3114836SGerry Liu installed_top_size_ex(bootops->boot_mem->rsvdmem, &rsvd_high_pfn, 10211de082f7SVikram Hegde &rsvd_pgcnt, &rsvdmemblocks); 10221de082f7SVikram Hegde PRM_DEBUG(rsvd_high_pfn); 10231de082f7SVikram Hegde PRM_DEBUG(rsvd_pgcnt); 10241de082f7SVikram Hegde PRM_DEBUG(rsvdmemblocks); 10251de082f7SVikram Hegde 10267c478bd9Sstevel@tonic-gate /* 10277c478bd9Sstevel@tonic-gate * Initialize hat's mmu parameters. 10287c478bd9Sstevel@tonic-gate * Check for enforce-prot-exec in boot environment. It's used to 10297c478bd9Sstevel@tonic-gate * enable/disable support for the page table entry NX bit. 10307c478bd9Sstevel@tonic-gate * The default is to enforce PROT_EXEC on processors that support NX. 10317c478bd9Sstevel@tonic-gate * Boot seems to round up the "len", but 8 seems to be big enough. 10327c478bd9Sstevel@tonic-gate */ 10337c478bd9Sstevel@tonic-gate mmu_init(); 10347c478bd9Sstevel@tonic-gate 10357c478bd9Sstevel@tonic-gate #ifdef __i386 10367c478bd9Sstevel@tonic-gate /* 10377c478bd9Sstevel@tonic-gate * physmax is lowered if there is more memory than can be 10387c478bd9Sstevel@tonic-gate * physically addressed in 32 bit (PAE/non-PAE) modes. 10397c478bd9Sstevel@tonic-gate */ 10407c478bd9Sstevel@tonic-gate if (mmu.pae_hat) { 10417c478bd9Sstevel@tonic-gate if (PFN_ABOVE64G(physmax)) { 10427c478bd9Sstevel@tonic-gate physinstalled -= (physmax - (PFN_64G - 1)); 10437c478bd9Sstevel@tonic-gate physmax = PFN_64G - 1; 10447c478bd9Sstevel@tonic-gate } 10457c478bd9Sstevel@tonic-gate } else { 10467c478bd9Sstevel@tonic-gate if (PFN_ABOVE4G(physmax)) { 10477c478bd9Sstevel@tonic-gate physinstalled -= (physmax - (PFN_4G - 1)); 10487c478bd9Sstevel@tonic-gate physmax = PFN_4G - 1; 10497c478bd9Sstevel@tonic-gate } 10507c478bd9Sstevel@tonic-gate } 10517c478bd9Sstevel@tonic-gate #endif 10527c478bd9Sstevel@tonic-gate 10537c478bd9Sstevel@tonic-gate startup_build_mem_nodes(bootops->boot_mem->physinstalled); 10547c478bd9Sstevel@tonic-gate 10557c478bd9Sstevel@tonic-gate if (BOP_GETPROPLEN(bootops, "enforce-prot-exec") >= 0) { 10567c478bd9Sstevel@tonic-gate int len = BOP_GETPROPLEN(bootops, "enforce-prot-exec"); 10577c478bd9Sstevel@tonic-gate char value[8]; 10587c478bd9Sstevel@tonic-gate 10597c478bd9Sstevel@tonic-gate if (len < 8) 10607c478bd9Sstevel@tonic-gate (void) BOP_GETPROP(bootops, "enforce-prot-exec", value); 10617c478bd9Sstevel@tonic-gate else 10627c478bd9Sstevel@tonic-gate (void) strcpy(value, ""); 10637c478bd9Sstevel@tonic-gate if (strcmp(value, "off") == 0) 10647c478bd9Sstevel@tonic-gate mmu.pt_nx = 0; 10657c478bd9Sstevel@tonic-gate } 10667c478bd9Sstevel@tonic-gate PRM_DEBUG(mmu.pt_nx); 10677c478bd9Sstevel@tonic-gate 10687c478bd9Sstevel@tonic-gate /* 10697c478bd9Sstevel@tonic-gate * We will need page_t's for every page in the system, except for 10707c478bd9Sstevel@tonic-gate * memory mapped at or above above the start of the kernel text segment. 10717c478bd9Sstevel@tonic-gate * 10727c478bd9Sstevel@tonic-gate * pages above e_modtext are attributed to kernel debugger (obp_pages) 10737c478bd9Sstevel@tonic-gate */ 10747c478bd9Sstevel@tonic-gate npages = physinstalled - 1; /* avail_filter() skips page 0, so "- 1" */ 10757c478bd9Sstevel@tonic-gate obp_pages = 0; 10767c478bd9Sstevel@tonic-gate va = KERNEL_TEXT; 1077ae115bc7Smrj while (kbm_probe(&va, &len, &pfn, &prot) != 0) { 10787c478bd9Sstevel@tonic-gate npages -= len >> MMU_PAGESHIFT; 10797c478bd9Sstevel@tonic-gate if (va >= (uintptr_t)e_moddata) 10807c478bd9Sstevel@tonic-gate obp_pages += len >> MMU_PAGESHIFT; 10817c478bd9Sstevel@tonic-gate va += len; 10827c478bd9Sstevel@tonic-gate } 10837c478bd9Sstevel@tonic-gate PRM_DEBUG(npages); 10847c478bd9Sstevel@tonic-gate PRM_DEBUG(obp_pages); 10857c478bd9Sstevel@tonic-gate 10867c478bd9Sstevel@tonic-gate /* 108735b1ab99Sjosephb * If physmem is patched to be non-zero, use it instead of the computed 108835b1ab99Sjosephb * value unless it is larger than the actual amount of memory on hand. 10897c478bd9Sstevel@tonic-gate */ 10904944b02eSkchow if (physmem == 0 || physmem > npages) { 10917c478bd9Sstevel@tonic-gate physmem = npages; 10924944b02eSkchow } else if (physmem < npages) { 1093f53ad214Skchow orig_npages = npages; 10947c478bd9Sstevel@tonic-gate npages = physmem; 10954944b02eSkchow } 10967c478bd9Sstevel@tonic-gate PRM_DEBUG(physmem); 10977c478bd9Sstevel@tonic-gate 10987c478bd9Sstevel@tonic-gate /* 10997c478bd9Sstevel@tonic-gate * We now compute the sizes of all the initial allocations for 11007c478bd9Sstevel@tonic-gate * structures the kernel needs in order do kmem_alloc(). These 11017c478bd9Sstevel@tonic-gate * include: 11027c478bd9Sstevel@tonic-gate * memsegs 11037c478bd9Sstevel@tonic-gate * memlists 11047c478bd9Sstevel@tonic-gate * page hash table 11057c478bd9Sstevel@tonic-gate * page_t's 11067c478bd9Sstevel@tonic-gate * page coloring data structs 11077c478bd9Sstevel@tonic-gate */ 11087c478bd9Sstevel@tonic-gate memseg_sz = sizeof (struct memseg) * (memblocks + POSS_NEW_FRAGMENTS); 11097c478bd9Sstevel@tonic-gate ADD_TO_ALLOCATIONS(memseg_base, memseg_sz); 11107c478bd9Sstevel@tonic-gate PRM_DEBUG(memseg_sz); 11117c478bd9Sstevel@tonic-gate 11127c478bd9Sstevel@tonic-gate /* 1113ae115bc7Smrj * Reserve space for memlists. There's no real good way to know exactly 1114ae115bc7Smrj * how much room we'll need, but this should be a good upper bound. 11157c478bd9Sstevel@tonic-gate */ 11167c478bd9Sstevel@tonic-gate memlist_sz = ROUND_UP_PAGE(2 * sizeof (struct memlist) * 11177c478bd9Sstevel@tonic-gate (memblocks + POSS_NEW_FRAGMENTS)); 11187c478bd9Sstevel@tonic-gate ADD_TO_ALLOCATIONS(memlist, memlist_sz); 11197c478bd9Sstevel@tonic-gate PRM_DEBUG(memlist_sz); 11207c478bd9Sstevel@tonic-gate 11211de082f7SVikram Hegde /* 11221de082f7SVikram Hegde * Reserve space for bios reserved memlists. 11231de082f7SVikram Hegde */ 11241de082f7SVikram Hegde rsvdmemlist_sz = ROUND_UP_PAGE(2 * sizeof (struct memlist) * 11251de082f7SVikram Hegde (rsvdmemblocks + POSS_NEW_FRAGMENTS)); 11261de082f7SVikram Hegde ADD_TO_ALLOCATIONS(bios_rsvd, rsvdmemlist_sz); 11271de082f7SVikram Hegde PRM_DEBUG(rsvdmemlist_sz); 11281de082f7SVikram Hegde 11297c478bd9Sstevel@tonic-gate /* 11307c478bd9Sstevel@tonic-gate * The page structure hash table size is a power of 2 11317c478bd9Sstevel@tonic-gate * such that the average hash chain length is PAGE_HASHAVELEN. 11327c478bd9Sstevel@tonic-gate */ 11337c478bd9Sstevel@tonic-gate page_hashsz = npages / PAGE_HASHAVELEN; 11347c478bd9Sstevel@tonic-gate page_hashsz = 1 << highbit(page_hashsz); 11357c478bd9Sstevel@tonic-gate pagehash_sz = sizeof (struct page *) * page_hashsz; 11367c478bd9Sstevel@tonic-gate ADD_TO_ALLOCATIONS(page_hash, pagehash_sz); 11377c478bd9Sstevel@tonic-gate PRM_DEBUG(pagehash_sz); 11387c478bd9Sstevel@tonic-gate 11397c478bd9Sstevel@tonic-gate /* 1140ae115bc7Smrj * Set aside room for the page structures themselves. 11417c478bd9Sstevel@tonic-gate */ 1142ae115bc7Smrj PRM_DEBUG(npages); 1143ae115bc7Smrj pp_sz = sizeof (struct page) * npages; 11447c478bd9Sstevel@tonic-gate ADD_TO_ALLOCATIONS(pp_base, pp_sz); 11457c478bd9Sstevel@tonic-gate PRM_DEBUG(pp_sz); 11467c478bd9Sstevel@tonic-gate 11477c478bd9Sstevel@tonic-gate /* 11487c478bd9Sstevel@tonic-gate * determine l2 cache info and memory size for page coloring 11497c478bd9Sstevel@tonic-gate */ 11507c478bd9Sstevel@tonic-gate (void) getl2cacheinfo(CPU, 11517c478bd9Sstevel@tonic-gate &l2cache_sz, &l2cache_linesz, &l2cache_assoc); 11527c478bd9Sstevel@tonic-gate pagecolor_memsz = 11537c478bd9Sstevel@tonic-gate page_coloring_init(l2cache_sz, l2cache_linesz, l2cache_assoc); 11547c478bd9Sstevel@tonic-gate ADD_TO_ALLOCATIONS(pagecolor_mem, pagecolor_memsz); 11557c478bd9Sstevel@tonic-gate PRM_DEBUG(pagecolor_memsz); 11567c478bd9Sstevel@tonic-gate 11577c478bd9Sstevel@tonic-gate page_ctrs_size = page_ctrs_sz(); 11587c478bd9Sstevel@tonic-gate ADD_TO_ALLOCATIONS(page_ctrs_mem, page_ctrs_size); 11597c478bd9Sstevel@tonic-gate PRM_DEBUG(page_ctrs_size); 11607c478bd9Sstevel@tonic-gate 1161d7d93655Sblakej /* 1162d7d93655Sblakej * Allocate the array that protects pp->p_selock. 1163d7d93655Sblakej */ 1164d7d93655Sblakej pse_shift = size_pse_array(physmem, max_ncpus); 1165d7d93655Sblakej pse_table_size = 1 << pse_shift; 1166d7d93655Sblakej pse_table_alloc_size = pse_table_size * sizeof (pad_mutex_t); 1167d7d93655Sblakej ADD_TO_ALLOCATIONS(pse_mutex, pse_table_alloc_size); 1168d7d93655Sblakej 1169ae115bc7Smrj #if defined(__amd64) 11707c478bd9Sstevel@tonic-gate valloc_sz = ROUND_UP_LPAGE(valloc_sz); 1171ae115bc7Smrj valloc_base = VALLOC_BASE; 117235b1ab99Sjosephb 117335b1ab99Sjosephb /* 117435b1ab99Sjosephb * The default values of VALLOC_BASE and SEGKPM_BASE should work 117535b1ab99Sjosephb * for values of physmax up to 1 Terabyte. They need adjusting when 11766c8c89eaSKit Chow * memory is at addresses above 1 TB. When adjusted, segkpm_base must 11776c8c89eaSKit Chow * be aligned on KERNEL_REDZONE_SIZE boundary (span of top level pte). 117835b1ab99Sjosephb */ 1179a3114836SGerry Liu if (physmax + 1 > mmu_btop(TERABYTE) || 1180a3114836SGerry Liu plat_dr_physmax > mmu_btop(TERABYTE)) { 118135b1ab99Sjosephb uint64_t kpm_resv_amount = mmu_ptob(physmax + 1); 118235b1ab99Sjosephb 1183a3114836SGerry Liu if (kpm_resv_amount < mmu_ptob(plat_dr_physmax)) { 1184a3114836SGerry Liu kpm_resv_amount = mmu_ptob(plat_dr_physmax); 1185a3114836SGerry Liu } 1186a3114836SGerry Liu 11876c8c89eaSKit Chow segkpm_base = -(P2ROUNDUP((2 * kpm_resv_amount), 11886c8c89eaSKit Chow KERNEL_REDZONE_SIZE)); /* down from top VA */ 118935b1ab99Sjosephb 119035b1ab99Sjosephb /* make sure we leave some space for user apps above hole */ 119135b1ab99Sjosephb segkpm_base = MAX(segkpm_base, AMD64_VA_HOLE_END + TERABYTE); 119235b1ab99Sjosephb if (segkpm_base > SEGKPM_BASE) 119335b1ab99Sjosephb segkpm_base = SEGKPM_BASE; 119435b1ab99Sjosephb PRM_DEBUG(segkpm_base); 119535b1ab99Sjosephb 11966c8c89eaSKit Chow valloc_base = segkpm_base + P2ROUNDUP(kpm_resv_amount, ONE_GIG); 11976c8c89eaSKit Chow if (valloc_base < segkpm_base) 11986c8c89eaSKit Chow panic("not enough kernel VA to support memory size"); 119935b1ab99Sjosephb PRM_DEBUG(valloc_base); 120035b1ab99Sjosephb } 1201ae115bc7Smrj #else /* __i386 */ 1202ae115bc7Smrj valloc_base = (uintptr_t)(MISC_VA_BASE - valloc_sz); 1203ae115bc7Smrj valloc_base = P2ALIGN(valloc_base, mmu.level_size[1]); 12047c478bd9Sstevel@tonic-gate PRM_DEBUG(valloc_base); 120535b1ab99Sjosephb #endif /* __i386 */ 1206ae115bc7Smrj 1207ae115bc7Smrj /* 1208ae115bc7Smrj * do all the initial allocations 1209ae115bc7Smrj */ 1210ae115bc7Smrj perform_allocations(); 1211ae115bc7Smrj 1212ae115bc7Smrj /* 1213ae115bc7Smrj * Build phys_install and phys_avail in kernel memspace. 1214ae115bc7Smrj * - phys_install should be all memory in the system. 1215ae115bc7Smrj * - phys_avail is phys_install minus any memory mapped before this 1216ae115bc7Smrj * point above KERNEL_TEXT. 1217ae115bc7Smrj */ 1218ae115bc7Smrj current = phys_install = memlist; 1219ae115bc7Smrj copy_memlist_filter(bootops->boot_mem->physinstalled, ¤t, NULL); 1220ae115bc7Smrj if ((caddr_t)current > (caddr_t)memlist + memlist_sz) 1221ae115bc7Smrj panic("physinstalled was too big!"); 1222ae115bc7Smrj if (prom_debug) 1223ae115bc7Smrj print_memlist("phys_install", phys_install); 1224ae115bc7Smrj 1225ae115bc7Smrj phys_avail = current; 1226ae115bc7Smrj PRM_POINT("Building phys_avail:\n"); 1227ae115bc7Smrj copy_memlist_filter(bootops->boot_mem->physinstalled, ¤t, 1228ae115bc7Smrj avail_filter); 1229ae115bc7Smrj if ((caddr_t)current > (caddr_t)memlist + memlist_sz) 1230ae115bc7Smrj panic("physavail was too big!"); 1231ae115bc7Smrj if (prom_debug) 1232ae115bc7Smrj print_memlist("phys_avail", phys_avail); 1233a3114836SGerry Liu #ifndef __xpv 1234a3114836SGerry Liu /* 1235a3114836SGerry Liu * Free unused memlist items, which may be used by memory DR driver 1236a3114836SGerry Liu * at runtime. 1237a3114836SGerry Liu */ 1238a3114836SGerry Liu if ((caddr_t)current < (caddr_t)memlist + memlist_sz) { 1239a3114836SGerry Liu memlist_free_block((caddr_t)current, 1240a3114836SGerry Liu (caddr_t)memlist + memlist_sz - (caddr_t)current); 1241a3114836SGerry Liu } 1242a3114836SGerry Liu #endif 1243ae115bc7Smrj 12441de082f7SVikram Hegde /* 12451de082f7SVikram Hegde * Build bios reserved memspace 12461de082f7SVikram Hegde */ 12471de082f7SVikram Hegde current = bios_rsvd; 12481de082f7SVikram Hegde copy_memlist_filter(bootops->boot_mem->rsvdmem, ¤t, NULL); 12491de082f7SVikram Hegde if ((caddr_t)current > (caddr_t)bios_rsvd + rsvdmemlist_sz) 12501de082f7SVikram Hegde panic("bios_rsvd was too big!"); 12511de082f7SVikram Hegde if (prom_debug) 12521de082f7SVikram Hegde print_memlist("bios_rsvd", bios_rsvd); 1253a3114836SGerry Liu #ifndef __xpv 1254a3114836SGerry Liu /* 1255a3114836SGerry Liu * Free unused memlist items, which may be used by memory DR driver 1256a3114836SGerry Liu * at runtime. 