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*5679c89fSjv * Copyright 2009 Sun Microsystems, Inc. All rights reserved. 237c478bd9Sstevel@tonic-gate * Use is subject to license terms. 247c478bd9Sstevel@tonic-gate */ 257c478bd9Sstevel@tonic-gate 267c478bd9Sstevel@tonic-gate #include <sys/types.h> 277c478bd9Sstevel@tonic-gate #include <sys/t_lock.h> 287c478bd9Sstevel@tonic-gate #include <sys/param.h> 297c478bd9Sstevel@tonic-gate #include <sys/sysmacros.h> 307c478bd9Sstevel@tonic-gate #include <sys/signal.h> 317c478bd9Sstevel@tonic-gate #include <sys/systm.h> 327c478bd9Sstevel@tonic-gate #include <sys/user.h> 337c478bd9Sstevel@tonic-gate #include <sys/mman.h> 347c478bd9Sstevel@tonic-gate #include <sys/vm.h> 357c478bd9Sstevel@tonic-gate #include <sys/conf.h> 367c478bd9Sstevel@tonic-gate #include <sys/avintr.h> 377c478bd9Sstevel@tonic-gate #include <sys/autoconf.h> 387c478bd9Sstevel@tonic-gate #include <sys/disp.h> 397c478bd9Sstevel@tonic-gate #include <sys/class.h> 407c478bd9Sstevel@tonic-gate #include <sys/bitmap.h> 417c478bd9Sstevel@tonic-gate 427c478bd9Sstevel@tonic-gate #include <sys/privregs.h> 437c478bd9Sstevel@tonic-gate 447c478bd9Sstevel@tonic-gate #include <sys/proc.h> 457c478bd9Sstevel@tonic-gate #include <sys/buf.h> 467c478bd9Sstevel@tonic-gate #include <sys/kmem.h> 47ae115bc7Smrj #include <sys/mem.h> 487c478bd9Sstevel@tonic-gate #include <sys/kstat.h> 497c478bd9Sstevel@tonic-gate 507c478bd9Sstevel@tonic-gate #include <sys/reboot.h> 517c478bd9Sstevel@tonic-gate 527c478bd9Sstevel@tonic-gate #include <sys/cred.h> 537c478bd9Sstevel@tonic-gate #include <sys/vnode.h> 547c478bd9Sstevel@tonic-gate #include <sys/file.h> 557c478bd9Sstevel@tonic-gate 567c478bd9Sstevel@tonic-gate #include <sys/procfs.h> 577c478bd9Sstevel@tonic-gate 587c478bd9Sstevel@tonic-gate #include <sys/vfs.h> 597c478bd9Sstevel@tonic-gate #include <sys/cmn_err.h> 607c478bd9Sstevel@tonic-gate #include <sys/utsname.h> 617c478bd9Sstevel@tonic-gate #include <sys/debug.h> 627c478bd9Sstevel@tonic-gate #include <sys/kdi.h> 637c478bd9Sstevel@tonic-gate 647c478bd9Sstevel@tonic-gate #include <sys/dumphdr.h> 657c478bd9Sstevel@tonic-gate #include <sys/bootconf.h> 667c478bd9Sstevel@tonic-gate #include <sys/varargs.h> 677c478bd9Sstevel@tonic-gate #include <sys/promif.h> 68843e1988Sjohnlev #include <sys/modctl.h> 697c478bd9Sstevel@tonic-gate 707c478bd9Sstevel@tonic-gate #include <sys/sunddi.h> 717c478bd9Sstevel@tonic-gate #include <sys/sunndi.h> 727c478bd9Sstevel@tonic-gate #include <sys/ndi_impldefs.h> 737c478bd9Sstevel@tonic-gate #include <sys/ddidmareq.h> 747c478bd9Sstevel@tonic-gate #include <sys/psw.h> 757c478bd9Sstevel@tonic-gate #include <sys/regset.h> 767c478bd9Sstevel@tonic-gate #include <sys/clock.h> 777c478bd9Sstevel@tonic-gate #include <sys/pte.h> 787c478bd9Sstevel@tonic-gate #include <sys/tss.h> 797c478bd9Sstevel@tonic-gate #include <sys/stack.h> 807c478bd9Sstevel@tonic-gate #include <sys/trap.h> 817c478bd9Sstevel@tonic-gate #include <sys/fp.h> 821d03c31eSjohnlev #include <vm/kboot_mmu.h> 837c478bd9Sstevel@tonic-gate #include <vm/anon.h> 847c478bd9Sstevel@tonic-gate #include <vm/as.h> 857c478bd9Sstevel@tonic-gate #include <vm/page.h> 867c478bd9Sstevel@tonic-gate #include <vm/seg.h> 877c478bd9Sstevel@tonic-gate #include <vm/seg_dev.h> 887c478bd9Sstevel@tonic-gate #include <vm/seg_kmem.h> 897c478bd9Sstevel@tonic-gate #include <vm/seg_kpm.h> 907c478bd9Sstevel@tonic-gate #include <vm/seg_map.h> 917c478bd9Sstevel@tonic-gate #include <vm/seg_vn.h> 927c478bd9Sstevel@tonic-gate #include <vm/seg_kp.h> 937c478bd9Sstevel@tonic-gate #include <sys/memnode.h> 947c478bd9Sstevel@tonic-gate #include <vm/vm_dep.h> 957c478bd9Sstevel@tonic-gate #include <sys/thread.h> 967c478bd9Sstevel@tonic-gate #include <sys/sysconf.h> 977c478bd9Sstevel@tonic-gate #include <sys/vm_machparam.h> 987c478bd9Sstevel@tonic-gate #include <sys/archsystm.h> 997c478bd9Sstevel@tonic-gate #include <sys/machsystm.h> 1007c478bd9Sstevel@tonic-gate #include <vm/hat.h> 1017c478bd9Sstevel@tonic-gate #include <vm/hat_i86.h> 1027c478bd9Sstevel@tonic-gate #include <sys/pmem.h> 1037c478bd9Sstevel@tonic-gate #include <sys/smp_impldefs.h> 1047c478bd9Sstevel@tonic-gate #include <sys/x86_archext.h> 105a563a037Sbholler #include <sys/cpuvar.h> 1067c478bd9Sstevel@tonic-gate #include <sys/segments.h> 1077c478bd9Sstevel@tonic-gate #include <sys/clconf.h> 1087c478bd9Sstevel@tonic-gate #include <sys/kobj.h> 1097c478bd9Sstevel@tonic-gate #include <sys/kobj_lex.h> 1107c478bd9Sstevel@tonic-gate #include <sys/cpc_impl.h> 1117c478bd9Sstevel@tonic-gate #include <sys/x86_archext.h> 1127aec1d6eScindi #include <sys/cpu_module.h> 11384ab085aSmws #include <sys/smbios.h> 114ae115bc7Smrj #include <sys/debug_info.h> 1151d03c31eSjohnlev #include <sys/bootinfo.h> 116dd4eeefdSeota #include <sys/ddi_timer.h> 117*5679c89fSjv #include <sys/systeminfo.h> 11819397407SSherry Moore #include <sys/multiboot.h> 119ae115bc7Smrj 120843e1988Sjohnlev #ifdef __xpv 1211d03c31eSjohnlev 122843e1988Sjohnlev #include <sys/hypervisor.h> 123843e1988Sjohnlev #include <sys/xen_mmu.h> 124843e1988Sjohnlev #include <sys/evtchn_impl.h> 125843e1988Sjohnlev #include <sys/gnttab.h> 126843e1988Sjohnlev #include <sys/xpv_panic.h> 127843e1988Sjohnlev #include <xen/sys/xenbus_comms.h> 128843e1988Sjohnlev #include <xen/public/physdev.h> 1291d03c31eSjohnlev 130843e1988Sjohnlev extern void xen_late_startup(void); 131ae115bc7Smrj 1321d03c31eSjohnlev struct xen_evt_data cpu0_evt_data; 1331d03c31eSjohnlev 1341d03c31eSjohnlev #endif /* __xpv */ 1357c478bd9Sstevel@tonic-gate 1367c478bd9Sstevel@tonic-gate extern void progressbar_init(void); 1377c478bd9Sstevel@tonic-gate extern void progressbar_start(void); 1389acbbeafSnn extern void brand_init(void); 13906fb6a36Sdv extern void pcf_init(void); 1407c478bd9Sstevel@tonic-gate 141d7d93655Sblakej extern int size_pse_array(pgcnt_t, int); 142d7d93655Sblakej 143ed5289f9SKen Erickson #if defined(_SOFT_HOSTID) 144ed5289f9SKen Erickson 145ed5289f9SKen Erickson #include <sys/rtc.h> 146ed5289f9SKen Erickson 147ed5289f9SKen Erickson static int32_t set_soft_hostid(void); 148ed5289f9SKen Erickson static char hostid_file[] = "/etc/hostid"; 149ed5289f9SKen Erickson 150ed5289f9SKen Erickson #endif 151ed5289f9SKen Erickson 15294f1124eSVikram Hegde void *gfx_devinfo_list; 153ed5289f9SKen Erickson 1547c478bd9Sstevel@tonic-gate /* 1557c478bd9Sstevel@tonic-gate * XXX make declaration below "static" when drivers no longer use this 1567c478bd9Sstevel@tonic-gate * interface. 1577c478bd9Sstevel@tonic-gate */ 1587c478bd9Sstevel@tonic-gate extern caddr_t p0_va; /* Virtual address for accessing physical page 0 */ 1597c478bd9Sstevel@tonic-gate 1607c478bd9Sstevel@tonic-gate /* 1617c478bd9Sstevel@tonic-gate * segkp 1627c478bd9Sstevel@tonic-gate */ 1637c478bd9Sstevel@tonic-gate extern int segkp_fromheap; 1647c478bd9Sstevel@tonic-gate 1657c478bd9Sstevel@tonic-gate static void kvm_init(void); 1667c478bd9Sstevel@tonic-gate static void startup_init(void); 1677c478bd9Sstevel@tonic-gate static void startup_memlist(void); 168ae115bc7Smrj static void startup_kmem(void); 1697c478bd9Sstevel@tonic-gate static void startup_modules(void); 1707c478bd9Sstevel@tonic-gate static void startup_vm(void); 1717c478bd9Sstevel@tonic-gate static void startup_end(void); 17235b1ab99Sjosephb static void layout_kernel_va(void); 1737c478bd9Sstevel@tonic-gate 1747c478bd9Sstevel@tonic-gate /* 1757c478bd9Sstevel@tonic-gate * Declare these as initialized data so we can patch them. 1767c478bd9Sstevel@tonic-gate */ 177f53ad214Skchow #ifdef __i386 17835b1ab99Sjosephb 179f53ad214Skchow /* 180f53ad214Skchow * Due to virtual address space limitations running in 32 bit mode, restrict 18135b1ab99Sjosephb * the amount of physical memory configured to a max of PHYSMEM pages (16g). 182f53ad214Skchow * 183f53ad214Skchow * If the physical max memory size of 64g were allowed to be configured, the 184f53ad214Skchow * size of user virtual address space will be less than 1g. A limited user 185f53ad214Skchow * address space greatly reduces the range of applications that can run. 186f53ad214Skchow * 18735b1ab99Sjosephb * If more physical memory than PHYSMEM is required, users should preferably 18835b1ab99Sjosephb * run in 64 bit mode which has far looser virtual address space limitations. 189f53ad214Skchow * 190f53ad214Skchow * If 64 bit mode is not available (as in IA32) and/or more physical memory 19135b1ab99Sjosephb * than PHYSMEM is required in 32 bit mode, physmem can be set to the desired 192f53ad214Skchow * value or to 0 (to configure all available memory) via eeprom(1M). kernelbase 193f53ad214Skchow * should also be carefully tuned to balance out the need of the user 194f53ad214Skchow * application while minimizing the risk of kernel heap exhaustion due to 195f53ad214Skchow * kernelbase being set too high. 196f53ad214Skchow */ 19735b1ab99Sjosephb #define PHYSMEM 0x400000 198f53ad214Skchow 19935b1ab99Sjosephb #else /* __amd64 */ 20035b1ab99Sjosephb 20135b1ab99Sjosephb /* 20235b1ab99Sjosephb * For now we can handle memory with physical addresses up to about 20335b1ab99Sjosephb * 64 Terabytes. This keeps the kernel above the VA hole, leaving roughly 20435b1ab99Sjosephb * half the VA space for seg_kpm. When systems get bigger than 64TB this 20535b1ab99Sjosephb * code will need revisiting. There is an implicit assumption that there 20635b1ab99Sjosephb * are no *huge* holes in the physical address space too. 20735b1ab99Sjosephb */ 20835b1ab99Sjosephb #define TERABYTE (1ul << 40) 20935b1ab99Sjosephb #define PHYSMEM_MAX64 mmu_btop(64 * TERABYTE) 21035b1ab99Sjosephb #define PHYSMEM PHYSMEM_MAX64 21135b1ab99Sjosephb #define AMD64_VA_HOLE_END 0xFFFF800000000000ul 21235b1ab99Sjosephb 21335b1ab99Sjosephb #endif /* __amd64 */ 21435b1ab99Sjosephb 21535b1ab99Sjosephb pgcnt_t physmem = PHYSMEM; 2167c478bd9Sstevel@tonic-gate pgcnt_t obp_pages; /* Memory used by PROM for its text and data */ 2177c478bd9Sstevel@tonic-gate 2187c478bd9Sstevel@tonic-gate char *kobj_file_buf; 2197c478bd9Sstevel@tonic-gate int kobj_file_bufsize; /* set in /etc/system */ 2207c478bd9Sstevel@tonic-gate 2217c478bd9Sstevel@tonic-gate /* Global variables for MP support. Used in mp_startup */ 22219397407SSherry Moore caddr_t rm_platter_va = 0; 2237c478bd9Sstevel@tonic-gate uint32_t rm_platter_pa; 2247c478bd9Sstevel@tonic-gate 22583f9b804Skchow int auto_lpg_disable = 1; 22683f9b804Skchow 2277c478bd9Sstevel@tonic-gate /* 2287c478bd9Sstevel@tonic-gate * Some CPUs have holes in the middle of the 64-bit virtual address range. 2297c478bd9Sstevel@tonic-gate */ 2307c478bd9Sstevel@tonic-gate uintptr_t hole_start, hole_end; 2317c478bd9Sstevel@tonic-gate 2327c478bd9Sstevel@tonic-gate /* 2337c478bd9Sstevel@tonic-gate * kpm mapping window 2347c478bd9Sstevel@tonic-gate */ 2357c478bd9Sstevel@tonic-gate caddr_t kpm_vbase; 2367c478bd9Sstevel@tonic-gate size_t kpm_size; 23735b1ab99Sjosephb static int kpm_desired; 23835b1ab99Sjosephb #ifdef __amd64 23935b1ab99Sjosephb static uintptr_t segkpm_base = (uintptr_t)SEGKPM_BASE; 24035b1ab99Sjosephb #endif 2417c478bd9Sstevel@tonic-gate 2427c478bd9Sstevel@tonic-gate /* 2437c478bd9Sstevel@tonic-gate * Configuration parameters set at boot time. 2447c478bd9Sstevel@tonic-gate */ 2457c478bd9Sstevel@tonic-gate 2467c478bd9Sstevel@tonic-gate caddr_t econtig; /* end of first block of contiguous kernel */ 2477c478bd9Sstevel@tonic-gate 2487c478bd9Sstevel@tonic-gate struct bootops *bootops = 0; /* passed in from boot */ 2497c478bd9Sstevel@tonic-gate struct bootops **bootopsp; 2507c478bd9Sstevel@tonic-gate struct boot_syscalls *sysp; /* passed in from boot */ 2517c478bd9Sstevel@tonic-gate 2527c478bd9Sstevel@tonic-gate char bootblock_fstype[16]; 2537c478bd9Sstevel@tonic-gate 2547c478bd9Sstevel@tonic-gate char kern_bootargs[OBP_MAXPATHLEN]; 25519397407SSherry Moore char kern_bootfile[OBP_MAXPATHLEN]; 2567c478bd9Sstevel@tonic-gate 257ad23a2dbSjohansen /* 258ad23a2dbSjohansen * ZFS zio segment. This allows us to exclude large portions of ZFS data that 259ad23a2dbSjohansen * gets cached in kmem caches on the heap. If this is set to zero, we allocate 260ad23a2dbSjohansen * zio buffers from their own segment, otherwise they are allocated from the 261ad23a2dbSjohansen * heap. The optimization of allocating zio buffers from their own segment is 262ad23a2dbSjohansen * only valid on 64-bit kernels. 263ad23a2dbSjohansen */ 264ad23a2dbSjohansen #if defined(__amd64) 265ad23a2dbSjohansen int segzio_fromheap = 0; 266ad23a2dbSjohansen #else 267ad23a2dbSjohansen int segzio_fromheap = 1; 268ad23a2dbSjohansen #endif 269ad23a2dbSjohansen 2707c478bd9Sstevel@tonic-gate /* 2717c478bd9Sstevel@tonic-gate * new memory fragmentations are possible in startup() due to BOP_ALLOCs. this 2727c478bd9Sstevel@tonic-gate * depends on number of BOP_ALLOC calls made and requested size, memory size 2737c478bd9Sstevel@tonic-gate * combination and whether boot.bin memory needs to be freed. 2747c478bd9Sstevel@tonic-gate */ 2757c478bd9Sstevel@tonic-gate #define POSS_NEW_FRAGMENTS 12 2767c478bd9Sstevel@tonic-gate 2777c478bd9Sstevel@tonic-gate /* 2787c478bd9Sstevel@tonic-gate * VM data structures 2797c478bd9Sstevel@tonic-gate */ 2807c478bd9Sstevel@tonic-gate long page_hashsz; /* Size of page hash table (power of two) */ 2817c478bd9Sstevel@tonic-gate struct page *pp_base; /* Base of initial system page struct array */ 2827c478bd9Sstevel@tonic-gate struct page **page_hash; /* Page hash table */ 283d7d93655Sblakej pad_mutex_t *pse_mutex; /* Locks protecting pp->p_selock */ 284d7d93655Sblakej size_t pse_table_size; /* Number of mutexes in pse_mutex[] */ 285d7d93655Sblakej int pse_shift; /* log2(pse_table_size) */ 2867c478bd9Sstevel@tonic-gate struct seg ktextseg; /* Segment used for kernel executable image */ 2877c478bd9Sstevel@tonic-gate struct seg kvalloc; /* Segment used for "valloc" mapping */ 2887c478bd9Sstevel@tonic-gate struct seg kpseg; /* Segment used for pageable kernel virt mem */ 2897c478bd9Sstevel@tonic-gate struct seg kmapseg; /* Segment used for generic kernel mappings */ 2907c478bd9Sstevel@tonic-gate struct seg kdebugseg; /* Segment used for the kernel debugger */ 2917c478bd9Sstevel@tonic-gate 2927c478bd9Sstevel@tonic-gate struct seg *segkmap = &kmapseg; /* Kernel generic mapping segment */ 293ae115bc7Smrj static struct seg *segmap = &kmapseg; /* easier to use name for in here */ 294ae115bc7Smrj 2957c478bd9Sstevel@tonic-gate struct seg *segkp = &kpseg; /* Pageable kernel virtual memory segment */ 2967c478bd9Sstevel@tonic-gate 2977c478bd9Sstevel@tonic-gate #if defined(__amd64) 2987c478bd9Sstevel@tonic-gate struct seg kvseg_core; /* Segment used for the core heap */ 2997c478bd9Sstevel@tonic-gate struct seg kpmseg; /* Segment used for physical mapping */ 3007c478bd9Sstevel@tonic-gate struct seg *segkpm = &kpmseg; /* 64bit kernel physical mapping segment */ 3017c478bd9Sstevel@tonic-gate #else 3027c478bd9Sstevel@tonic-gate struct seg *segkpm = NULL; /* Unused on IA32 */ 3037c478bd9Sstevel@tonic-gate #endif 3047c478bd9Sstevel@tonic-gate 3057c478bd9Sstevel@tonic-gate caddr_t segkp_base; /* Base address of segkp */ 306ad23a2dbSjohansen caddr_t segzio_base; /* Base address of segzio */ 3077c478bd9Sstevel@tonic-gate #if defined(__amd64) 3087c478bd9Sstevel@tonic-gate pgcnt_t segkpsize = btop(SEGKPDEFSIZE); /* size of segkp segment in pages */ 3097c478bd9Sstevel@tonic-gate #else 3107c478bd9Sstevel@tonic-gate pgcnt_t segkpsize = 0; 3117c478bd9Sstevel@tonic-gate #endif 312ad23a2dbSjohansen pgcnt_t segziosize = 0; /* size of zio segment in pages */ 3137c478bd9Sstevel@tonic-gate 314ae115bc7Smrj /* 315ae115bc7Smrj * VA range available to the debugger 316ae115bc7Smrj */ 317ae115bc7Smrj const caddr_t kdi_segdebugbase = (const caddr_t)SEGDEBUGBASE; 318ae115bc7Smrj const size_t kdi_segdebugsize = SEGDEBUGSIZE; 319ae115bc7Smrj 3207c478bd9Sstevel@tonic-gate struct memseg *memseg_base; 3217c478bd9Sstevel@tonic-gate struct vnode unused_pages_vp; 3227c478bd9Sstevel@tonic-gate 3237c478bd9Sstevel@tonic-gate #define FOURGB 0x100000000LL 3247c478bd9Sstevel@tonic-gate 3257c478bd9Sstevel@tonic-gate struct memlist *memlist; 3267c478bd9Sstevel@tonic-gate 3277c478bd9Sstevel@tonic-gate caddr_t s_text; /* start of kernel text segment */ 3287c478bd9Sstevel@tonic-gate caddr_t e_text; /* end of kernel text segment */ 3297c478bd9Sstevel@tonic-gate caddr_t s_data; /* start of kernel data segment */ 3307c478bd9Sstevel@tonic-gate caddr_t e_data; /* end of kernel data segment */ 3317c478bd9Sstevel@tonic-gate caddr_t modtext; /* start of loadable module text reserved */ 3327c478bd9Sstevel@tonic-gate caddr_t e_modtext; /* end of loadable module text reserved */ 3337c478bd9Sstevel@tonic-gate caddr_t moddata; /* start of loadable module data reserved */ 3347c478bd9Sstevel@tonic-gate caddr_t e_moddata; /* end of loadable module data reserved */ 3357c478bd9Sstevel@tonic-gate 3367c478bd9Sstevel@tonic-gate struct memlist *phys_install; /* Total installed physical memory */ 3377c478bd9Sstevel@tonic-gate struct memlist *phys_avail; /* Total available physical memory */ 3387c478bd9Sstevel@tonic-gate 3397c478bd9Sstevel@tonic-gate /* 3407c478bd9Sstevel@tonic-gate * kphysm_init returns the number of pages that were processed 3417c478bd9Sstevel@tonic-gate */ 342ae115bc7Smrj static pgcnt_t kphysm_init(page_t *, pgcnt_t); 3437c478bd9Sstevel@tonic-gate 3447c478bd9Sstevel@tonic-gate #define IO_PROP_SIZE 64 /* device property size */ 3457c478bd9Sstevel@tonic-gate 3467c478bd9Sstevel@tonic-gate /* 3477c478bd9Sstevel@tonic-gate * a couple useful roundup macros 3487c478bd9Sstevel@tonic-gate */ 3497c478bd9Sstevel@tonic-gate #define ROUND_UP_PAGE(x) \ 3507c478bd9Sstevel@tonic-gate ((uintptr_t)P2ROUNDUP((uintptr_t)(x), (uintptr_t)MMU_PAGESIZE)) 3517c478bd9Sstevel@tonic-gate #define ROUND_UP_LPAGE(x) \ 3527c478bd9Sstevel@tonic-gate ((uintptr_t)P2ROUNDUP((uintptr_t)(x), mmu.level_size[1])) 3537c478bd9Sstevel@tonic-gate #define ROUND_UP_4MEG(x) \ 354ae115bc7Smrj ((uintptr_t)P2ROUNDUP((uintptr_t)(x), (uintptr_t)FOUR_MEG)) 3557c478bd9Sstevel@tonic-gate #define ROUND_UP_TOPLEVEL(x) \ 3567c478bd9Sstevel@tonic-gate ((uintptr_t)P2ROUNDUP((uintptr_t)(x), mmu.level_size[mmu.max_level])) 3577c478bd9Sstevel@tonic-gate 3587c478bd9Sstevel@tonic-gate /* 3597c478bd9Sstevel@tonic-gate * 32-bit Kernel's Virtual memory layout. 