1257a3114836SGerry Liu */ 1258a3114836SGerry Liu if ((caddr_t)current < (caddr_t)bios_rsvd + rsvdmemlist_sz) { 1259a3114836SGerry Liu memlist_free_block((caddr_t)current, 1260a3114836SGerry Liu (caddr_t)bios_rsvd + rsvdmemlist_sz - (caddr_t)current); 1261a3114836SGerry Liu } 1262a3114836SGerry Liu #endif 12631de082f7SVikram Hegde 1264ae115bc7Smrj /* 1265ae115bc7Smrj * setup page coloring 1266ae115bc7Smrj */ 1267ae115bc7Smrj page_coloring_setup(pagecolor_mem); 1268ae115bc7Smrj page_lock_init(); /* currently a no-op */ 1269ae115bc7Smrj 1270ae115bc7Smrj /* 1271ae115bc7Smrj * free page list counters 1272ae115bc7Smrj */ 1273ae115bc7Smrj (void) page_ctrs_alloc(page_ctrs_mem); 1274ae115bc7Smrj 127506fb6a36Sdv /* 127606fb6a36Sdv * Size the pcf array based on the number of cpus in the box at 127706fb6a36Sdv * boot time. 127806fb6a36Sdv */ 127906fb6a36Sdv 128006fb6a36Sdv pcf_init(); 128106fb6a36Sdv 1282ae115bc7Smrj /* 1283ae115bc7Smrj * Initialize the page structures from the memory lists. 1284ae115bc7Smrj */ 1285ae115bc7Smrj availrmem_initial = availrmem = freemem = 0; 1286ae115bc7Smrj PRM_POINT("Calling kphysm_init()..."); 1287ae115bc7Smrj npages = kphysm_init(pp_base, npages); 1288ae115bc7Smrj PRM_POINT("kphysm_init() done"); 1289ae115bc7Smrj PRM_DEBUG(npages); 1290ae115bc7Smrj 1291ae115bc7Smrj init_debug_info(); 1292ae115bc7Smrj 1293ae115bc7Smrj /* 1294ae115bc7Smrj * Now that page_t's have been initialized, remove all the 1295ae115bc7Smrj * initial allocation pages from the kernel free page lists. 1296ae115bc7Smrj */ 1297ae115bc7Smrj boot_mapin((caddr_t)valloc_base, valloc_sz); 12980cfdb603Sjosephb boot_mapin((caddr_t)MISC_VA_BASE, MISC_VA_SIZE); 1299ae115bc7Smrj PRM_POINT("startup_memlist() done"); 1300ae115bc7Smrj 1301ae115bc7Smrj PRM_DEBUG(valloc_sz); 13022cb27123Saguzovsk 1303567d55e1Saguzovsk #if defined(__amd64) 1304567d55e1Saguzovsk if ((availrmem >> (30 - MMU_PAGESHIFT)) >= 1305567d55e1Saguzovsk textrepl_min_gb && l2cache_sz <= 2 << 20) { 1306567d55e1Saguzovsk extern size_t textrepl_size_thresh; 13072cb27123Saguzovsk textrepl_size_thresh = (16 << 20) - 1; 13082cb27123Saguzovsk } 1309567d55e1Saguzovsk #endif 1310ae115bc7Smrj } 1311ae115bc7Smrj 1312ae115bc7Smrj /* 1313ae115bc7Smrj * Layout the kernel's part of address space and initialize kmem allocator. 1314ae115bc7Smrj */ 1315ae115bc7Smrj static void 1316ae115bc7Smrj startup_kmem(void) 1317ae115bc7Smrj { 1318932dc8e5Sdp extern void page_set_colorequiv_arr(void); 1319a50a8b93SKuriakose Kuruvilla const char *fmt = "?features: %b\n"; 1320932dc8e5Sdp 1321ae115bc7Smrj PRM_POINT("startup_kmem() starting..."); 13227c478bd9Sstevel@tonic-gate 13237c478bd9Sstevel@tonic-gate #if defined(__amd64) 13247c478bd9Sstevel@tonic-gate if (eprom_kernelbase && eprom_kernelbase != KERNELBASE) 13257c478bd9Sstevel@tonic-gate cmn_err(CE_NOTE, "!kernelbase cannot be changed on 64-bit " 13267c478bd9Sstevel@tonic-gate "systems."); 132735b1ab99Sjosephb kernelbase = segkpm_base - KERNEL_REDZONE_SIZE; 13287c478bd9Sstevel@tonic-gate core_base = (uintptr_t)COREHEAP_BASE; 1329ae115bc7Smrj core_size = (size_t)MISC_VA_BASE - COREHEAP_BASE; 13307c478bd9Sstevel@tonic-gate #else /* __i386 */ 13317c478bd9Sstevel@tonic-gate /* 13327c478bd9Sstevel@tonic-gate * We configure kernelbase based on: 13337c478bd9Sstevel@tonic-gate * 13347c478bd9Sstevel@tonic-gate * 1. user specified kernelbase via eeprom command. Value cannot exceed 13357c478bd9Sstevel@tonic-gate * KERNELBASE_MAX. we large page align eprom_kernelbase 13367c478bd9Sstevel@tonic-gate * 13377c478bd9Sstevel@tonic-gate * 2. Default to KERNELBASE and adjust to 2X less the size for page_t. 13387c478bd9Sstevel@tonic-gate * On large memory systems we must lower kernelbase to allow 13397c478bd9Sstevel@tonic-gate * enough room for page_t's for all of memory. 13407c478bd9Sstevel@tonic-gate * 13417c478bd9Sstevel@tonic-gate * The value set here, might be changed a little later. 13427c478bd9Sstevel@tonic-gate */ 13437c478bd9Sstevel@tonic-gate if (eprom_kernelbase) { 13447c478bd9Sstevel@tonic-gate kernelbase = eprom_kernelbase & mmu.level_mask[1]; 13457c478bd9Sstevel@tonic-gate if (kernelbase > KERNELBASE_MAX) 13467c478bd9Sstevel@tonic-gate kernelbase = KERNELBASE_MAX; 13477c478bd9Sstevel@tonic-gate } else { 13487c478bd9Sstevel@tonic-gate kernelbase = (uintptr_t)KERNELBASE; 13497c478bd9Sstevel@tonic-gate kernelbase -= ROUND_UP_4MEG(2 * valloc_sz); 13507c478bd9Sstevel@tonic-gate } 13517c478bd9Sstevel@tonic-gate ASSERT((kernelbase & mmu.level_offset[1]) == 0); 1352ae115bc7Smrj core_base = valloc_base; 13537c478bd9Sstevel@tonic-gate core_size = 0; 1354ae115bc7Smrj #endif /* __i386 */ 13557c478bd9Sstevel@tonic-gate 13567c478bd9Sstevel@tonic-gate PRM_DEBUG(core_base); 13577c478bd9Sstevel@tonic-gate PRM_DEBUG(core_size); 1358ae115bc7Smrj PRM_DEBUG(kernelbase); 13597c478bd9Sstevel@tonic-gate 136035b1ab99Sjosephb #if defined(__i386) 13617c478bd9Sstevel@tonic-gate segkp_fromheap = 1; 1362ae115bc7Smrj #endif /* __i386 */ 136335b1ab99Sjosephb 13647c478bd9Sstevel@tonic-gate ekernelheap = (char *)core_base; 1365ae115bc7Smrj PRM_DEBUG(ekernelheap); 13667c478bd9Sstevel@tonic-gate 13677c478bd9Sstevel@tonic-gate /* 13687c478bd9Sstevel@tonic-gate * Now that we know the real value of kernelbase, 13697c478bd9Sstevel@tonic-gate * update variables that were initialized with a value of 13707c478bd9Sstevel@tonic-gate * KERNELBASE (in common/conf/param.c). 13717c478bd9Sstevel@tonic-gate * 13727c478bd9Sstevel@tonic-gate * XXX The problem with this sort of hackery is that the 13737c478bd9Sstevel@tonic-gate * compiler just may feel like putting the const declarations 13747c478bd9Sstevel@tonic-gate * (in param.c) into the .text section. Perhaps they should 13757c478bd9Sstevel@tonic-gate * just be declared as variables there? 13767c478bd9Sstevel@tonic-gate */ 13777c478bd9Sstevel@tonic-gate 13787c478bd9Sstevel@tonic-gate *(uintptr_t *)&_kernelbase = kernelbase; 13797c478bd9Sstevel@tonic-gate *(uintptr_t *)&_userlimit = kernelbase; 1380ccbaea4fSjosephb #if defined(__amd64) 1381ccbaea4fSjosephb *(uintptr_t *)&_userlimit -= KERNELBASE - USERLIMIT; 1382ccbaea4fSjosephb #else 13837c478bd9Sstevel@tonic-gate *(uintptr_t *)&_userlimit32 = _userlimit; 13847c478bd9Sstevel@tonic-gate #endif 13857c478bd9Sstevel@tonic-gate PRM_DEBUG(_kernelbase); 13867c478bd9Sstevel@tonic-gate PRM_DEBUG(_userlimit); 13877c478bd9Sstevel@tonic-gate PRM_DEBUG(_userlimit32); 13887c478bd9Sstevel@tonic-gate 138935b1ab99Sjosephb layout_kernel_va(); 139035b1ab99Sjosephb 139135b1ab99Sjosephb #if defined(__i386) 139235b1ab99Sjosephb /* 139335b1ab99Sjosephb * If segmap is too large we can push the bottom of the kernel heap 139435b1ab99Sjosephb * higher than the base. Or worse, it could exceed the top of the 139535b1ab99Sjosephb * VA space entirely, causing it to wrap around. 139635b1ab99Sjosephb */ 139735b1ab99Sjosephb if (kernelheap >= ekernelheap || (uintptr_t)kernelheap < kernelbase) 139835b1ab99Sjosephb panic("too little address space available for kernelheap," 139935b1ab99Sjosephb " use eeprom for lower kernelbase or smaller segmapsize"); 140035b1ab99Sjosephb #endif /* __i386 */ 140135b1ab99Sjosephb 14027c478bd9Sstevel@tonic-gate /* 14037c478bd9Sstevel@tonic-gate * Initialize the kernel heap. Note 3rd argument must be > 1st. 14047c478bd9Sstevel@tonic-gate */ 140535b1ab99Sjosephb kernelheap_init(kernelheap, ekernelheap, 140635b1ab99Sjosephb kernelheap + MMU_PAGESIZE, 1407ae115bc7Smrj (void *)core_base, (void *)(core_base + core_size)); 14087c478bd9Sstevel@tonic-gate 1409843e1988Sjohnlev #if defined(__xpv) 1410843e1988Sjohnlev /* 1411843e1988Sjohnlev * Link pending events struct into cpu struct 1412843e1988Sjohnlev */ 1413843e1988Sjohnlev CPU->cpu_m.mcpu_evt_pend = &cpu0_evt_data; 1414843e1988Sjohnlev #endif 14157c478bd9Sstevel@tonic-gate /* 14167c478bd9Sstevel@tonic-gate * Initialize kernel memory allocator. 14177c478bd9Sstevel@tonic-gate */ 14187c478bd9Sstevel@tonic-gate kmem_init(); 14197c478bd9Sstevel@tonic-gate 1420932dc8e5Sdp /* 1421932dc8e5Sdp * Factor in colorequiv to check additional 'equivalent' bins 1422932dc8e5Sdp */ 1423932dc8e5Sdp page_set_colorequiv_arr(); 1424932dc8e5Sdp 14257c478bd9Sstevel@tonic-gate /* 14267c478bd9Sstevel@tonic-gate * print this out early so that we know what's going on 14277c478bd9Sstevel@tonic-gate */ 1428a50a8b93SKuriakose Kuruvilla cmn_err(CE_CONT, fmt, x86_feature, FMT_X86_FEATURE); 14297c478bd9Sstevel@tonic-gate 14307c478bd9Sstevel@tonic-gate /* 14317c478bd9Sstevel@tonic-gate * Initialize bp_mapin(). 14327c478bd9Sstevel@tonic-gate */ 14337c478bd9Sstevel@tonic-gate bp_init(MMU_PAGESIZE, HAT_STORECACHING_OK); 14347c478bd9Sstevel@tonic-gate 1435f53ad214Skchow /* 1436f53ad214Skchow * orig_npages is non-zero if physmem has been configured for less 1437f53ad214Skchow * than the available memory. 1438f53ad214Skchow */ 1439f53ad214Skchow if (orig_npages) { 144035b1ab99Sjosephb cmn_err(CE_WARN, "!%slimiting physmem to 0x%lx of 0x%lx pages", 144135b1ab99Sjosephb (npages == PHYSMEM ? "Due to virtual address space " : ""), 144235b1ab99Sjosephb npages, orig_npages); 1443f53ad214Skchow } 14447c478bd9Sstevel@tonic-gate #if defined(__i386) 14457c478bd9Sstevel@tonic-gate if (eprom_kernelbase && (eprom_kernelbase != kernelbase)) 14467c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "kernelbase value, User specified 0x%lx, " 14477c478bd9Sstevel@tonic-gate "System using 0x%lx", 14487c478bd9Sstevel@tonic-gate (uintptr_t)eprom_kernelbase, (uintptr_t)kernelbase); 14497c478bd9Sstevel@tonic-gate #endif 14507c478bd9Sstevel@tonic-gate 14517c478bd9Sstevel@tonic-gate #ifdef KERNELBASE_ABI_MIN 14527c478bd9Sstevel@tonic-gate if (kernelbase < (uintptr_t)KERNELBASE_ABI_MIN) { 14537c478bd9Sstevel@tonic-gate cmn_err(CE_NOTE, "!kernelbase set to 0x%lx, system is not " 14547c478bd9Sstevel@tonic-gate "i386 ABI compliant.", (uintptr_t)kernelbase); 14557c478bd9Sstevel@tonic-gate } 14567c478bd9Sstevel@tonic-gate #endif 14577c478bd9Sstevel@tonic-gate 1458a3114836SGerry Liu #ifndef __xpv 1459a3114836SGerry Liu if (plat_dr_support_memory()) { 1460a3114836SGerry Liu mem_config_init(); 1461a3114836SGerry Liu } 1462a3114836SGerry Liu #else /* __xpv */ 1463843e1988Sjohnlev /* 1464843e1988Sjohnlev * Some of the xen start information has to be relocated up 1465843e1988Sjohnlev * into the kernel's permanent address space. 1466843e1988Sjohnlev */ 1467843e1988Sjohnlev PRM_POINT("calling xen_relocate_start_info()"); 1468843e1988Sjohnlev xen_relocate_start_info(); 1469843e1988Sjohnlev PRM_POINT("xen_relocate_start_info() done"); 1470843e1988Sjohnlev 1471843e1988Sjohnlev /* 1472843e1988Sjohnlev * (Update the vcpu pointer in our cpu structure to point into 1473843e1988Sjohnlev * the relocated shared info.) 1474843e1988Sjohnlev */ 1475843e1988Sjohnlev CPU->cpu_m.mcpu_vcpu_info = 1476843e1988Sjohnlev &HYPERVISOR_shared_info->vcpu_info[CPU->cpu_id]; 1477a3114836SGerry Liu #endif /* __xpv */ 1478843e1988Sjohnlev 1479ae115bc7Smrj PRM_POINT("startup_kmem() done"); 14807c478bd9Sstevel@tonic-gate } 14817c478bd9Sstevel@tonic-gate 1482551bc2a6Smrj #ifndef __xpv 1483551bc2a6Smrj /* 1484551bc2a6Smrj * If we have detected that we are running in an HVM environment, we need 1485551bc2a6Smrj * to prepend the PV driver directory to the module search path. 1486551bc2a6Smrj */ 1487551bc2a6Smrj #define HVM_MOD_DIR "/platform/i86hvm/kernel" 1488551bc2a6Smrj static void 1489551bc2a6Smrj update_default_path() 1490551bc2a6Smrj { 1491551bc2a6Smrj char *current, *newpath; 1492551bc2a6Smrj int newlen; 1493551bc2a6Smrj 1494551bc2a6Smrj /* 1495551bc2a6Smrj * We are about to resync with krtld. krtld will reset its 1496551bc2a6Smrj * internal module search path iff Solaris has set default_path. 1497551bc2a6Smrj * We want to be sure we're prepending this new directory to the 1498551bc2a6Smrj * right search path. 1499551bc2a6Smrj */ 1500551bc2a6Smrj current = (default_path == NULL) ? kobj_module_path : default_path; 1501551bc2a6Smrj 1502b6611e3bSStuart Maybee newlen = strlen(HVM_MOD_DIR) + strlen(current) + 2; 1503551bc2a6Smrj newpath = kmem_alloc(newlen, KM_SLEEP); 1504551bc2a6Smrj (void) strcpy(newpath, HVM_MOD_DIR); 1505551bc2a6Smrj (void) strcat(newpath, " "); 1506551bc2a6Smrj (void) strcat(newpath, current); 1507551bc2a6Smrj 1508551bc2a6Smrj default_path = newpath; 1509551bc2a6Smrj } 1510551bc2a6Smrj #endif 1511551bc2a6Smrj 15127c478bd9Sstevel@tonic-gate static void 15137c478bd9Sstevel@tonic-gate startup_modules(void) 15147c478bd9Sstevel@tonic-gate { 1515ed5289f9SKen Erickson int cnt; 15167c478bd9Sstevel@tonic-gate extern void prom_setup(void); 1517ed5289f9SKen Erickson int32_t v, h; 1518ed5289f9SKen Erickson char d[11]; 1519ed5289f9SKen Erickson char *cp; 1520e4b86885SCheng Sean Ye cmi_hdl_t hdl; 15217c478bd9Sstevel@tonic-gate 15227c478bd9Sstevel@tonic-gate PRM_POINT("startup_modules() starting..."); 1523843e1988Sjohnlev 1524843e1988Sjohnlev #ifndef __xpv 15257c478bd9Sstevel@tonic-gate /* 15267c478bd9Sstevel@tonic-gate * Initialize ten-micro second timer so that drivers will 15277c478bd9Sstevel@tonic-gate * not get short changed in their init phase. This was 15287c478bd9Sstevel@tonic-gate * not getting called until clkinit which, on fast cpu's 15297c478bd9Sstevel@tonic-gate * caused the drv_usecwait to be way too short. 15307c478bd9Sstevel@tonic-gate */ 15317c478bd9Sstevel@tonic-gate microfind(); 1532551bc2a6Smrj 1533b9bfdccdSStuart Maybee if (get_hwenv() == HW_XEN_HVM) 1534551bc2a6Smrj update_default_path(); 1535843e1988Sjohnlev #endif 15367c478bd9Sstevel@tonic-gate 15377c478bd9Sstevel@tonic-gate /* 15387c478bd9Sstevel@tonic-gate * Read the GMT lag from /etc/rtc_config. 15397c478bd9Sstevel@tonic-gate */ 1540ae115bc7Smrj sgmtl(process_rtc_config_file()); 15417c478bd9Sstevel@tonic-gate 15427c478bd9Sstevel@tonic-gate /* 15437c478bd9Sstevel@tonic-gate * Calculate default settings of system parameters based upon 15447c478bd9Sstevel@tonic-gate * maxusers, yet allow to be overridden via the /etc/system file. 15457c478bd9Sstevel@tonic-gate */ 15467c478bd9Sstevel@tonic-gate param_calc(0); 15477c478bd9Sstevel@tonic-gate 15487c478bd9Sstevel@tonic-gate mod_setup(); 15497c478bd9Sstevel@tonic-gate 15507c478bd9Sstevel@tonic-gate /* 15517c478bd9Sstevel@tonic-gate * Initialize system parameters. 15527c478bd9Sstevel@tonic-gate */ 15537c478bd9Sstevel@tonic-gate param_init(); 15547c478bd9Sstevel@tonic-gate 15559acbbeafSnn /* 15569acbbeafSnn * Initialize the default brands 15579acbbeafSnn */ 15589acbbeafSnn brand_init(); 15599acbbeafSnn 15607c478bd9Sstevel@tonic-gate /* 15617c478bd9Sstevel@tonic-gate * maxmem is the amount of physical memory we're playing with. 15627c478bd9Sstevel@tonic-gate */ 15637c478bd9Sstevel@tonic-gate maxmem = physmem; 15647c478bd9Sstevel@tonic-gate 15657c478bd9Sstevel@tonic-gate /* 15667c478bd9Sstevel@tonic-gate * Initialize segment management stuff. 15677c478bd9Sstevel@tonic-gate */ 15687c478bd9Sstevel@tonic-gate seg_init(); 15697c478bd9Sstevel@tonic-gate 15707c478bd9Sstevel@tonic-gate if (modload("fs", "specfs") == -1) 15717c478bd9Sstevel@tonic-gate halt("Can't load specfs"); 15727c478bd9Sstevel@tonic-gate 15737c478bd9Sstevel@tonic-gate if (modload("fs", "devfs") == -1) 15747c478bd9Sstevel@tonic-gate halt("Can't load devfs"); 15757c478bd9Sstevel@tonic-gate 1576facf4a8dSllai if (modload("fs", "dev") == -1) 1577facf4a8dSllai halt("Can't load dev"); 1578facf4a8dSllai 157935a5a358SJonathan Adams if (modload("fs", "procfs") == -1) 158035a5a358SJonathan Adams halt("Can't load procfs"); 158135a5a358SJonathan Adams 158245916cd2Sjpk (void) modloadonly("sys", "lbl_edition"); 158345916cd2Sjpk 15847c478bd9Sstevel@tonic-gate dispinit(); 15857c478bd9Sstevel@tonic-gate 15867c478bd9Sstevel@tonic-gate /* 15877c478bd9Sstevel@tonic-gate * This is needed here to initialize hw_serial[] for cluster booting. 15887c478bd9Sstevel@tonic-gate */ 15895679c89fSjv if ((h = set_soft_hostid()) == HW_INVALID_HOSTID) { 1590ed5289f9SKen Erickson cmn_err(CE_WARN, "Unable to set hostid"); 15915679c89fSjv } else { 1592ed5289f9SKen Erickson for (v = h, cnt = 0; cnt < 10; cnt++) { 15935679c89fSjv d[cnt] = (char)(v % 10); 1594ed5289f9SKen Erickson v /= 10; 1595ed5289f9SKen Erickson if (v == 0) 1596ed5289f9SKen Erickson break; 1597ed5289f9SKen Erickson } 1598ed5289f9SKen Erickson for (cp = hw_serial; cnt >= 0; cnt--) 1599ed5289f9SKen Erickson *cp++ = d[cnt] + '0'; 1600ed5289f9SKen Erickson *cp = 0; 1601ed5289f9SKen Erickson } 16027c478bd9Sstevel@tonic-gate 16037c478bd9Sstevel@tonic-gate /* Read cluster configuration data. */ 16047c478bd9Sstevel@tonic-gate clconf_init(); 16057c478bd9Sstevel@tonic-gate 1606843e1988Sjohnlev #if defined(__xpv) 1607349b53ddSStuart Maybee (void) ec_init(); 1608843e1988Sjohnlev gnttab_init(); 1609843e1988Sjohnlev (void) xs_early_init(); 1610843e1988Sjohnlev #endif /* __xpv */ 1611843e1988Sjohnlev 16127c478bd9Sstevel@tonic-gate /* 16137c478bd9Sstevel@tonic-gate * Create a kernel device tree. First, create rootnex and 16147c478bd9Sstevel@tonic-gate * then invoke bus specific code to probe devices. 16157c478bd9Sstevel@tonic-gate */ 16167c478bd9Sstevel@tonic-gate setup_ddi(); 16177aec1d6eScindi 1618074bb90dSTom Pothier #ifdef __xpv 1619074bb90dSTom Pothier if (DOMAIN_IS_INITDOMAIN(xen_info)) 1620074bb90dSTom Pothier #endif 1621074bb90dSTom Pothier { 1622074bb90dSTom Pothier /* 1623074bb90dSTom Pothier * Load the System Management BIOS into the global ksmbios 1624074bb90dSTom Pothier * handle, if an SMBIOS is present on this system. 1625074bb90dSTom Pothier */ 1626074bb90dSTom Pothier ksmbios = smbios_open(NULL, SMB_VERSION, ksmbios_flags, NULL); 1627074bb90dSTom Pothier } 1628074bb90dSTom Pothier 1629074bb90dSTom Pothier 1630e4b86885SCheng Sean Ye /* 1631e4b86885SCheng Sean Ye * Set up the CPU module subsystem for the boot cpu in the native 1632e4b86885SCheng Sean Ye * case, and all physical cpu resource in the xpv dom0 case. 1633e4b86885SCheng Sean Ye * Modifies the device tree, so this must be done after 1634e4b86885SCheng Sean Ye * setup_ddi(). 1635e4b86885SCheng Sean Ye */ 1636e4b86885SCheng Sean Ye #ifdef __xpv 1637e4b86885SCheng Sean Ye /* 1638e4b86885SCheng Sean Ye * If paravirtualized and on dom0 then we initialize all physical 1639e4b86885SCheng Sean Ye * cpu handles now; if paravirtualized on a domU then do not 1640e4b86885SCheng Sean Ye * initialize. 1641e4b86885SCheng Sean Ye */ 1642e4b86885SCheng Sean Ye if (DOMAIN_IS_INITDOMAIN(xen_info)) { 1643e4b86885SCheng Sean Ye xen_mc_lcpu_cookie_t cpi; 1644e4b86885SCheng Sean Ye 1645e4b86885SCheng Sean Ye for (cpi = xen_physcpu_next(NULL); cpi != NULL; 1646e4b86885SCheng Sean Ye cpi = xen_physcpu_next(cpi)) { 1647e4b86885SCheng Sean Ye if ((hdl = cmi_init(CMI_HDL_SOLARIS_xVM_MCA, 1648e4b86885SCheng Sean Ye xen_physcpu_chipid(cpi), xen_physcpu_coreid(cpi), 1649e4b86885SCheng Sean Ye xen_physcpu_strandid(cpi))) != NULL && 1650e4b86885SCheng Sean Ye (x86_feature & X86_MCA)) 165120c794b3Sgavinm cmi_mca_init(hdl); 165220c794b3Sgavinm } 165320c794b3Sgavinm } 1654e4b86885SCheng Sean Ye #else 1655e4b86885SCheng Sean Ye /* 1656e4b86885SCheng Sean Ye * Initialize a handle for the boot cpu - others will initialize 1657e4b86885SCheng Sean Ye * as they startup. Do not do this if we know we are in an HVM domU. 1658e4b86885SCheng Sean Ye */ 1659b9bfdccdSStuart Maybee if ((get_hwenv() != HW_XEN_HVM) && 1660e4b86885SCheng Sean Ye (hdl = cmi_init(CMI_HDL_NATIVE, cmi_ntv_hwchipid(CPU), 1661e4b86885SCheng Sean Ye cmi_ntv_hwcoreid(CPU), cmi_ntv_hwstrandid(CPU))) != NULL && 1662a3114836SGerry Liu (x86_feature & X86_MCA)) { 1663e4b86885SCheng Sean Ye cmi_mca_init(hdl); 1664a3114836SGerry Liu CPU->cpu_m.mcpu_cmi_hdl = hdl; 1665a3114836SGerry Liu } 166620c794b3Sgavinm #endif /* __xpv */ 16677aec1d6eScindi 16687c478bd9Sstevel@tonic-gate /* 16697c478bd9Sstevel@tonic-gate * Fake a prom tree such that /dev/openprom continues to work 16707c478bd9Sstevel@tonic-gate */ 1671ae115bc7Smrj PRM_POINT("startup_modules: calling prom_setup..."); 16727c478bd9Sstevel@tonic-gate prom_setup(); 1673ae115bc7Smrj PRM_POINT("startup_modules: done"); 16747c478bd9Sstevel@tonic-gate 16757c478bd9Sstevel@tonic-gate /* 16767c478bd9Sstevel@tonic-gate * Load all platform specific modules 16777c478bd9Sstevel@tonic-gate */ 1678ae115bc7Smrj PRM_POINT("startup_modules: calling psm_modload..."); 16797c478bd9Sstevel@tonic-gate psm_modload(); 16807c478bd9Sstevel@tonic-gate 16817c478bd9Sstevel@tonic-gate PRM_POINT("startup_modules() done"); 16827c478bd9Sstevel@tonic-gate } 16837c478bd9Sstevel@tonic-gate 1684ae115bc7Smrj /* 1685ae115bc7Smrj * claim a "setaside" boot page for use in the kernel 1686ae115bc7Smrj */ 1687ae115bc7Smrj page_t * 1688ae115bc7Smrj boot_claim_page(pfn_t pfn) 16897c478bd9Sstevel@tonic-gate { 1690ae115bc7Smrj page_t *pp; 16917c478bd9Sstevel@tonic-gate 1692ae115bc7Smrj pp = page_numtopp_nolock(pfn); 1693ae115bc7Smrj ASSERT(pp != NULL); 16947c478bd9Sstevel@tonic-gate 1695ae115bc7Smrj if (PP_ISBOOTPAGES(pp)) { 1696ae115bc7Smrj if (pp->p_next != NULL) 1697ae115bc7Smrj pp->p_next->p_prev = pp->p_prev; 1698ae115bc7Smrj if (pp->p_prev == NULL) 1699ae115bc7Smrj bootpages = pp->p_next; 1700ae115bc7Smrj else 1701ae115bc7Smrj pp->p_prev->p_next = pp->p_next; 1702ae115bc7Smrj } else { 1703ae115bc7Smrj /* 1704ae115bc7Smrj * htable_attach() expects a base pagesize page 1705ae115bc7Smrj */ 1706ae115bc7Smrj if (pp->p_szc != 0) 1707ae115bc7Smrj page_boot_demote(pp); 1708ae115bc7Smrj pp = page_numtopp(pfn, SE_EXCL); 1709ae115bc7Smrj } 1710ae115bc7Smrj return (pp); 17117c478bd9Sstevel@tonic-gate } 17127c478bd9Sstevel@tonic-gate 17137c478bd9Sstevel@tonic-gate /* 17147c478bd9Sstevel@tonic-gate * Walk through the pagetables looking for pages mapped in by boot. If the 17157c478bd9Sstevel@tonic-gate * setaside flag is set the pages are expected to be returned to the 17167c478bd9Sstevel@tonic-gate * kernel later in boot, so we add them to the bootpages list. 17177c478bd9Sstevel@tonic-gate */ 17187c478bd9Sstevel@tonic-gate static void 17197c478bd9Sstevel@tonic-gate protect_boot_range(uintptr_t low, uintptr_t high, int setaside) 17207c478bd9Sstevel@tonic-gate { 17217c478bd9Sstevel@tonic-gate uintptr_t va = low; 17227c478bd9Sstevel@tonic-gate size_t len; 17237c478bd9Sstevel@tonic-gate uint_t prot; 17247c478bd9Sstevel@tonic-gate pfn_t pfn; 17257c478bd9Sstevel@tonic-gate page_t *pp; 17267c478bd9Sstevel@tonic-gate pgcnt_t boot_protect_cnt = 0; 17277c478bd9Sstevel@tonic-gate 1728ae115bc7Smrj while (kbm_probe(&va, &len, &pfn, &prot) != 0 && va < high) { 17297c478bd9Sstevel@tonic-gate if (va + len >= high) 17307c478bd9Sstevel@tonic-gate panic("0x%lx byte mapping at 0x%p exceeds boot's " 17317c478bd9Sstevel@tonic-gate "legal range.", len, (void *)va); 17327c478bd9Sstevel@tonic-gate 17337c478bd9Sstevel@tonic-gate while (len > 0) { 17347c478bd9Sstevel@tonic-gate pp = page_numtopp_alloc(pfn); 17357c478bd9Sstevel@tonic-gate if (pp != NULL) { 17367c478bd9Sstevel@tonic-gate if (setaside == 0) 17377c478bd9Sstevel@tonic-gate panic("Unexpected mapping by boot. " 17387c478bd9Sstevel@tonic-gate "addr=%p pfn=%lx\n", 17397c478bd9Sstevel@tonic-gate (void *)va, pfn); 17407c478bd9Sstevel@tonic-gate 17417c478bd9Sstevel@tonic-gate pp->p_next = bootpages; 1742ae115bc7Smrj pp->p_prev = NULL; 1743ae115bc7Smrj PP_SETBOOTPAGES(pp); 1744ae115bc7Smrj if (bootpages != NULL) { 1745ae115bc7Smrj bootpages->p_prev = pp; 1746ae115bc7Smrj } 17477c478bd9Sstevel@tonic-gate bootpages = pp; 17487c478bd9Sstevel@tonic-gate ++boot_protect_cnt; 17497c478bd9Sstevel@tonic-gate } 17507c478bd9Sstevel@tonic-gate 17517c478bd9Sstevel@tonic-gate ++pfn; 17527c478bd9Sstevel@tonic-gate len -= MMU_PAGESIZE; 17537c478bd9Sstevel@tonic-gate va += MMU_PAGESIZE; 17547c478bd9Sstevel@tonic-gate } 17557c478bd9Sstevel@tonic-gate } 17567c478bd9Sstevel@tonic-gate PRM_DEBUG(boot_protect_cnt); 17577c478bd9Sstevel@tonic-gate } 17587c478bd9Sstevel@tonic-gate 1759ae115bc7Smrj /* 176035b1ab99Sjosephb * 1761ae115bc7Smrj */ 17627c478bd9Sstevel@tonic-gate static void 176335b1ab99Sjosephb layout_kernel_va(void) 17647c478bd9Sstevel@tonic-gate { 176535b1ab99Sjosephb PRM_POINT("layout_kernel_va() starting..."); 17667c478bd9Sstevel@tonic-gate /* 1767ae115bc7Smrj * Establish the final size of the kernel's heap, size of segmap, 1768ae115bc7Smrj * segkp, etc. 17697c478bd9Sstevel@tonic-gate */ 17707c478bd9Sstevel@tonic-gate 17717c478bd9Sstevel@tonic-gate #if defined(__amd64) 17727c478bd9Sstevel@tonic-gate 177335b1ab99Sjosephb kpm_vbase = (caddr_t)segkpm_base; 1774a3114836SGerry Liu if (physmax + 1 < plat_dr_physmax) { 1775a3114836SGerry Liu kpm_size = ROUND_UP_LPAGE(mmu_ptob(plat_dr_physmax)); 1776a3114836SGerry Liu } else { 1777a3114836SGerry Liu kpm_size = ROUND_UP_LPAGE(mmu_ptob(physmax + 1)); 1778a3114836SGerry Liu } 177935b1ab99Sjosephb if ((uintptr_t)kpm_vbase + kpm_size > (uintptr_t)valloc_base) 178035b1ab99Sjosephb panic("not enough room for kpm!"); 1781ae115bc7Smrj PRM_DEBUG(kpm_size); 1782ae115bc7Smrj PRM_DEBUG(kpm_vbase); 17837c478bd9Sstevel@tonic-gate 17847c478bd9Sstevel@tonic-gate /* 1785ae115bc7Smrj * By default we create a seg_kp in 64 bit kernels, it's a little 1786ae115bc7Smrj * faster to access than embedding it in the heap. 17877c478bd9Sstevel@tonic-gate */ 1788ae115bc7Smrj segkp_base = (caddr_t)valloc_base + valloc_sz; 17897c478bd9Sstevel@tonic-gate if (!segkp_fromheap) { 17907c478bd9Sstevel@tonic-gate size_t sz = mmu_ptob(segkpsize); 17917c478bd9Sstevel@tonic-gate 17927c478bd9Sstevel@tonic-gate /* 1793ae115bc7Smrj * determine size of segkp 17947c478bd9Sstevel@tonic-gate */ 17957c478bd9Sstevel@tonic-gate if (sz < SEGKPMINSIZE || sz > SEGKPMAXSIZE) { 17967c478bd9Sstevel@tonic-gate sz = SEGKPDEFSIZE; 17977c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "!Illegal value for segkpsize. " 17987c478bd9Sstevel@tonic-gate "segkpsize has been reset to %ld pages", 17997c478bd9Sstevel@tonic-gate mmu_btop(sz)); 18007c478bd9Sstevel@tonic-gate } 18017c478bd9Sstevel@tonic-gate sz = MIN(sz, MAX(SEGKPMINSIZE, mmu_ptob(physmem))); 18027c478bd9Sstevel@tonic-gate 18037c478bd9Sstevel@tonic-gate segkpsize = mmu_btop(ROUND_UP_LPAGE(sz)); 18047c478bd9Sstevel@tonic-gate } 1805ae115bc7Smrj PRM_DEBUG(segkp_base); 1806ae115bc7Smrj PRM_DEBUG(segkpsize); 18077c478bd9Sstevel@tonic-gate 180835b1ab99Sjosephb /* 180935b1ab99Sjosephb * segzio is used for ZFS cached data. It uses a distinct VA 181035b1ab99Sjosephb * segment (from kernel heap) so that we can easily tell not to 181135b1ab99Sjosephb * include it in kernel crash dumps on 64 bit kernels. The trick is 181235b1ab99Sjosephb * to give it lots of VA, but not constrain the kernel heap. 181335b1ab99Sjosephb * We scale the size of segzio linearly with physmem up to 181435b1ab99Sjosephb * SEGZIOMAXSIZE. Above that amount it scales at 50% of physmem. 