3607c478bd9Sstevel@tonic-gate * +-----------------------+ 361ae115bc7Smrj * | | 3627c478bd9Sstevel@tonic-gate * 0xFFC00000 -|-----------------------|- ARGSBASE 3637c478bd9Sstevel@tonic-gate * | debugger | 3647c478bd9Sstevel@tonic-gate * 0xFF800000 -|-----------------------|- SEGDEBUGBASE 3657c478bd9Sstevel@tonic-gate * | Kernel Data | 3667c478bd9Sstevel@tonic-gate * 0xFEC00000 -|-----------------------| 3677c478bd9Sstevel@tonic-gate * | Kernel Text | 368843e1988Sjohnlev * 0xFE800000 -|-----------------------|- KERNEL_TEXT (0xFB400000 on Xen) 369ae115bc7Smrj * |--- GDT ---|- GDT page (GDT_VA) 370ae115bc7Smrj * |--- debug info ---|- debug info (DEBUG_INFO_VA) 371ae115bc7Smrj * | | 372ae115bc7Smrj * | page_t structures | 3737c478bd9Sstevel@tonic-gate * | memsegs, memlists, | 3747c478bd9Sstevel@tonic-gate * | page hash, etc. | 375ae115bc7Smrj * --- -|-----------------------|- ekernelheap, valloc_base (floating) 376ae115bc7Smrj * | | (segkp is just an arena in the heap) 3777c478bd9Sstevel@tonic-gate * | | 3787c478bd9Sstevel@tonic-gate * | kvseg | 3797c478bd9Sstevel@tonic-gate * | | 3807c478bd9Sstevel@tonic-gate * | | 3817c478bd9Sstevel@tonic-gate * --- -|-----------------------|- kernelheap (floating) 3827c478bd9Sstevel@tonic-gate * | Segkmap | 383ae115bc7Smrj * 0xC3002000 -|-----------------------|- segmap_start (floating) 3847c478bd9Sstevel@tonic-gate * | Red Zone | 3857c478bd9Sstevel@tonic-gate * 0xC3000000 -|-----------------------|- kernelbase / userlimit (floating) 3867c478bd9Sstevel@tonic-gate * | | || 3877c478bd9Sstevel@tonic-gate * | Shared objects | \/ 3887c478bd9Sstevel@tonic-gate * | | 3897c478bd9Sstevel@tonic-gate * : : 3907c478bd9Sstevel@tonic-gate * | user data | 3917c478bd9Sstevel@tonic-gate * |-----------------------| 3927c478bd9Sstevel@tonic-gate * | user text | 3937c478bd9Sstevel@tonic-gate * 0x08048000 -|-----------------------| 3947c478bd9Sstevel@tonic-gate * | user stack | 3957c478bd9Sstevel@tonic-gate * : : 3967c478bd9Sstevel@tonic-gate * | invalid | 3977c478bd9Sstevel@tonic-gate * 0x00000000 +-----------------------+ 3987c478bd9Sstevel@tonic-gate * 3997c478bd9Sstevel@tonic-gate * 4007c478bd9Sstevel@tonic-gate * 64-bit Kernel's Virtual memory layout. (assuming 64 bit app) 4017c478bd9Sstevel@tonic-gate * +-----------------------+ 402ae115bc7Smrj * | | 4037c478bd9Sstevel@tonic-gate * 0xFFFFFFFF.FFC00000 |-----------------------|- ARGSBASE 4047c478bd9Sstevel@tonic-gate * | debugger (?) | 4057c478bd9Sstevel@tonic-gate * 0xFFFFFFFF.FF800000 |-----------------------|- SEGDEBUGBASE 4067c478bd9Sstevel@tonic-gate * | unused | 4077c478bd9Sstevel@tonic-gate * +-----------------------+ 4087c478bd9Sstevel@tonic-gate * | Kernel Data | 4097c478bd9Sstevel@tonic-gate * 0xFFFFFFFF.FBC00000 |-----------------------| 4107c478bd9Sstevel@tonic-gate * | Kernel Text | 4117c478bd9Sstevel@tonic-gate * 0xFFFFFFFF.FB800000 |-----------------------|- KERNEL_TEXT 412ae115bc7Smrj * |--- GDT ---|- GDT page (GDT_VA) 413ae115bc7Smrj * |--- debug info ---|- debug info (DEBUG_INFO_VA) 414ae115bc7Smrj * | | 4157c478bd9Sstevel@tonic-gate * | Core heap | (used for loadable modules) 4167c478bd9Sstevel@tonic-gate * 0xFFFFFFFF.C0000000 |-----------------------|- core_base / ekernelheap 4177c478bd9Sstevel@tonic-gate * | Kernel | 4187c478bd9Sstevel@tonic-gate * | heap | 4197c478bd9Sstevel@tonic-gate * 0xFFFFFXXX.XXX00000 |-----------------------|- kernelheap (floating) 420ae115bc7Smrj * | segmap | 421ae115bc7Smrj * 0xFFFFFXXX.XXX00000 |-----------------------|- segmap_start (floating) 4227c478bd9Sstevel@tonic-gate * | device mappings | 4237c478bd9Sstevel@tonic-gate * 0xFFFFFXXX.XXX00000 |-----------------------|- toxic_addr (floating) 424ad23a2dbSjohansen * | segzio | 425ad23a2dbSjohansen * 0xFFFFFXXX.XXX00000 |-----------------------|- segzio_base (floating) 4267c478bd9Sstevel@tonic-gate * | segkp | 427ae115bc7Smrj * --- |-----------------------|- segkp_base (floating) 428ae115bc7Smrj * | page_t structures | valloc_base + valloc_sz 429ae115bc7Smrj * | memsegs, memlists, | 430ae115bc7Smrj * | page hash, etc. | 43135b1ab99Sjosephb * 0xFFFFFF00.00000000 |-----------------------|- valloc_base (lower if > 1TB) 4327c478bd9Sstevel@tonic-gate * | segkpm | 4337c478bd9Sstevel@tonic-gate * 0xFFFFFE00.00000000 |-----------------------| 4347c478bd9Sstevel@tonic-gate * | Red Zone | 43535b1ab99Sjosephb * 0xFFFFFD80.00000000 |-----------------------|- KERNELBASE (lower if > 1TB) 4367c478bd9Sstevel@tonic-gate * | User stack |- User space memory 4377c478bd9Sstevel@tonic-gate * | | 4387c478bd9Sstevel@tonic-gate * | shared objects, etc | (grows downwards) 4397c478bd9Sstevel@tonic-gate * : : 4407c478bd9Sstevel@tonic-gate * | | 4417c478bd9Sstevel@tonic-gate * 0xFFFF8000.00000000 |-----------------------| 4427c478bd9Sstevel@tonic-gate * | | 4437c478bd9Sstevel@tonic-gate * | VA Hole / unused | 4447c478bd9Sstevel@tonic-gate * | | 4457c478bd9Sstevel@tonic-gate * 0x00008000.00000000 |-----------------------| 4467c478bd9Sstevel@tonic-gate * | | 4477c478bd9Sstevel@tonic-gate * | | 4487c478bd9Sstevel@tonic-gate * : : 4497c478bd9Sstevel@tonic-gate * | user heap | (grows upwards) 4507c478bd9Sstevel@tonic-gate * | | 4517c478bd9Sstevel@tonic-gate * | user data | 4527c478bd9Sstevel@tonic-gate * |-----------------------| 4537c478bd9Sstevel@tonic-gate * | user text | 4547c478bd9Sstevel@tonic-gate * 0x00000000.04000000 |-----------------------| 4557c478bd9Sstevel@tonic-gate * | invalid | 4567c478bd9Sstevel@tonic-gate * 0x00000000.00000000 +-----------------------+ 4577c478bd9Sstevel@tonic-gate * 4587c478bd9Sstevel@tonic-gate * A 32 bit app on the 64 bit kernel sees the same layout as on the 32 bit 4597c478bd9Sstevel@tonic-gate * kernel, except that userlimit is raised to 0xfe000000 4607c478bd9Sstevel@tonic-gate * 4617c478bd9Sstevel@tonic-gate * Floating values: 4627c478bd9Sstevel@tonic-gate * 4637c478bd9Sstevel@tonic-gate * valloc_base: start of the kernel's memory management/tracking data 464ae115bc7Smrj * structures. This region contains page_t structures for 465ae115bc7Smrj * physical memory, memsegs, memlists, and the page hash. 4667c478bd9Sstevel@tonic-gate * 4677c478bd9Sstevel@tonic-gate * core_base: start of the kernel's "core" heap area on 64-bit systems. 4687c478bd9Sstevel@tonic-gate * This area is intended to be used for global data as well as for module 4697c478bd9Sstevel@tonic-gate * text/data that does not fit into the nucleus pages. The core heap is 4707c478bd9Sstevel@tonic-gate * restricted to a 2GB range, allowing every address within it to be 4717c478bd9Sstevel@tonic-gate * accessed using rip-relative addressing 4727c478bd9Sstevel@tonic-gate * 4737c478bd9Sstevel@tonic-gate * ekernelheap: end of kernelheap and start of segmap. 4747c478bd9Sstevel@tonic-gate * 4757c478bd9Sstevel@tonic-gate * kernelheap: start of kernel heap. On 32-bit systems, this starts right 4767c478bd9Sstevel@tonic-gate * above a red zone that separates the user's address space from the 4777c478bd9Sstevel@tonic-gate * kernel's. On 64-bit systems, it sits above segkp and segkpm. 4787c478bd9Sstevel@tonic-gate * 479ae115bc7Smrj * segmap_start: start of segmap. The length of segmap can be modified 48052eef812SAmrita Sadhukhan * through eeprom. The default length is 16MB on 32-bit systems and 64MB 48152eef812SAmrita Sadhukhan * on 64-bit systems. 4827c478bd9Sstevel@tonic-gate * 4837c478bd9Sstevel@tonic-gate * kernelbase: On a 32-bit kernel the default value of 0xd4000000 will be 4847c478bd9Sstevel@tonic-gate * decreased by 2X the size required for page_t. This allows the kernel 4857c478bd9Sstevel@tonic-gate * heap to grow in size with physical memory. With sizeof(page_t) == 80 4867c478bd9Sstevel@tonic-gate * bytes, the following shows the values of kernelbase and kernel heap 4877c478bd9Sstevel@tonic-gate * sizes for different memory configurations (assuming default segmap and 4887c478bd9Sstevel@tonic-gate * segkp sizes). 4897c478bd9Sstevel@tonic-gate * 4907c478bd9Sstevel@tonic-gate * mem size for kernelbase kernel heap 4917c478bd9Sstevel@tonic-gate * size page_t's size 4927c478bd9Sstevel@tonic-gate * ---- --------- ---------- ----------- 4937c478bd9Sstevel@tonic-gate * 1gb 0x01400000 0xd1800000 684MB 4947c478bd9Sstevel@tonic-gate * 2gb 0x02800000 0xcf000000 704MB 4957c478bd9Sstevel@tonic-gate * 4gb 0x05000000 0xca000000 744MB 4967c478bd9Sstevel@tonic-gate * 6gb 0x07800000 0xc5000000 784MB 4977c478bd9Sstevel@tonic-gate * 8gb 0x0a000000 0xc0000000 824MB 4987c478bd9Sstevel@tonic-gate * 16gb 0x14000000 0xac000000 984MB 4997c478bd9Sstevel@tonic-gate * 32gb 0x28000000 0x84000000 1304MB 5007c478bd9Sstevel@tonic-gate * 64gb 0x50000000 0x34000000 1944MB (*) 5017c478bd9Sstevel@tonic-gate * 5027c478bd9Sstevel@tonic-gate * kernelbase is less than the abi minimum of 0xc0000000 for memory 5037c478bd9Sstevel@tonic-gate * configurations above 8gb. 5047c478bd9Sstevel@tonic-gate * 5057c478bd9Sstevel@tonic-gate * (*) support for memory configurations above 32gb will require manual tuning 5067c478bd9Sstevel@tonic-gate * of kernelbase to balance out the need of user applications. 5077c478bd9Sstevel@tonic-gate */ 5087c478bd9Sstevel@tonic-gate 5097c478bd9Sstevel@tonic-gate /* real-time-clock initialization parameters */ 510ae115bc7Smrj extern time_t process_rtc_config_file(void); 5117c478bd9Sstevel@tonic-gate 5127c478bd9Sstevel@tonic-gate uintptr_t kernelbase; 513ae115bc7Smrj uintptr_t postbootkernelbase; /* not set till boot loader is gone */ 5147c478bd9Sstevel@tonic-gate uintptr_t eprom_kernelbase; 5157c478bd9Sstevel@tonic-gate size_t segmapsize; 516ae115bc7Smrj uintptr_t segmap_start; 5177c478bd9Sstevel@tonic-gate int segmapfreelists; 5187c478bd9Sstevel@tonic-gate pgcnt_t npages; 519ae115bc7Smrj pgcnt_t orig_npages; 5207c478bd9Sstevel@tonic-gate size_t core_size; /* size of "core" heap */ 5217c478bd9Sstevel@tonic-gate uintptr_t core_base; /* base address of "core" heap */ 5227c478bd9Sstevel@tonic-gate 5237c478bd9Sstevel@tonic-gate /* 5247c478bd9Sstevel@tonic-gate * List of bootstrap pages. We mark these as allocated in startup. 5257c478bd9Sstevel@tonic-gate * release_bootstrap() will free them when we're completely done with 5267c478bd9Sstevel@tonic-gate * the bootstrap. 5277c478bd9Sstevel@tonic-gate */ 528ae115bc7Smrj static page_t *bootpages; 529ae115bc7Smrj 530ae115bc7Smrj /* 531ae115bc7Smrj * boot time pages that have a vnode from the ramdisk will keep that forever. 532ae115bc7Smrj */ 533ae115bc7Smrj static page_t *rd_pages; 5347c478bd9Sstevel@tonic-gate 53519397407SSherry Moore /* 53619397407SSherry Moore * Lower 64K 53719397407SSherry Moore */ 53819397407SSherry Moore static page_t *lower_pages = NULL; 53919397407SSherry Moore static int lower_pages_count = 0; 54019397407SSherry Moore 5417c478bd9Sstevel@tonic-gate struct system_hardware system_hardware; 5427c478bd9Sstevel@tonic-gate 543551bc2a6Smrj /* 544551bc2a6Smrj * Is this Solaris instance running in a fully virtualized xVM domain? 545551bc2a6Smrj */ 546551bc2a6Smrj int xpv_is_hvm = 0; 547551bc2a6Smrj 5487c478bd9Sstevel@tonic-gate /* 5497c478bd9Sstevel@tonic-gate * Enable some debugging messages concerning memory usage... 5507c478bd9Sstevel@tonic-gate */ 5517c478bd9Sstevel@tonic-gate static void 552ae115bc7Smrj print_memlist(char *title, struct memlist *mp) 5537c478bd9Sstevel@tonic-gate { 5547c478bd9Sstevel@tonic-gate prom_printf("MEMLIST: %s:\n", title); 5557c478bd9Sstevel@tonic-gate while (mp != NULL) { 5567c478bd9Sstevel@tonic-gate prom_printf("\tAddress 0x%" PRIx64 ", size 0x%" PRIx64 "\n", 5577c478bd9Sstevel@tonic-gate mp->address, mp->size); 5587c478bd9Sstevel@tonic-gate mp = mp->next; 5597c478bd9Sstevel@tonic-gate } 5607c478bd9Sstevel@tonic-gate } 5617c478bd9Sstevel@tonic-gate 5627c478bd9Sstevel@tonic-gate /* 5637c478bd9Sstevel@tonic-gate * XX64 need a comment here.. are these just default values, surely 5647c478bd9Sstevel@tonic-gate * we read the "cpuid" type information to figure this out. 5657c478bd9Sstevel@tonic-gate */ 5667c478bd9Sstevel@tonic-gate int l2cache_sz = 0x80000; 5677c478bd9Sstevel@tonic-gate int l2cache_linesz = 0x40; 5687c478bd9Sstevel@tonic-gate int l2cache_assoc = 1; 5697c478bd9Sstevel@tonic-gate 5702cb27123Saguzovsk static size_t textrepl_min_gb = 10; 5712cb27123Saguzovsk 5727c478bd9Sstevel@tonic-gate /* 5737c478bd9Sstevel@tonic-gate * on 64 bit we use a predifined VA range for mapping devices in the kernel 5747c478bd9Sstevel@tonic-gate * on 32 bit the mappings are intermixed in the heap, so we use a bit map 5757c478bd9Sstevel@tonic-gate */ 5767c478bd9Sstevel@tonic-gate #ifdef __amd64 5777c478bd9Sstevel@tonic-gate 5787c478bd9Sstevel@tonic-gate vmem_t *device_arena; 5797c478bd9Sstevel@tonic-gate uintptr_t toxic_addr = (uintptr_t)NULL; 580ae115bc7Smrj size_t toxic_size = 1024 * 1024 * 1024; /* Sparc uses 1 gig too */ 5817c478bd9Sstevel@tonic-gate 5827c478bd9Sstevel@tonic-gate #else /* __i386 */ 5837c478bd9Sstevel@tonic-gate 5847c478bd9Sstevel@tonic-gate ulong_t *toxic_bit_map; /* one bit for each 4k of VA in heap_arena */ 5857c478bd9Sstevel@tonic-gate size_t toxic_bit_map_len = 0; /* in bits */ 5867c478bd9Sstevel@tonic-gate 5877c478bd9Sstevel@tonic-gate #endif /* __i386 */ 5887c478bd9Sstevel@tonic-gate 5897c478bd9Sstevel@tonic-gate /* 5907c478bd9Sstevel@tonic-gate * Simple boot time debug facilities 5917c478bd9Sstevel@tonic-gate */ 5927c478bd9Sstevel@tonic-gate static char *prm_dbg_str[] = { 5937c478bd9Sstevel@tonic-gate "%s:%d: '%s' is 0x%x\n", 5947c478bd9Sstevel@tonic-gate "%s:%d: '%s' is 0x%llx\n" 5957c478bd9Sstevel@tonic-gate }; 5967c478bd9Sstevel@tonic-gate 5977c478bd9Sstevel@tonic-gate int prom_debug; 5987c478bd9Sstevel@tonic-gate 5997c478bd9Sstevel@tonic-gate #define PRM_DEBUG(q) if (prom_debug) \ 6007c478bd9Sstevel@tonic-gate prom_printf(prm_dbg_str[sizeof (q) >> 3], "startup.c", __LINE__, #q, q); 6017c478bd9Sstevel@tonic-gate #define PRM_POINT(q) if (prom_debug) \ 6027c478bd9Sstevel@tonic-gate prom_printf("%s:%d: %s\n", "startup.c", __LINE__, q); 6037c478bd9Sstevel@tonic-gate 6047c478bd9Sstevel@tonic-gate /* 6057c478bd9Sstevel@tonic-gate * This structure is used to keep track of the intial allocations 6067c478bd9Sstevel@tonic-gate * done in startup_memlist(). The value of NUM_ALLOCATIONS needs to 6077c478bd9Sstevel@tonic-gate * be >= the number of ADD_TO_ALLOCATIONS() executed in the code. 6087c478bd9Sstevel@tonic-gate */ 6097c478bd9Sstevel@tonic-gate #define NUM_ALLOCATIONS 7 6107c478bd9Sstevel@tonic-gate int num_allocations = 0; 6117c478bd9Sstevel@tonic-gate struct { 6127c478bd9Sstevel@tonic-gate void **al_ptr; 6137c478bd9Sstevel@tonic-gate size_t al_size; 6147c478bd9Sstevel@tonic-gate } allocations[NUM_ALLOCATIONS]; 6157c478bd9Sstevel@tonic-gate size_t valloc_sz = 0; 6167c478bd9Sstevel@tonic-gate uintptr_t valloc_base; 6177c478bd9Sstevel@tonic-gate 6187c478bd9Sstevel@tonic-gate #define ADD_TO_ALLOCATIONS(ptr, size) { \ 6197c478bd9Sstevel@tonic-gate size = ROUND_UP_PAGE(size); \ 6207c478bd9Sstevel@tonic-gate if (num_allocations == NUM_ALLOCATIONS) \ 6217c478bd9Sstevel@tonic-gate panic("too many ADD_TO_ALLOCATIONS()"); \ 6227c478bd9Sstevel@tonic-gate allocations[num_allocations].al_ptr = (void**)&ptr; \ 6237c478bd9Sstevel@tonic-gate allocations[num_allocations].al_size = size; \ 6247c478bd9Sstevel@tonic-gate valloc_sz += size; \ 6257c478bd9Sstevel@tonic-gate ++num_allocations; \ 6267c478bd9Sstevel@tonic-gate } 6277c478bd9Sstevel@tonic-gate 628ae115bc7Smrj /* 629ae115bc7Smrj * Allocate all the initial memory needed by the page allocator. 630ae115bc7Smrj */ 6317c478bd9Sstevel@tonic-gate static void 6327c478bd9Sstevel@tonic-gate perform_allocations(void) 6337c478bd9Sstevel@tonic-gate { 6347c478bd9Sstevel@tonic-gate caddr_t mem; 6357c478bd9Sstevel@tonic-gate int i; 636ae115bc7Smrj int valloc_align; 6377c478bd9Sstevel@tonic-gate 638ae115bc7Smrj PRM_DEBUG(valloc_base); 639ae115bc7Smrj PRM_DEBUG(valloc_sz); 640ae115bc7Smrj valloc_align = mmu.level_size[mmu.max_page_level > 0]; 641ae115bc7Smrj mem = BOP_ALLOC(bootops, (caddr_t)valloc_base, valloc_sz, valloc_align); 6427c478bd9Sstevel@tonic-gate if (mem != (caddr_t)valloc_base) 6437c478bd9Sstevel@tonic-gate panic("BOP_ALLOC() failed"); 6447c478bd9Sstevel@tonic-gate bzero(mem, valloc_sz); 6457c478bd9Sstevel@tonic-gate for (i = 0; i < num_allocations; ++i) { 6467c478bd9Sstevel@tonic-gate *allocations[i].al_ptr = (void *)mem; 6477c478bd9Sstevel@tonic-gate mem += allocations[i].al_size; 6487c478bd9Sstevel@tonic-gate } 6497c478bd9Sstevel@tonic-gate } 6507c478bd9Sstevel@tonic-gate 6517c478bd9Sstevel@tonic-gate /* 6527c478bd9Sstevel@tonic-gate * Our world looks like this at startup time. 6537c478bd9Sstevel@tonic-gate * 6547c478bd9Sstevel@tonic-gate * In a 32-bit OS, boot loads the kernel text at 0xfe800000 and kernel data 6557c478bd9Sstevel@tonic-gate * at 0xfec00000. On a 64-bit OS, kernel text and data are loaded at 6567c478bd9Sstevel@tonic-gate * 0xffffffff.fe800000 and 0xffffffff.fec00000 respectively. Those 6577c478bd9Sstevel@tonic-gate * addresses are fixed in the binary at link time. 6587c478bd9Sstevel@tonic-gate * 6597c478bd9Sstevel@tonic-gate * On the text page: 6607c478bd9Sstevel@tonic-gate * unix/genunix/krtld/module text loads. 