181535b1ab99Sjosephb */ 1816ae115bc7Smrj segzio_base = segkp_base + mmu_ptob(segkpsize); 1817ae115bc7Smrj if (segzio_fromheap) { 1818ae115bc7Smrj segziosize = 0; 1819ae115bc7Smrj } else { 182035b1ab99Sjosephb size_t physmem_size = mmu_ptob(physmem); 182135b1ab99Sjosephb size_t size = (segziosize == 0) ? 182235b1ab99Sjosephb physmem_size : mmu_ptob(segziosize); 1823ad23a2dbSjohansen 182435b1ab99Sjosephb if (size < SEGZIOMINSIZE) 1825ad23a2dbSjohansen size = SEGZIOMINSIZE; 182635b1ab99Sjosephb if (size > SEGZIOMAXSIZE) { 1827a778305aSjohansen size = SEGZIOMAXSIZE; 182835b1ab99Sjosephb if (physmem_size > size) 182935b1ab99Sjosephb size += (physmem_size - size) / 2; 1830ad23a2dbSjohansen } 1831ad23a2dbSjohansen segziosize = mmu_btop(ROUND_UP_LPAGE(size)); 1832ad23a2dbSjohansen } 1833ae115bc7Smrj PRM_DEBUG(segziosize); 1834ae115bc7Smrj PRM_DEBUG(segzio_base); 1835ad23a2dbSjohansen 18367c478bd9Sstevel@tonic-gate /* 1837ae115bc7Smrj * Put the range of VA for device mappings next, kmdb knows to not 1838ae115bc7Smrj * grep in this range of addresses. 18397c478bd9Sstevel@tonic-gate */ 1840ae115bc7Smrj toxic_addr = 1841ae115bc7Smrj ROUND_UP_LPAGE((uintptr_t)segzio_base + mmu_ptob(segziosize)); 18427c478bd9Sstevel@tonic-gate PRM_DEBUG(toxic_addr); 1843ae115bc7Smrj segmap_start = ROUND_UP_LPAGE(toxic_addr + toxic_size); 1844ae115bc7Smrj #else /* __i386 */ 1845ae115bc7Smrj segmap_start = ROUND_UP_LPAGE(kernelbase); 1846ae115bc7Smrj #endif /* __i386 */ 1847ae115bc7Smrj PRM_DEBUG(segmap_start); 18487c478bd9Sstevel@tonic-gate 18497c478bd9Sstevel@tonic-gate /* 185052eef812SAmrita Sadhukhan * Users can change segmapsize through eeprom. If the variable 185152eef812SAmrita Sadhukhan * is tuned through eeprom, there is no upper bound on the 1852ed5289f9SKen Erickson * size of segmap. 18537c478bd9Sstevel@tonic-gate */ 18547c478bd9Sstevel@tonic-gate segmapsize = MAX(ROUND_UP_LPAGE(segmapsize), SEGMAPDEFAULT); 18557c478bd9Sstevel@tonic-gate 18567c478bd9Sstevel@tonic-gate #if defined(__i386) 18577c478bd9Sstevel@tonic-gate /* 18587c478bd9Sstevel@tonic-gate * 32-bit systems don't have segkpm or segkp, so segmap appears at 18597c478bd9Sstevel@tonic-gate * the bottom of the kernel's address range. Set aside space for a 1860ae115bc7Smrj * small red zone just below the start of segmap. 18617c478bd9Sstevel@tonic-gate */ 1862ae115bc7Smrj segmap_start += KERNEL_REDZONE_SIZE; 18637c478bd9Sstevel@tonic-gate segmapsize -= KERNEL_REDZONE_SIZE; 18647c478bd9Sstevel@tonic-gate #endif 18657c478bd9Sstevel@tonic-gate 1866ae115bc7Smrj PRM_DEBUG(segmap_start); 18677c478bd9Sstevel@tonic-gate PRM_DEBUG(segmapsize); 186835b1ab99Sjosephb kernelheap = (caddr_t)ROUND_UP_LPAGE(segmap_start + segmapsize); 186935b1ab99Sjosephb PRM_DEBUG(kernelheap); 187035b1ab99Sjosephb PRM_POINT("layout_kernel_va() done..."); 187135b1ab99Sjosephb } 187235b1ab99Sjosephb 187335b1ab99Sjosephb /* 187435b1ab99Sjosephb * Finish initializing the VM system, now that we are no longer 187535b1ab99Sjosephb * relying on the boot time memory allocators. 187635b1ab99Sjosephb */ 187735b1ab99Sjosephb static void 187835b1ab99Sjosephb startup_vm(void) 187935b1ab99Sjosephb { 188035b1ab99Sjosephb struct segmap_crargs a; 188135b1ab99Sjosephb 188235b1ab99Sjosephb extern int use_brk_lpg, use_stk_lpg; 188335b1ab99Sjosephb 188435b1ab99Sjosephb PRM_POINT("startup_vm() starting..."); 188535b1ab99Sjosephb 188635b1ab99Sjosephb /* 188735b1ab99Sjosephb * Initialize the hat layer. 188835b1ab99Sjosephb */ 188935b1ab99Sjosephb hat_init(); 18907c478bd9Sstevel@tonic-gate 1891ae115bc7Smrj /* 1892ae115bc7Smrj * Do final allocations of HAT data structures that need to 1893ae115bc7Smrj * be allocated before quiescing the boot loader. 1894ae115bc7Smrj */ 1895ae115bc7Smrj PRM_POINT("Calling hat_kern_alloc()..."); 1896ae115bc7Smrj hat_kern_alloc((caddr_t)segmap_start, segmapsize, ekernelheap); 1897ae115bc7Smrj PRM_POINT("hat_kern_alloc() done"); 1898ae115bc7Smrj 1899843e1988Sjohnlev #ifndef __xpv 1900ae115bc7Smrj /* 19011d03c31eSjohnlev * Setup Page Attribute Table 1902ae115bc7Smrj */ 19031d03c31eSjohnlev pat_sync(); 1904843e1988Sjohnlev #endif 1905ae115bc7Smrj 1906ae115bc7Smrj /* 1907ae115bc7Smrj * The next two loops are done in distinct steps in order 1908ae115bc7Smrj * to be sure that any page that is doubly mapped (both above 1909ae115bc7Smrj * KERNEL_TEXT and below kernelbase) is dealt with correctly. 1910ae115bc7Smrj * Note this may never happen, but it might someday. 1911ae115bc7Smrj */ 1912ae115bc7Smrj bootpages = NULL; 1913ae115bc7Smrj PRM_POINT("Protecting boot pages"); 1914ae115bc7Smrj 1915ae115bc7Smrj /* 1916ae115bc7Smrj * Protect any pages mapped above KERNEL_TEXT that somehow have 1917ae115bc7Smrj * page_t's. This can only happen if something weird allocated 1918ae115bc7Smrj * in this range (like kadb/kmdb). 1919ae115bc7Smrj */ 1920ae115bc7Smrj protect_boot_range(KERNEL_TEXT, (uintptr_t)-1, 0); 1921ae115bc7Smrj 1922ae115bc7Smrj /* 1923ae115bc7Smrj * Before we can take over memory allocation/mapping from the boot 1924ae115bc7Smrj * loader we must remove from our free page lists any boot allocated 1925ae115bc7Smrj * pages that stay mapped until release_bootstrap(). 1926ae115bc7Smrj */ 1927ae115bc7Smrj protect_boot_range(0, kernelbase, 1); 1928ae115bc7Smrj 1929843e1988Sjohnlev 1930ae115bc7Smrj /* 1931ae115bc7Smrj * Switch to running on regular HAT (not boot_mmu) 1932ae115bc7Smrj */ 1933ae115bc7Smrj PRM_POINT("Calling hat_kern_setup()..."); 1934ae115bc7Smrj hat_kern_setup(); 1935ae115bc7Smrj 1936ae115bc7Smrj /* 1937ae115bc7Smrj * It is no longer safe to call BOP_ALLOC(), so make sure we don't. 1938ae115bc7Smrj */ 1939ae115bc7Smrj bop_no_more_mem(); 1940ae115bc7Smrj 1941ae115bc7Smrj PRM_POINT("hat_kern_setup() done"); 1942ae115bc7Smrj 1943ae115bc7Smrj hat_cpu_online(CPU); 1944ae115bc7Smrj 19457c478bd9Sstevel@tonic-gate /* 19467c478bd9Sstevel@tonic-gate * Initialize VM system 19477c478bd9Sstevel@tonic-gate */ 19487c478bd9Sstevel@tonic-gate PRM_POINT("Calling kvm_init()..."); 19497c478bd9Sstevel@tonic-gate kvm_init(); 19507c478bd9Sstevel@tonic-gate PRM_POINT("kvm_init() done"); 19517c478bd9Sstevel@tonic-gate 19527c478bd9Sstevel@tonic-gate /* 19537c478bd9Sstevel@tonic-gate * Tell kmdb that the VM system is now working 19547c478bd9Sstevel@tonic-gate */ 19557c478bd9Sstevel@tonic-gate if (boothowto & RB_DEBUG) 19567c478bd9Sstevel@tonic-gate kdi_dvec_vmready(); 19577c478bd9Sstevel@tonic-gate 1958843e1988Sjohnlev #if defined(__xpv) 1959843e1988Sjohnlev /* 1960843e1988Sjohnlev * Populate the I/O pool on domain 0 1961843e1988Sjohnlev */ 1962843e1988Sjohnlev if (DOMAIN_IS_INITDOMAIN(xen_info)) { 1963843e1988Sjohnlev extern long populate_io_pool(void); 1964843e1988Sjohnlev long init_io_pool_cnt; 1965843e1988Sjohnlev 1966843e1988Sjohnlev PRM_POINT("Populating reserve I/O page pool"); 1967843e1988Sjohnlev init_io_pool_cnt = populate_io_pool(); 1968843e1988Sjohnlev PRM_DEBUG(init_io_pool_cnt); 1969843e1988Sjohnlev } 1970843e1988Sjohnlev #endif 19717c478bd9Sstevel@tonic-gate /* 19727c478bd9Sstevel@tonic-gate * Mangle the brand string etc. 19737c478bd9Sstevel@tonic-gate */ 19747c478bd9Sstevel@tonic-gate cpuid_pass3(CPU); 19757c478bd9Sstevel@tonic-gate 19767c478bd9Sstevel@tonic-gate #if defined(__amd64) 19777c478bd9Sstevel@tonic-gate 19787c478bd9Sstevel@tonic-gate /* 19797c478bd9Sstevel@tonic-gate * Create the device arena for toxic (to dtrace/kmdb) mappings. 19807c478bd9Sstevel@tonic-gate */ 19817c478bd9Sstevel@tonic-gate device_arena = vmem_create("device", (void *)toxic_addr, 19827c478bd9Sstevel@tonic-gate toxic_size, MMU_PAGESIZE, NULL, NULL, NULL, 0, VM_SLEEP); 19837c478bd9Sstevel@tonic-gate 19847c478bd9Sstevel@tonic-gate #else /* __i386 */ 19857c478bd9Sstevel@tonic-gate 19867c478bd9Sstevel@tonic-gate /* 19877c478bd9Sstevel@tonic-gate * allocate the bit map that tracks toxic pages 19887c478bd9Sstevel@tonic-gate */ 1989ae115bc7Smrj toxic_bit_map_len = btop((ulong_t)(valloc_base - kernelbase)); 19907c478bd9Sstevel@tonic-gate PRM_DEBUG(toxic_bit_map_len); 19917c478bd9Sstevel@tonic-gate toxic_bit_map = 19927c478bd9Sstevel@tonic-gate kmem_zalloc(BT_SIZEOFMAP(toxic_bit_map_len), KM_NOSLEEP); 19937c478bd9Sstevel@tonic-gate ASSERT(toxic_bit_map != NULL); 19947c478bd9Sstevel@tonic-gate PRM_DEBUG(toxic_bit_map); 19957c478bd9Sstevel@tonic-gate 19967c478bd9Sstevel@tonic-gate #endif /* __i386 */ 19977c478bd9Sstevel@tonic-gate 19987c478bd9Sstevel@tonic-gate 19997c478bd9Sstevel@tonic-gate /* 20007c478bd9Sstevel@tonic-gate * Now that we've got more VA, as well as the ability to allocate from 20017c478bd9Sstevel@tonic-gate * it, tell the debugger. 20027c478bd9Sstevel@tonic-gate */ 20037c478bd9Sstevel@tonic-gate if (boothowto & RB_DEBUG) 20047c478bd9Sstevel@tonic-gate kdi_dvec_memavail(); 20057c478bd9Sstevel@tonic-gate 20067c478bd9Sstevel@tonic-gate /* 20077c478bd9Sstevel@tonic-gate * The following code installs a special page fault handler (#pf) 20087c478bd9Sstevel@tonic-gate * to work around a pentium bug. 20097c478bd9Sstevel@tonic-gate */ 2010843e1988Sjohnlev #if !defined(__amd64) && !defined(__xpv) 20117c478bd9Sstevel@tonic-gate if (x86_type == X86_TYPE_P5) { 2012ae115bc7Smrj desctbr_t idtr; 20137c478bd9Sstevel@tonic-gate gate_desc_t *newidt; 20147c478bd9Sstevel@tonic-gate 20157c478bd9Sstevel@tonic-gate if ((newidt = kmem_zalloc(MMU_PAGESIZE, KM_NOSLEEP)) == NULL) 20167c478bd9Sstevel@tonic-gate panic("failed to install pentium_pftrap"); 20177c478bd9Sstevel@tonic-gate 20180cfdb603Sjosephb bcopy(idt0, newidt, NIDT * sizeof (*idt0)); 20197c478bd9Sstevel@tonic-gate set_gatesegd(&newidt[T_PGFLT], &pentium_pftrap, 20209844da31SSeth Goldberg KCS_SEL, SDT_SYSIGT, TRP_KPL, 0); 20217c478bd9Sstevel@tonic-gate 20227c478bd9Sstevel@tonic-gate (void) as_setprot(&kas, (caddr_t)newidt, MMU_PAGESIZE, 20230cfdb603Sjosephb PROT_READ | PROT_EXEC); 20247c478bd9Sstevel@tonic-gate 20250cfdb603Sjosephb CPU->cpu_idt = newidt; 20260cfdb603Sjosephb idtr.dtr_base = (uintptr_t)CPU->cpu_idt; 20270cfdb603Sjosephb idtr.dtr_limit = (NIDT * sizeof (*idt0)) - 1; 2028ae115bc7Smrj wr_idtr(&idtr); 20297c478bd9Sstevel@tonic-gate } 20307c478bd9Sstevel@tonic-gate #endif /* !__amd64 */ 20317c478bd9Sstevel@tonic-gate 2032843e1988Sjohnlev #if !defined(__xpv) 20337c478bd9Sstevel@tonic-gate /* 20347c478bd9Sstevel@tonic-gate * Map page pfn=0 for drivers, such as kd, that need to pick up 20357c478bd9Sstevel@tonic-gate * parameters left there by controllers/BIOS. 20367c478bd9Sstevel@tonic-gate */ 20377c478bd9Sstevel@tonic-gate PRM_POINT("setup up p0_va"); 20387c478bd9Sstevel@tonic-gate p0_va = i86devmap(0, 1, PROT_READ); 20397c478bd9Sstevel@tonic-gate PRM_DEBUG(p0_va); 2040843e1988Sjohnlev #endif 20417c478bd9Sstevel@tonic-gate 20427c478bd9Sstevel@tonic-gate cmn_err(CE_CONT, "?mem = %luK (0x%lx)\n", 20437c478bd9Sstevel@tonic-gate physinstalled << (MMU_PAGESHIFT - 10), ptob(physinstalled)); 20447c478bd9Sstevel@tonic-gate 204583f9b804Skchow /* 204683f9b804Skchow * disable automatic large pages for small memory systems or 204783f9b804Skchow * when the disable flag is set. 204802bc52beSkchow * 204902bc52beSkchow * Do not yet consider page sizes larger than 2m/4m. 205083f9b804Skchow */ 2051ec25b48fSsusans if (!auto_lpg_disable && mmu.max_page_level > 0) { 2052ec25b48fSsusans max_uheap_lpsize = LEVEL_SIZE(1); 2053ec25b48fSsusans max_ustack_lpsize = LEVEL_SIZE(1); 2054ec25b48fSsusans max_privmap_lpsize = LEVEL_SIZE(1); 2055ec25b48fSsusans max_uidata_lpsize = LEVEL_SIZE(1); 2056ec25b48fSsusans max_utext_lpsize = LEVEL_SIZE(1); 2057ec25b48fSsusans max_shm_lpsize = LEVEL_SIZE(1); 2058ec25b48fSsusans } 2059ec25b48fSsusans if (physmem < privm_lpg_min_physmem || mmu.max_page_level == 0 || 2060ec25b48fSsusans auto_lpg_disable) { 2061beb1bda0Sdavemq use_brk_lpg = 0; 2062beb1bda0Sdavemq use_stk_lpg = 0; 2063ec25b48fSsusans } 206402bc52beSkchow mcntl0_lpsize = LEVEL_SIZE(mmu.umax_page_level); 2065beb1bda0Sdavemq 20667c478bd9Sstevel@tonic-gate PRM_POINT("Calling hat_init_finish()..."); 20677c478bd9Sstevel@tonic-gate hat_init_finish(); 20687c478bd9Sstevel@tonic-gate PRM_POINT("hat_init_finish() done"); 20697c478bd9Sstevel@tonic-gate 20707c478bd9Sstevel@tonic-gate /* 20717c478bd9Sstevel@tonic-gate * Initialize the segkp segment type. 20727c478bd9Sstevel@tonic-gate */ 20737c478bd9Sstevel@tonic-gate rw_enter(&kas.a_lock, RW_WRITER); 2074ae115bc7Smrj PRM_POINT("Attaching segkp"); 2075ae115bc7Smrj if (segkp_fromheap) { 20767c478bd9Sstevel@tonic-gate segkp->s_as = &kas; 2077ae115bc7Smrj } else if (seg_attach(&kas, (caddr_t)segkp_base, mmu_ptob(segkpsize), 2078ae115bc7Smrj segkp) < 0) { 2079ae115bc7Smrj panic("startup: cannot attach segkp"); 2080ae115bc7Smrj /*NOTREACHED*/ 20817c478bd9Sstevel@tonic-gate } 2082ae115bc7Smrj PRM_POINT("Doing segkp_create()"); 20837c478bd9Sstevel@tonic-gate if (segkp_create(segkp) != 0) { 20847c478bd9Sstevel@tonic-gate panic("startup: segkp_create failed"); 20857c478bd9Sstevel@tonic-gate /*NOTREACHED*/ 20867c478bd9Sstevel@tonic-gate } 20877c478bd9Sstevel@tonic-gate PRM_DEBUG(segkp); 20887c478bd9Sstevel@tonic-gate rw_exit(&kas.a_lock); 20897c478bd9Sstevel@tonic-gate 20907c478bd9Sstevel@tonic-gate /* 20917c478bd9Sstevel@tonic-gate * kpm segment 20927c478bd9Sstevel@tonic-gate */ 20937c478bd9Sstevel@tonic-gate segmap_kpm = 0; 20947c478bd9Sstevel@tonic-gate if (kpm_desired) { 20957c478bd9Sstevel@tonic-gate kpm_init(); 20967c478bd9Sstevel@tonic-gate kpm_enable = 1; 20977c478bd9Sstevel@tonic-gate } 20987c478bd9Sstevel@tonic-gate 20997c478bd9Sstevel@tonic-gate /* 21007c478bd9Sstevel@tonic-gate * Now create segmap segment. 21017c478bd9Sstevel@tonic-gate */ 21027c478bd9Sstevel@tonic-gate rw_enter(&kas.a_lock, RW_WRITER); 2103ae115bc7Smrj if (seg_attach(&kas, (caddr_t)segmap_start, segmapsize, segmap) < 0) { 2104ae115bc7Smrj panic("cannot attach segmap"); 21057c478bd9Sstevel@tonic-gate /*NOTREACHED*/ 21067c478bd9Sstevel@tonic-gate } 2107ae115bc7Smrj PRM_DEBUG(segmap); 21087c478bd9Sstevel@tonic-gate 21097c478bd9Sstevel@tonic-gate a.prot = PROT_READ | PROT_WRITE; 21107c478bd9Sstevel@tonic-gate a.shmsize = 0; 21117c478bd9Sstevel@tonic-gate a.nfreelist = segmapfreelists; 21127c478bd9Sstevel@tonic-gate 2113ae115bc7Smrj if (segmap_create(segmap, (caddr_t)&a) != 0) 2114ae115bc7Smrj panic("segmap_create segmap"); 21157c478bd9Sstevel@tonic-gate rw_exit(&kas.a_lock); 21167c478bd9Sstevel@tonic-gate 21177c478bd9Sstevel@tonic-gate setup_vaddr_for_ppcopy(CPU); 21187c478bd9Sstevel@tonic-gate 21197c478bd9Sstevel@tonic-gate segdev_init(); 2120843e1988Sjohnlev #if defined(__xpv) 2121843e1988Sjohnlev if (DOMAIN_IS_INITDOMAIN(xen_info)) 2122843e1988Sjohnlev #endif 2123843e1988Sjohnlev pmem_init(); 2124ae115bc7Smrj 21257c478bd9Sstevel@tonic-gate PRM_POINT("startup_vm() done"); 21267c478bd9Sstevel@tonic-gate } 21277c478bd9Sstevel@tonic-gate 2128ae115bc7Smrj /* 2129ae115bc7Smrj * Load a tod module for the non-standard tod part found on this system. 2130ae115bc7Smrj */ 2131ae115bc7Smrj static void 2132ae115bc7Smrj load_tod_module(char *todmod) 2133ae115bc7Smrj { 2134ae115bc7Smrj if (modload("tod", todmod) == -1) 2135ae115bc7Smrj halt("Can't load TOD module"); 2136ae115bc7Smrj } 2137ae115bc7Smrj 21387c478bd9Sstevel@tonic-gate static void 21397c478bd9Sstevel@tonic-gate startup_end(void) 21407c478bd9Sstevel@tonic-gate { 2141dd4eeefdSeota int i; 21427c478bd9Sstevel@tonic-gate extern void setx86isalist(void); 2143fb2caebeSRandy Fishel extern void cpu_event_init(void); 21447c478bd9Sstevel@tonic-gate 21457c478bd9Sstevel@tonic-gate PRM_POINT("startup_end() starting..."); 21467c478bd9Sstevel@tonic-gate 21477c478bd9Sstevel@tonic-gate /* 21487c478bd9Sstevel@tonic-gate * Perform tasks that get done after most of the VM 21497c478bd9Sstevel@tonic-gate * initialization has been done but before the clock 21507c478bd9Sstevel@tonic-gate * and other devices get started. 21517c478bd9Sstevel@tonic-gate */ 21527c478bd9Sstevel@tonic-gate kern_setup1(); 21537c478bd9Sstevel@tonic-gate 21547c478bd9Sstevel@tonic-gate /* 21557c478bd9Sstevel@tonic-gate * Perform CPC initialization for this CPU. 21567c478bd9Sstevel@tonic-gate */ 21577c478bd9Sstevel@tonic-gate kcpc_hw_init(CPU); 21587c478bd9Sstevel@tonic-gate 2159fb2caebeSRandy Fishel /* 2160fb2caebeSRandy Fishel * Initialize cpu event framework. 2161fb2caebeSRandy Fishel */ 2162fb2caebeSRandy Fishel cpu_event_init(); 2163fb2caebeSRandy Fishel 2164ee88d2b9Skchow #if defined(OPTERON_WORKAROUND_6323525) 2165ee88d2b9Skchow if (opteron_workaround_6323525) 2166ee88d2b9Skchow patch_workaround_6323525(); 2167ee88d2b9Skchow #endif 2168ae115bc7Smrj /* 2169ae115bc7Smrj * If needed, load TOD module now so that ddi_get_time(9F) etc. work 2170ae115bc7Smrj * (For now, "needed" is defined as set tod_module_name in /etc/system) 2171ae115bc7Smrj */ 2172ae115bc7Smrj if (tod_module_name != NULL) { 2173ae115bc7Smrj PRM_POINT("load_tod_module()"); 2174ae115bc7Smrj load_tod_module(tod_module_name); 2175ae115bc7Smrj } 2176ae115bc7Smrj 2177843e1988Sjohnlev #if defined(__xpv) 2178843e1988Sjohnlev /* 2179843e1988Sjohnlev * Forceload interposing TOD module for the hypervisor. 2180843e1988Sjohnlev */ 2181843e1988Sjohnlev PRM_POINT("load_tod_module()"); 2182843e1988Sjohnlev load_tod_module("xpvtod"); 2183843e1988Sjohnlev #endif 2184843e1988Sjohnlev 21857c478bd9Sstevel@tonic-gate /* 21867c478bd9Sstevel@tonic-gate * Configure the system. 21877c478bd9Sstevel@tonic-gate */ 21887c478bd9Sstevel@tonic-gate PRM_POINT("Calling configure()..."); 21897c478bd9Sstevel@tonic-gate configure(); /* set up devices */ 21907c478bd9Sstevel@tonic-gate PRM_POINT("configure() done"); 21917c478bd9Sstevel@tonic-gate 21927c478bd9Sstevel@tonic-gate /* 21937c478bd9Sstevel@tonic-gate * Set the isa_list string to the defined instruction sets we 21947c478bd9Sstevel@tonic-gate * support. 21957c478bd9Sstevel@tonic-gate */ 21967c478bd9Sstevel@tonic-gate setx86isalist(); 2197100b72f4Sandrei cpu_intr_alloc(CPU, NINTR_THREADS); 21987c478bd9Sstevel@tonic-gate psm_install(); 21997c478bd9Sstevel@tonic-gate 22007c478bd9Sstevel@tonic-gate /* 22017c478bd9Sstevel@tonic-gate * We're done with bootops. We don't unmap the bootstrap yet because 22027c478bd9Sstevel@tonic-gate * we're still using bootsvcs. 22037c478bd9Sstevel@tonic-gate */ 2204ae115bc7Smrj PRM_POINT("NULLing out bootops"); 2205ae115bc7Smrj *bootopsp = (struct bootops *)NULL; 22067c478bd9Sstevel@tonic-gate bootops = (struct bootops *)NULL; 22077c478bd9Sstevel@tonic-gate 2208843e1988Sjohnlev #if defined(__xpv) 2209843e1988Sjohnlev ec_init_debug_irq(); 2210843e1988Sjohnlev xs_domu_init(); 2211843e1988Sjohnlev #endif 22121de082f7SVikram Hegde 22133a634bfcSVikram Hegde #if defined(__amd64) && !defined(__xpv) 22143a634bfcSVikram Hegde /* 22153a634bfcSVikram Hegde * Intel IOMMU has been setup/initialized in ddi_impl.c 22163a634bfcSVikram Hegde * Start it up now. 22173a634bfcSVikram Hegde */ 22183a634bfcSVikram Hegde immu_startup(); 22191de082f7SVikram Hegde #endif 22203a634bfcSVikram Hegde 22217c478bd9Sstevel@tonic-gate PRM_POINT("Enabling interrupts"); 22227c478bd9Sstevel@tonic-gate (*picinitf)(); 22237c478bd9Sstevel@tonic-gate sti(); 2224843e1988Sjohnlev #if defined(__xpv) 2225843e1988Sjohnlev ASSERT(CPU->cpu_m.mcpu_vcpu_info->evtchn_upcall_mask == 0); 2226843e1988Sjohnlev xen_late_startup(); 2227843e1988Sjohnlev #endif 22287c478bd9Sstevel@tonic-gate 22297c478bd9Sstevel@tonic-gate (void) add_avsoftintr((void *)&softlevel1_hdl, 1, softlevel1, 223035b1ab99Sjosephb "softlevel1", NULL, NULL); /* XXX to be moved later */ 22317c478bd9Sstevel@tonic-gate 2232dd4eeefdSeota /* 2233dd4eeefdSeota * Register these software interrupts for ddi timer. 2234dd4eeefdSeota * Software interrupts up to the level 10 are supported. 2235dd4eeefdSeota */ 2236dd4eeefdSeota for (i = DDI_IPL_1; i <= DDI_IPL_10; i++) { 2237dd4eeefdSeota char name[sizeof ("timer_softintr") + 2]; 2238dd4eeefdSeota (void) sprintf(name, "timer_softintr%02d", i); 2239dd4eeefdSeota (void) add_avsoftintr((void *)&softlevel_hdl[i-1], i, 2240dd4eeefdSeota (avfunc)timer_softintr, name, (caddr_t)(uintptr_t)i, NULL); 2241dd4eeefdSeota } 2242dd4eeefdSeota 22437125fcbdSVikram Hegde #if !defined(__xpv) 22447125fcbdSVikram Hegde if (modload("drv", "amd_iommu") < 0) { 22457125fcbdSVikram Hegde PRM_POINT("No AMD IOMMU present\n"); 224694f1124eSVikram Hegde } else if (ddi_hold_installed_driver(ddi_name_to_major( 224794f1124eSVikram Hegde "amd_iommu")) == NULL) { 22487125fcbdSVikram Hegde prom_printf("ERROR: failed to attach AMD IOMMU\n"); 22497125fcbdSVikram Hegde } 22507125fcbdSVikram Hegde #endif 2251e774b42bSBill Holler post_startup_cpu_fixups(); 22527125fcbdSVikram Hegde 22537c478bd9Sstevel@tonic-gate PRM_POINT("startup_end() done"); 22547c478bd9Sstevel@tonic-gate } 22557c478bd9Sstevel@tonic-gate 225696d008c2SKen Erickson /* 225796d008c2SKen Erickson * Don't remove the following 2 variables. They are necessary 225896d008c2SKen Erickson * for reading the hostid from the legacy file (/kernel/misc/sysinit). 225996d008c2SKen Erickson */ 22607c478bd9Sstevel@tonic-gate char *_hs1107 = hw_serial; 22617c478bd9Sstevel@tonic-gate ulong_t _bdhs34; 22627c478bd9Sstevel@tonic-gate 22637c478bd9Sstevel@tonic-gate void 22647c478bd9Sstevel@tonic-gate post_startup(void) 22657c478bd9Sstevel@tonic-gate { 22660e751525SEric Saxe extern void cpupm_init(cpu_t *); 2267fb2caebeSRandy Fishel extern void cpu_event_init_cpu(cpu_t *); 22680e751525SEric Saxe 22697c478bd9Sstevel@tonic-gate /* 22707c478bd9Sstevel@tonic-gate * Set the system wide, processor-specific flags to be passed 22717c478bd9Sstevel@tonic-gate * to userland via the aux vector for performance hints and 22727c478bd9Sstevel@tonic-gate * instruction set extensions. 22737c478bd9Sstevel@tonic-gate */ 22747c478bd9Sstevel@tonic-gate bind_hwcap(); 22757c478bd9Sstevel@tonic-gate 2276843e1988Sjohnlev #ifdef __xpv 2277843e1988Sjohnlev if (DOMAIN_IS_INITDOMAIN(xen_info)) 2278843e1988Sjohnlev #endif 2279843e1988Sjohnlev { 2280843e1988Sjohnlev #if defined(__xpv) 2281843e1988Sjohnlev xpv_panic_init(); 2282843e1988Sjohnlev #else 2283843e1988Sjohnlev /* 2284843e1988Sjohnlev * Startup the memory scrubber. 2285843e1988Sjohnlev * XXPV This should be running somewhere .. 2286843e1988Sjohnlev */ 2287b9bfdccdSStuart Maybee if (get_hwenv() != HW_XEN_HVM) 2288e4b86885SCheng Sean Ye memscrub_init(); 2289843e1988Sjohnlev #endif 2290843e1988Sjohnlev } 22917c478bd9Sstevel@tonic-gate 22927aec1d6eScindi /* 22937aec1d6eScindi * Complete CPU module initialization 22947aec1d6eScindi */ 229520c794b3Sgavinm cmi_post_startup(); 22967aec1d6eScindi 22977c478bd9Sstevel@tonic-gate /* 229884ab085aSmws * Perform forceloading tasks for /etc/system. 22997c478bd9Sstevel@tonic-gate */ 230084ab085aSmws (void) mod_sysctl(SYS_FORCELOAD, NULL); 23017c478bd9Sstevel@tonic-gate 23027c478bd9Sstevel@tonic-gate /* 23037c478bd9Sstevel@tonic-gate * ON4.0: Force /proc module in until clock interrupt handle fixed 23047c478bd9Sstevel@tonic-gate * ON4.0: This must be fixed or restated in /etc/systems. 23057c478bd9Sstevel@tonic-gate */ 23067c478bd9Sstevel@tonic-gate (void) modload("fs", "procfs"); 23077c478bd9Sstevel@tonic-gate 2308c35aa225Smarx (void) i_ddi_attach_hw_nodes("pit_beep"); 2309c35aa225Smarx 23107c478bd9Sstevel@tonic-gate #if defined(__i386) 23117c478bd9Sstevel@tonic-gate /* 23127c478bd9Sstevel@tonic-gate * Check for required functional Floating Point hardware, 23137c478bd9Sstevel@tonic-gate * unless FP hardware explicitly disabled. 23147c478bd9Sstevel@tonic-gate */ 23157c478bd9Sstevel@tonic-gate if (fpu_exists && (fpu_pentium_fdivbug || fp_kind == FP_NO)) 23167c478bd9Sstevel@tonic-gate halt("No working FP hardware found"); 23177c478bd9Sstevel@tonic-gate #endif 23187c478bd9Sstevel@tonic-gate 23197c478bd9Sstevel@tonic-gate maxmem = freemem; 23207c478bd9Sstevel@tonic-gate 2321fb2caebeSRandy Fishel cpu_event_init_cpu(CPU); 23220e751525SEric Saxe cpupm_init(CPU); 2323fa96bd91SMichael Corcoran (void) mach_cpu_create_device_node(CPU, NULL); 23240e751525SEric Saxe 23250e751525SEric Saxe pg_init(); 23267c478bd9Sstevel@tonic-gate } 23277c478bd9Sstevel@tonic-gate 23287c478bd9Sstevel@tonic-gate static int 232919397407SSherry Moore pp_in_range(page_t *pp, uint64_t low_addr, uint64_t high_addr) 23307c478bd9Sstevel@tonic-gate { 233119397407SSherry Moore return ((pp->p_pagenum >= btop(low_addr)) && 233219397407SSherry Moore (pp->p_pagenum < btopr(high_addr))); 23337c478bd9Sstevel@tonic-gate } 23347c478bd9Sstevel@tonic-gate 23357c478bd9Sstevel@tonic-gate void 23367c478bd9Sstevel@tonic-gate release_bootstrap(void) 23377c478bd9Sstevel@tonic-gate { 23387c478bd9Sstevel@tonic-gate int root_is_ramdisk; 23397c478bd9Sstevel@tonic-gate page_t *pp; 23407c478bd9Sstevel@tonic-gate extern void kobj_boot_unmountroot(void); 23417c478bd9Sstevel@tonic-gate extern dev_t rootdev; 234255d507a9SSeth Goldberg #if !defined(__xpv) 234355d507a9SSeth Goldberg pfn_t pfn; 234455d507a9SSeth Goldberg #endif 23457c478bd9Sstevel@tonic-gate 23467c478bd9Sstevel@tonic-gate /* unmount boot ramdisk and release kmem usage */ 23477c478bd9Sstevel@tonic-gate kobj_boot_unmountroot(); 23487c478bd9Sstevel@tonic-gate 23497c478bd9Sstevel@tonic-gate /* 23507c478bd9Sstevel@tonic-gate * We're finished using the boot loader so free its pages. 