6617c478bd9Sstevel@tonic-gate * 6627c478bd9Sstevel@tonic-gate * On the data page: 663ae115bc7Smrj * unix/genunix/krtld/module data loads. 664ae115bc7Smrj * 6657c478bd9Sstevel@tonic-gate * Machine-dependent startup code 6667c478bd9Sstevel@tonic-gate */ 6677c478bd9Sstevel@tonic-gate void 6687c478bd9Sstevel@tonic-gate startup(void) 6697c478bd9Sstevel@tonic-gate { 670843e1988Sjohnlev #if !defined(__xpv) 67175bcd456Sjg extern void startup_bios_disk(void); 67275bcd456Sjg extern void startup_pci_bios(void); 67319397407SSherry Moore extern int post_fastreboot; 674843e1988Sjohnlev #endif 675a563a037Sbholler extern cpuset_t cpu_ready_set; 676a563a037Sbholler 677843e1988Sjohnlev /* 678843e1988Sjohnlev * Make sure that nobody tries to use sekpm until we have 679843e1988Sjohnlev * initialized it properly. 680843e1988Sjohnlev */ 6817c478bd9Sstevel@tonic-gate #if defined(__amd64) 68235b1ab99Sjosephb kpm_desired = 1; 6837c478bd9Sstevel@tonic-gate #endif 6847c478bd9Sstevel@tonic-gate kpm_enable = 0; 685a563a037Sbholler CPUSET_ONLY(cpu_ready_set, 0); /* cpu 0 is boot cpu */ 6867c478bd9Sstevel@tonic-gate 687843e1988Sjohnlev #if defined(__xpv) /* XXPV fix me! */ 688843e1988Sjohnlev { 689843e1988Sjohnlev extern int segvn_use_regions; 690843e1988Sjohnlev segvn_use_regions = 0; 691843e1988Sjohnlev } 692843e1988Sjohnlev #endif 6937c478bd9Sstevel@tonic-gate progressbar_init(); 6947c478bd9Sstevel@tonic-gate startup_init(); 695ab4a9bebSjohnlev #if defined(__xpv) 696ab4a9bebSjohnlev startup_xen_version(); 697ab4a9bebSjohnlev #endif 6987c478bd9Sstevel@tonic-gate startup_memlist(); 699ae115bc7Smrj startup_kmem(); 70035b1ab99Sjosephb startup_vm(); 701843e1988Sjohnlev #if !defined(__xpv) 70219397407SSherry Moore if (!post_fastreboot) 70319397407SSherry Moore startup_pci_bios(); 704e4b86885SCheng Sean Ye #endif 705e4b86885SCheng Sean Ye #if defined(__xpv) 706e4b86885SCheng Sean Ye startup_xen_mca(); 707843e1988Sjohnlev #endif 7087c478bd9Sstevel@tonic-gate startup_modules(); 709843e1988Sjohnlev #if !defined(__xpv) 71019397407SSherry Moore if (!post_fastreboot) 71119397407SSherry Moore startup_bios_disk(); 712843e1988Sjohnlev #endif 7137c478bd9Sstevel@tonic-gate startup_end(); 7147c478bd9Sstevel@tonic-gate progressbar_start(); 7157c478bd9Sstevel@tonic-gate } 7167c478bd9Sstevel@tonic-gate 7177c478bd9Sstevel@tonic-gate static void 7187c478bd9Sstevel@tonic-gate startup_init() 7197c478bd9Sstevel@tonic-gate { 7207c478bd9Sstevel@tonic-gate PRM_POINT("startup_init() starting..."); 7217c478bd9Sstevel@tonic-gate 7227c478bd9Sstevel@tonic-gate /* 7237c478bd9Sstevel@tonic-gate * Complete the extraction of cpuid data 7247c478bd9Sstevel@tonic-gate */ 7257c478bd9Sstevel@tonic-gate cpuid_pass2(CPU); 7267c478bd9Sstevel@tonic-gate 7277c478bd9Sstevel@tonic-gate (void) check_boot_version(BOP_GETVERSION(bootops)); 7287c478bd9Sstevel@tonic-gate 7297c478bd9Sstevel@tonic-gate /* 7307c478bd9Sstevel@tonic-gate * Check for prom_debug in boot environment 7317c478bd9Sstevel@tonic-gate */ 7327c478bd9Sstevel@tonic-gate if (BOP_GETPROPLEN(bootops, "prom_debug") >= 0) { 7337c478bd9Sstevel@tonic-gate ++prom_debug; 7347c478bd9Sstevel@tonic-gate PRM_POINT("prom_debug found in boot enviroment"); 7357c478bd9Sstevel@tonic-gate } 7367c478bd9Sstevel@tonic-gate 7377c478bd9Sstevel@tonic-gate /* 7387c478bd9Sstevel@tonic-gate * Collect node, cpu and memory configuration information. 7397c478bd9Sstevel@tonic-gate */ 7407c478bd9Sstevel@tonic-gate get_system_configuration(); 7417c478bd9Sstevel@tonic-gate 7427c478bd9Sstevel@tonic-gate /* 7437c478bd9Sstevel@tonic-gate * Halt if this is an unsupported processor. 7447c478bd9Sstevel@tonic-gate */ 7457c478bd9Sstevel@tonic-gate if (x86_type == X86_TYPE_486 || x86_type == X86_TYPE_CYRIX_486) { 7467c478bd9Sstevel@tonic-gate printf("\n486 processor (\"%s\") detected.\n", 7477c478bd9Sstevel@tonic-gate CPU->cpu_brandstr); 7487c478bd9Sstevel@tonic-gate halt("This processor is not supported by this release " 7497c478bd9Sstevel@tonic-gate "of Solaris."); 7507c478bd9Sstevel@tonic-gate } 7517c478bd9Sstevel@tonic-gate 7527c478bd9Sstevel@tonic-gate PRM_POINT("startup_init() done"); 7537c478bd9Sstevel@tonic-gate } 7547c478bd9Sstevel@tonic-gate 7557c478bd9Sstevel@tonic-gate /* 7567c478bd9Sstevel@tonic-gate * Callback for copy_memlist_filter() to filter nucleus, kadb/kmdb, (ie. 7577c478bd9Sstevel@tonic-gate * everything mapped above KERNEL_TEXT) pages from phys_avail. Note it 7587c478bd9Sstevel@tonic-gate * also filters out physical page zero. There is some reliance on the 7597c478bd9Sstevel@tonic-gate * boot loader allocating only a few contiguous physical memory chunks. 7607c478bd9Sstevel@tonic-gate */ 7617c478bd9Sstevel@tonic-gate static void 7627c478bd9Sstevel@tonic-gate avail_filter(uint64_t *addr, uint64_t *size) 7637c478bd9Sstevel@tonic-gate { 7647c478bd9Sstevel@tonic-gate uintptr_t va; 7657c478bd9Sstevel@tonic-gate uintptr_t next_va; 7667c478bd9Sstevel@tonic-gate pfn_t pfn; 7677c478bd9Sstevel@tonic-gate uint64_t pfn_addr; 7687c478bd9Sstevel@tonic-gate uint64_t pfn_eaddr; 7697c478bd9Sstevel@tonic-gate uint_t prot; 7707c478bd9Sstevel@tonic-gate size_t len; 7717c478bd9Sstevel@tonic-gate uint_t change; 7727c478bd9Sstevel@tonic-gate 7737c478bd9Sstevel@tonic-gate if (prom_debug) 7747c478bd9Sstevel@tonic-gate prom_printf("\tFilter: in: a=%" PRIx64 ", s=%" PRIx64 "\n", 7757c478bd9Sstevel@tonic-gate *addr, *size); 7767c478bd9Sstevel@tonic-gate 7777c478bd9Sstevel@tonic-gate /* 7787c478bd9Sstevel@tonic-gate * page zero is required for BIOS.. never make it available 7797c478bd9Sstevel@tonic-gate */ 7807c478bd9Sstevel@tonic-gate if (*addr == 0) { 7817c478bd9Sstevel@tonic-gate *addr += MMU_PAGESIZE; 7827c478bd9Sstevel@tonic-gate *size -= MMU_PAGESIZE; 7837c478bd9Sstevel@tonic-gate } 7847c478bd9Sstevel@tonic-gate 7857c478bd9Sstevel@tonic-gate /* 786ae115bc7Smrj * First we trim from the front of the range. Since kbm_probe() 7877c478bd9Sstevel@tonic-gate * walks ranges in virtual order, but addr/size are physical, we need 7887c478bd9Sstevel@tonic-gate * to the list until no changes are seen. This deals with the case 7897c478bd9Sstevel@tonic-gate * where page "p" is mapped at v, page "p + PAGESIZE" is mapped at w 7907c478bd9Sstevel@tonic-gate * but w < v. 7917c478bd9Sstevel@tonic-gate */ 7927c478bd9Sstevel@tonic-gate do { 7937c478bd9Sstevel@tonic-gate change = 0; 7947c478bd9Sstevel@tonic-gate for (va = KERNEL_TEXT; 795ae115bc7Smrj *size > 0 && kbm_probe(&va, &len, &pfn, &prot) != 0; 7967c478bd9Sstevel@tonic-gate va = next_va) { 7977c478bd9Sstevel@tonic-gate 7987c478bd9Sstevel@tonic-gate next_va = va + len; 799ae115bc7Smrj pfn_addr = pfn_to_pa(pfn); 8007c478bd9Sstevel@tonic-gate pfn_eaddr = pfn_addr + len; 8017c478bd9Sstevel@tonic-gate 8027c478bd9Sstevel@tonic-gate if (pfn_addr <= *addr && pfn_eaddr > *addr) { 8037c478bd9Sstevel@tonic-gate change = 1; 8047c478bd9Sstevel@tonic-gate while (*size > 0 && len > 0) { 8057c478bd9Sstevel@tonic-gate *addr += MMU_PAGESIZE; 8067c478bd9Sstevel@tonic-gate *size -= MMU_PAGESIZE; 8077c478bd9Sstevel@tonic-gate len -= MMU_PAGESIZE; 8087c478bd9Sstevel@tonic-gate } 8097c478bd9Sstevel@tonic-gate } 8107c478bd9Sstevel@tonic-gate } 8117c478bd9Sstevel@tonic-gate if (change && prom_debug) 8127c478bd9Sstevel@tonic-gate prom_printf("\t\ttrim: a=%" PRIx64 ", s=%" PRIx64 "\n", 8137c478bd9Sstevel@tonic-gate *addr, *size); 8147c478bd9Sstevel@tonic-gate } while (change); 8157c478bd9Sstevel@tonic-gate 8167c478bd9Sstevel@tonic-gate /* 8177c478bd9Sstevel@tonic-gate * Trim pages from the end of the range. 8187c478bd9Sstevel@tonic-gate */ 8197c478bd9Sstevel@tonic-gate for (va = KERNEL_TEXT; 820ae115bc7Smrj *size > 0 && kbm_probe(&va, &len, &pfn, &prot) != 0; 8217c478bd9Sstevel@tonic-gate va = next_va) { 8227c478bd9Sstevel@tonic-gate 8237c478bd9Sstevel@tonic-gate next_va = va + len; 824ae115bc7Smrj pfn_addr = pfn_to_pa(pfn); 8257c478bd9Sstevel@tonic-gate 8267c478bd9Sstevel@tonic-gate if (pfn_addr >= *addr && pfn_addr < *addr + *size) 8277c478bd9Sstevel@tonic-gate *size = pfn_addr - *addr; 8287c478bd9Sstevel@tonic-gate } 8297c478bd9Sstevel@tonic-gate 8307c478bd9Sstevel@tonic-gate if (prom_debug) 8317c478bd9Sstevel@tonic-gate prom_printf("\tFilter out: a=%" PRIx64 ", s=%" PRIx64 "\n", 8327c478bd9Sstevel@tonic-gate *addr, *size); 8337c478bd9Sstevel@tonic-gate } 8347c478bd9Sstevel@tonic-gate 8357c478bd9Sstevel@tonic-gate static void 8367c478bd9Sstevel@tonic-gate kpm_init() 8377c478bd9Sstevel@tonic-gate { 8387c478bd9Sstevel@tonic-gate struct segkpm_crargs b; 8397c478bd9Sstevel@tonic-gate 8407c478bd9Sstevel@tonic-gate /* 8417c478bd9Sstevel@tonic-gate * These variables were all designed for sfmmu in which segkpm is 8427c478bd9Sstevel@tonic-gate * mapped using a single pagesize - either 8KB or 4MB. On x86, we 8437c478bd9Sstevel@tonic-gate * might use 2+ page sizes on a single machine, so none of these 8447c478bd9Sstevel@tonic-gate * variables have a single correct value. They are set up as if we 8457c478bd9Sstevel@tonic-gate * always use a 4KB pagesize, which should do no harm. In the long 8467c478bd9Sstevel@tonic-gate * run, we should get rid of KPM's assumption that only a single 8477c478bd9Sstevel@tonic-gate * pagesize is used. 8487c478bd9Sstevel@tonic-gate */ 8497c478bd9Sstevel@tonic-gate kpm_pgshft = MMU_PAGESHIFT; 8507c478bd9Sstevel@tonic-gate kpm_pgsz = MMU_PAGESIZE; 8517c478bd9Sstevel@tonic-gate kpm_pgoff = MMU_PAGEOFFSET; 8527c478bd9Sstevel@tonic-gate kpmp2pshft = 0; 8537c478bd9Sstevel@tonic-gate kpmpnpgs = 1; 8547c478bd9Sstevel@tonic-gate ASSERT(((uintptr_t)kpm_vbase & (kpm_pgsz - 1)) == 0); 8557c478bd9Sstevel@tonic-gate 8567c478bd9Sstevel@tonic-gate PRM_POINT("about to create segkpm"); 8577c478bd9Sstevel@tonic-gate rw_enter(&kas.a_lock, RW_WRITER); 8587c478bd9Sstevel@tonic-gate 8597c478bd9Sstevel@tonic-gate if (seg_attach(&kas, kpm_vbase, kpm_size, segkpm) < 0) 8607c478bd9Sstevel@tonic-gate panic("cannot attach segkpm"); 8617c478bd9Sstevel@tonic-gate 8627c478bd9Sstevel@tonic-gate b.prot = PROT_READ | PROT_WRITE; 8637c478bd9Sstevel@tonic-gate b.nvcolors = 1; 8647c478bd9Sstevel@tonic-gate 8657c478bd9Sstevel@tonic-gate if (segkpm_create(segkpm, (caddr_t)&b) != 0) 8667c478bd9Sstevel@tonic-gate panic("segkpm_create segkpm"); 8677c478bd9Sstevel@tonic-gate 8687c478bd9Sstevel@tonic-gate rw_exit(&kas.a_lock); 869ae115bc7Smrj } 8707c478bd9Sstevel@tonic-gate 871ae115bc7Smrj /* 872ae115bc7Smrj * The debug info page provides enough information to allow external 873ae115bc7Smrj * inspectors (e.g. when running under a hypervisor) to bootstrap 874ae115bc7Smrj * themselves into allowing full-blown kernel debugging. 875ae115bc7Smrj */ 876ae115bc7Smrj static void 877ae115bc7Smrj init_debug_info(void) 878ae115bc7Smrj { 879ae115bc7Smrj caddr_t mem; 880ae115bc7Smrj debug_info_t *di; 881ae115bc7Smrj 882ae115bc7Smrj #ifndef __lint 883ae115bc7Smrj ASSERT(sizeof (debug_info_t) < MMU_PAGESIZE); 884ae115bc7Smrj #endif 885ae115bc7Smrj 886ae115bc7Smrj mem = BOP_ALLOC(bootops, (caddr_t)DEBUG_INFO_VA, MMU_PAGESIZE, 887ae115bc7Smrj MMU_PAGESIZE); 888ae115bc7Smrj 889ae115bc7Smrj if (mem != (caddr_t)DEBUG_INFO_VA) 890ae115bc7Smrj panic("BOP_ALLOC() failed"); 891ae115bc7Smrj bzero(mem, MMU_PAGESIZE); 892ae115bc7Smrj 893ae115bc7Smrj di = (debug_info_t *)mem; 894ae115bc7Smrj 895ae115bc7Smrj di->di_magic = DEBUG_INFO_MAGIC; 896ae115bc7Smrj di->di_version = DEBUG_INFO_VERSION; 897843e1988Sjohnlev di->di_modules = (uintptr_t)&modules; 898843e1988Sjohnlev di->di_s_text = (uintptr_t)s_text; 899843e1988Sjohnlev di->di_e_text = (uintptr_t)e_text; 900843e1988Sjohnlev di->di_s_data = (uintptr_t)s_data; 901843e1988Sjohnlev di->di_e_data = (uintptr_t)e_data; 902843e1988Sjohnlev di->di_hat_htable_off = offsetof(hat_t, hat_htable); 903843e1988Sjohnlev di->di_ht_pfn_off = offsetof(htable_t, ht_pfn); 9047c478bd9Sstevel@tonic-gate } 9057c478bd9Sstevel@tonic-gate 9067c478bd9Sstevel@tonic-gate /* 907ae115bc7Smrj * Build the memlists and other kernel essential memory system data structures. 908ae115bc7Smrj * This is everything at valloc_base. 9097c478bd9Sstevel@tonic-gate */ 9107c478bd9Sstevel@tonic-gate static void 9117c478bd9Sstevel@tonic-gate startup_memlist(void) 9127c478bd9Sstevel@tonic-gate { 9137c478bd9Sstevel@tonic-gate size_t memlist_sz; 9147c478bd9Sstevel@tonic-gate size_t memseg_sz; 9157c478bd9Sstevel@tonic-gate size_t pagehash_sz; 9167c478bd9Sstevel@tonic-gate size_t pp_sz; 9177c478bd9Sstevel@tonic-gate uintptr_t va; 9187c478bd9Sstevel@tonic-gate size_t len; 9197c478bd9Sstevel@tonic-gate uint_t prot; 9207c478bd9Sstevel@tonic-gate pfn_t pfn; 9217c478bd9Sstevel@tonic-gate int memblocks; 9227c478bd9Sstevel@tonic-gate caddr_t pagecolor_mem; 9237c478bd9Sstevel@tonic-gate size_t pagecolor_memsz; 9247c478bd9Sstevel@tonic-gate caddr_t page_ctrs_mem; 9257c478bd9Sstevel@tonic-gate size_t page_ctrs_size; 926d7d93655Sblakej size_t pse_table_alloc_size; 9277c478bd9Sstevel@tonic-gate struct memlist *current; 9287c478bd9Sstevel@tonic-gate extern void startup_build_mem_nodes(struct memlist *); 9297c478bd9Sstevel@tonic-gate 9307c478bd9Sstevel@tonic-gate /* XX64 fix these - they should be in include files */ 9317c478bd9Sstevel@tonic-gate extern size_t page_coloring_init(uint_t, int, int); 9327c478bd9Sstevel@tonic-gate extern void page_coloring_setup(caddr_t); 9337c478bd9Sstevel@tonic-gate 9347c478bd9Sstevel@tonic-gate PRM_POINT("startup_memlist() starting..."); 9357c478bd9Sstevel@tonic-gate 9367c478bd9Sstevel@tonic-gate /* 9377c478bd9Sstevel@tonic-gate * Use leftover large page nucleus text/data space for loadable modules. 9387c478bd9Sstevel@tonic-gate * Use at most MODTEXT/MODDATA. 9397c478bd9Sstevel@tonic-gate */ 940ae115bc7Smrj len = kbm_nucleus_size; 941ae115bc7Smrj ASSERT(len > MMU_PAGESIZE); 9427c478bd9Sstevel@tonic-gate 943ae115bc7Smrj moddata = (caddr_t)ROUND_UP_PAGE(e_data); 944ae115bc7Smrj e_moddata = (caddr_t)P2ROUNDUP((uintptr_t)e_data, (uintptr_t)len); 945ae115bc7Smrj if (e_moddata - moddata > MODDATA) 946ae115bc7Smrj e_moddata = moddata + MODDATA; 9477c478bd9Sstevel@tonic-gate 948ae115bc7Smrj modtext = (caddr_t)ROUND_UP_PAGE(e_text); 949ae115bc7Smrj e_modtext = (caddr_t)P2ROUNDUP((uintptr_t)e_text, (uintptr_t)len); 950ae115bc7Smrj if (e_modtext - modtext > MODTEXT) 951ae115bc7Smrj e_modtext = modtext + MODTEXT; 9527c478bd9Sstevel@tonic-gate 9537c478bd9Sstevel@tonic-gate econtig = e_moddata; 9547c478bd9Sstevel@tonic-gate 9557c478bd9Sstevel@tonic-gate PRM_DEBUG(modtext); 9567c478bd9Sstevel@tonic-gate PRM_DEBUG(e_modtext); 9577c478bd9Sstevel@tonic-gate PRM_DEBUG(moddata); 9587c478bd9Sstevel@tonic-gate PRM_DEBUG(e_moddata); 9597c478bd9Sstevel@tonic-gate PRM_DEBUG(econtig); 9607c478bd9Sstevel@tonic-gate 9617c478bd9Sstevel@tonic-gate /* 962ae115bc7Smrj * Examine the boot loader physical memory map to find out: 9637c478bd9Sstevel@tonic-gate * - total memory in system - physinstalled 9647c478bd9Sstevel@tonic-gate * - the max physical address - physmax 965ae115bc7Smrj * - the number of discontiguous segments of memory. 9667c478bd9Sstevel@tonic-gate */ 9677c478bd9Sstevel@tonic-gate if (prom_debug) 968ae115bc7Smrj print_memlist("boot physinstalled", 9697c478bd9Sstevel@tonic-gate bootops->boot_mem->physinstalled); 9707c478bd9Sstevel@tonic-gate installed_top_size(bootops->boot_mem->physinstalled, &physmax, 9717c478bd9Sstevel@tonic-gate &physinstalled, &memblocks); 9727c478bd9Sstevel@tonic-gate PRM_DEBUG(physmax); 9737c478bd9Sstevel@tonic-gate PRM_DEBUG(physinstalled); 9747c478bd9Sstevel@tonic-gate PRM_DEBUG(memblocks); 9757c478bd9Sstevel@tonic-gate 9767c478bd9Sstevel@tonic-gate /* 9777c478bd9Sstevel@tonic-gate * Initialize hat's mmu parameters. 9787c478bd9Sstevel@tonic-gate * Check for enforce-prot-exec in boot environment. It's used to 9797c478bd9Sstevel@tonic-gate * enable/disable support for the page table entry NX bit. 9807c478bd9Sstevel@tonic-gate * The default is to enforce PROT_EXEC on processors that support NX. 9817c478bd9Sstevel@tonic-gate * Boot seems to round up the "len", but 8 seems to be big enough. 9827c478bd9Sstevel@tonic-gate */ 9837c478bd9Sstevel@tonic-gate mmu_init(); 9847c478bd9Sstevel@tonic-gate 9857c478bd9Sstevel@tonic-gate #ifdef __i386 9867c478bd9Sstevel@tonic-gate /* 9877c478bd9Sstevel@tonic-gate * physmax is lowered if there is more memory than can be 9887c478bd9Sstevel@tonic-gate * physically addressed in 32 bit (PAE/non-PAE) modes. 