23517c478bd9Sstevel@tonic-gate */ 23527c478bd9Sstevel@tonic-gate PRM_POINT("Unmapping lower boot pages"); 235319397407SSherry Moore 2354ae115bc7Smrj clear_boot_mappings(0, _userlimit); 235519397407SSherry Moore 2356ae115bc7Smrj postbootkernelbase = kernelbase; 23577c478bd9Sstevel@tonic-gate 23587c478bd9Sstevel@tonic-gate /* 23597c478bd9Sstevel@tonic-gate * If root isn't on ramdisk, destroy the hardcoded 23607c478bd9Sstevel@tonic-gate * ramdisk node now and release the memory. Else, 23617c478bd9Sstevel@tonic-gate * ramdisk memory is kept in rd_pages. 23627c478bd9Sstevel@tonic-gate */ 23637c478bd9Sstevel@tonic-gate root_is_ramdisk = (getmajor(rootdev) == ddi_name_to_major("ramdisk")); 23647c478bd9Sstevel@tonic-gate if (!root_is_ramdisk) { 23657c478bd9Sstevel@tonic-gate dev_info_t *dip = ddi_find_devinfo("ramdisk", -1, 0); 23667c478bd9Sstevel@tonic-gate ASSERT(dip && ddi_get_parent(dip) == ddi_root_node()); 23677c478bd9Sstevel@tonic-gate ndi_rele_devi(dip); /* held from ddi_find_devinfo */ 23687c478bd9Sstevel@tonic-gate (void) ddi_remove_child(dip, 0); 23697c478bd9Sstevel@tonic-gate } 23707c478bd9Sstevel@tonic-gate 23717c478bd9Sstevel@tonic-gate PRM_POINT("Releasing boot pages"); 23727c478bd9Sstevel@tonic-gate while (bootpages) { 237319397407SSherry Moore extern uint64_t ramdisk_start, ramdisk_end; 23747c478bd9Sstevel@tonic-gate pp = bootpages; 23757c478bd9Sstevel@tonic-gate bootpages = pp->p_next; 237619397407SSherry Moore 237719397407SSherry Moore 237819397407SSherry Moore /* Keep pages for the lower 64K */ 237919397407SSherry Moore if (pp_in_range(pp, 0, 0x40000)) { 238019397407SSherry Moore pp->p_next = lower_pages; 238119397407SSherry Moore lower_pages = pp; 238219397407SSherry Moore lower_pages_count++; 238319397407SSherry Moore continue; 238419397407SSherry Moore } 238519397407SSherry Moore 238619397407SSherry Moore 238719397407SSherry Moore if (root_is_ramdisk && pp_in_range(pp, ramdisk_start, 238819397407SSherry Moore ramdisk_end)) { 23897c478bd9Sstevel@tonic-gate pp->p_next = rd_pages; 23907c478bd9Sstevel@tonic-gate rd_pages = pp; 23917c478bd9Sstevel@tonic-gate continue; 23927c478bd9Sstevel@tonic-gate } 23937c478bd9Sstevel@tonic-gate pp->p_next = (struct page *)0; 2394ae115bc7Smrj pp->p_prev = (struct page *)0; 2395ae115bc7Smrj PP_CLRBOOTPAGES(pp); 23967c478bd9Sstevel@tonic-gate page_free(pp, 1); 23977c478bd9Sstevel@tonic-gate } 2398ae115bc7Smrj PRM_POINT("Boot pages released"); 23997c478bd9Sstevel@tonic-gate 2400843e1988Sjohnlev #if !defined(__xpv) 2401843e1988Sjohnlev /* XXPV -- note this following bunch of code needs to be revisited in Xen 3.0 */ 24027c478bd9Sstevel@tonic-gate /* 240355d507a9SSeth Goldberg * Find 1 page below 1 MB so that other processors can boot up or 240455d507a9SSeth Goldberg * so that any processor can resume. 24057c478bd9Sstevel@tonic-gate * Make sure it has a kernel VA as well as a 1:1 mapping. 24067c478bd9Sstevel@tonic-gate * We should have just free'd one up. 24077c478bd9Sstevel@tonic-gate */ 240819397407SSherry Moore 240919397407SSherry Moore /* 241019397407SSherry Moore * 0x10 pages is 64K. Leave the bottom 64K alone 241119397407SSherry Moore * for BIOS. 241219397407SSherry Moore */ 241319397407SSherry Moore for (pfn = 0x10; pfn < btop(1*1024*1024); pfn++) { 241455d507a9SSeth Goldberg if (page_numtopp_alloc(pfn) == NULL) 241555d507a9SSeth Goldberg continue; 241655d507a9SSeth Goldberg rm_platter_va = i86devmap(pfn, 1, 241755d507a9SSeth Goldberg PROT_READ | PROT_WRITE | PROT_EXEC); 241855d507a9SSeth Goldberg rm_platter_pa = ptob(pfn); 241955d507a9SSeth Goldberg break; 24207c478bd9Sstevel@tonic-gate } 242155d507a9SSeth Goldberg if (pfn == btop(1*1024*1024) && use_mp) 242255d507a9SSeth Goldberg panic("No page below 1M available for starting " 242355d507a9SSeth Goldberg "other processors or for resuming from system-suspend"); 2424843e1988Sjohnlev #endif /* !__xpv */ 24257c478bd9Sstevel@tonic-gate } 24267c478bd9Sstevel@tonic-gate 24277c478bd9Sstevel@tonic-gate /* 24287c478bd9Sstevel@tonic-gate * Initialize the platform-specific parts of a page_t. 24297c478bd9Sstevel@tonic-gate */ 24307c478bd9Sstevel@tonic-gate void 24317c478bd9Sstevel@tonic-gate add_physmem_cb(page_t *pp, pfn_t pnum) 24327c478bd9Sstevel@tonic-gate { 24337c478bd9Sstevel@tonic-gate pp->p_pagenum = pnum; 24347c478bd9Sstevel@tonic-gate pp->p_mapping = NULL; 24357c478bd9Sstevel@tonic-gate pp->p_embed = 0; 24367c478bd9Sstevel@tonic-gate pp->p_share = 0; 24377c478bd9Sstevel@tonic-gate pp->p_mlentry = 0; 24387c478bd9Sstevel@tonic-gate } 24397c478bd9Sstevel@tonic-gate 24407c478bd9Sstevel@tonic-gate /* 24417c478bd9Sstevel@tonic-gate * kphysm_init() initializes physical memory. 24427c478bd9Sstevel@tonic-gate */ 24437c478bd9Sstevel@tonic-gate static pgcnt_t 24447c478bd9Sstevel@tonic-gate kphysm_init( 2445ae115bc7Smrj page_t *pp, 24467c478bd9Sstevel@tonic-gate pgcnt_t npages) 24477c478bd9Sstevel@tonic-gate { 24487c478bd9Sstevel@tonic-gate struct memlist *pmem; 24497c478bd9Sstevel@tonic-gate struct memseg *cur_memseg; 24507c478bd9Sstevel@tonic-gate pfn_t base_pfn; 245118968004SKit Chow pfn_t end_pfn; 24527c478bd9Sstevel@tonic-gate pgcnt_t num; 24537c478bd9Sstevel@tonic-gate pgcnt_t pages_done = 0; 24547c478bd9Sstevel@tonic-gate uint64_t addr; 24557c478bd9Sstevel@tonic-gate uint64_t size; 24567c478bd9Sstevel@tonic-gate extern pfn_t ddiphysmin; 245718968004SKit Chow extern int mnode_xwa; 245818968004SKit Chow int ms = 0, me = 0; 24597c478bd9Sstevel@tonic-gate 24607c478bd9Sstevel@tonic-gate ASSERT(page_hash != NULL && page_hashsz != 0); 24617c478bd9Sstevel@tonic-gate 2462ae115bc7Smrj cur_memseg = memseg_base; 246356f33205SJonathan Adams for (pmem = phys_avail; pmem && npages; pmem = pmem->ml_next) { 24647c478bd9Sstevel@tonic-gate /* 24657c478bd9Sstevel@tonic-gate * In a 32 bit kernel can't use higher memory if we're 24667c478bd9Sstevel@tonic-gate * not booting in PAE mode. This check takes care of that. 24677c478bd9Sstevel@tonic-gate */ 246856f33205SJonathan Adams addr = pmem->ml_address; 246956f33205SJonathan Adams size = pmem->ml_size; 24707c478bd9Sstevel@tonic-gate if (btop(addr) > physmax) 24717c478bd9Sstevel@tonic-gate continue; 24727c478bd9Sstevel@tonic-gate 24737c478bd9Sstevel@tonic-gate /* 24747c478bd9Sstevel@tonic-gate * align addr and size - they may not be at page boundaries 24757c478bd9Sstevel@tonic-gate */ 24767c478bd9Sstevel@tonic-gate if ((addr & MMU_PAGEOFFSET) != 0) { 24777c478bd9Sstevel@tonic-gate addr += MMU_PAGEOFFSET; 24787c478bd9Sstevel@tonic-gate addr &= ~(uint64_t)MMU_PAGEOFFSET; 247956f33205SJonathan Adams size -= addr - pmem->ml_address; 24807c478bd9Sstevel@tonic-gate } 24817c478bd9Sstevel@tonic-gate 24826bb54764Skchow /* only process pages below or equal to physmax */ 24836bb54764Skchow if ((btop(addr + size) - 1) > physmax) 24846bb54764Skchow size = ptob(physmax - btop(addr) + 1); 24857c478bd9Sstevel@tonic-gate 24867c478bd9Sstevel@tonic-gate num = btop(size); 24877c478bd9Sstevel@tonic-gate if (num == 0) 24887c478bd9Sstevel@tonic-gate continue; 24897c478bd9Sstevel@tonic-gate 24907c478bd9Sstevel@tonic-gate if (num > npages) 24917c478bd9Sstevel@tonic-gate num = npages; 24927c478bd9Sstevel@tonic-gate 24937c478bd9Sstevel@tonic-gate npages -= num; 24947c478bd9Sstevel@tonic-gate pages_done += num; 24957c478bd9Sstevel@tonic-gate base_pfn = btop(addr); 24967c478bd9Sstevel@tonic-gate 24977c478bd9Sstevel@tonic-gate if (prom_debug) 24987c478bd9Sstevel@tonic-gate prom_printf("MEMSEG addr=0x%" PRIx64 24997c478bd9Sstevel@tonic-gate " pgs=0x%lx pfn 0x%lx-0x%lx\n", 25007c478bd9Sstevel@tonic-gate addr, num, base_pfn, base_pfn + num); 25017c478bd9Sstevel@tonic-gate 25027c478bd9Sstevel@tonic-gate /* 2503ae115bc7Smrj * Ignore pages below ddiphysmin to simplify ddi memory 25047c478bd9Sstevel@tonic-gate * allocation with non-zero addr_lo requests. 25057c478bd9Sstevel@tonic-gate */ 25067c478bd9Sstevel@tonic-gate if (base_pfn < ddiphysmin) { 2507ae115bc7Smrj if (base_pfn + num <= ddiphysmin) 25087c478bd9Sstevel@tonic-gate continue; 25097c478bd9Sstevel@tonic-gate pp += (ddiphysmin - base_pfn); 25107c478bd9Sstevel@tonic-gate num -= (ddiphysmin - base_pfn); 25117c478bd9Sstevel@tonic-gate base_pfn = ddiphysmin; 25127c478bd9Sstevel@tonic-gate } 2513ae115bc7Smrj 25147c478bd9Sstevel@tonic-gate /* 251518968004SKit Chow * mnode_xwa is greater than 1 when large pages regions can 251618968004SKit Chow * cross memory node boundaries. To prevent the formation 251718968004SKit Chow * of these large pages, configure the memsegs based on the 251818968004SKit Chow * memory node ranges which had been made non-contiguous. 25197c478bd9Sstevel@tonic-gate */ 252018968004SKit Chow if (mnode_xwa > 1) { 252118968004SKit Chow 252218968004SKit Chow end_pfn = base_pfn + num - 1; 252318968004SKit Chow ms = PFN_2_MEM_NODE(base_pfn); 252418968004SKit Chow me = PFN_2_MEM_NODE(end_pfn); 252518968004SKit Chow 252618968004SKit Chow if (ms != me) { 252718968004SKit Chow /* 252818968004SKit Chow * current range spans more than 1 memory node. 252918968004SKit Chow * Set num to only the pfn range in the start 253018968004SKit Chow * memory node. 253118968004SKit Chow */ 253218968004SKit Chow num = mem_node_config[ms].physmax - base_pfn 253318968004SKit Chow + 1; 253418968004SKit Chow ASSERT(end_pfn > mem_node_config[ms].physmax); 253518968004SKit Chow } 25367c478bd9Sstevel@tonic-gate } 25377c478bd9Sstevel@tonic-gate 253818968004SKit Chow for (;;) { 253918968004SKit Chow /* 254018968004SKit Chow * Build the memsegs entry 254118968004SKit Chow */ 254218968004SKit Chow cur_memseg->pages = pp; 254318968004SKit Chow cur_memseg->epages = pp + num; 254418968004SKit Chow cur_memseg->pages_base = base_pfn; 254518968004SKit Chow cur_memseg->pages_end = base_pfn + num; 254618968004SKit Chow 254718968004SKit Chow /* 254818968004SKit Chow * Insert into memseg list in decreasing pfn range 254918968004SKit Chow * order. Low memory is typically more fragmented such 255018968004SKit Chow * that this ordering keeps the larger ranges at the 255118968004SKit Chow * front of the list for code that searches memseg. 255218968004SKit Chow * This ASSERTS that the memsegs coming in from boot 255318968004SKit Chow * are in increasing physical address order and not 255418968004SKit Chow * contiguous. 255518968004SKit Chow */ 255618968004SKit Chow if (memsegs != NULL) { 255718968004SKit Chow ASSERT(cur_memseg->pages_base >= 255818968004SKit Chow memsegs->pages_end); 255918968004SKit Chow cur_memseg->next = memsegs; 256018968004SKit Chow } 256118968004SKit Chow memsegs = cur_memseg; 256218968004SKit Chow 256318968004SKit Chow /* 256418968004SKit Chow * add_physmem() initializes the PSM part of the page 256518968004SKit Chow * struct by calling the PSM back with add_physmem_cb(). 256618968004SKit Chow * In addition it coalesces pages into larger pages as 256718968004SKit Chow * it initializes them. 256818968004SKit Chow */ 256918968004SKit Chow add_physmem(pp, num, base_pfn); 257018968004SKit Chow cur_memseg++; 257118968004SKit Chow availrmem_initial += num; 257218968004SKit Chow availrmem += num; 257318968004SKit Chow 257418968004SKit Chow pp += num; 257518968004SKit Chow if (ms >= me) 257618968004SKit Chow break; 25777c478bd9Sstevel@tonic-gate 257818968004SKit Chow /* process next memory node range */ 257918968004SKit Chow ms++; 258018968004SKit Chow base_pfn = mem_node_config[ms].physbase; 258118968004SKit Chow num = MIN(mem_node_config[ms].physmax, 258218968004SKit Chow end_pfn) - base_pfn + 1; 258318968004SKit Chow } 25847c478bd9Sstevel@tonic-gate } 25857c478bd9Sstevel@tonic-gate 25867c478bd9Sstevel@tonic-gate PRM_DEBUG(availrmem_initial); 25877c478bd9Sstevel@tonic-gate PRM_DEBUG(availrmem); 25887c478bd9Sstevel@tonic-gate PRM_DEBUG(freemem); 25897c478bd9Sstevel@tonic-gate build_pfn_hash(); 25907c478bd9Sstevel@tonic-gate return (pages_done); 25917c478bd9Sstevel@tonic-gate } 25927c478bd9Sstevel@tonic-gate 25937c478bd9Sstevel@tonic-gate /* 25947c478bd9Sstevel@tonic-gate * Kernel VM initialization. 25957c478bd9Sstevel@tonic-gate */ 25967c478bd9Sstevel@tonic-gate static void 25977c478bd9Sstevel@tonic-gate kvm_init(void) 25987c478bd9Sstevel@tonic-gate { 25997c478bd9Sstevel@tonic-gate ASSERT((((uintptr_t)s_text) & MMU_PAGEOFFSET) == 0); 26007c478bd9Sstevel@tonic-gate 26017c478bd9Sstevel@tonic-gate /* 26027c478bd9Sstevel@tonic-gate * Put the kernel segments in kernel address space. 26037c478bd9Sstevel@tonic-gate */ 26047c478bd9Sstevel@tonic-gate rw_enter(&kas.a_lock, RW_WRITER); 26057c478bd9Sstevel@tonic-gate as_avlinit(&kas); 26067c478bd9Sstevel@tonic-gate 26077c478bd9Sstevel@tonic-gate (void) seg_attach(&kas, s_text, e_moddata - s_text, &ktextseg); 26087c478bd9Sstevel@tonic-gate (void) segkmem_create(&ktextseg); 26097c478bd9Sstevel@tonic-gate 26107c478bd9Sstevel@tonic-gate (void) seg_attach(&kas, (caddr_t)valloc_base, valloc_sz, &kvalloc); 26117c478bd9Sstevel@tonic-gate (void) segkmem_create(&kvalloc); 26127c478bd9Sstevel@tonic-gate 261335b1ab99Sjosephb (void) seg_attach(&kas, kernelheap, 261435b1ab99Sjosephb ekernelheap - kernelheap, &kvseg); 26157c478bd9Sstevel@tonic-gate (void) segkmem_create(&kvseg); 26167c478bd9Sstevel@tonic-gate 2617ae115bc7Smrj if (core_size > 0) { 2618ae115bc7Smrj PRM_POINT("attaching kvseg_core"); 2619ae115bc7Smrj (void) seg_attach(&kas, (caddr_t)core_base, core_size, 2620ae115bc7Smrj &kvseg_core); 2621ae115bc7Smrj (void) segkmem_create(&kvseg_core); 2622ae115bc7Smrj } 2623ad23a2dbSjohansen 2624ae115bc7Smrj if (segziosize > 0) { 2625ae115bc7Smrj PRM_POINT("attaching segzio"); 2626ad23a2dbSjohansen (void) seg_attach(&kas, segzio_base, mmu_ptob(segziosize), 2627ad23a2dbSjohansen &kzioseg); 2628ad23a2dbSjohansen (void) segkmem_zio_create(&kzioseg); 2629ad23a2dbSjohansen 2630ad23a2dbSjohansen /* create zio area covering new segment */ 2631ad23a2dbSjohansen segkmem_zio_init(segzio_base, mmu_ptob(segziosize)); 2632ad23a2dbSjohansen } 26337c478bd9Sstevel@tonic-gate 2634ae115bc7Smrj (void) seg_attach(&kas, kdi_segdebugbase, kdi_segdebugsize, &kdebugseg); 26357c478bd9Sstevel@tonic-gate (void) segkmem_create(&kdebugseg); 26367c478bd9Sstevel@tonic-gate 26377c478bd9Sstevel@tonic-gate rw_exit(&kas.a_lock); 26387c478bd9Sstevel@tonic-gate 26397c478bd9Sstevel@tonic-gate /* 26407c478bd9Sstevel@tonic-gate * Ensure that the red zone at kernelbase is never accessible. 26417c478bd9Sstevel@tonic-gate */ 2642ae115bc7Smrj PRM_POINT("protecting redzone"); 26437c478bd9Sstevel@tonic-gate (void) as_setprot(&kas, (caddr_t)kernelbase, KERNEL_REDZONE_SIZE, 0); 26447c478bd9Sstevel@tonic-gate 26457c478bd9Sstevel@tonic-gate /* 26467c478bd9Sstevel@tonic-gate * Make the text writable so that it can be hot patched by DTrace. 26477c478bd9Sstevel@tonic-gate */ 26487c478bd9Sstevel@tonic-gate (void) as_setprot(&kas, s_text, e_modtext - s_text, 26497c478bd9Sstevel@tonic-gate PROT_READ | PROT_WRITE | PROT_EXEC); 26507c478bd9Sstevel@tonic-gate 26517c478bd9Sstevel@tonic-gate /* 26527c478bd9Sstevel@tonic-gate * Make data writable until end. 26537c478bd9Sstevel@tonic-gate */ 26547c478bd9Sstevel@tonic-gate (void) as_setprot(&kas, s_data, e_moddata - s_data, 26557c478bd9Sstevel@tonic-gate PROT_READ | PROT_WRITE | PROT_EXEC); 26567c478bd9Sstevel@tonic-gate } 26577c478bd9Sstevel@tonic-gate 2658843e1988Sjohnlev #ifndef __xpv 26597c478bd9Sstevel@tonic-gate /* 26601d03c31eSjohnlev * Solaris adds an entry for Write Combining caching to the PAT 26617c478bd9Sstevel@tonic-gate */ 26621d03c31eSjohnlev static uint64_t pat_attr_reg = PAT_DEFAULT_ATTRIBUTE; 26637c478bd9Sstevel@tonic-gate 26647c478bd9Sstevel@tonic-gate void 26651d03c31eSjohnlev pat_sync(void) 26667c478bd9Sstevel@tonic-gate { 26671d03c31eSjohnlev ulong_t cr0, cr0_orig, cr4; 26687c478bd9Sstevel@tonic-gate 26691d03c31eSjohnlev if (!(x86_feature & X86_PAT)) 26707c478bd9Sstevel@tonic-gate return; 26711d03c31eSjohnlev cr0_orig = cr0 = getcr0(); 26721d03c31eSjohnlev cr4 = getcr4(); 26737c478bd9Sstevel@tonic-gate 26741d03c31eSjohnlev /* disable caching and flush all caches and TLBs */ 26751d03c31eSjohnlev cr0 |= CR0_CD; 26761d03c31eSjohnlev cr0 &= ~CR0_NW; 26771d03c31eSjohnlev setcr0(cr0); 26781d03c31eSjohnlev invalidate_cache(); 26791d03c31eSjohnlev if (cr4 & CR4_PGE) { 26801d03c31eSjohnlev setcr4(cr4 & ~(ulong_t)CR4_PGE); 26811d03c31eSjohnlev setcr4(cr4); 26821d03c31eSjohnlev } else { 26831d03c31eSjohnlev reload_cr3(); 26847c478bd9Sstevel@tonic-gate } 26857c478bd9Sstevel@tonic-gate 26861d03c31eSjohnlev /* add our entry to the PAT */ 26871d03c31eSjohnlev wrmsr(REG_PAT, pat_attr_reg); 26887c478bd9Sstevel@tonic-gate 26891d03c31eSjohnlev /* flush TLBs and cache again, then reenable cr0 caching */ 26901d03c31eSjohnlev if (cr4 & CR4_PGE) { 26911d03c31eSjohnlev setcr4(cr4 & ~(ulong_t)CR4_PGE); 26921d03c31eSjohnlev setcr4(cr4); 26931d03c31eSjohnlev } else { 26941d03c31eSjohnlev reload_cr3(); 26957c478bd9Sstevel@tonic-gate } 26967c478bd9Sstevel@tonic-gate invalidate_cache(); 26977c478bd9Sstevel@tonic-gate setcr0(cr0_orig); 26987c478bd9Sstevel@tonic-gate } 26997c478bd9Sstevel@tonic-gate 27001d03c31eSjohnlev #endif /* !__xpv */ 27017c478bd9Sstevel@tonic-gate 2702ed5289f9SKen Erickson #if defined(_SOFT_HOSTID) 2703ed5289f9SKen Erickson /* 2704ed5289f9SKen Erickson * On platforms that do not have a hardware serial number, attempt 2705ed5289f9SKen Erickson * to set one based on the contents of /etc/hostid. If this file does 2706ed5289f9SKen Erickson * not exist, assume that we are to generate a new hostid and set 2707ed5289f9SKen Erickson * it in the kernel, for subsequent saving by a userland process 2708ed5289f9SKen Erickson * once the system is up and the root filesystem is mounted r/w. 2709ed5289f9SKen Erickson * 271096d008c2SKen Erickson * In order to gracefully support upgrade on OpenSolaris, if 271196d008c2SKen Erickson * /etc/hostid does not exist, we will attempt to get a serial number 271296d008c2SKen Erickson * using the legacy method (/kernel/misc/sysinit). 271396d008c2SKen Erickson * 2714ed5289f9SKen Erickson * In an attempt to make the hostid less prone to abuse 2715ed5289f9SKen Erickson * (for license circumvention, etc), we store it in /etc/hostid 2716ed5289f9SKen Erickson * in rot47 format. 2717ed5289f9SKen Erickson */ 2718ed5289f9SKen Erickson extern volatile unsigned long tenmicrodata; 271996d008c2SKen Erickson static int atoi(char *); 2720ed5289f9SKen Erickson 2721ed5289f9SKen Erickson static int32_t 2722ed5289f9SKen Erickson set_soft_hostid(void) 2723ed5289f9SKen Erickson { 2724ed5289f9SKen Erickson struct _buf *file; 2725ed5289f9SKen Erickson char tokbuf[MAXNAMELEN]; 2726ed5289f9SKen Erickson token_t token; 2727ed5289f9SKen Erickson int done = 0; 2728ed5289f9SKen Erickson u_longlong_t tmp; 272996d008c2SKen Erickson int i; 27305679c89fSjv int32_t hostid = (int32_t)HW_INVALID_HOSTID; 2731ed5289f9SKen Erickson unsigned char *c; 2732ed5289f9SKen Erickson hrtime_t tsc; 2733ed5289f9SKen Erickson 2734ed5289f9SKen Erickson /* 2735ed5289f9SKen Erickson * If /etc/hostid file not found, we'd like to get a pseudo 2736ed5289f9SKen Erickson * random number to use at the hostid. A nice way to do this 2737ed5289f9SKen Erickson * is to read the real time clock. To remain xen-compatible, 2738ed5289f9SKen Erickson * we can't poke the real hardware, so we use tsc_read() to 2739ed5289f9SKen Erickson * read the real time clock. However, there is an ominous 2740ed5289f9SKen Erickson * warning in tsc_read that says it can return zero, so we 2741ed5289f9SKen Erickson * deal with that possibility by falling back to using the 2742ed5289f9SKen Erickson * (hopefully random enough) value in tenmicrodata. 2743ed5289f9SKen Erickson */ 2744ed5289f9SKen Erickson 2745ed5289f9SKen Erickson if ((file = kobj_open_file(hostid_file)) == (struct _buf *)-1) { 274696d008c2SKen Erickson /* 274796d008c2SKen Erickson * hostid file not found - try to load sysinit module 274896d008c2SKen Erickson * and see if it has a nonzero hostid value...use that 274996d008c2SKen Erickson * instead of generating a new hostid here if so. 275096d008c2SKen Erickson */ 275196d008c2SKen Erickson if ((i = modload("misc", "sysinit")) != -1) { 275296d008c2SKen Erickson if (strlen(hw_serial) > 0) 275396d008c2SKen Erickson hostid = (int32_t)atoi(hw_serial); 275496d008c2SKen Erickson (void) modunload(i); 275596d008c2SKen Erickson } 27565679c89fSjv if (hostid == HW_INVALID_HOSTID) { 275796d008c2SKen Erickson tsc = tsc_read(); 275896d008c2SKen Erickson if (tsc == 0) /* tsc_read can return zero sometimes */ 275996d008c2SKen Erickson hostid = (int32_t)tenmicrodata & 0x0CFFFFF; 276096d008c2SKen Erickson else 276196d008c2SKen Erickson hostid = (int32_t)tsc & 0x0CFFFFF; 276296d008c2SKen Erickson } 2763ed5289f9SKen Erickson } else { 2764ed5289f9SKen Erickson /* hostid file found */ 2765ed5289f9SKen Erickson while (!done) { 2766ed5289f9SKen Erickson token = kobj_lex(file, tokbuf, sizeof (tokbuf)); 2767ed5289f9SKen Erickson 2768ed5289f9SKen Erickson switch (token) { 2769ed5289f9SKen Erickson case POUND: 2770ed5289f9SKen Erickson /* 2771ed5289f9SKen Erickson * skip comments 2772ed5289f9SKen Erickson */ 2773ed5289f9SKen Erickson kobj_find_eol(file); 2774ed5289f9SKen Erickson break; 2775ed5289f9SKen Erickson case STRING: 2776ed5289f9SKen Erickson /* 2777ed5289f9SKen Erickson * un-rot47 - obviously this 2778ed5289f9SKen Erickson * nonsense is ascii-specific 2779ed5289f9SKen Erickson */ 2780ed5289f9SKen Erickson for (c = (unsigned char *)tokbuf; 2781ed5289f9SKen Erickson *c != '\0'; c++) { 2782ed5289f9SKen Erickson *c += 47; 2783ed5289f9SKen Erickson if (*c > '~') 2784ed5289f9SKen Erickson *c -= 94; 2785ed5289f9SKen Erickson else if (*c < '!') 2786ed5289f9SKen Erickson *c += 94; 2787ed5289f9SKen Erickson } 2788ed5289f9SKen Erickson /* 2789ed5289f9SKen Erickson * now we should have a real number 2790ed5289f9SKen Erickson */ 2791ed5289f9SKen Erickson 2792ed5289f9SKen Erickson if (kobj_getvalue(tokbuf, &tmp) != 0) 2793ed5289f9SKen Erickson kobj_file_err(CE_WARN, file, 2794ed5289f9SKen Erickson "Bad value %s for hostid", 2795ed5289f9SKen Erickson tokbuf); 2796ed5289f9SKen Erickson else 2797ed5289f9SKen Erickson hostid = (int32_t)tmp; 2798ed5289f9SKen Erickson 2799ed5289f9SKen Erickson break; 2800ed5289f9SKen Erickson case EOF: 2801ed5289f9SKen Erickson done = 1; 2802ed5289f9SKen Erickson /* FALLTHROUGH */ 2803ed5289f9SKen Erickson case NEWLINE: 2804ed5289f9SKen Erickson kobj_newline(file); 2805ed5289f9SKen Erickson break; 2806ed5289f9SKen Erickson default: 2807ed5289f9SKen Erickson break; 2808ed5289f9SKen Erickson 2809ed5289f9SKen Erickson } 2810ed5289f9SKen Erickson } 28115679c89fSjv if (hostid == HW_INVALID_HOSTID) /* didn't find a hostid */ 2812ed5289f9SKen Erickson kobj_file_err(CE_WARN, file, 2813ed5289f9SKen Erickson "hostid missing or corrupt"); 2814ed5289f9SKen Erickson 2815ed5289f9SKen Erickson kobj_close_file(file); 2816ed5289f9SKen Erickson } 2817ed5289f9SKen Erickson /* 2818ed5289f9SKen Erickson * hostid is now the value read from /etc/hostid, or the 28195679c89fSjv * new hostid we generated in this routine or HW_INVALID_HOSTID if not 28205679c89fSjv * set. 2821ed5289f9SKen Erickson */ 2822ed5289f9SKen Erickson return (hostid); 2823ed5289f9SKen Erickson } 282496d008c2SKen Erickson 282596d008c2SKen Erickson static int 282696d008c2SKen Erickson atoi(char *p) 282796d008c2SKen Erickson { 282896d008c2SKen Erickson int i = 0; 282996d008c2SKen Erickson 283096d008c2SKen Erickson while (*p != '\0') 283196d008c2SKen Erickson i = 10 * i + (*p++ - '0'); 283296d008c2SKen Erickson 283396d008c2SKen Erickson return (i); 283496d008c2SKen Erickson } 283596d008c2SKen Erickson 2836ed5289f9SKen Erickson #endif /* _SOFT_HOSTID */ 2837ed5289f9SKen Erickson 28387c478bd9Sstevel@tonic-gate void 283945916cd2Sjpk get_system_configuration(void) 28407c478bd9Sstevel@tonic-gate { 28417c478bd9Sstevel@tonic-gate char prop[32]; 28427c478bd9Sstevel@tonic-gate u_longlong_t nodes_ll, cpus_pernode_ll, lvalue; 28437c478bd9Sstevel@tonic-gate 284435b1ab99Sjosephb if (BOP_GETPROPLEN(bootops, "nodes") > sizeof (prop) || 284535b1ab99Sjosephb BOP_GETPROP(bootops, "nodes", prop) < 0 || 284635b1ab99Sjosephb kobj_getvalue(prop, &nodes_ll) == -1 || 284735b1ab99Sjosephb nodes_ll > MAXNODES || 284835b1ab99Sjosephb BOP_GETPROPLEN(bootops, "cpus_pernode") > sizeof (prop) || 284935b1ab99Sjosephb BOP_GETPROP(bootops, "cpus_pernode", prop) < 0 || 285035b1ab99Sjosephb kobj_getvalue(prop, &cpus_pernode_ll) == -1) { 28517c478bd9Sstevel@tonic-gate system_hardware.hd_nodes = 1; 28527c478bd9Sstevel@tonic-gate system_hardware.hd_cpus_per_node = 0; 28537c478bd9Sstevel@tonic-gate } else { 28547c478bd9Sstevel@tonic-gate system_hardware.hd_nodes = (int)nodes_ll; 28557c478bd9Sstevel@tonic-gate system_hardware.hd_cpus_per_node = (int)cpus_pernode_ll; 28567c478bd9Sstevel@tonic-gate } 285735b1ab99Sjosephb 285835b1ab99Sjosephb if (BOP_GETPROPLEN(bootops, "kernelbase") > sizeof (prop) || 285935b1ab99Sjosephb BOP_GETPROP(bootops, "kernelbase", prop) < 0 || 286035b1ab99Sjosephb kobj_getvalue(prop, &lvalue) == -1) 286135b1ab99Sjosephb eprom_kernelbase = NULL; 28627c478bd9Sstevel@tonic-gate else 286335b1ab99Sjosephb eprom_kernelbase = (uintptr_t)lvalue; 28647c478bd9Sstevel@tonic-gate 286535b1ab99Sjosephb if (BOP_GETPROPLEN(bootops, "segmapsize") > sizeof (prop) || 286635b1ab99Sjosephb BOP_GETPROP(bootops, "segmapsize", prop) < 0 || 286735b1ab99Sjosephb kobj_getvalue(prop, &lvalue) == -1) 28687c478bd9Sstevel@tonic-gate segmapsize = SEGMAPDEFAULT; 286935b1ab99Sjosephb else 28707c478bd9Sstevel@tonic-gate segmapsize = (uintptr_t)lvalue; 28717c478bd9Sstevel@tonic-gate 287235b1ab99Sjosephb if (BOP_GETPROPLEN(bootops, "segmapfreelists") > sizeof (prop) || 287335b1ab99Sjosephb BOP_GETPROP(bootops, "segmapfreelists", prop) < 0 || 287435b1ab99Sjosephb kobj_getvalue(prop, &lvalue) == -1) 28757c478bd9Sstevel@tonic-gate segmapfreelists = 0; /* use segmap driver default */ 287635b1ab99Sjosephb else 28777c478bd9Sstevel@tonic-gate segmapfreelists = (int)lvalue; 28784944b02eSkchow 2879aac11643Sjosephb /* physmem used to be here, but moved much earlier to fakebop.c */ 28807c478bd9Sstevel@tonic-gate } 28817c478bd9Sstevel@tonic-gate 28827c478bd9Sstevel@tonic-gate /* 28837c478bd9Sstevel@tonic-gate * Add to a memory list. 28847c478bd9Sstevel@tonic-gate * start = start of new memory segment 28857c478bd9Sstevel@tonic-gate * len = length of new memory segment in bytes 28867c478bd9Sstevel@tonic-gate * new = pointer to a new struct memlist 28877c478bd9Sstevel@tonic-gate * memlistp = memory list to which to add segment. 