9897c478bd9Sstevel@tonic-gate */ 9907c478bd9Sstevel@tonic-gate if (mmu.pae_hat) { 9917c478bd9Sstevel@tonic-gate if (PFN_ABOVE64G(physmax)) { 9927c478bd9Sstevel@tonic-gate physinstalled -= (physmax - (PFN_64G - 1)); 9937c478bd9Sstevel@tonic-gate physmax = PFN_64G - 1; 9947c478bd9Sstevel@tonic-gate } 9957c478bd9Sstevel@tonic-gate } else { 9967c478bd9Sstevel@tonic-gate if (PFN_ABOVE4G(physmax)) { 9977c478bd9Sstevel@tonic-gate physinstalled -= (physmax - (PFN_4G - 1)); 9987c478bd9Sstevel@tonic-gate physmax = PFN_4G - 1; 9997c478bd9Sstevel@tonic-gate } 10007c478bd9Sstevel@tonic-gate } 10017c478bd9Sstevel@tonic-gate #endif 10027c478bd9Sstevel@tonic-gate 10037c478bd9Sstevel@tonic-gate startup_build_mem_nodes(bootops->boot_mem->physinstalled); 10047c478bd9Sstevel@tonic-gate 10057c478bd9Sstevel@tonic-gate if (BOP_GETPROPLEN(bootops, "enforce-prot-exec") >= 0) { 10067c478bd9Sstevel@tonic-gate int len = BOP_GETPROPLEN(bootops, "enforce-prot-exec"); 10077c478bd9Sstevel@tonic-gate char value[8]; 10087c478bd9Sstevel@tonic-gate 10097c478bd9Sstevel@tonic-gate if (len < 8) 10107c478bd9Sstevel@tonic-gate (void) BOP_GETPROP(bootops, "enforce-prot-exec", value); 10117c478bd9Sstevel@tonic-gate else 10127c478bd9Sstevel@tonic-gate (void) strcpy(value, ""); 10137c478bd9Sstevel@tonic-gate if (strcmp(value, "off") == 0) 10147c478bd9Sstevel@tonic-gate mmu.pt_nx = 0; 10157c478bd9Sstevel@tonic-gate } 10167c478bd9Sstevel@tonic-gate PRM_DEBUG(mmu.pt_nx); 10177c478bd9Sstevel@tonic-gate 10187c478bd9Sstevel@tonic-gate /* 10197c478bd9Sstevel@tonic-gate * We will need page_t's for every page in the system, except for 10207c478bd9Sstevel@tonic-gate * memory mapped at or above above the start of the kernel text segment. 10217c478bd9Sstevel@tonic-gate * 10227c478bd9Sstevel@tonic-gate * pages above e_modtext are attributed to kernel debugger (obp_pages) 10237c478bd9Sstevel@tonic-gate */ 10247c478bd9Sstevel@tonic-gate npages = physinstalled - 1; /* avail_filter() skips page 0, so "- 1" */ 10257c478bd9Sstevel@tonic-gate obp_pages = 0; 10267c478bd9Sstevel@tonic-gate va = KERNEL_TEXT; 1027ae115bc7Smrj while (kbm_probe(&va, &len, &pfn, &prot) != 0) { 10287c478bd9Sstevel@tonic-gate npages -= len >> MMU_PAGESHIFT; 10297c478bd9Sstevel@tonic-gate if (va >= (uintptr_t)e_moddata) 10307c478bd9Sstevel@tonic-gate obp_pages += len >> MMU_PAGESHIFT; 10317c478bd9Sstevel@tonic-gate va += len; 10327c478bd9Sstevel@tonic-gate } 10337c478bd9Sstevel@tonic-gate PRM_DEBUG(npages); 10347c478bd9Sstevel@tonic-gate PRM_DEBUG(obp_pages); 10357c478bd9Sstevel@tonic-gate 10367c478bd9Sstevel@tonic-gate /* 103735b1ab99Sjosephb * If physmem is patched to be non-zero, use it instead of the computed 103835b1ab99Sjosephb * value unless it is larger than the actual amount of memory on hand. 10397c478bd9Sstevel@tonic-gate */ 10404944b02eSkchow if (physmem == 0 || physmem > npages) { 10417c478bd9Sstevel@tonic-gate physmem = npages; 10424944b02eSkchow } else if (physmem < npages) { 1043f53ad214Skchow orig_npages = npages; 10447c478bd9Sstevel@tonic-gate npages = physmem; 10454944b02eSkchow } 10467c478bd9Sstevel@tonic-gate PRM_DEBUG(physmem); 10477c478bd9Sstevel@tonic-gate 10487c478bd9Sstevel@tonic-gate /* 10497c478bd9Sstevel@tonic-gate * We now compute the sizes of all the initial allocations for 10507c478bd9Sstevel@tonic-gate * structures the kernel needs in order do kmem_alloc(). These 10517c478bd9Sstevel@tonic-gate * include: 10527c478bd9Sstevel@tonic-gate * memsegs 10537c478bd9Sstevel@tonic-gate * memlists 10547c478bd9Sstevel@tonic-gate * page hash table 10557c478bd9Sstevel@tonic-gate * page_t's 10567c478bd9Sstevel@tonic-gate * page coloring data structs 10577c478bd9Sstevel@tonic-gate */ 10587c478bd9Sstevel@tonic-gate memseg_sz = sizeof (struct memseg) * (memblocks + POSS_NEW_FRAGMENTS); 10597c478bd9Sstevel@tonic-gate ADD_TO_ALLOCATIONS(memseg_base, memseg_sz); 10607c478bd9Sstevel@tonic-gate PRM_DEBUG(memseg_sz); 10617c478bd9Sstevel@tonic-gate 10627c478bd9Sstevel@tonic-gate /* 1063ae115bc7Smrj * Reserve space for memlists. There's no real good way to know exactly 1064ae115bc7Smrj * how much room we'll need, but this should be a good upper bound. 10657c478bd9Sstevel@tonic-gate */ 10667c478bd9Sstevel@tonic-gate memlist_sz = ROUND_UP_PAGE(2 * sizeof (struct memlist) * 10677c478bd9Sstevel@tonic-gate (memblocks + POSS_NEW_FRAGMENTS)); 10687c478bd9Sstevel@tonic-gate ADD_TO_ALLOCATIONS(memlist, memlist_sz); 10697c478bd9Sstevel@tonic-gate PRM_DEBUG(memlist_sz); 10707c478bd9Sstevel@tonic-gate 10717c478bd9Sstevel@tonic-gate /* 10727c478bd9Sstevel@tonic-gate * The page structure hash table size is a power of 2 10737c478bd9Sstevel@tonic-gate * such that the average hash chain length is PAGE_HASHAVELEN. 10747c478bd9Sstevel@tonic-gate */ 10757c478bd9Sstevel@tonic-gate page_hashsz = npages / PAGE_HASHAVELEN; 10767c478bd9Sstevel@tonic-gate page_hashsz = 1 << highbit(page_hashsz); 10777c478bd9Sstevel@tonic-gate pagehash_sz = sizeof (struct page *) * page_hashsz; 10787c478bd9Sstevel@tonic-gate ADD_TO_ALLOCATIONS(page_hash, pagehash_sz); 10797c478bd9Sstevel@tonic-gate PRM_DEBUG(pagehash_sz); 10807c478bd9Sstevel@tonic-gate 10817c478bd9Sstevel@tonic-gate /* 1082ae115bc7Smrj * Set aside room for the page structures themselves. 10837c478bd9Sstevel@tonic-gate */ 1084ae115bc7Smrj PRM_DEBUG(npages); 1085ae115bc7Smrj pp_sz = sizeof (struct page) * npages; 10867c478bd9Sstevel@tonic-gate ADD_TO_ALLOCATIONS(pp_base, pp_sz); 10877c478bd9Sstevel@tonic-gate PRM_DEBUG(pp_sz); 10887c478bd9Sstevel@tonic-gate 10897c478bd9Sstevel@tonic-gate /* 10907c478bd9Sstevel@tonic-gate * determine l2 cache info and memory size for page coloring 10917c478bd9Sstevel@tonic-gate */ 10927c478bd9Sstevel@tonic-gate (void) getl2cacheinfo(CPU, 10937c478bd9Sstevel@tonic-gate &l2cache_sz, &l2cache_linesz, &l2cache_assoc); 10947c478bd9Sstevel@tonic-gate pagecolor_memsz = 10957c478bd9Sstevel@tonic-gate page_coloring_init(l2cache_sz, l2cache_linesz, l2cache_assoc); 10967c478bd9Sstevel@tonic-gate ADD_TO_ALLOCATIONS(pagecolor_mem, pagecolor_memsz); 10977c478bd9Sstevel@tonic-gate PRM_DEBUG(pagecolor_memsz); 10987c478bd9Sstevel@tonic-gate 10997c478bd9Sstevel@tonic-gate page_ctrs_size = page_ctrs_sz(); 11007c478bd9Sstevel@tonic-gate ADD_TO_ALLOCATIONS(page_ctrs_mem, page_ctrs_size); 11017c478bd9Sstevel@tonic-gate PRM_DEBUG(page_ctrs_size); 11027c478bd9Sstevel@tonic-gate 1103d7d93655Sblakej /* 1104d7d93655Sblakej * Allocate the array that protects pp->p_selock. 1105d7d93655Sblakej */ 1106d7d93655Sblakej pse_shift = size_pse_array(physmem, max_ncpus); 1107d7d93655Sblakej pse_table_size = 1 << pse_shift; 1108d7d93655Sblakej pse_table_alloc_size = pse_table_size * sizeof (pad_mutex_t); 1109d7d93655Sblakej ADD_TO_ALLOCATIONS(pse_mutex, pse_table_alloc_size); 1110d7d93655Sblakej 1111ae115bc7Smrj #if defined(__amd64) 11127c478bd9Sstevel@tonic-gate valloc_sz = ROUND_UP_LPAGE(valloc_sz); 1113ae115bc7Smrj valloc_base = VALLOC_BASE; 111435b1ab99Sjosephb 111535b1ab99Sjosephb /* 111635b1ab99Sjosephb * The default values of VALLOC_BASE and SEGKPM_BASE should work 111735b1ab99Sjosephb * for values of physmax up to 1 Terabyte. They need adjusting when 111835b1ab99Sjosephb * memory is at addresses above 1 TB. 111935b1ab99Sjosephb */ 112035b1ab99Sjosephb if (physmax + 1 > mmu_btop(TERABYTE)) { 112135b1ab99Sjosephb uint64_t kpm_resv_amount = mmu_ptob(physmax + 1); 112235b1ab99Sjosephb 112335b1ab99Sjosephb /* Round to largest possible pagesize for now */ 112435b1ab99Sjosephb kpm_resv_amount = P2ROUNDUP(kpm_resv_amount, ONE_GIG); 112535b1ab99Sjosephb 112635b1ab99Sjosephb segkpm_base = -(2 * kpm_resv_amount); /* down from top VA */ 112735b1ab99Sjosephb 112835b1ab99Sjosephb /* make sure we leave some space for user apps above hole */ 112935b1ab99Sjosephb segkpm_base = MAX(segkpm_base, AMD64_VA_HOLE_END + TERABYTE); 113035b1ab99Sjosephb if (segkpm_base > SEGKPM_BASE) 113135b1ab99Sjosephb segkpm_base = SEGKPM_BASE; 113235b1ab99Sjosephb PRM_DEBUG(segkpm_base); 113335b1ab99Sjosephb 113435b1ab99Sjosephb valloc_base = segkpm_base + kpm_resv_amount; 113535b1ab99Sjosephb PRM_DEBUG(valloc_base); 113635b1ab99Sjosephb } 1137ae115bc7Smrj #else /* __i386 */ 1138ae115bc7Smrj valloc_base = (uintptr_t)(MISC_VA_BASE - valloc_sz); 1139ae115bc7Smrj valloc_base = P2ALIGN(valloc_base, mmu.level_size[1]); 11407c478bd9Sstevel@tonic-gate PRM_DEBUG(valloc_base); 114135b1ab99Sjosephb #endif /* __i386 */ 1142ae115bc7Smrj 1143ae115bc7Smrj /* 1144ae115bc7Smrj * do all the initial allocations 1145ae115bc7Smrj */ 1146ae115bc7Smrj perform_allocations(); 1147ae115bc7Smrj 1148ae115bc7Smrj /* 1149ae115bc7Smrj * Build phys_install and phys_avail in kernel memspace. 1150ae115bc7Smrj * - phys_install should be all memory in the system. 1151ae115bc7Smrj * - phys_avail is phys_install minus any memory mapped before this 1152ae115bc7Smrj * point above KERNEL_TEXT. 1153ae115bc7Smrj */ 1154ae115bc7Smrj current = phys_install = memlist; 1155ae115bc7Smrj copy_memlist_filter(bootops->boot_mem->physinstalled, ¤t, NULL); 1156ae115bc7Smrj if ((caddr_t)current > (caddr_t)memlist + memlist_sz) 1157ae115bc7Smrj panic("physinstalled was too big!"); 1158ae115bc7Smrj if (prom_debug) 1159ae115bc7Smrj print_memlist("phys_install", phys_install); 1160ae115bc7Smrj 1161ae115bc7Smrj phys_avail = current; 1162ae115bc7Smrj PRM_POINT("Building phys_avail:\n"); 1163ae115bc7Smrj copy_memlist_filter(bootops->boot_mem->physinstalled, ¤t, 1164ae115bc7Smrj avail_filter); 1165ae115bc7Smrj if ((caddr_t)current > (caddr_t)memlist + memlist_sz) 1166ae115bc7Smrj panic("physavail was too big!"); 1167ae115bc7Smrj if (prom_debug) 1168ae115bc7Smrj print_memlist("phys_avail", phys_avail); 1169ae115bc7Smrj 1170ae115bc7Smrj /* 1171ae115bc7Smrj * setup page coloring 1172ae115bc7Smrj */ 1173ae115bc7Smrj page_coloring_setup(pagecolor_mem); 1174ae115bc7Smrj page_lock_init(); /* currently a no-op */ 1175ae115bc7Smrj 1176ae115bc7Smrj /* 1177ae115bc7Smrj * free page list counters 1178ae115bc7Smrj */ 1179ae115bc7Smrj (void) page_ctrs_alloc(page_ctrs_mem); 1180ae115bc7Smrj 118106fb6a36Sdv /* 118206fb6a36Sdv * Size the pcf array based on the number of cpus in the box at 118306fb6a36Sdv * boot time. 118406fb6a36Sdv */ 118506fb6a36Sdv 118606fb6a36Sdv pcf_init(); 118706fb6a36Sdv 1188ae115bc7Smrj /* 1189ae115bc7Smrj * Initialize the page structures from the memory lists. 1190ae115bc7Smrj */ 1191ae115bc7Smrj availrmem_initial = availrmem = freemem = 0; 1192ae115bc7Smrj PRM_POINT("Calling kphysm_init()..."); 1193ae115bc7Smrj npages = kphysm_init(pp_base, npages); 1194ae115bc7Smrj PRM_POINT("kphysm_init() done"); 1195ae115bc7Smrj PRM_DEBUG(npages); 1196ae115bc7Smrj 1197ae115bc7Smrj init_debug_info(); 1198ae115bc7Smrj 1199ae115bc7Smrj /* 1200ae115bc7Smrj * Now that page_t's have been initialized, remove all the 1201ae115bc7Smrj * initial allocation pages from the kernel free page lists. 1202ae115bc7Smrj */ 1203ae115bc7Smrj boot_mapin((caddr_t)valloc_base, valloc_sz); 12040cfdb603Sjosephb boot_mapin((caddr_t)MISC_VA_BASE, MISC_VA_SIZE); 1205ae115bc7Smrj PRM_POINT("startup_memlist() done"); 1206ae115bc7Smrj 1207ae115bc7Smrj PRM_DEBUG(valloc_sz); 12082cb27123Saguzovsk 1209567d55e1Saguzovsk #if defined(__amd64) 1210567d55e1Saguzovsk if ((availrmem >> (30 - MMU_PAGESHIFT)) >= 1211567d55e1Saguzovsk textrepl_min_gb && l2cache_sz <= 2 << 20) { 1212567d55e1Saguzovsk extern size_t textrepl_size_thresh; 12132cb27123Saguzovsk textrepl_size_thresh = (16 << 20) - 1; 12142cb27123Saguzovsk } 1215567d55e1Saguzovsk #endif 1216ae115bc7Smrj } 1217ae115bc7Smrj 1218ae115bc7Smrj /* 1219ae115bc7Smrj * Layout the kernel's part of address space and initialize kmem allocator. 1220ae115bc7Smrj */ 1221ae115bc7Smrj static void 1222ae115bc7Smrj startup_kmem(void) 1223ae115bc7Smrj { 1224932dc8e5Sdp extern void page_set_colorequiv_arr(void); 1225932dc8e5Sdp 1226ae115bc7Smrj PRM_POINT("startup_kmem() starting..."); 12277c478bd9Sstevel@tonic-gate 12287c478bd9Sstevel@tonic-gate #if defined(__amd64) 12297c478bd9Sstevel@tonic-gate if (eprom_kernelbase && eprom_kernelbase != KERNELBASE) 12307c478bd9Sstevel@tonic-gate cmn_err(CE_NOTE, "!kernelbase cannot be changed on 64-bit " 12317c478bd9Sstevel@tonic-gate "systems."); 123235b1ab99Sjosephb kernelbase = segkpm_base - KERNEL_REDZONE_SIZE; 12337c478bd9Sstevel@tonic-gate core_base = (uintptr_t)COREHEAP_BASE; 1234ae115bc7Smrj core_size = (size_t)MISC_VA_BASE - COREHEAP_BASE; 12357c478bd9Sstevel@tonic-gate #else /* __i386 */ 12367c478bd9Sstevel@tonic-gate /* 12377c478bd9Sstevel@tonic-gate * We configure kernelbase based on: 12387c478bd9Sstevel@tonic-gate * 12397c478bd9Sstevel@tonic-gate * 1. user specified kernelbase via eeprom command. Value cannot exceed 12407c478bd9Sstevel@tonic-gate * KERNELBASE_MAX. we large page align eprom_kernelbase 12417c478bd9Sstevel@tonic-gate * 12427c478bd9Sstevel@tonic-gate * 2. Default to KERNELBASE and adjust to 2X less the size for page_t. 12437c478bd9Sstevel@tonic-gate * On large memory systems we must lower kernelbase to allow 12447c478bd9Sstevel@tonic-gate * enough room for page_t's for all of memory. 12457c478bd9Sstevel@tonic-gate * 12467c478bd9Sstevel@tonic-gate * The value set here, might be changed a little later. 12477c478bd9Sstevel@tonic-gate */ 12487c478bd9Sstevel@tonic-gate if (eprom_kernelbase) { 12497c478bd9Sstevel@tonic-gate kernelbase = eprom_kernelbase & mmu.level_mask[1]; 12507c478bd9Sstevel@tonic-gate if (kernelbase > KERNELBASE_MAX) 12517c478bd9Sstevel@tonic-gate kernelbase = KERNELBASE_MAX; 12527c478bd9Sstevel@tonic-gate } else { 12537c478bd9Sstevel@tonic-gate kernelbase = (uintptr_t)KERNELBASE; 12547c478bd9Sstevel@tonic-gate kernelbase -= ROUND_UP_4MEG(2 * valloc_sz); 12557c478bd9Sstevel@tonic-gate } 12567c478bd9Sstevel@tonic-gate ASSERT((kernelbase & mmu.level_offset[1]) == 0); 1257ae115bc7Smrj core_base = valloc_base; 12587c478bd9Sstevel@tonic-gate core_size = 0; 1259ae115bc7Smrj #endif /* __i386 */ 12607c478bd9Sstevel@tonic-gate 12617c478bd9Sstevel@tonic-gate PRM_DEBUG(core_base); 12627c478bd9Sstevel@tonic-gate PRM_DEBUG(core_size); 1263ae115bc7Smrj PRM_DEBUG(kernelbase); 12647c478bd9Sstevel@tonic-gate 126535b1ab99Sjosephb #if defined(__i386) 12667c478bd9Sstevel@tonic-gate segkp_fromheap = 1; 1267ae115bc7Smrj #endif /* __i386 */ 126835b1ab99Sjosephb 12697c478bd9Sstevel@tonic-gate ekernelheap = (char *)core_base; 1270ae115bc7Smrj PRM_DEBUG(ekernelheap); 12717c478bd9Sstevel@tonic-gate 12727c478bd9Sstevel@tonic-gate /* 12737c478bd9Sstevel@tonic-gate * Now that we know the real value of kernelbase, 12747c478bd9Sstevel@tonic-gate * update variables that were initialized with a value of 12757c478bd9Sstevel@tonic-gate * KERNELBASE (in common/conf/param.c). 12767c478bd9Sstevel@tonic-gate * 12777c478bd9Sstevel@tonic-gate * XXX The problem with this sort of hackery is that the 12787c478bd9Sstevel@tonic-gate * compiler just may feel like putting the const declarations 12797c478bd9Sstevel@tonic-gate * (in param.c) into the .text section. Perhaps they should 12807c478bd9Sstevel@tonic-gate * just be declared as variables there? 12817c478bd9Sstevel@tonic-gate */ 12827c478bd9Sstevel@tonic-gate 12837c478bd9Sstevel@tonic-gate *(uintptr_t *)&_kernelbase = kernelbase; 12847c478bd9Sstevel@tonic-gate *(uintptr_t *)&_userlimit = kernelbase; 1285ccbaea4fSjosephb #if defined(__amd64) 1286ccbaea4fSjosephb *(uintptr_t *)&_userlimit -= KERNELBASE - USERLIMIT; 1287ccbaea4fSjosephb #else 12887c478bd9Sstevel@tonic-gate *(uintptr_t *)&_userlimit32 = _userlimit; 12897c478bd9Sstevel@tonic-gate #endif 12907c478bd9Sstevel@tonic-gate PRM_DEBUG(_kernelbase); 12917c478bd9Sstevel@tonic-gate PRM_DEBUG(_userlimit); 12927c478bd9Sstevel@tonic-gate PRM_DEBUG(_userlimit32); 12937c478bd9Sstevel@tonic-gate 129435b1ab99Sjosephb layout_kernel_va(); 129535b1ab99Sjosephb 129635b1ab99Sjosephb #if defined(__i386) 129735b1ab99Sjosephb /* 129835b1ab99Sjosephb * If segmap is too large we can push the bottom of the kernel heap 129935b1ab99Sjosephb * higher than the base. Or worse, it could exceed the top of the 130035b1ab99Sjosephb * VA space entirely, causing it to wrap around. 130135b1ab99Sjosephb */ 130235b1ab99Sjosephb if (kernelheap >= ekernelheap || (uintptr_t)kernelheap < kernelbase) 130335b1ab99Sjosephb panic("too little address space available for kernelheap," 130435b1ab99Sjosephb " use eeprom for lower kernelbase or smaller segmapsize"); 130535b1ab99Sjosephb #endif /* __i386 */ 130635b1ab99Sjosephb 13077c478bd9Sstevel@tonic-gate /* 13087c478bd9Sstevel@tonic-gate * Initialize the kernel heap. Note 3rd argument must be > 1st. 13097c478bd9Sstevel@tonic-gate */ 131035b1ab99Sjosephb kernelheap_init(kernelheap, ekernelheap, 131135b1ab99Sjosephb kernelheap + MMU_PAGESIZE, 1312ae115bc7Smrj (void *)core_base, (void *)(core_base + core_size)); 13137c478bd9Sstevel@tonic-gate 1314843e1988Sjohnlev #if defined(__xpv) 1315843e1988Sjohnlev /* 1316843e1988Sjohnlev * Link pending events struct into cpu struct 1317843e1988Sjohnlev */ 1318843e1988Sjohnlev CPU->cpu_m.mcpu_evt_pend = &cpu0_evt_data; 1319843e1988Sjohnlev #endif 13207c478bd9Sstevel@tonic-gate /* 13217c478bd9Sstevel@tonic-gate * Initialize kernel memory allocator. 13227c478bd9Sstevel@tonic-gate */ 13237c478bd9Sstevel@tonic-gate kmem_init(); 13247c478bd9Sstevel@tonic-gate 1325932dc8e5Sdp /* 1326932dc8e5Sdp * Factor in colorequiv to check additional 'equivalent' bins 1327932dc8e5Sdp */ 1328932dc8e5Sdp page_set_colorequiv_arr(); 1329932dc8e5Sdp 13307c478bd9Sstevel@tonic-gate /* 13317c478bd9Sstevel@tonic-gate * print this out early so that we know what's going on 13327c478bd9Sstevel@tonic-gate */ 13337c478bd9Sstevel@tonic-gate cmn_err(CE_CONT, "?features: %b\n", x86_feature, FMT_X86_FEATURE); 13347c478bd9Sstevel@tonic-gate 13357c478bd9Sstevel@tonic-gate /* 13367c478bd9Sstevel@tonic-gate * Initialize bp_mapin(). 13377c478bd9Sstevel@tonic-gate */ 13387c478bd9Sstevel@tonic-gate bp_init(MMU_PAGESIZE, HAT_STORECACHING_OK); 13397c478bd9Sstevel@tonic-gate 1340f53ad214Skchow /* 1341f53ad214Skchow * orig_npages is non-zero if physmem has been configured for less 1342f53ad214Skchow * than the available memory. 