28887c478bd9Sstevel@tonic-gate */ 2889ae115bc7Smrj void 28907c478bd9Sstevel@tonic-gate memlist_add( 28917c478bd9Sstevel@tonic-gate uint64_t start, 28927c478bd9Sstevel@tonic-gate uint64_t len, 28937c478bd9Sstevel@tonic-gate struct memlist *new, 28947c478bd9Sstevel@tonic-gate struct memlist **memlistp) 28957c478bd9Sstevel@tonic-gate { 28967c478bd9Sstevel@tonic-gate struct memlist *cur; 28977c478bd9Sstevel@tonic-gate uint64_t end = start + len; 28987c478bd9Sstevel@tonic-gate 289956f33205SJonathan Adams new->ml_address = start; 290056f33205SJonathan Adams new->ml_size = len; 29017c478bd9Sstevel@tonic-gate 29027c478bd9Sstevel@tonic-gate cur = *memlistp; 29037c478bd9Sstevel@tonic-gate 29047c478bd9Sstevel@tonic-gate while (cur) { 290556f33205SJonathan Adams if (cur->ml_address >= end) { 290656f33205SJonathan Adams new->ml_next = cur; 29077c478bd9Sstevel@tonic-gate *memlistp = new; 290856f33205SJonathan Adams new->ml_prev = cur->ml_prev; 290956f33205SJonathan Adams cur->ml_prev = new; 29107c478bd9Sstevel@tonic-gate return; 29117c478bd9Sstevel@tonic-gate } 291256f33205SJonathan Adams ASSERT(cur->ml_address + cur->ml_size <= start); 291356f33205SJonathan Adams if (cur->ml_next == NULL) { 291456f33205SJonathan Adams cur->ml_next = new; 291556f33205SJonathan Adams new->ml_prev = cur; 291656f33205SJonathan Adams new->ml_next = NULL; 29177c478bd9Sstevel@tonic-gate return; 29187c478bd9Sstevel@tonic-gate } 291956f33205SJonathan Adams memlistp = &cur->ml_next; 292056f33205SJonathan Adams cur = cur->ml_next; 29217c478bd9Sstevel@tonic-gate } 29227c478bd9Sstevel@tonic-gate } 29237c478bd9Sstevel@tonic-gate 29247c478bd9Sstevel@tonic-gate void 29257c478bd9Sstevel@tonic-gate kobj_vmem_init(vmem_t **text_arena, vmem_t **data_arena) 29267c478bd9Sstevel@tonic-gate { 29277c478bd9Sstevel@tonic-gate size_t tsize = e_modtext - modtext; 29287c478bd9Sstevel@tonic-gate size_t dsize = e_moddata - moddata; 29297c478bd9Sstevel@tonic-gate 29307c478bd9Sstevel@tonic-gate *text_arena = vmem_create("module_text", tsize ? modtext : NULL, tsize, 29317c478bd9Sstevel@tonic-gate 1, segkmem_alloc, segkmem_free, heaptext_arena, 0, VM_SLEEP); 29327c478bd9Sstevel@tonic-gate *data_arena = vmem_create("module_data", dsize ? moddata : NULL, dsize, 29337c478bd9Sstevel@tonic-gate 1, segkmem_alloc, segkmem_free, heap32_arena, 0, VM_SLEEP); 29347c478bd9Sstevel@tonic-gate } 29357c478bd9Sstevel@tonic-gate 29367c478bd9Sstevel@tonic-gate caddr_t 29377c478bd9Sstevel@tonic-gate kobj_text_alloc(vmem_t *arena, size_t size) 29387c478bd9Sstevel@tonic-gate { 29397c478bd9Sstevel@tonic-gate return (vmem_alloc(arena, size, VM_SLEEP | VM_BESTFIT)); 29407c478bd9Sstevel@tonic-gate } 29417c478bd9Sstevel@tonic-gate 29427c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 29437c478bd9Sstevel@tonic-gate caddr_t 29447c478bd9Sstevel@tonic-gate kobj_texthole_alloc(caddr_t addr, size_t size) 29457c478bd9Sstevel@tonic-gate { 29467c478bd9Sstevel@tonic-gate panic("unexpected call to kobj_texthole_alloc()"); 29477c478bd9Sstevel@tonic-gate /*NOTREACHED*/ 29487c478bd9Sstevel@tonic-gate return (0); 29497c478bd9Sstevel@tonic-gate } 29507c478bd9Sstevel@tonic-gate 29517c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 29527c478bd9Sstevel@tonic-gate void 29537c478bd9Sstevel@tonic-gate kobj_texthole_free(caddr_t addr, size_t size) 29547c478bd9Sstevel@tonic-gate { 29557c478bd9Sstevel@tonic-gate panic("unexpected call to kobj_texthole_free()"); 29567c478bd9Sstevel@tonic-gate } 29577c478bd9Sstevel@tonic-gate 29587c478bd9Sstevel@tonic-gate /* 29597c478bd9Sstevel@tonic-gate * This is called just after configure() in startup(). 29607c478bd9Sstevel@tonic-gate * 29617c478bd9Sstevel@tonic-gate * The ISALIST concept is a bit hopeless on Intel, because 29627c478bd9Sstevel@tonic-gate * there's no guarantee of an ever-more-capable processor 29637c478bd9Sstevel@tonic-gate * given that various parts of the instruction set may appear 29647c478bd9Sstevel@tonic-gate * and disappear between different implementations. 29657c478bd9Sstevel@tonic-gate * 29667c478bd9Sstevel@tonic-gate * While it would be possible to correct it and even enhance 29677c478bd9Sstevel@tonic-gate * it somewhat, the explicit hardware capability bitmask allows 29687c478bd9Sstevel@tonic-gate * more flexibility. 29697c478bd9Sstevel@tonic-gate * 29707c478bd9Sstevel@tonic-gate * So, we just leave this alone. 29717c478bd9Sstevel@tonic-gate */ 29727c478bd9Sstevel@tonic-gate void 29737c478bd9Sstevel@tonic-gate setx86isalist(void) 29747c478bd9Sstevel@tonic-gate { 29757c478bd9Sstevel@tonic-gate char *tp; 29767c478bd9Sstevel@tonic-gate size_t len; 29777c478bd9Sstevel@tonic-gate extern char *isa_list; 29787c478bd9Sstevel@tonic-gate 29797c478bd9Sstevel@tonic-gate #define TBUFSIZE 1024 29807c478bd9Sstevel@tonic-gate 29817c478bd9Sstevel@tonic-gate tp = kmem_alloc(TBUFSIZE, KM_SLEEP); 29827c478bd9Sstevel@tonic-gate *tp = '\0'; 29837c478bd9Sstevel@tonic-gate 29847c478bd9Sstevel@tonic-gate #if defined(__amd64) 29857c478bd9Sstevel@tonic-gate (void) strcpy(tp, "amd64 "); 29867c478bd9Sstevel@tonic-gate #endif 29877c478bd9Sstevel@tonic-gate 29887c478bd9Sstevel@tonic-gate switch (x86_vendor) { 29897c478bd9Sstevel@tonic-gate case X86_VENDOR_Intel: 29907c478bd9Sstevel@tonic-gate case X86_VENDOR_AMD: 29917c478bd9Sstevel@tonic-gate case X86_VENDOR_TM: 29927c478bd9Sstevel@tonic-gate if (x86_feature & X86_CMOV) { 29937c478bd9Sstevel@tonic-gate /* 29947c478bd9Sstevel@tonic-gate * Pentium Pro or later 29957c478bd9Sstevel@tonic-gate */ 29967c478bd9Sstevel@tonic-gate (void) strcat(tp, "pentium_pro"); 29977c478bd9Sstevel@tonic-gate (void) strcat(tp, x86_feature & X86_MMX ? 29987c478bd9Sstevel@tonic-gate "+mmx pentium_pro " : " "); 29997c478bd9Sstevel@tonic-gate } 30007c478bd9Sstevel@tonic-gate /*FALLTHROUGH*/ 30017c478bd9Sstevel@tonic-gate case X86_VENDOR_Cyrix: 30027c478bd9Sstevel@tonic-gate /* 30037c478bd9Sstevel@tonic-gate * The Cyrix 6x86 does not have any Pentium features 30047c478bd9Sstevel@tonic-gate * accessible while not at privilege level 0. 30057c478bd9Sstevel@tonic-gate */ 30067c478bd9Sstevel@tonic-gate if (x86_feature & X86_CPUID) { 30077c478bd9Sstevel@tonic-gate (void) strcat(tp, "pentium"); 30087c478bd9Sstevel@tonic-gate (void) strcat(tp, x86_feature & X86_MMX ? 30097c478bd9Sstevel@tonic-gate "+mmx pentium " : " "); 30107c478bd9Sstevel@tonic-gate } 30117c478bd9Sstevel@tonic-gate break; 30127c478bd9Sstevel@tonic-gate default: 30137c478bd9Sstevel@tonic-gate break; 30147c478bd9Sstevel@tonic-gate } 30157c478bd9Sstevel@tonic-gate (void) strcat(tp, "i486 i386 i86"); 30167c478bd9Sstevel@tonic-gate len = strlen(tp) + 1; /* account for NULL at end of string */ 30177c478bd9Sstevel@tonic-gate isa_list = strcpy(kmem_alloc(len, KM_SLEEP), tp); 30187c478bd9Sstevel@tonic-gate kmem_free(tp, TBUFSIZE); 30197c478bd9Sstevel@tonic-gate 30207c478bd9Sstevel@tonic-gate #undef TBUFSIZE 30217c478bd9Sstevel@tonic-gate } 30227c478bd9Sstevel@tonic-gate 30237c478bd9Sstevel@tonic-gate 30247c478bd9Sstevel@tonic-gate #ifdef __amd64 30257c478bd9Sstevel@tonic-gate 30267c478bd9Sstevel@tonic-gate void * 30277c478bd9Sstevel@tonic-gate device_arena_alloc(size_t size, int vm_flag) 30287c478bd9Sstevel@tonic-gate { 30297c478bd9Sstevel@tonic-gate return (vmem_alloc(device_arena, size, vm_flag)); 30307c478bd9Sstevel@tonic-gate } 30317c478bd9Sstevel@tonic-gate 30327c478bd9Sstevel@tonic-gate void 30337c478bd9Sstevel@tonic-gate device_arena_free(void *vaddr, size_t size) 30347c478bd9Sstevel@tonic-gate { 30357c478bd9Sstevel@tonic-gate vmem_free(device_arena, vaddr, size); 30367c478bd9Sstevel@tonic-gate } 30377c478bd9Sstevel@tonic-gate 3038ae115bc7Smrj #else /* __i386 */ 30397c478bd9Sstevel@tonic-gate 30407c478bd9Sstevel@tonic-gate void * 30417c478bd9Sstevel@tonic-gate device_arena_alloc(size_t size, int vm_flag) 30427c478bd9Sstevel@tonic-gate { 30437c478bd9Sstevel@tonic-gate caddr_t vaddr; 30447c478bd9Sstevel@tonic-gate uintptr_t v; 30457c478bd9Sstevel@tonic-gate size_t start; 30467c478bd9Sstevel@tonic-gate size_t end; 30477c478bd9Sstevel@tonic-gate 30487c478bd9Sstevel@tonic-gate vaddr = vmem_alloc(heap_arena, size, vm_flag); 30497c478bd9Sstevel@tonic-gate if (vaddr == NULL) 30507c478bd9Sstevel@tonic-gate return (NULL); 30517c478bd9Sstevel@tonic-gate 30527c478bd9Sstevel@tonic-gate v = (uintptr_t)vaddr; 30537c478bd9Sstevel@tonic-gate ASSERT(v >= kernelbase); 3054ae115bc7Smrj ASSERT(v + size <= valloc_base); 30557c478bd9Sstevel@tonic-gate 30567c478bd9Sstevel@tonic-gate start = btop(v - kernelbase); 30577c478bd9Sstevel@tonic-gate end = btop(v + size - 1 - kernelbase); 30587c478bd9Sstevel@tonic-gate ASSERT(start < toxic_bit_map_len); 30597c478bd9Sstevel@tonic-gate ASSERT(end < toxic_bit_map_len); 30607c478bd9Sstevel@tonic-gate 30617c478bd9Sstevel@tonic-gate while (start <= end) { 30627c478bd9Sstevel@tonic-gate BT_ATOMIC_SET(toxic_bit_map, start); 30637c478bd9Sstevel@tonic-gate ++start; 30647c478bd9Sstevel@tonic-gate } 30657c478bd9Sstevel@tonic-gate return (vaddr); 30667c478bd9Sstevel@tonic-gate } 30677c478bd9Sstevel@tonic-gate 30687c478bd9Sstevel@tonic-gate void 30697c478bd9Sstevel@tonic-gate device_arena_free(void *vaddr, size_t size) 30707c478bd9Sstevel@tonic-gate { 30717c478bd9Sstevel@tonic-gate uintptr_t v = (uintptr_t)vaddr; 30727c478bd9Sstevel@tonic-gate size_t start; 30737c478bd9Sstevel@tonic-gate size_t end; 30747c478bd9Sstevel@tonic-gate 30757c478bd9Sstevel@tonic-gate ASSERT(v >= kernelbase); 3076ae115bc7Smrj ASSERT(v + size <= valloc_base); 30777c478bd9Sstevel@tonic-gate 30787c478bd9Sstevel@tonic-gate start = btop(v - kernelbase); 30797c478bd9Sstevel@tonic-gate end = btop(v + size - 1 - kernelbase); 30807c478bd9Sstevel@tonic-gate ASSERT(start < toxic_bit_map_len); 30817c478bd9Sstevel@tonic-gate ASSERT(end < toxic_bit_map_len); 30827c478bd9Sstevel@tonic-gate 30837c478bd9Sstevel@tonic-gate while (start <= end) { 30847c478bd9Sstevel@tonic-gate ASSERT(BT_TEST(toxic_bit_map, start) != 0); 30857c478bd9Sstevel@tonic-gate BT_ATOMIC_CLEAR(toxic_bit_map, start); 30867c478bd9Sstevel@tonic-gate ++start; 30877c478bd9Sstevel@tonic-gate } 30887c478bd9Sstevel@tonic-gate vmem_free(heap_arena, vaddr, size); 30897c478bd9Sstevel@tonic-gate } 30907c478bd9Sstevel@tonic-gate 30917c478bd9Sstevel@tonic-gate /* 30927c478bd9Sstevel@tonic-gate * returns 1st address in range that is in device arena, or NULL 30937c478bd9Sstevel@tonic-gate * if len is not NULL it returns the length of the toxic range 30947c478bd9Sstevel@tonic-gate */ 30957c478bd9Sstevel@tonic-gate void * 30967c478bd9Sstevel@tonic-gate device_arena_contains(void *vaddr, size_t size, size_t *len) 30977c478bd9Sstevel@tonic-gate { 30987c478bd9Sstevel@tonic-gate uintptr_t v = (uintptr_t)vaddr; 30997c478bd9Sstevel@tonic-gate uintptr_t eaddr = v + size; 31007c478bd9Sstevel@tonic-gate size_t start; 31017c478bd9Sstevel@tonic-gate size_t end; 31027c478bd9Sstevel@tonic-gate 31037c478bd9Sstevel@tonic-gate /* 31047c478bd9Sstevel@tonic-gate * if called very early by kmdb, just return NULL 31057c478bd9Sstevel@tonic-gate */ 31067c478bd9Sstevel@tonic-gate if (toxic_bit_map == NULL) 31077c478bd9Sstevel@tonic-gate return (NULL); 31087c478bd9Sstevel@tonic-gate 31097c478bd9Sstevel@tonic-gate /* 31107c478bd9Sstevel@tonic-gate * First check if we're completely outside the bitmap range. 31117c478bd9Sstevel@tonic-gate */ 3112ae115bc7Smrj if (v >= valloc_base || eaddr < kernelbase) 31137c478bd9Sstevel@tonic-gate return (NULL); 31147c478bd9Sstevel@tonic-gate 31157c478bd9Sstevel@tonic-gate /* 31167c478bd9Sstevel@tonic-gate * Trim ends of search to look at only what the bitmap covers. 31177c478bd9Sstevel@tonic-gate */ 31187c478bd9Sstevel@tonic-gate if (v < kernelbase) 31197c478bd9Sstevel@tonic-gate v = kernelbase; 31207c478bd9Sstevel@tonic-gate start = btop(v - kernelbase); 31217c478bd9Sstevel@tonic-gate end = btop(eaddr - kernelbase); 31227c478bd9Sstevel@tonic-gate if (end >= toxic_bit_map_len) 31237c478bd9Sstevel@tonic-gate end = toxic_bit_map_len; 31247c478bd9Sstevel@tonic-gate 31257c478bd9Sstevel@tonic-gate if (bt_range(toxic_bit_map, &start, &end, end) == 0) 31267c478bd9Sstevel@tonic-gate return (NULL); 31277c478bd9Sstevel@tonic-gate 31287c478bd9Sstevel@tonic-gate v = kernelbase + ptob(start); 31297c478bd9Sstevel@tonic-gate if (len != NULL) 31307c478bd9Sstevel@tonic-gate *len = ptob(end - start); 31317c478bd9Sstevel@tonic-gate return ((void *)v); 31327c478bd9Sstevel@tonic-gate } 31337c478bd9Sstevel@tonic-gate 3134ae115bc7Smrj #endif /* __i386 */ 3135