1343f53ad214Skchow */ 1344f53ad214Skchow if (orig_npages) { 134535b1ab99Sjosephb cmn_err(CE_WARN, "!%slimiting physmem to 0x%lx of 0x%lx pages", 134635b1ab99Sjosephb (npages == PHYSMEM ? "Due to virtual address space " : ""), 134735b1ab99Sjosephb npages, orig_npages); 1348f53ad214Skchow } 13497c478bd9Sstevel@tonic-gate #if defined(__i386) 13507c478bd9Sstevel@tonic-gate if (eprom_kernelbase && (eprom_kernelbase != kernelbase)) 13517c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "kernelbase value, User specified 0x%lx, " 13527c478bd9Sstevel@tonic-gate "System using 0x%lx", 13537c478bd9Sstevel@tonic-gate (uintptr_t)eprom_kernelbase, (uintptr_t)kernelbase); 13547c478bd9Sstevel@tonic-gate #endif 13557c478bd9Sstevel@tonic-gate 13567c478bd9Sstevel@tonic-gate #ifdef KERNELBASE_ABI_MIN 13577c478bd9Sstevel@tonic-gate if (kernelbase < (uintptr_t)KERNELBASE_ABI_MIN) { 13587c478bd9Sstevel@tonic-gate cmn_err(CE_NOTE, "!kernelbase set to 0x%lx, system is not " 13597c478bd9Sstevel@tonic-gate "i386 ABI compliant.", (uintptr_t)kernelbase); 13607c478bd9Sstevel@tonic-gate } 13617c478bd9Sstevel@tonic-gate #endif 13627c478bd9Sstevel@tonic-gate 1363843e1988Sjohnlev #ifdef __xpv 1364843e1988Sjohnlev /* 1365843e1988Sjohnlev * Some of the xen start information has to be relocated up 1366843e1988Sjohnlev * into the kernel's permanent address space. 1367843e1988Sjohnlev */ 1368843e1988Sjohnlev PRM_POINT("calling xen_relocate_start_info()"); 1369843e1988Sjohnlev xen_relocate_start_info(); 1370843e1988Sjohnlev PRM_POINT("xen_relocate_start_info() done"); 1371843e1988Sjohnlev 1372843e1988Sjohnlev /* 1373843e1988Sjohnlev * (Update the vcpu pointer in our cpu structure to point into 1374843e1988Sjohnlev * the relocated shared info.) 1375843e1988Sjohnlev */ 1376843e1988Sjohnlev CPU->cpu_m.mcpu_vcpu_info = 1377843e1988Sjohnlev &HYPERVISOR_shared_info->vcpu_info[CPU->cpu_id]; 1378843e1988Sjohnlev #endif 1379843e1988Sjohnlev 1380ae115bc7Smrj PRM_POINT("startup_kmem() done"); 13817c478bd9Sstevel@tonic-gate } 13827c478bd9Sstevel@tonic-gate 1383551bc2a6Smrj #ifndef __xpv 1384551bc2a6Smrj /* 1385551bc2a6Smrj * If we have detected that we are running in an HVM environment, we need 1386551bc2a6Smrj * to prepend the PV driver directory to the module search path. 1387551bc2a6Smrj */ 1388551bc2a6Smrj #define HVM_MOD_DIR "/platform/i86hvm/kernel" 1389551bc2a6Smrj static void 1390551bc2a6Smrj update_default_path() 1391551bc2a6Smrj { 1392551bc2a6Smrj char *current, *newpath; 1393551bc2a6Smrj int newlen; 1394551bc2a6Smrj 1395551bc2a6Smrj /* 1396551bc2a6Smrj * We are about to resync with krtld. krtld will reset its 1397551bc2a6Smrj * internal module search path iff Solaris has set default_path. 1398551bc2a6Smrj * We want to be sure we're prepending this new directory to the 1399551bc2a6Smrj * right search path. 1400551bc2a6Smrj */ 1401551bc2a6Smrj current = (default_path == NULL) ? kobj_module_path : default_path; 1402551bc2a6Smrj 1403551bc2a6Smrj newlen = strlen(HVM_MOD_DIR) + strlen(current) + 1; 1404551bc2a6Smrj newpath = kmem_alloc(newlen, KM_SLEEP); 1405551bc2a6Smrj (void) strcpy(newpath, HVM_MOD_DIR); 1406551bc2a6Smrj (void) strcat(newpath, " "); 1407551bc2a6Smrj (void) strcat(newpath, current); 1408551bc2a6Smrj 1409551bc2a6Smrj default_path = newpath; 1410551bc2a6Smrj } 1411551bc2a6Smrj #endif 1412551bc2a6Smrj 14137c478bd9Sstevel@tonic-gate static void 14147c478bd9Sstevel@tonic-gate startup_modules(void) 14157c478bd9Sstevel@tonic-gate { 1416ed5289f9SKen Erickson int cnt; 14177c478bd9Sstevel@tonic-gate extern void prom_setup(void); 1418ed5289f9SKen Erickson int32_t v, h; 1419ed5289f9SKen Erickson char d[11]; 1420ed5289f9SKen Erickson char *cp; 1421e4b86885SCheng Sean Ye cmi_hdl_t hdl; 14227c478bd9Sstevel@tonic-gate 14237c478bd9Sstevel@tonic-gate PRM_POINT("startup_modules() starting..."); 1424843e1988Sjohnlev 1425843e1988Sjohnlev #ifndef __xpv 14267c478bd9Sstevel@tonic-gate /* 14277c478bd9Sstevel@tonic-gate * Initialize ten-micro second timer so that drivers will 14287c478bd9Sstevel@tonic-gate * not get short changed in their init phase. This was 14297c478bd9Sstevel@tonic-gate * not getting called until clkinit which, on fast cpu's 14307c478bd9Sstevel@tonic-gate * caused the drv_usecwait to be way too short. 14317c478bd9Sstevel@tonic-gate */ 14327c478bd9Sstevel@tonic-gate microfind(); 1433551bc2a6Smrj 1434551bc2a6Smrj if (xpv_is_hvm) 1435551bc2a6Smrj update_default_path(); 1436843e1988Sjohnlev #endif 14377c478bd9Sstevel@tonic-gate 14387c478bd9Sstevel@tonic-gate /* 14397c478bd9Sstevel@tonic-gate * Read the GMT lag from /etc/rtc_config. 14407c478bd9Sstevel@tonic-gate */ 1441ae115bc7Smrj sgmtl(process_rtc_config_file()); 14427c478bd9Sstevel@tonic-gate 14437c478bd9Sstevel@tonic-gate /* 14447c478bd9Sstevel@tonic-gate * Calculate default settings of system parameters based upon 14457c478bd9Sstevel@tonic-gate * maxusers, yet allow to be overridden via the /etc/system file. 14467c478bd9Sstevel@tonic-gate */ 14477c478bd9Sstevel@tonic-gate param_calc(0); 14487c478bd9Sstevel@tonic-gate 14497c478bd9Sstevel@tonic-gate mod_setup(); 14507c478bd9Sstevel@tonic-gate 14517c478bd9Sstevel@tonic-gate /* 14527c478bd9Sstevel@tonic-gate * Initialize system parameters. 14537c478bd9Sstevel@tonic-gate */ 14547c478bd9Sstevel@tonic-gate param_init(); 14557c478bd9Sstevel@tonic-gate 14569acbbeafSnn /* 14579acbbeafSnn * Initialize the default brands 14589acbbeafSnn */ 14599acbbeafSnn brand_init(); 14609acbbeafSnn 14617c478bd9Sstevel@tonic-gate /* 14627c478bd9Sstevel@tonic-gate * maxmem is the amount of physical memory we're playing with. 14637c478bd9Sstevel@tonic-gate */ 14647c478bd9Sstevel@tonic-gate maxmem = physmem; 14657c478bd9Sstevel@tonic-gate 14667c478bd9Sstevel@tonic-gate /* 14677c478bd9Sstevel@tonic-gate * Initialize segment management stuff. 14687c478bd9Sstevel@tonic-gate */ 14697c478bd9Sstevel@tonic-gate seg_init(); 14707c478bd9Sstevel@tonic-gate 14717c478bd9Sstevel@tonic-gate if (modload("fs", "specfs") == -1) 14727c478bd9Sstevel@tonic-gate halt("Can't load specfs"); 14737c478bd9Sstevel@tonic-gate 14747c478bd9Sstevel@tonic-gate if (modload("fs", "devfs") == -1) 14757c478bd9Sstevel@tonic-gate halt("Can't load devfs"); 14767c478bd9Sstevel@tonic-gate 1477facf4a8dSllai if (modload("fs", "dev") == -1) 1478facf4a8dSllai halt("Can't load dev"); 1479facf4a8dSllai 148045916cd2Sjpk (void) modloadonly("sys", "lbl_edition"); 148145916cd2Sjpk 14827c478bd9Sstevel@tonic-gate dispinit(); 14837c478bd9Sstevel@tonic-gate 14847c478bd9Sstevel@tonic-gate /* 14857c478bd9Sstevel@tonic-gate * This is needed here to initialize hw_serial[] for cluster booting. 14867c478bd9Sstevel@tonic-gate */ 1487*5679c89fSjv if ((h = set_soft_hostid()) == HW_INVALID_HOSTID) { 1488ed5289f9SKen Erickson cmn_err(CE_WARN, "Unable to set hostid"); 1489*5679c89fSjv } else { 1490ed5289f9SKen Erickson for (v = h, cnt = 0; cnt < 10; cnt++) { 1491*5679c89fSjv d[cnt] = (char)(v % 10); 1492ed5289f9SKen Erickson v /= 10; 1493ed5289f9SKen Erickson if (v == 0) 1494ed5289f9SKen Erickson break; 1495ed5289f9SKen Erickson } 1496ed5289f9SKen Erickson for (cp = hw_serial; cnt >= 0; cnt--) 1497ed5289f9SKen Erickson *cp++ = d[cnt] + '0'; 1498ed5289f9SKen Erickson *cp = 0; 1499ed5289f9SKen Erickson } 15007c478bd9Sstevel@tonic-gate 15017c478bd9Sstevel@tonic-gate /* Read cluster configuration data. */ 15027c478bd9Sstevel@tonic-gate clconf_init(); 15037c478bd9Sstevel@tonic-gate 1504843e1988Sjohnlev #if defined(__xpv) 1505843e1988Sjohnlev ec_init(); 1506843e1988Sjohnlev gnttab_init(); 1507843e1988Sjohnlev (void) xs_early_init(); 1508843e1988Sjohnlev #endif /* __xpv */ 1509843e1988Sjohnlev 15107c478bd9Sstevel@tonic-gate /* 15117c478bd9Sstevel@tonic-gate * Create a kernel device tree. First, create rootnex and 15127c478bd9Sstevel@tonic-gate * then invoke bus specific code to probe devices. 15137c478bd9Sstevel@tonic-gate */ 15147c478bd9Sstevel@tonic-gate setup_ddi(); 15157aec1d6eScindi 1516e4b86885SCheng Sean Ye /* 1517e4b86885SCheng Sean Ye * Set up the CPU module subsystem for the boot cpu in the native 1518e4b86885SCheng Sean Ye * case, and all physical cpu resource in the xpv dom0 case. 1519e4b86885SCheng Sean Ye * Modifies the device tree, so this must be done after 1520e4b86885SCheng Sean Ye * setup_ddi(). 1521e4b86885SCheng Sean Ye */ 1522e4b86885SCheng Sean Ye #ifdef __xpv 1523e4b86885SCheng Sean Ye /* 1524e4b86885SCheng Sean Ye * If paravirtualized and on dom0 then we initialize all physical 1525e4b86885SCheng Sean Ye * cpu handles now; if paravirtualized on a domU then do not 1526e4b86885SCheng Sean Ye * initialize. 1527e4b86885SCheng Sean Ye */ 1528e4b86885SCheng Sean Ye if (DOMAIN_IS_INITDOMAIN(xen_info)) { 1529e4b86885SCheng Sean Ye xen_mc_lcpu_cookie_t cpi; 1530e4b86885SCheng Sean Ye 1531e4b86885SCheng Sean Ye for (cpi = xen_physcpu_next(NULL); cpi != NULL; 1532e4b86885SCheng Sean Ye cpi = xen_physcpu_next(cpi)) { 1533e4b86885SCheng Sean Ye if ((hdl = cmi_init(CMI_HDL_SOLARIS_xVM_MCA, 1534e4b86885SCheng Sean Ye xen_physcpu_chipid(cpi), xen_physcpu_coreid(cpi), 1535e4b86885SCheng Sean Ye xen_physcpu_strandid(cpi))) != NULL && 1536e4b86885SCheng Sean Ye (x86_feature & X86_MCA)) 153720c794b3Sgavinm cmi_mca_init(hdl); 153820c794b3Sgavinm } 153920c794b3Sgavinm } 1540e4b86885SCheng Sean Ye #else 1541e4b86885SCheng Sean Ye /* 1542e4b86885SCheng Sean Ye * Initialize a handle for the boot cpu - others will initialize 1543e4b86885SCheng Sean Ye * as they startup. Do not do this if we know we are in an HVM domU. 1544e4b86885SCheng Sean Ye */ 1545e4b86885SCheng Sean Ye if (!xpv_is_hvm && 1546e4b86885SCheng Sean Ye (hdl = cmi_init(CMI_HDL_NATIVE, cmi_ntv_hwchipid(CPU), 1547e4b86885SCheng Sean Ye cmi_ntv_hwcoreid(CPU), cmi_ntv_hwstrandid(CPU))) != NULL && 1548e4b86885SCheng Sean Ye (x86_feature & X86_MCA)) 1549e4b86885SCheng Sean Ye cmi_mca_init(hdl); 155020c794b3Sgavinm #endif /* __xpv */ 15517aec1d6eScindi 15527c478bd9Sstevel@tonic-gate /* 15537c478bd9Sstevel@tonic-gate * Fake a prom tree such that /dev/openprom continues to work 15547c478bd9Sstevel@tonic-gate */ 1555ae115bc7Smrj PRM_POINT("startup_modules: calling prom_setup..."); 15567c478bd9Sstevel@tonic-gate prom_setup(); 1557ae115bc7Smrj PRM_POINT("startup_modules: done"); 15587c478bd9Sstevel@tonic-gate 15597c478bd9Sstevel@tonic-gate /* 15607c478bd9Sstevel@tonic-gate * Load all platform specific modules 15617c478bd9Sstevel@tonic-gate */ 1562ae115bc7Smrj PRM_POINT("startup_modules: calling psm_modload..."); 15637c478bd9Sstevel@tonic-gate psm_modload(); 15647c478bd9Sstevel@tonic-gate 15657c478bd9Sstevel@tonic-gate PRM_POINT("startup_modules() done"); 15667c478bd9Sstevel@tonic-gate } 15677c478bd9Sstevel@tonic-gate 1568ae115bc7Smrj /* 1569ae115bc7Smrj * claim a "setaside" boot page for use in the kernel 1570ae115bc7Smrj */ 1571ae115bc7Smrj page_t * 1572ae115bc7Smrj boot_claim_page(pfn_t pfn) 15737c478bd9Sstevel@tonic-gate { 1574ae115bc7Smrj page_t *pp; 15757c478bd9Sstevel@tonic-gate 1576ae115bc7Smrj pp = page_numtopp_nolock(pfn); 1577ae115bc7Smrj ASSERT(pp != NULL); 15787c478bd9Sstevel@tonic-gate 1579ae115bc7Smrj if (PP_ISBOOTPAGES(pp)) { 1580ae115bc7Smrj if (pp->p_next != NULL) 1581ae115bc7Smrj pp->p_next->p_prev = pp->p_prev; 1582ae115bc7Smrj if (pp->p_prev == NULL) 1583ae115bc7Smrj bootpages = pp->p_next; 1584ae115bc7Smrj else 1585ae115bc7Smrj pp->p_prev->p_next = pp->p_next; 1586ae115bc7Smrj } else { 1587ae115bc7Smrj /* 1588ae115bc7Smrj * htable_attach() expects a base pagesize page 1589ae115bc7Smrj */ 1590ae115bc7Smrj if (pp->p_szc != 0) 1591ae115bc7Smrj page_boot_demote(pp); 1592ae115bc7Smrj pp = page_numtopp(pfn, SE_EXCL); 1593ae115bc7Smrj } 1594ae115bc7Smrj return (pp); 15957c478bd9Sstevel@tonic-gate } 15967c478bd9Sstevel@tonic-gate 15977c478bd9Sstevel@tonic-gate /* 15987c478bd9Sstevel@tonic-gate * Walk through the pagetables looking for pages mapped in by boot. If the 15997c478bd9Sstevel@tonic-gate * setaside flag is set the pages are expected to be returned to the 16007c478bd9Sstevel@tonic-gate * kernel later in boot, so we add them to the bootpages list. 16017c478bd9Sstevel@tonic-gate */ 16027c478bd9Sstevel@tonic-gate static void 16037c478bd9Sstevel@tonic-gate protect_boot_range(uintptr_t low, uintptr_t high, int setaside) 16047c478bd9Sstevel@tonic-gate { 16057c478bd9Sstevel@tonic-gate uintptr_t va = low; 16067c478bd9Sstevel@tonic-gate size_t len; 16077c478bd9Sstevel@tonic-gate uint_t prot; 16087c478bd9Sstevel@tonic-gate pfn_t pfn; 16097c478bd9Sstevel@tonic-gate page_t *pp; 16107c478bd9Sstevel@tonic-gate pgcnt_t boot_protect_cnt = 0; 16117c478bd9Sstevel@tonic-gate 1612ae115bc7Smrj while (kbm_probe(&va, &len, &pfn, &prot) != 0 && va < high) { 16137c478bd9Sstevel@tonic-gate if (va + len >= high) 16147c478bd9Sstevel@tonic-gate panic("0x%lx byte mapping at 0x%p exceeds boot's " 16157c478bd9Sstevel@tonic-gate "legal range.", len, (void *)va); 16167c478bd9Sstevel@tonic-gate 16177c478bd9Sstevel@tonic-gate while (len > 0) { 16187c478bd9Sstevel@tonic-gate pp = page_numtopp_alloc(pfn); 16197c478bd9Sstevel@tonic-gate if (pp != NULL) { 16207c478bd9Sstevel@tonic-gate if (setaside == 0) 16217c478bd9Sstevel@tonic-gate panic("Unexpected mapping by boot. " 16227c478bd9Sstevel@tonic-gate "addr=%p pfn=%lx\n", 16237c478bd9Sstevel@tonic-gate (void *)va, pfn); 16247c478bd9Sstevel@tonic-gate 16257c478bd9Sstevel@tonic-gate pp->p_next = bootpages; 1626ae115bc7Smrj pp->p_prev = NULL; 1627ae115bc7Smrj PP_SETBOOTPAGES(pp); 1628ae115bc7Smrj if (bootpages != NULL) { 1629ae115bc7Smrj bootpages->p_prev = pp; 1630ae115bc7Smrj } 16317c478bd9Sstevel@tonic-gate bootpages = pp; 16327c478bd9Sstevel@tonic-gate ++boot_protect_cnt; 16337c478bd9Sstevel@tonic-gate } 16347c478bd9Sstevel@tonic-gate 16357c478bd9Sstevel@tonic-gate ++pfn; 16367c478bd9Sstevel@tonic-gate len -= MMU_PAGESIZE; 16377c478bd9Sstevel@tonic-gate va += MMU_PAGESIZE; 16387c478bd9Sstevel@tonic-gate } 16397c478bd9Sstevel@tonic-gate } 16407c478bd9Sstevel@tonic-gate PRM_DEBUG(boot_protect_cnt); 16417c478bd9Sstevel@tonic-gate } 16427c478bd9Sstevel@tonic-gate 1643ae115bc7Smrj /* 164435b1ab99Sjosephb * 1645ae115bc7Smrj */ 16467c478bd9Sstevel@tonic-gate static void 164735b1ab99Sjosephb layout_kernel_va(void) 16487c478bd9Sstevel@tonic-gate { 164935b1ab99Sjosephb PRM_POINT("layout_kernel_va() starting..."); 16507c478bd9Sstevel@tonic-gate /* 1651ae115bc7Smrj * Establish the final size of the kernel's heap, size of segmap, 1652ae115bc7Smrj * segkp, etc. 16537c478bd9Sstevel@tonic-gate */ 16547c478bd9Sstevel@tonic-gate 16557c478bd9Sstevel@tonic-gate #if defined(__amd64) 16567c478bd9Sstevel@tonic-gate 165735b1ab99Sjosephb kpm_vbase = (caddr_t)segkpm_base; 165835b1ab99Sjosephb kpm_size = ROUND_UP_LPAGE(mmu_ptob(physmax + 1)); 165935b1ab99Sjosephb if ((uintptr_t)kpm_vbase + kpm_size > (uintptr_t)valloc_base) 166035b1ab99Sjosephb panic("not enough room for kpm!"); 1661ae115bc7Smrj PRM_DEBUG(kpm_size); 1662ae115bc7Smrj PRM_DEBUG(kpm_vbase); 16637c478bd9Sstevel@tonic-gate 16647c478bd9Sstevel@tonic-gate /* 1665ae115bc7Smrj * By default we create a seg_kp in 64 bit kernels, it's a little 1666ae115bc7Smrj * faster to access than embedding it in the heap. 16677c478bd9Sstevel@tonic-gate */ 1668ae115bc7Smrj segkp_base = (caddr_t)valloc_base + valloc_sz; 16697c478bd9Sstevel@tonic-gate if (!segkp_fromheap) { 16707c478bd9Sstevel@tonic-gate size_t sz = mmu_ptob(segkpsize); 16717c478bd9Sstevel@tonic-gate 16727c478bd9Sstevel@tonic-gate /* 1673ae115bc7Smrj * determine size of segkp 16747c478bd9Sstevel@tonic-gate */ 16757c478bd9Sstevel@tonic-gate if (sz < SEGKPMINSIZE || sz > SEGKPMAXSIZE) { 16767c478bd9Sstevel@tonic-gate sz = SEGKPDEFSIZE; 16777c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "!Illegal value for segkpsize. " 16787c478bd9Sstevel@tonic-gate "segkpsize has been reset to %ld pages", 16797c478bd9Sstevel@tonic-gate mmu_btop(sz)); 16807c478bd9Sstevel@tonic-gate } 16817c478bd9Sstevel@tonic-gate sz = MIN(sz, MAX(SEGKPMINSIZE, mmu_ptob(physmem))); 16827c478bd9Sstevel@tonic-gate 16837c478bd9Sstevel@tonic-gate segkpsize = mmu_btop(ROUND_UP_LPAGE(sz)); 16847c478bd9Sstevel@tonic-gate } 1685ae115bc7Smrj PRM_DEBUG(segkp_base); 1686ae115bc7Smrj PRM_DEBUG(segkpsize); 16877c478bd9Sstevel@tonic-gate 168835b1ab99Sjosephb /* 168935b1ab99Sjosephb * segzio is used for ZFS cached data. It uses a distinct VA 169035b1ab99Sjosephb * segment (from kernel heap) so that we can easily tell not to 169135b1ab99Sjosephb * include it in kernel crash dumps on 64 bit kernels. The trick is 169235b1ab99Sjosephb * to give it lots of VA, but not constrain the kernel heap. 169335b1ab99Sjosephb * We scale the size of segzio linearly with physmem up to 169435b1ab99Sjosephb * SEGZIOMAXSIZE. Above that amount it scales at 50% of physmem. 169535b1ab99Sjosephb */ 1696ae115bc7Smrj segzio_base = segkp_base + mmu_ptob(segkpsize); 1697ae115bc7Smrj if (segzio_fromheap) { 1698ae115bc7Smrj segziosize = 0; 1699ae115bc7Smrj } else { 170035b1ab99Sjosephb size_t physmem_size = mmu_ptob(physmem); 170135b1ab99Sjosephb size_t size = (segziosize == 0) ? 170235b1ab99Sjosephb physmem_size : mmu_ptob(segziosize); 1703ad23a2dbSjohansen 170435b1ab99Sjosephb if (size < SEGZIOMINSIZE) 1705ad23a2dbSjohansen size = SEGZIOMINSIZE; 170635b1ab99Sjosephb if (size > SEGZIOMAXSIZE) { 1707a778305aSjohansen size = SEGZIOMAXSIZE; 170835b1ab99Sjosephb if (physmem_size > size) 170935b1ab99Sjosephb size += (physmem_size - size) / 2; 1710ad23a2dbSjohansen } 1711ad23a2dbSjohansen segziosize = mmu_btop(ROUND_UP_LPAGE(size)); 1712ad23a2dbSjohansen } 1713ae115bc7Smrj PRM_DEBUG(segziosize); 1714ae115bc7Smrj PRM_DEBUG(segzio_base); 1715ad23a2dbSjohansen 17167c478bd9Sstevel@tonic-gate /* 1717ae115bc7Smrj * Put the range of VA for device mappings next, kmdb knows to not 1718ae115bc7Smrj * grep in this range of addresses. 17197c478bd9Sstevel@tonic-gate */ 1720ae115bc7Smrj toxic_addr = 1721ae115bc7Smrj ROUND_UP_LPAGE((uintptr_t)segzio_base + mmu_ptob(segziosize)); 17227c478bd9Sstevel@tonic-gate PRM_DEBUG(toxic_addr); 1723ae115bc7Smrj segmap_start = ROUND_UP_LPAGE(toxic_addr + toxic_size); 1724ae115bc7Smrj #else /* __i386 */ 1725ae115bc7Smrj segmap_start = ROUND_UP_LPAGE(kernelbase); 1726ae115bc7Smrj #endif /* __i386 */ 1727ae115bc7Smrj PRM_DEBUG(segmap_start); 17287c478bd9Sstevel@tonic-gate 17297c478bd9Sstevel@tonic-gate /* 173052eef812SAmrita Sadhukhan * Users can change segmapsize through eeprom. If the variable 173152eef812SAmrita Sadhukhan * is tuned through eeprom, there is no upper bound on the 1732ed5289f9SKen Erickson * size of segmap. 17337c478bd9Sstevel@tonic-gate */ 17347c478bd9Sstevel@tonic-gate segmapsize = MAX(ROUND_UP_LPAGE(segmapsize), SEGMAPDEFAULT); 17357c478bd9Sstevel@tonic-gate 17367c478bd9Sstevel@tonic-gate #if defined(__i386) 17377c478bd9Sstevel@tonic-gate /* 17387c478bd9Sstevel@tonic-gate * 32-bit systems don't have segkpm or segkp, so segmap appears at 17397c478bd9Sstevel@tonic-gate * the bottom of the kernel's address range. Set aside space for a 1740ae115bc7Smrj * small red zone just below the start of segmap. 17417c478bd9Sstevel@tonic-gate */ 1742ae115bc7Smrj segmap_start += KERNEL_REDZONE_SIZE; 17437c478bd9Sstevel@tonic-gate segmapsize -= KERNEL_REDZONE_SIZE; 17447c478bd9Sstevel@tonic-gate #endif 17457c478bd9Sstevel@tonic-gate 1746ae115bc7Smrj PRM_DEBUG(segmap_start); 17477c478bd9Sstevel@tonic-gate PRM_DEBUG(segmapsize); 174835b1ab99Sjosephb kernelheap = (caddr_t)ROUND_UP_LPAGE(segmap_start + segmapsize); 174935b1ab99Sjosephb PRM_DEBUG(kernelheap); 175035b1ab99Sjosephb PRM_POINT("layout_kernel_va() done..."); 175135b1ab99Sjosephb } 175235b1ab99Sjosephb 175335b1ab99Sjosephb /* 175435b1ab99Sjosephb * Finish initializing the VM system, now that we are no longer 175535b1ab99Sjosephb * relying on the boot time memory allocators. 175635b1ab99Sjosephb */ 175735b1ab99Sjosephb static void 175835b1ab99Sjosephb startup_vm(void) 175935b1ab99Sjosephb { 176035b1ab99Sjosephb struct segmap_crargs a; 176135b1ab99Sjosephb 176235b1ab99Sjosephb extern int use_brk_lpg, use_stk_lpg; 176335b1ab99Sjosephb 176435b1ab99Sjosephb PRM_POINT("startup_vm() starting..."); 176535b1ab99Sjosephb 176635b1ab99Sjosephb /* 176735b1ab99Sjosephb * Initialize the hat layer. 176835b1ab99Sjosephb */ 176935b1ab99Sjosephb hat_init(); 17707c478bd9Sstevel@tonic-gate 1771ae115bc7Smrj /* 1772ae115bc7Smrj * Do final allocations of HAT data structures that need to 1773ae115bc7Smrj * be allocated before quiescing the boot loader. 1774ae115bc7Smrj */ 1775ae115bc7Smrj PRM_POINT("Calling hat_kern_alloc()..."); 1776ae115bc7Smrj hat_kern_alloc((caddr_t)segmap_start, segmapsize, ekernelheap); 1777ae115bc7Smrj PRM_POINT("hat_kern_alloc() done"); 1778ae115bc7Smrj 1779843e1988Sjohnlev #ifndef __xpv 1780ae115bc7Smrj /* 17811d03c31eSjohnlev * Setup Page Attribute Table 1782ae115bc7Smrj */ 17831d03c31eSjohnlev pat_sync(); 1784843e1988Sjohnlev #endif 1785ae115bc7Smrj 1786ae115bc7Smrj /* 1787ae115bc7Smrj * The next two loops are done in distinct steps in order 1788ae115bc7Smrj * to be sure that any page that is doubly mapped (both above 1789ae115bc7Smrj * KERNEL_TEXT and below kernelbase) is dealt with correctly. 1790ae115bc7Smrj * Note this may never happen, but it might someday. 1791ae115bc7Smrj */ 1792ae115bc7Smrj bootpages = NULL; 1793ae115bc7Smrj PRM_POINT("Protecting boot pages"); 1794ae115bc7Smrj 1795ae115bc7Smrj /* 1796ae115bc7Smrj * Protect any pages mapped above KERNEL_TEXT that somehow have 1797ae115bc7Smrj * page_t's. This can only happen if something weird allocated 1798ae115bc7Smrj * in this range (like kadb/kmdb). 1799ae115bc7Smrj */ 1800ae115bc7Smrj protect_boot_range(KERNEL_TEXT, (uintptr_t)-1, 0); 1801ae115bc7Smrj 1802ae115bc7Smrj /* 1803ae115bc7Smrj * Before we can take over memory allocation/mapping from the boot 1804ae115bc7Smrj * loader we must remove from our free page lists any boot allocated 1805ae115bc7Smrj * pages that stay mapped until release_bootstrap(). 1806ae115bc7Smrj */ 1807ae115bc7Smrj protect_boot_range(0, kernelbase, 1); 1808ae115bc7Smrj 1809843e1988Sjohnlev 1810ae115bc7Smrj /* 1811ae115bc7Smrj * Switch to running on regular HAT (not boot_mmu) 1812ae115bc7Smrj */ 1813ae115bc7Smrj PRM_POINT("Calling hat_kern_setup()..."); 1814ae115bc7Smrj hat_kern_setup(); 1815ae115bc7Smrj 1816ae115bc7Smrj /* 1817ae115bc7Smrj * It is no longer safe to call BOP_ALLOC(), so make sure we don't. 1818ae115bc7Smrj */ 1819ae115bc7Smrj bop_no_more_mem(); 1820ae115bc7Smrj 1821ae115bc7Smrj PRM_POINT("hat_kern_setup() done"); 1822ae115bc7Smrj 1823ae115bc7Smrj hat_cpu_online(CPU); 1824ae115bc7Smrj 18257c478bd9Sstevel@tonic-gate /* 18267c478bd9Sstevel@tonic-gate * Initialize VM system 18277c478bd9Sstevel@tonic-gate */ 18287c478bd9Sstevel@tonic-gate PRM_POINT("Calling kvm_init()..."); 18297c478bd9Sstevel@tonic-gate kvm_init(); 18307c478bd9Sstevel@tonic-gate PRM_POINT("kvm_init() done"); 18317c478bd9Sstevel@tonic-gate 18327c478bd9Sstevel@tonic-gate /* 18337c478bd9Sstevel@tonic-gate * Tell kmdb that the VM system is now working 18347c478bd9Sstevel@tonic-gate */ 18357c478bd9Sstevel@tonic-gate if (boothowto & RB_DEBUG) 18367c478bd9Sstevel@tonic-gate kdi_dvec_vmready(); 18377c478bd9Sstevel@tonic-gate 1838843e1988Sjohnlev #if defined(__xpv) 1839843e1988Sjohnlev /* 1840843e1988Sjohnlev * Populate the I/O pool on domain 0 1841843e1988Sjohnlev */ 1842843e1988Sjohnlev if (DOMAIN_IS_INITDOMAIN(xen_info)) { 1843843e1988Sjohnlev extern long populate_io_pool(void); 1844843e1988Sjohnlev long init_io_pool_cnt; 1845843e1988Sjohnlev 1846843e1988Sjohnlev PRM_POINT("Populating reserve I/O page pool"); 1847843e1988Sjohnlev init_io_pool_cnt = populate_io_pool(); 1848843e1988Sjohnlev PRM_DEBUG(init_io_pool_cnt); 1849843e1988Sjohnlev } 1850843e1988Sjohnlev #endif 18517c478bd9Sstevel@tonic-gate /* 18527c478bd9Sstevel@tonic-gate * Mangle the brand string etc. 18537c478bd9Sstevel@tonic-gate */ 18547c478bd9Sstevel@tonic-gate cpuid_pass3(CPU); 18557c478bd9Sstevel@tonic-gate 18567c478bd9Sstevel@tonic-gate #if defined(__amd64) 18577c478bd9Sstevel@tonic-gate 18587c478bd9Sstevel@tonic-gate /* 18597c478bd9Sstevel@tonic-gate * Create the device arena for toxic (to dtrace/kmdb) mappings. 18607c478bd9Sstevel@tonic-gate */ 18617c478bd9Sstevel@tonic-gate device_arena = vmem_create("device", (void *)toxic_addr, 18627c478bd9Sstevel@tonic-gate toxic_size, MMU_PAGESIZE, NULL, NULL, NULL, 0, VM_SLEEP); 18637c478bd9Sstevel@tonic-gate 18647c478bd9Sstevel@tonic-gate #else /* __i386 */ 18657c478bd9Sstevel@tonic-gate 18667c478bd9Sstevel@tonic-gate /* 18677c478bd9Sstevel@tonic-gate * allocate the bit map that tracks toxic pages 18687c478bd9Sstevel@tonic-gate */ 1869ae115bc7Smrj toxic_bit_map_len = btop((ulong_t)(valloc_base - kernelbase)); 18707c478bd9Sstevel@tonic-gate PRM_DEBUG(toxic_bit_map_len); 18717c478bd9Sstevel@tonic-gate toxic_bit_map = 18727c478bd9Sstevel@tonic-gate kmem_zalloc(BT_SIZEOFMAP(toxic_bit_map_len), KM_NOSLEEP); 18737c478bd9Sstevel@tonic-gate ASSERT(toxic_bit_map != NULL); 18747c478bd9Sstevel@tonic-gate PRM_DEBUG(toxic_bit_map); 18757c478bd9Sstevel@tonic-gate 18767c478bd9Sstevel@tonic-gate #endif /* __i386 */ 18777c478bd9Sstevel@tonic-gate 18787c478bd9Sstevel@tonic-gate 18797c478bd9Sstevel@tonic-gate /* 18807c478bd9Sstevel@tonic-gate * Now that we've got more VA, as well as the ability to allocate from 18817c478bd9Sstevel@tonic-gate * it, tell the debugger. 18827c478bd9Sstevel@tonic-gate */ 18837c478bd9Sstevel@tonic-gate if (boothowto & RB_DEBUG) 18847c478bd9Sstevel@tonic-gate kdi_dvec_memavail(); 18857c478bd9Sstevel@tonic-gate 18867c478bd9Sstevel@tonic-gate /* 18877c478bd9Sstevel@tonic-gate * The following code installs a special page fault handler (#pf) 18887c478bd9Sstevel@tonic-gate * to work around a pentium bug. 18897c478bd9Sstevel@tonic-gate */ 1890843e1988Sjohnlev #if !defined(__amd64) && !defined(__xpv) 18917c478bd9Sstevel@tonic-gate if (x86_type == X86_TYPE_P5) { 1892ae115bc7Smrj desctbr_t idtr; 18937c478bd9Sstevel@tonic-gate gate_desc_t *newidt; 18947c478bd9Sstevel@tonic-gate 18957c478bd9Sstevel@tonic-gate if ((newidt = kmem_zalloc(MMU_PAGESIZE, KM_NOSLEEP)) == NULL) 18967c478bd9Sstevel@tonic-gate panic("failed to install pentium_pftrap"); 18977c478bd9Sstevel@tonic-gate 18980cfdb603Sjosephb bcopy(idt0, newidt, NIDT * sizeof (*idt0)); 18997c478bd9Sstevel@tonic-gate set_gatesegd(&newidt[T_PGFLT], &pentium_pftrap, 1900843e1988Sjohnlev KCS_SEL, SDT_SYSIGT, TRP_KPL); 19017c478bd9Sstevel@tonic-gate 19027c478bd9Sstevel@tonic-gate (void) as_setprot(&kas, (caddr_t)newidt, MMU_PAGESIZE, 19030cfdb603Sjosephb PROT_READ | PROT_EXEC); 19047c478bd9Sstevel@tonic-gate 19050cfdb603Sjosephb CPU->cpu_idt = newidt; 19060cfdb603Sjosephb idtr.dtr_base = (uintptr_t)CPU->cpu_idt; 19070cfdb603Sjosephb idtr.dtr_limit = (NIDT * sizeof (*idt0)) - 1; 1908ae115bc7Smrj wr_idtr(&idtr); 19097c478bd9Sstevel@tonic-gate } 19107c478bd9Sstevel@tonic-gate #endif /* !__amd64 */ 19117c478bd9Sstevel@tonic-gate 1912843e1988Sjohnlev #if !defined(__xpv) 19137c478bd9Sstevel@tonic-gate /* 19147c478bd9Sstevel@tonic-gate * Map page pfn=0 for drivers, such as kd, that need to pick up 19157c478bd9Sstevel@tonic-gate * parameters left there by controllers/BIOS. 19167c478bd9Sstevel@tonic-gate */ 19177c478bd9Sstevel@tonic-gate PRM_POINT("setup up p0_va"); 19187c478bd9Sstevel@tonic-gate p0_va = i86devmap(0, 1, PROT_READ); 19197c478bd9Sstevel@tonic-gate PRM_DEBUG(p0_va); 1920843e1988Sjohnlev #endif 19217c478bd9Sstevel@tonic-gate 19227c478bd9Sstevel@tonic-gate cmn_err(CE_CONT, "?mem = %luK (0x%lx)\n", 19237c478bd9Sstevel@tonic-gate physinstalled << (MMU_PAGESHIFT - 10), ptob(physinstalled)); 19247c478bd9Sstevel@tonic-gate 192583f9b804Skchow /* 192683f9b804Skchow * disable automatic large pages for small memory systems or 192783f9b804Skchow * when the disable flag is set. 192802bc52beSkchow * 192902bc52beSkchow * Do not yet consider page sizes larger than 2m/4m. 193083f9b804Skchow */ 1931ec25b48fSsusans if (!auto_lpg_disable && mmu.max_page_level > 0) { 1932ec25b48fSsusans max_uheap_lpsize = LEVEL_SIZE(1); 1933ec25b48fSsusans max_ustack_lpsize = LEVEL_SIZE(1); 1934ec25b48fSsusans max_privmap_lpsize = LEVEL_SIZE(1); 1935ec25b48fSsusans max_uidata_lpsize = LEVEL_SIZE(1); 1936ec25b48fSsusans max_utext_lpsize = LEVEL_SIZE(1); 1937ec25b48fSsusans max_shm_lpsize = LEVEL_SIZE(1); 1938ec25b48fSsusans } 1939ec25b48fSsusans if (physmem < privm_lpg_min_physmem || mmu.max_page_level == 0 || 1940ec25b48fSsusans auto_lpg_disable) { 1941beb1bda0Sdavemq use_brk_lpg = 0; 1942beb1bda0Sdavemq use_stk_lpg = 0; 1943ec25b48fSsusans } 194402bc52beSkchow mcntl0_lpsize = LEVEL_SIZE(mmu.umax_page_level); 1945beb1bda0Sdavemq 19467c478bd9Sstevel@tonic-gate PRM_POINT("Calling hat_init_finish()..."); 19477c478bd9Sstevel@tonic-gate hat_init_finish(); 19487c478bd9Sstevel@tonic-gate PRM_POINT("hat_init_finish() done"); 19497c478bd9Sstevel@tonic-gate 19507c478bd9Sstevel@tonic-gate /* 19517c478bd9Sstevel@tonic-gate * Initialize the segkp segment type. 19527c478bd9Sstevel@tonic-gate */ 19537c478bd9Sstevel@tonic-gate rw_enter(&kas.a_lock, RW_WRITER); 1954ae115bc7Smrj PRM_POINT("Attaching segkp"); 1955ae115bc7Smrj if (segkp_fromheap) { 19567c478bd9Sstevel@tonic-gate segkp->s_as = &kas; 1957ae115bc7Smrj } else if (seg_attach(&kas, (caddr_t)segkp_base, mmu_ptob(segkpsize), 1958ae115bc7Smrj segkp) < 0) { 1959ae115bc7Smrj panic("startup: cannot attach segkp"); 1960ae115bc7Smrj /*NOTREACHED*/ 19617c478bd9Sstevel@tonic-gate } 1962ae115bc7Smrj PRM_POINT("Doing segkp_create()"); 19637c478bd9Sstevel@tonic-gate if (segkp_create(segkp) != 0) { 19647c478bd9Sstevel@tonic-gate panic("startup: segkp_create failed"); 19657c478bd9Sstevel@tonic-gate /*NOTREACHED*/ 19667c478bd9Sstevel@tonic-gate } 19677c478bd9Sstevel@tonic-gate PRM_DEBUG(segkp); 19687c478bd9Sstevel@tonic-gate rw_exit(&kas.a_lock); 19697c478bd9Sstevel@tonic-gate 19707c478bd9Sstevel@tonic-gate /* 19717c478bd9Sstevel@tonic-gate * kpm segment 19727c478bd9Sstevel@tonic-gate */ 19737c478bd9Sstevel@tonic-gate segmap_kpm = 0; 19747c478bd9Sstevel@tonic-gate if (kpm_desired) { 19757c478bd9Sstevel@tonic-gate kpm_init(); 19767c478bd9Sstevel@tonic-gate kpm_enable = 1; 1977a5652762Spraks vpm_enable = 1; 19787c478bd9Sstevel@tonic-gate } 19797c478bd9Sstevel@tonic-gate 19807c478bd9Sstevel@tonic-gate /* 19817c478bd9Sstevel@tonic-gate * Now create segmap segment. 19827c478bd9Sstevel@tonic-gate */ 19837c478bd9Sstevel@tonic-gate rw_enter(&kas.a_lock, RW_WRITER); 1984ae115bc7Smrj if (seg_attach(&kas, (caddr_t)segmap_start, segmapsize, segmap) < 0) { 1985ae115bc7Smrj panic("cannot attach segmap"); 19867c478bd9Sstevel@tonic-gate /*NOTREACHED*/ 19877c478bd9Sstevel@tonic-gate } 1988ae115bc7Smrj PRM_DEBUG(segmap); 19897c478bd9Sstevel@tonic-gate 19907c478bd9Sstevel@tonic-gate a.prot = PROT_READ | PROT_WRITE; 19917c478bd9Sstevel@tonic-gate a.shmsize = 0; 19927c478bd9Sstevel@tonic-gate a.nfreelist = segmapfreelists; 19937c478bd9Sstevel@tonic-gate 1994ae115bc7Smrj if (segmap_create(segmap, (caddr_t)&a) != 0) 1995ae115bc7Smrj panic("segmap_create segmap"); 19967c478bd9Sstevel@tonic-gate rw_exit(&kas.a_lock); 19977c478bd9Sstevel@tonic-gate 19987c478bd9Sstevel@tonic-gate setup_vaddr_for_ppcopy(CPU); 19997c478bd9Sstevel@tonic-gate 20007c478bd9Sstevel@tonic-gate segdev_init(); 2001843e1988Sjohnlev #if defined(__xpv) 2002843e1988Sjohnlev if (DOMAIN_IS_INITDOMAIN(xen_info)) 2003843e1988Sjohnlev #endif 2004843e1988Sjohnlev pmem_init(); 2005ae115bc7Smrj 20067c478bd9Sstevel@tonic-gate PRM_POINT("startup_vm() done"); 20077c478bd9Sstevel@tonic-gate } 20087c478bd9Sstevel@tonic-gate 2009ae115bc7Smrj /* 2010ae115bc7Smrj * Load a tod module for the non-standard tod part found on this system. 2011ae115bc7Smrj */ 2012ae115bc7Smrj static void 2013ae115bc7Smrj load_tod_module(char *todmod) 2014ae115bc7Smrj { 2015ae115bc7Smrj if (modload("tod", todmod) == -1) 2016ae115bc7Smrj halt("Can't load TOD module"); 2017ae115bc7Smrj } 2018ae115bc7Smrj 20197c478bd9Sstevel@tonic-gate static void 20207c478bd9Sstevel@tonic-gate startup_end(void) 20217c478bd9Sstevel@tonic-gate { 2022dd4eeefdSeota int i; 20237c478bd9Sstevel@tonic-gate extern void setx86isalist(void); 20247c478bd9Sstevel@tonic-gate 20257c478bd9Sstevel@tonic-gate PRM_POINT("startup_end() starting..."); 20267c478bd9Sstevel@tonic-gate 20277c478bd9Sstevel@tonic-gate /* 20287c478bd9Sstevel@tonic-gate * Perform tasks that get done after most of the VM 20297c478bd9Sstevel@tonic-gate * initialization has been done but before the clock 20307c478bd9Sstevel@tonic-gate * and other devices get started. 20317c478bd9Sstevel@tonic-gate */ 20327c478bd9Sstevel@tonic-gate kern_setup1(); 20337c478bd9Sstevel@tonic-gate 20347c478bd9Sstevel@tonic-gate /* 20357c478bd9Sstevel@tonic-gate * Perform CPC initialization for this CPU. 20367c478bd9Sstevel@tonic-gate */ 20377c478bd9Sstevel@tonic-gate kcpc_hw_init(CPU); 20387c478bd9Sstevel@tonic-gate 2039ee88d2b9Skchow #if defined(OPTERON_WORKAROUND_6323525) 2040ee88d2b9Skchow if (opteron_workaround_6323525) 2041ee88d2b9Skchow patch_workaround_6323525(); 2042ee88d2b9Skchow #endif 2043ae115bc7Smrj /* 2044ae115bc7Smrj * If needed, load TOD module now so that ddi_get_time(9F) etc. work 2045ae115bc7Smrj * (For now, "needed" is defined as set tod_module_name in /etc/system) 2046ae115bc7Smrj */ 2047ae115bc7Smrj if (tod_module_name != NULL) { 2048ae115bc7Smrj PRM_POINT("load_tod_module()"); 2049ae115bc7Smrj load_tod_module(tod_module_name); 2050ae115bc7Smrj } 2051ae115bc7Smrj 2052843e1988Sjohnlev #if defined(__xpv) 2053843e1988Sjohnlev /* 2054843e1988Sjohnlev * Forceload interposing TOD module for the hypervisor. 2055843e1988Sjohnlev */ 2056843e1988Sjohnlev PRM_POINT("load_tod_module()"); 2057843e1988Sjohnlev load_tod_module("xpvtod"); 2058843e1988Sjohnlev #endif 2059843e1988Sjohnlev 20607c478bd9Sstevel@tonic-gate /* 20617c478bd9Sstevel@tonic-gate * Configure the system. 20627c478bd9Sstevel@tonic-gate */ 20637c478bd9Sstevel@tonic-gate PRM_POINT("Calling configure()..."); 20647c478bd9Sstevel@tonic-gate configure(); /* set up devices */ 20657c478bd9Sstevel@tonic-gate PRM_POINT("configure() done"); 20667c478bd9Sstevel@tonic-gate 20677c478bd9Sstevel@tonic-gate /* 20687c478bd9Sstevel@tonic-gate * Set the isa_list string to the defined instruction sets we 20697c478bd9Sstevel@tonic-gate * support. 20707c478bd9Sstevel@tonic-gate */ 20717c478bd9Sstevel@tonic-gate setx86isalist(); 2072100b72f4Sandrei cpu_intr_alloc(CPU, NINTR_THREADS); 20737c478bd9Sstevel@tonic-gate psm_install(); 20747c478bd9Sstevel@tonic-gate 20757c478bd9Sstevel@tonic-gate /* 20767c478bd9Sstevel@tonic-gate * We're done with bootops. We don't unmap the bootstrap yet because 20777c478bd9Sstevel@tonic-gate * we're still using bootsvcs. 20787c478bd9Sstevel@tonic-gate */ 2079ae115bc7Smrj PRM_POINT("NULLing out bootops"); 2080ae115bc7Smrj *bootopsp = (struct bootops *)NULL; 20817c478bd9Sstevel@tonic-gate bootops = (struct bootops *)NULL; 20827c478bd9Sstevel@tonic-gate 2083843e1988Sjohnlev #if defined(__xpv) 2084843e1988Sjohnlev ec_init_debug_irq(); 2085843e1988Sjohnlev xs_domu_init(); 2086843e1988Sjohnlev #endif 20877c478bd9Sstevel@tonic-gate PRM_POINT("Enabling interrupts"); 20887c478bd9Sstevel@tonic-gate (*picinitf)(); 20897c478bd9Sstevel@tonic-gate sti(); 2090843e1988Sjohnlev #if defined(__xpv) 2091843e1988Sjohnlev ASSERT(CPU->cpu_m.mcpu_vcpu_info->evtchn_upcall_mask == 0); 2092843e1988Sjohnlev xen_late_startup(); 2093843e1988Sjohnlev #endif 20947c478bd9Sstevel@tonic-gate 20957c478bd9Sstevel@tonic-gate (void) add_avsoftintr((void *)&softlevel1_hdl, 1, softlevel1, 209635b1ab99Sjosephb "softlevel1", NULL, NULL); /* XXX to be moved later */ 20977c478bd9Sstevel@tonic-gate 2098dd4eeefdSeota /* 2099dd4eeefdSeota * Register these software interrupts for ddi timer. 2100dd4eeefdSeota * Software interrupts up to the level 10 are supported. 2101dd4eeefdSeota */ 2102dd4eeefdSeota for (i = DDI_IPL_1; i <= DDI_IPL_10; i++) { 2103dd4eeefdSeota char name[sizeof ("timer_softintr") + 2]; 2104dd4eeefdSeota (void) sprintf(name, "timer_softintr%02d", i); 2105dd4eeefdSeota (void) add_avsoftintr((void *)&softlevel_hdl[i-1], i, 2106dd4eeefdSeota (avfunc)timer_softintr, name, (caddr_t)(uintptr_t)i, NULL); 2107dd4eeefdSeota } 2108dd4eeefdSeota 21097125fcbdSVikram Hegde #if !defined(__xpv) 21107125fcbdSVikram Hegde if (modload("drv", "amd_iommu") < 0) { 21117125fcbdSVikram Hegde PRM_POINT("No AMD IOMMU present\n"); 211294f1124eSVikram Hegde } else if (ddi_hold_installed_driver(ddi_name_to_major( 211394f1124eSVikram Hegde "amd_iommu")) == NULL) { 21147125fcbdSVikram Hegde prom_printf("ERROR: failed to attach AMD IOMMU\n"); 21157125fcbdSVikram Hegde } 21167125fcbdSVikram Hegde #endif 21177125fcbdSVikram Hegde 21187c478bd9Sstevel@tonic-gate PRM_POINT("startup_end() done"); 21197c478bd9Sstevel@tonic-gate } 21207c478bd9Sstevel@tonic-gate 212196d008c2SKen Erickson /* 212296d008c2SKen Erickson * Don't remove the following 2 variables. They are necessary 212396d008c2SKen Erickson * for reading the hostid from the legacy file (/kernel/misc/sysinit). 212496d008c2SKen Erickson */ 21257c478bd9Sstevel@tonic-gate char *_hs1107 = hw_serial; 21267c478bd9Sstevel@tonic-gate ulong_t _bdhs34; 21277c478bd9Sstevel@tonic-gate 21287c478bd9Sstevel@tonic-gate void 21297c478bd9Sstevel@tonic-gate post_startup(void) 21307c478bd9Sstevel@tonic-gate { 21317c478bd9Sstevel@tonic-gate /* 21327c478bd9Sstevel@tonic-gate * Set the system wide, processor-specific flags to be passed 21337c478bd9Sstevel@tonic-gate * to userland via the aux vector for performance hints and 21347c478bd9Sstevel@tonic-gate * instruction set extensions. 21357c478bd9Sstevel@tonic-gate */ 21367c478bd9Sstevel@tonic-gate bind_hwcap(); 21377c478bd9Sstevel@tonic-gate 2138843e1988Sjohnlev #ifdef __xpv 2139843e1988Sjohnlev if (DOMAIN_IS_INITDOMAIN(xen_info)) 2140843e1988Sjohnlev #endif 2141843e1988Sjohnlev { 2142843e1988Sjohnlev /* 2143843e1988Sjohnlev * Load the System Management BIOS into the global ksmbios 2144843e1988Sjohnlev * handle, if an SMBIOS is present on this system. 2145843e1988Sjohnlev */ 2146843e1988Sjohnlev ksmbios = smbios_open(NULL, SMB_VERSION, ksmbios_flags, NULL); 21477c478bd9Sstevel@tonic-gate 2148843e1988Sjohnlev #if defined(__xpv) 2149843e1988Sjohnlev xpv_panic_init(); 2150843e1988Sjohnlev #else 2151843e1988Sjohnlev /* 2152843e1988Sjohnlev * Startup the memory scrubber. 2153843e1988Sjohnlev * XXPV This should be running somewhere .. 2154843e1988Sjohnlev */ 2155e4b86885SCheng Sean Ye if (!xpv_is_hvm) 2156e4b86885SCheng Sean Ye memscrub_init(); 2157843e1988Sjohnlev #endif 2158843e1988Sjohnlev } 21597c478bd9Sstevel@tonic-gate 21607aec1d6eScindi /* 21617aec1d6eScindi * Complete CPU module initialization 21627aec1d6eScindi */ 216320c794b3Sgavinm cmi_post_startup(); 21647aec1d6eScindi 21657c478bd9Sstevel@tonic-gate /* 216684ab085aSmws * Perform forceloading tasks for /etc/system. 21677c478bd9Sstevel@tonic-gate */ 216884ab085aSmws (void) mod_sysctl(SYS_FORCELOAD, NULL); 21697c478bd9Sstevel@tonic-gate 21707c478bd9Sstevel@tonic-gate /* 21717c478bd9Sstevel@tonic-gate * ON4.0: Force /proc module in until clock interrupt handle fixed 21727c478bd9Sstevel@tonic-gate * ON4.0: This must be fixed or restated in /etc/systems. 21737c478bd9Sstevel@tonic-gate */ 21747c478bd9Sstevel@tonic-gate (void) modload("fs", "procfs"); 21757c478bd9Sstevel@tonic-gate 2176c35aa225Smarx (void) i_ddi_attach_hw_nodes("pit_beep"); 2177c35aa225Smarx 21787c478bd9Sstevel@tonic-gate #if defined(__i386) 21797c478bd9Sstevel@tonic-gate /* 21807c478bd9Sstevel@tonic-gate * Check for required functional Floating Point hardware, 21817c478bd9Sstevel@tonic-gate * unless FP hardware explicitly disabled. 21827c478bd9Sstevel@tonic-gate */ 21837c478bd9Sstevel@tonic-gate if (fpu_exists && (fpu_pentium_fdivbug || fp_kind == FP_NO)) 21847c478bd9Sstevel@tonic-gate halt("No working FP hardware found"); 21857c478bd9Sstevel@tonic-gate #endif 21867c478bd9Sstevel@tonic-gate 21877c478bd9Sstevel@tonic-gate maxmem = freemem; 21887c478bd9Sstevel@tonic-gate 21897c478bd9Sstevel@tonic-gate add_cpunode2devtree(CPU->cpu_id, CPU->cpu_m.mcpu_cpi); 21907c478bd9Sstevel@tonic-gate } 21917c478bd9Sstevel@tonic-gate 21927c478bd9Sstevel@tonic-gate static int 219319397407SSherry Moore pp_in_range(page_t *pp, uint64_t low_addr, uint64_t high_addr) 21947c478bd9Sstevel@tonic-gate { 219519397407SSherry Moore return ((pp->p_pagenum >= btop(low_addr)) && 219619397407SSherry Moore (pp->p_pagenum < btopr(high_addr))); 21977c478bd9Sstevel@tonic-gate } 21987c478bd9Sstevel@tonic-gate 21997c478bd9Sstevel@tonic-gate void 22007c478bd9Sstevel@tonic-gate release_bootstrap(void) 22017c478bd9Sstevel@tonic-gate { 22027c478bd9Sstevel@tonic-gate int root_is_ramdisk; 22037c478bd9Sstevel@tonic-gate page_t *pp; 22047c478bd9Sstevel@tonic-gate extern void kobj_boot_unmountroot(void); 22057c478bd9Sstevel@tonic-gate extern dev_t rootdev; 220655d507a9SSeth Goldberg #if !defined(__xpv) 220755d507a9SSeth Goldberg pfn_t pfn; 220855d507a9SSeth Goldberg #endif 22097c478bd9Sstevel@tonic-gate 22107c478bd9Sstevel@tonic-gate /* unmount boot ramdisk and release kmem usage */ 22117c478bd9Sstevel@tonic-gate kobj_boot_unmountroot(); 22127c478bd9Sstevel@tonic-gate 22137c478bd9Sstevel@tonic-gate /* 22147c478bd9Sstevel@tonic-gate * We're finished using the boot loader so free its pages. 22157c478bd9Sstevel@tonic-gate */ 22167c478bd9Sstevel@tonic-gate PRM_POINT("Unmapping lower boot pages"); 221719397407SSherry Moore 2218ae115bc7Smrj clear_boot_mappings(0, _userlimit); 221919397407SSherry Moore 2220ae115bc7Smrj postbootkernelbase = kernelbase; 22217c478bd9Sstevel@tonic-gate 22227c478bd9Sstevel@tonic-gate /* 22237c478bd9Sstevel@tonic-gate * If root isn't on ramdisk, destroy the hardcoded 22247c478bd9Sstevel@tonic-gate * ramdisk node now and release the memory. Else, 22257c478bd9Sstevel@tonic-gate * ramdisk memory is kept in rd_pages. 22267c478bd9Sstevel@tonic-gate */ 22277c478bd9Sstevel@tonic-gate root_is_ramdisk = (getmajor(rootdev) == ddi_name_to_major("ramdisk")); 22287c478bd9Sstevel@tonic-gate if (!root_is_ramdisk) { 22297c478bd9Sstevel@tonic-gate dev_info_t *dip = ddi_find_devinfo("ramdisk", -1, 0); 22307c478bd9Sstevel@tonic-gate ASSERT(dip && ddi_get_parent(dip) == ddi_root_node()); 22317c478bd9Sstevel@tonic-gate ndi_rele_devi(dip); /* held from ddi_find_devinfo */ 22327c478bd9Sstevel@tonic-gate (void) ddi_remove_child(dip, 0); 22337c478bd9Sstevel@tonic-gate } 22347c478bd9Sstevel@tonic-gate 22357c478bd9Sstevel@tonic-gate PRM_POINT("Releasing boot pages"); 22367c478bd9Sstevel@tonic-gate while (bootpages) { 223719397407SSherry Moore extern uint64_t ramdisk_start, ramdisk_end; 22387c478bd9Sstevel@tonic-gate pp = bootpages; 22397c478bd9Sstevel@tonic-gate bootpages = pp->p_next; 224019397407SSherry Moore 224119397407SSherry Moore 224219397407SSherry Moore /* Keep pages for the lower 64K */ 224319397407SSherry Moore if (pp_in_range(pp, 0, 0x40000)) { 224419397407SSherry Moore pp->p_next = lower_pages; 224519397407SSherry Moore lower_pages = pp; 224619397407SSherry Moore lower_pages_count++; 224719397407SSherry Moore continue; 224819397407SSherry Moore } 224919397407SSherry Moore 225019397407SSherry Moore 225119397407SSherry Moore if (root_is_ramdisk && pp_in_range(pp, ramdisk_start, 225219397407SSherry Moore ramdisk_end)) { 22537c478bd9Sstevel@tonic-gate pp->p_next = rd_pages; 22547c478bd9Sstevel@tonic-gate rd_pages = pp; 22557c478bd9Sstevel@tonic-gate continue; 22567c478bd9Sstevel@tonic-gate } 22577c478bd9Sstevel@tonic-gate pp->p_next = (struct page *)0; 2258ae115bc7Smrj pp->p_prev = (struct page *)0; 2259ae115bc7Smrj PP_CLRBOOTPAGES(pp); 22607c478bd9Sstevel@tonic-gate page_free(pp, 1); 22617c478bd9Sstevel@tonic-gate } 2262ae115bc7Smrj PRM_POINT("Boot pages released"); 22637c478bd9Sstevel@tonic-gate 2264843e1988Sjohnlev #if !defined(__xpv) 2265843e1988Sjohnlev /* XXPV -- note this following bunch of code needs to be revisited in Xen 3.0 */ 22667c478bd9Sstevel@tonic-gate /* 226755d507a9SSeth Goldberg * Find 1 page below 1 MB so that other processors can boot up or 226855d507a9SSeth Goldberg * so that any processor can resume. 22697c478bd9Sstevel@tonic-gate * Make sure it has a kernel VA as well as a 1:1 mapping. 22707c478bd9Sstevel@tonic-gate * We should have just free'd one up. 22717c478bd9Sstevel@tonic-gate */ 227219397407SSherry Moore 227319397407SSherry Moore /* 227419397407SSherry Moore * 0x10 pages is 64K. Leave the bottom 64K alone 227519397407SSherry Moore * for BIOS. 227619397407SSherry Moore */ 227719397407SSherry Moore for (pfn = 0x10; pfn < btop(1*1024*1024); pfn++) { 227855d507a9SSeth Goldberg if (page_numtopp_alloc(pfn) == NULL) 227955d507a9SSeth Goldberg continue; 228055d507a9SSeth Goldberg rm_platter_va = i86devmap(pfn, 1, 228155d507a9SSeth Goldberg PROT_READ | PROT_WRITE | PROT_EXEC); 228255d507a9SSeth Goldberg rm_platter_pa = ptob(pfn); 228355d507a9SSeth Goldberg hat_devload(kas.a_hat, 228455d507a9SSeth Goldberg (caddr_t)(uintptr_t)rm_platter_pa, MMU_PAGESIZE, 228555d507a9SSeth Goldberg pfn, PROT_READ | PROT_WRITE | PROT_EXEC, 228655d507a9SSeth Goldberg HAT_LOAD_NOCONSIST); 228755d507a9SSeth Goldberg break; 22887c478bd9Sstevel@tonic-gate } 228955d507a9SSeth Goldberg if (pfn == btop(1*1024*1024) && use_mp) 229055d507a9SSeth Goldberg panic("No page below 1M available for starting " 229155d507a9SSeth Goldberg "other processors or for resuming from system-suspend"); 2292843e1988Sjohnlev #endif /* !__xpv */ 22937c478bd9Sstevel@tonic-gate } 22947c478bd9Sstevel@tonic-gate 22957c478bd9Sstevel@tonic-gate /* 22967c478bd9Sstevel@tonic-gate * Initialize the platform-specific parts of a page_t. 22977c478bd9Sstevel@tonic-gate */ 22987c478bd9Sstevel@tonic-gate void 22997c478bd9Sstevel@tonic-gate add_physmem_cb(page_t *pp, pfn_t pnum) 23007c478bd9Sstevel@tonic-gate { 23017c478bd9Sstevel@tonic-gate pp->p_pagenum = pnum; 23027c478bd9Sstevel@tonic-gate pp->p_mapping = NULL; 23037c478bd9Sstevel@tonic-gate pp->p_embed = 0; 23047c478bd9Sstevel@tonic-gate pp->p_share = 0; 23057c478bd9Sstevel@tonic-gate pp->p_mlentry = 0; 23067c478bd9Sstevel@tonic-gate } 23077c478bd9Sstevel@tonic-gate 23087c478bd9Sstevel@tonic-gate /* 23097c478bd9Sstevel@tonic-gate * kphysm_init() initializes physical memory. 23107c478bd9Sstevel@tonic-gate */ 23117c478bd9Sstevel@tonic-gate static pgcnt_t 23127c478bd9Sstevel@tonic-gate kphysm_init( 2313ae115bc7Smrj page_t *pp, 23147c478bd9Sstevel@tonic-gate pgcnt_t npages) 23157c478bd9Sstevel@tonic-gate { 23167c478bd9Sstevel@tonic-gate struct memlist *pmem; 23177c478bd9Sstevel@tonic-gate struct memseg *cur_memseg; 23187c478bd9Sstevel@tonic-gate pfn_t base_pfn; 23197c478bd9Sstevel@tonic-gate pgcnt_t num; 23207c478bd9Sstevel@tonic-gate pgcnt_t pages_done = 0; 23217c478bd9Sstevel@tonic-gate uint64_t addr; 23227c478bd9Sstevel@tonic-gate uint64_t size; 23237c478bd9Sstevel@tonic-gate extern pfn_t ddiphysmin; 23247c478bd9Sstevel@tonic-gate 23257c478bd9Sstevel@tonic-gate ASSERT(page_hash != NULL && page_hashsz != 0); 23267c478bd9Sstevel@tonic-gate 2327ae115bc7Smrj cur_memseg = memseg_base; 23287c478bd9Sstevel@tonic-gate for (pmem = phys_avail; pmem && npages; pmem = pmem->next) { 23297c478bd9Sstevel@tonic-gate /* 23307c478bd9Sstevel@tonic-gate * In a 32 bit kernel can't use higher memory if we're 23317c478bd9Sstevel@tonic-gate * not booting in PAE mode. This check takes care of that. 23327c478bd9Sstevel@tonic-gate */ 23337c478bd9Sstevel@tonic-gate addr = pmem->address; 23347c478bd9Sstevel@tonic-gate size = pmem->size; 23357c478bd9Sstevel@tonic-gate if (btop(addr) > physmax) 23367c478bd9Sstevel@tonic-gate continue; 23377c478bd9Sstevel@tonic-gate 23387c478bd9Sstevel@tonic-gate /* 23397c478bd9Sstevel@tonic-gate * align addr and size - they may not be at page boundaries 23407c478bd9Sstevel@tonic-gate */ 23417c478bd9Sstevel@tonic-gate if ((addr & MMU_PAGEOFFSET) != 0) { 23427c478bd9Sstevel@tonic-gate addr += MMU_PAGEOFFSET; 23437c478bd9Sstevel@tonic-gate addr &= ~(uint64_t)MMU_PAGEOFFSET; 23447c478bd9Sstevel@tonic-gate size -= addr - pmem->address; 23457c478bd9Sstevel@tonic-gate } 23467c478bd9Sstevel@tonic-gate 23476bb54764Skchow /* only process pages below or equal to physmax */ 23486bb54764Skchow if ((btop(addr + size) - 1) > physmax) 23496bb54764Skchow size = ptob(physmax - btop(addr) + 1); 23507c478bd9Sstevel@tonic-gate 23517c478bd9Sstevel@tonic-gate num = btop(size); 23527c478bd9Sstevel@tonic-gate if (num == 0) 23537c478bd9Sstevel@tonic-gate continue; 23547c478bd9Sstevel@tonic-gate 23557c478bd9Sstevel@tonic-gate if (num > npages) 23567c478bd9Sstevel@tonic-gate num = npages; 23577c478bd9Sstevel@tonic-gate 23587c478bd9Sstevel@tonic-gate npages -= num; 23597c478bd9Sstevel@tonic-gate pages_done += num; 23607c478bd9Sstevel@tonic-gate base_pfn = btop(addr); 23617c478bd9Sstevel@tonic-gate 23627c478bd9Sstevel@tonic-gate if (prom_debug) 23637c478bd9Sstevel@tonic-gate prom_printf("MEMSEG addr=0x%" PRIx64 23647c478bd9Sstevel@tonic-gate " pgs=0x%lx pfn 0x%lx-0x%lx\n", 23657c478bd9Sstevel@tonic-gate addr, num, base_pfn, base_pfn + num); 23667c478bd9Sstevel@tonic-gate 23677c478bd9Sstevel@tonic-gate /* 2368ae115bc7Smrj * Ignore pages below ddiphysmin to simplify ddi memory 23697c478bd9Sstevel@tonic-gate * allocation with non-zero addr_lo requests. 23707c478bd9Sstevel@tonic-gate */ 23717c478bd9Sstevel@tonic-gate if (base_pfn < ddiphysmin) { 2372ae115bc7Smrj if (base_pfn + num <= ddiphysmin) 23737c478bd9Sstevel@tonic-gate continue; 23747c478bd9Sstevel@tonic-gate pp += (ddiphysmin - base_pfn); 23757c478bd9Sstevel@tonic-gate num -= (ddiphysmin - base_pfn); 23767c478bd9Sstevel@tonic-gate base_pfn = ddiphysmin; 23777c478bd9Sstevel@tonic-gate } 2378ae115bc7Smrj 23797c478bd9Sstevel@tonic-gate /* 23807c478bd9Sstevel@tonic-gate * Build the memsegs entry 23817c478bd9Sstevel@tonic-gate */ 23827c478bd9Sstevel@tonic-gate cur_memseg->pages = pp; 23837c478bd9Sstevel@tonic-gate cur_memseg->epages = pp + num; 23847c478bd9Sstevel@tonic-gate cur_memseg->pages_base = base_pfn; 23857c478bd9Sstevel@tonic-gate cur_memseg->pages_end = base_pfn + num; 23867c478bd9Sstevel@tonic-gate 23877c478bd9Sstevel@tonic-gate /* 2388ae115bc7Smrj * Insert into memseg list in decreasing pfn range order. 23897c478bd9Sstevel@tonic-gate * Low memory is typically more fragmented such that this 23907c478bd9Sstevel@tonic-gate * ordering keeps the larger ranges at the front of the list 23917c478bd9Sstevel@tonic-gate * for code that searches memseg. 2392ae115bc7Smrj * This ASSERTS that the memsegs coming in from boot are in 2393ae115bc7Smrj * increasing physical address order and not contiguous. 23947c478bd9Sstevel@tonic-gate */ 2395ae115bc7Smrj if (memsegs != NULL) { 2396ae115bc7Smrj ASSERT(cur_memseg->pages_base >= memsegs->pages_end); 2397ae115bc7Smrj cur_memseg->next = memsegs; 23987c478bd9Sstevel@tonic-gate } 2399ae115bc7Smrj memsegs = cur_memseg; 24007c478bd9Sstevel@tonic-gate 24017c478bd9Sstevel@tonic-gate /* 24027c478bd9Sstevel@tonic-gate * add_physmem() initializes the PSM part of the page 24037c478bd9Sstevel@tonic-gate * struct by calling the PSM back with add_physmem_cb(). 24047c478bd9Sstevel@tonic-gate * In addition it coalesces pages into larger pages as 24057c478bd9Sstevel@tonic-gate * it initializes them. 24067c478bd9Sstevel@tonic-gate */ 24077c478bd9Sstevel@tonic-gate add_physmem(pp, num, base_pfn); 24087c478bd9Sstevel@tonic-gate cur_memseg++; 24097c478bd9Sstevel@tonic-gate availrmem_initial += num; 24107c478bd9Sstevel@tonic-gate availrmem += num; 24117c478bd9Sstevel@tonic-gate 2412ae115bc7Smrj pp += num; 24137c478bd9Sstevel@tonic-gate } 24147c478bd9Sstevel@tonic-gate 24157c478bd9Sstevel@tonic-gate PRM_DEBUG(availrmem_initial); 24167c478bd9Sstevel@tonic-gate PRM_DEBUG(availrmem); 24177c478bd9Sstevel@tonic-gate PRM_DEBUG(freemem); 24187c478bd9Sstevel@tonic-gate build_pfn_hash(); 24197c478bd9Sstevel@tonic-gate return (pages_done); 24207c478bd9Sstevel@tonic-gate } 24217c478bd9Sstevel@tonic-gate 24227c478bd9Sstevel@tonic-gate /* 24237c478bd9Sstevel@tonic-gate * Kernel VM initialization. 24247c478bd9Sstevel@tonic-gate */ 24257c478bd9Sstevel@tonic-gate static void 24267c478bd9Sstevel@tonic-gate kvm_init(void) 24277c478bd9Sstevel@tonic-gate { 24287c478bd9Sstevel@tonic-gate ASSERT((((uintptr_t)s_text) & MMU_PAGEOFFSET) == 0); 24297c478bd9Sstevel@tonic-gate 24307c478bd9Sstevel@tonic-gate /* 24317c478bd9Sstevel@tonic-gate * Put the kernel segments in kernel address space. 24327c478bd9Sstevel@tonic-gate */ 24337c478bd9Sstevel@tonic-gate rw_enter(&kas.a_lock, RW_WRITER); 24347c478bd9Sstevel@tonic-gate as_avlinit(&kas); 24357c478bd9Sstevel@tonic-gate 24367c478bd9Sstevel@tonic-gate (void) seg_attach(&kas, s_text, e_moddata - s_text, &ktextseg); 24377c478bd9Sstevel@tonic-gate (void) segkmem_create(&ktextseg); 24387c478bd9Sstevel@tonic-gate 24397c478bd9Sstevel@tonic-gate (void) seg_attach(&kas, (caddr_t)valloc_base, valloc_sz, &kvalloc); 24407c478bd9Sstevel@tonic-gate (void) segkmem_create(&kvalloc); 24417c478bd9Sstevel@tonic-gate 244235b1ab99Sjosephb (void) seg_attach(&kas, kernelheap, 244335b1ab99Sjosephb ekernelheap - kernelheap, &kvseg); 24447c478bd9Sstevel@tonic-gate (void) segkmem_create(&kvseg); 24457c478bd9Sstevel@tonic-gate 2446ae115bc7Smrj if (core_size > 0) { 2447ae115bc7Smrj PRM_POINT("attaching kvseg_core"); 2448ae115bc7Smrj (void) seg_attach(&kas, (caddr_t)core_base, core_size, 2449ae115bc7Smrj &kvseg_core); 2450ae115bc7Smrj (void) segkmem_create(&kvseg_core); 2451ae115bc7Smrj } 2452ad23a2dbSjohansen 2453ae115bc7Smrj if (segziosize > 0) { 2454ae115bc7Smrj PRM_POINT("attaching segzio"); 2455ad23a2dbSjohansen (void) seg_attach(&kas, segzio_base, mmu_ptob(segziosize), 2456ad23a2dbSjohansen &kzioseg); 2457ad23a2dbSjohansen (void) segkmem_zio_create(&kzioseg); 2458ad23a2dbSjohansen 2459ad23a2dbSjohansen /* create zio area covering new segment */ 2460ad23a2dbSjohansen segkmem_zio_init(segzio_base, mmu_ptob(segziosize)); 2461ad23a2dbSjohansen } 24627c478bd9Sstevel@tonic-gate 2463ae115bc7Smrj (void) seg_attach(&kas, kdi_segdebugbase, kdi_segdebugsize, &kdebugseg); 24647c478bd9Sstevel@tonic-gate (void) segkmem_create(&kdebugseg); 24657c478bd9Sstevel@tonic-gate 24667c478bd9Sstevel@tonic-gate rw_exit(&kas.a_lock); 24677c478bd9Sstevel@tonic-gate 24687c478bd9Sstevel@tonic-gate /* 24697c478bd9Sstevel@tonic-gate * Ensure that the red zone at kernelbase is never accessible. 24707c478bd9Sstevel@tonic-gate */ 2471ae115bc7Smrj PRM_POINT("protecting redzone"); 24727c478bd9Sstevel@tonic-gate (void) as_setprot(&kas, (caddr_t)kernelbase, KERNEL_REDZONE_SIZE, 0); 24737c478bd9Sstevel@tonic-gate 24747c478bd9Sstevel@tonic-gate /* 24757c478bd9Sstevel@tonic-gate * Make the text writable so that it can be hot patched by DTrace. 24767c478bd9Sstevel@tonic-gate */ 24777c478bd9Sstevel@tonic-gate (void) as_setprot(&kas, s_text, e_modtext - s_text, 24787c478bd9Sstevel@tonic-gate PROT_READ | PROT_WRITE | PROT_EXEC); 24797c478bd9Sstevel@tonic-gate 24807c478bd9Sstevel@tonic-gate /* 24817c478bd9Sstevel@tonic-gate * Make data writable until end. 24827c478bd9Sstevel@tonic-gate */ 24837c478bd9Sstevel@tonic-gate (void) as_setprot(&kas, s_data, e_moddata - s_data, 24847c478bd9Sstevel@tonic-gate PROT_READ | PROT_WRITE | PROT_EXEC); 24857c478bd9Sstevel@tonic-gate } 24867c478bd9Sstevel@tonic-gate 2487843e1988Sjohnlev #ifndef __xpv 24887c478bd9Sstevel@tonic-gate /* 24891d03c31eSjohnlev * Solaris adds an entry for Write Combining caching to the PAT 24907c478bd9Sstevel@tonic-gate */ 24911d03c31eSjohnlev static uint64_t pat_attr_reg = PAT_DEFAULT_ATTRIBUTE; 24927c478bd9Sstevel@tonic-gate 24937c478bd9Sstevel@tonic-gate void 24941d03c31eSjohnlev pat_sync(void) 24957c478bd9Sstevel@tonic-gate { 24961d03c31eSjohnlev ulong_t cr0, cr0_orig, cr4; 24977c478bd9Sstevel@tonic-gate 24981d03c31eSjohnlev if (!(x86_feature & X86_PAT)) 24997c478bd9Sstevel@tonic-gate return; 25001d03c31eSjohnlev cr0_orig = cr0 = getcr0(); 25011d03c31eSjohnlev cr4 = getcr4(); 25027c478bd9Sstevel@tonic-gate 25031d03c31eSjohnlev /* disable caching and flush all caches and TLBs */ 25041d03c31eSjohnlev cr0 |= CR0_CD; 25051d03c31eSjohnlev cr0 &= ~CR0_NW; 25061d03c31eSjohnlev setcr0(cr0); 25071d03c31eSjohnlev invalidate_cache(); 25081d03c31eSjohnlev if (cr4 & CR4_PGE) { 25091d03c31eSjohnlev setcr4(cr4 & ~(ulong_t)CR4_PGE); 25101d03c31eSjohnlev setcr4(cr4); 25111d03c31eSjohnlev } else { 25121d03c31eSjohnlev reload_cr3(); 25137c478bd9Sstevel@tonic-gate } 25147c478bd9Sstevel@tonic-gate 25151d03c31eSjohnlev /* add our entry to the PAT */ 25161d03c31eSjohnlev wrmsr(REG_PAT, pat_attr_reg); 25177c478bd9Sstevel@tonic-gate 25181d03c31eSjohnlev /* flush TLBs and cache again, then reenable cr0 caching */ 25191d03c31eSjohnlev if (cr4 & CR4_PGE) { 25201d03c31eSjohnlev setcr4(cr4 & ~(ulong_t)CR4_PGE); 25211d03c31eSjohnlev setcr4(cr4); 25221d03c31eSjohnlev } else { 25231d03c31eSjohnlev reload_cr3(); 25247c478bd9Sstevel@tonic-gate } 25257c478bd9Sstevel@tonic-gate invalidate_cache(); 25267c478bd9Sstevel@tonic-gate setcr0(cr0_orig); 25277c478bd9Sstevel@tonic-gate } 25287c478bd9Sstevel@tonic-gate 25291d03c31eSjohnlev #endif /* !__xpv */ 25307c478bd9Sstevel@tonic-gate 2531ed5289f9SKen Erickson #if defined(_SOFT_HOSTID) 2532ed5289f9SKen Erickson /* 2533ed5289f9SKen Erickson * On platforms that do not have a hardware serial number, attempt 2534ed5289f9SKen Erickson * to set one based on the contents of /etc/hostid. If this file does 2535ed5289f9SKen Erickson * not exist, assume that we are to generate a new hostid and set 2536ed5289f9SKen Erickson * it in the kernel, for subsequent saving by a userland process 2537ed5289f9SKen Erickson * once the system is up and the root filesystem is mounted r/w. 2538ed5289f9SKen Erickson * 253996d008c2SKen Erickson * In order to gracefully support upgrade on OpenSolaris, if 254096d008c2SKen Erickson * /etc/hostid does not exist, we will attempt to get a serial number 254196d008c2SKen Erickson * using the legacy method (/kernel/misc/sysinit). 254296d008c2SKen Erickson * 2543ed5289f9SKen Erickson * In an attempt to make the hostid less prone to abuse 2544ed5289f9SKen Erickson * (for license circumvention, etc), we store it in /etc/hostid 2545ed5289f9SKen Erickson * in rot47 format. 2546ed5289f9SKen Erickson */ 2547ed5289f9SKen Erickson extern volatile unsigned long tenmicrodata; 254896d008c2SKen Erickson static int atoi(char *); 2549ed5289f9SKen Erickson 2550ed5289f9SKen Erickson static int32_t 2551ed5289f9SKen Erickson set_soft_hostid(void) 2552ed5289f9SKen Erickson { 2553ed5289f9SKen Erickson struct _buf *file; 2554ed5289f9SKen Erickson char tokbuf[MAXNAMELEN]; 2555ed5289f9SKen Erickson token_t token; 2556ed5289f9SKen Erickson int done = 0; 2557ed5289f9SKen Erickson u_longlong_t tmp; 255896d008c2SKen Erickson int i; 2559*5679c89fSjv int32_t hostid = (int32_t)HW_INVALID_HOSTID; 2560ed5289f9SKen Erickson unsigned char *c; 2561ed5289f9SKen Erickson hrtime_t tsc; 2562ed5289f9SKen Erickson 2563ed5289f9SKen Erickson /* 2564ed5289f9SKen Erickson * If /etc/hostid file not found, we'd like to get a pseudo 2565ed5289f9SKen Erickson * random number to use at the hostid. A nice way to do this 2566ed5289f9SKen Erickson * is to read the real time clock. To remain xen-compatible, 2567ed5289f9SKen Erickson * we can't poke the real hardware, so we use tsc_read() to 2568ed5289f9SKen Erickson * read the real time clock. However, there is an ominous 2569ed5289f9SKen Erickson * warning in tsc_read that says it can return zero, so we 2570ed5289f9SKen Erickson * deal with that possibility by falling back to using the 2571ed5289f9SKen Erickson * (hopefully random enough) value in tenmicrodata. 2572ed5289f9SKen Erickson */ 2573ed5289f9SKen Erickson 2574ed5289f9SKen Erickson if ((file = kobj_open_file(hostid_file)) == (struct _buf *)-1) { 257596d008c2SKen Erickson /* 257696d008c2SKen Erickson * hostid file not found - try to load sysinit module 257796d008c2SKen Erickson * and see if it has a nonzero hostid value...use that 257896d008c2SKen Erickson * instead of generating a new hostid here if so. 257996d008c2SKen Erickson */ 258096d008c2SKen Erickson if ((i = modload("misc", "sysinit")) != -1) { 258196d008c2SKen Erickson if (strlen(hw_serial) > 0) 258296d008c2SKen Erickson hostid = (int32_t)atoi(hw_serial); 258396d008c2SKen Erickson (void) modunload(i); 258496d008c2SKen Erickson } 2585*5679c89fSjv if (hostid == HW_INVALID_HOSTID) { 258696d008c2SKen Erickson tsc = tsc_read(); 258796d008c2SKen Erickson if (tsc == 0) /* tsc_read can return zero sometimes */ 258896d008c2SKen Erickson hostid = (int32_t)tenmicrodata & 0x0CFFFFF; 258996d008c2SKen Erickson else 259096d008c2SKen Erickson hostid = (int32_t)tsc & 0x0CFFFFF; 259196d008c2SKen Erickson } 2592ed5289f9SKen Erickson } else { 2593ed5289f9SKen Erickson /* hostid file found */ 2594ed5289f9SKen Erickson while (!done) { 2595ed5289f9SKen Erickson token = kobj_lex(file, tokbuf, sizeof (tokbuf)); 2596ed5289f9SKen Erickson 2597ed5289f9SKen Erickson switch (token) { 2598ed5289f9SKen Erickson case POUND: 2599ed5289f9SKen Erickson /* 2600ed5289f9SKen Erickson * skip comments 2601ed5289f9SKen Erickson */ 2602ed5289f9SKen Erickson kobj_find_eol(file); 2603ed5289f9SKen Erickson break; 2604ed5289f9SKen Erickson case STRING: 2605ed5289f9SKen Erickson /* 2606ed5289f9SKen Erickson * un-rot47 - obviously this 2607ed5289f9SKen Erickson * nonsense is ascii-specific 2608ed5289f9SKen Erickson */ 2609ed5289f9SKen Erickson for (c = (unsigned char *)tokbuf; 2610ed5289f9SKen Erickson *c != '\0'; c++) { 2611ed5289f9SKen Erickson *c += 47; 2612ed5289f9SKen Erickson if (*c > '~') 2613ed5289f9SKen Erickson *c -= 94; 2614ed5289f9SKen Erickson else if (*c < '!') 2615ed5289f9SKen Erickson *c += 94; 2616ed5289f9SKen Erickson } 2617ed5289f9SKen Erickson /* 2618ed5289f9SKen Erickson * now we should have a real number 2619ed5289f9SKen Erickson */ 2620ed5289f9SKen Erickson 2621ed5289f9SKen Erickson if (kobj_getvalue(tokbuf, &tmp) != 0) 2622ed5289f9SKen Erickson kobj_file_err(CE_WARN, file, 2623ed5289f9SKen Erickson "Bad value %s for hostid", 2624ed5289f9SKen Erickson tokbuf); 2625ed5289f9SKen Erickson else 2626ed5289f9SKen Erickson hostid = (int32_t)tmp; 2627ed5289f9SKen Erickson 2628ed5289f9SKen Erickson break; 2629ed5289f9SKen Erickson case EOF: 2630ed5289f9SKen Erickson done = 1; 2631ed5289f9SKen Erickson /* FALLTHROUGH */ 2632ed5289f9SKen Erickson case NEWLINE: 2633ed5289f9SKen Erickson kobj_newline(file); 2634ed5289f9SKen Erickson break; 2635ed5289f9SKen Erickson default: 2636ed5289f9SKen Erickson break; 2637ed5289f9SKen Erickson 2638ed5289f9SKen Erickson } 2639ed5289f9SKen Erickson } 2640*5679c89fSjv if (hostid == HW_INVALID_HOSTID) /* didn't find a hostid */ 2641ed5289f9SKen Erickson kobj_file_err(CE_WARN, file, 2642ed5289f9SKen Erickson "hostid missing or corrupt"); 2643ed5289f9SKen Erickson 2644ed5289f9SKen Erickson kobj_close_file(file); 2645ed5289f9SKen Erickson } 2646ed5289f9SKen Erickson /* 2647ed5289f9SKen Erickson * hostid is now the value read from /etc/hostid, or the 2648*5679c89fSjv * new hostid we generated in this routine or HW_INVALID_HOSTID if not 2649*5679c89fSjv * set. 2650ed5289f9SKen Erickson */ 2651ed5289f9SKen Erickson return (hostid); 2652ed5289f9SKen Erickson } 265396d008c2SKen Erickson 265496d008c2SKen Erickson static int 265596d008c2SKen Erickson atoi(char *p) 265696d008c2SKen Erickson { 265796d008c2SKen Erickson int i = 0; 265896d008c2SKen Erickson 265996d008c2SKen Erickson while (*p != '\0') 266096d008c2SKen Erickson i = 10 * i + (*p++ - '0'); 266196d008c2SKen Erickson 266296d008c2SKen Erickson return (i); 266396d008c2SKen Erickson } 266496d008c2SKen Erickson 2665ed5289f9SKen Erickson #endif /* _SOFT_HOSTID */ 2666ed5289f9SKen Erickson 26677c478bd9Sstevel@tonic-gate void 266845916cd2Sjpk get_system_configuration(void) 26697c478bd9Sstevel@tonic-gate { 26707c478bd9Sstevel@tonic-gate char prop[32]; 26717c478bd9Sstevel@tonic-gate u_longlong_t nodes_ll, cpus_pernode_ll, lvalue; 26727c478bd9Sstevel@tonic-gate 267335b1ab99Sjosephb if (BOP_GETPROPLEN(bootops, "nodes") > sizeof (prop) || 267435b1ab99Sjosephb BOP_GETPROP(bootops, "nodes", prop) < 0 || 267535b1ab99Sjosephb kobj_getvalue(prop, &nodes_ll) == -1 || 267635b1ab99Sjosephb nodes_ll > MAXNODES || 267735b1ab99Sjosephb BOP_GETPROPLEN(bootops, "cpus_pernode") > sizeof (prop) || 267835b1ab99Sjosephb BOP_GETPROP(bootops, "cpus_pernode", prop) < 0 || 267935b1ab99Sjosephb kobj_getvalue(prop, &cpus_pernode_ll) == -1) { 26807c478bd9Sstevel@tonic-gate system_hardware.hd_nodes = 1; 26817c478bd9Sstevel@tonic-gate system_hardware.hd_cpus_per_node = 0; 26827c478bd9Sstevel@tonic-gate } else { 26837c478bd9Sstevel@tonic-gate system_hardware.hd_nodes = (int)nodes_ll; 26847c478bd9Sstevel@tonic-gate system_hardware.hd_cpus_per_node = (int)cpus_pernode_ll; 26857c478bd9Sstevel@tonic-gate } 268635b1ab99Sjosephb 268735b1ab99Sjosephb if (BOP_GETPROPLEN(bootops, "kernelbase") > sizeof (prop) || 268835b1ab99Sjosephb BOP_GETPROP(bootops, "kernelbase", prop) < 0 || 268935b1ab99Sjosephb kobj_getvalue(prop, &lvalue) == -1) 269035b1ab99Sjosephb eprom_kernelbase = NULL; 26917c478bd9Sstevel@tonic-gate else 269235b1ab99Sjosephb eprom_kernelbase = (uintptr_t)lvalue; 26937c478bd9Sstevel@tonic-gate 269435b1ab99Sjosephb if (BOP_GETPROPLEN(bootops, "segmapsize") > sizeof (prop) || 269535b1ab99Sjosephb BOP_GETPROP(bootops, "segmapsize", prop) < 0 || 269635b1ab99Sjosephb kobj_getvalue(prop, &lvalue) == -1) 26977c478bd9Sstevel@tonic-gate segmapsize = SEGMAPDEFAULT; 269835b1ab99Sjosephb else 26997c478bd9Sstevel@tonic-gate segmapsize = (uintptr_t)lvalue; 27007c478bd9Sstevel@tonic-gate 270135b1ab99Sjosephb if (BOP_GETPROPLEN(bootops, "segmapfreelists") > sizeof (prop) || 270235b1ab99Sjosephb BOP_GETPROP(bootops, "segmapfreelists", prop) < 0 || 270335b1ab99Sjosephb kobj_getvalue(prop, &lvalue) == -1) 27047c478bd9Sstevel@tonic-gate segmapfreelists = 0; /* use segmap driver default */ 270535b1ab99Sjosephb else 27067c478bd9Sstevel@tonic-gate segmapfreelists = (int)lvalue; 27074944b02eSkchow 2708aac11643Sjosephb /* physmem used to be here, but moved much earlier to fakebop.c */ 27097c478bd9Sstevel@tonic-gate } 27107c478bd9Sstevel@tonic-gate 27117c478bd9Sstevel@tonic-gate /* 27127c478bd9Sstevel@tonic-gate * Add to a memory list. 27137c478bd9Sstevel@tonic-gate * start = start of new memory segment 27147c478bd9Sstevel@tonic-gate * len = length of new memory segment in bytes 27157c478bd9Sstevel@tonic-gate * new = pointer to a new struct memlist 27167c478bd9Sstevel@tonic-gate * memlistp = memory list to which to add segment. 27177c478bd9Sstevel@tonic-gate */ 2718ae115bc7Smrj void 27197c478bd9Sstevel@tonic-gate memlist_add( 27207c478bd9Sstevel@tonic-gate uint64_t start, 27217c478bd9Sstevel@tonic-gate uint64_t len, 27227c478bd9Sstevel@tonic-gate struct memlist *new, 27237c478bd9Sstevel@tonic-gate struct memlist **memlistp) 27247c478bd9Sstevel@tonic-gate { 27257c478bd9Sstevel@tonic-gate struct memlist *cur; 27267c478bd9Sstevel@tonic-gate uint64_t end = start + len; 27277c478bd9Sstevel@tonic-gate 27287c478bd9Sstevel@tonic-gate new->address = start; 27297c478bd9Sstevel@tonic-gate new->size = len; 27307c478bd9Sstevel@tonic-gate 27317c478bd9Sstevel@tonic-gate cur = *memlistp; 27327c478bd9Sstevel@tonic-gate 27337c478bd9Sstevel@tonic-gate while (cur) { 27347c478bd9Sstevel@tonic-gate if (cur->address >= end) { 27357c478bd9Sstevel@tonic-gate new->next = cur; 27367c478bd9Sstevel@tonic-gate *memlistp = new; 27377c478bd9Sstevel@tonic-gate new->prev = cur->prev; 27387c478bd9Sstevel@tonic-gate cur->prev = new; 27397c478bd9Sstevel@tonic-gate return; 27407c478bd9Sstevel@tonic-gate } 27417c478bd9Sstevel@tonic-gate ASSERT(cur->address + cur->size <= start); 27427c478bd9Sstevel@tonic-gate if (cur->next == NULL) { 27437c478bd9Sstevel@tonic-gate cur->next = new; 27447c478bd9Sstevel@tonic-gate new->prev = cur; 27457c478bd9Sstevel@tonic-gate new->next = NULL; 27467c478bd9Sstevel@tonic-gate return; 27477c478bd9Sstevel@tonic-gate } 27487c478bd9Sstevel@tonic-gate memlistp = &cur->next; 27497c478bd9Sstevel@tonic-gate cur = cur->next; 27507c478bd9Sstevel@tonic-gate } 27517c478bd9Sstevel@tonic-gate } 27527c478bd9Sstevel@tonic-gate 27537c478bd9Sstevel@tonic-gate void 27547c478bd9Sstevel@tonic-gate kobj_vmem_init(vmem_t **text_arena, vmem_t **data_arena) 27557c478bd9Sstevel@tonic-gate { 27567c478bd9Sstevel@tonic-gate size_t tsize = e_modtext - modtext; 27577c478bd9Sstevel@tonic-gate size_t dsize = e_moddata - moddata; 27587c478bd9Sstevel@tonic-gate 27597c478bd9Sstevel@tonic-gate *text_arena = vmem_create("module_text", tsize ? modtext : NULL, tsize, 27607c478bd9Sstevel@tonic-gate 1, segkmem_alloc, segkmem_free, heaptext_arena, 0, VM_SLEEP); 27617c478bd9Sstevel@tonic-gate *data_arena = vmem_create("module_data", dsize ? moddata : NULL, dsize, 27627c478bd9Sstevel@tonic-gate 1, segkmem_alloc, segkmem_free, heap32_arena, 0, VM_SLEEP); 27637c478bd9Sstevel@tonic-gate } 27647c478bd9Sstevel@tonic-gate 27657c478bd9Sstevel@tonic-gate caddr_t 27667c478bd9Sstevel@tonic-gate kobj_text_alloc(vmem_t *arena, size_t size) 27677c478bd9Sstevel@tonic-gate { 27687c478bd9Sstevel@tonic-gate return (vmem_alloc(arena, size, VM_SLEEP | VM_BESTFIT)); 27697c478bd9Sstevel@tonic-gate } 27707c478bd9Sstevel@tonic-gate 27717c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 27727c478bd9Sstevel@tonic-gate caddr_t 27737c478bd9Sstevel@tonic-gate kobj_texthole_alloc(caddr_t addr, size_t size) 27747c478bd9Sstevel@tonic-gate { 27757c478bd9Sstevel@tonic-gate panic("unexpected call to kobj_texthole_alloc()"); 27767c478bd9Sstevel@tonic-gate /*NOTREACHED*/ 27777c478bd9Sstevel@tonic-gate return (0); 27787c478bd9Sstevel@tonic-gate } 27797c478bd9Sstevel@tonic-gate 27807c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 27817c478bd9Sstevel@tonic-gate void 27827c478bd9Sstevel@tonic-gate kobj_texthole_free(caddr_t addr, size_t size) 27837c478bd9Sstevel@tonic-gate { 27847c478bd9Sstevel@tonic-gate panic("unexpected call to kobj_texthole_free()"); 27857c478bd9Sstevel@tonic-gate } 27867c478bd9Sstevel@tonic-gate 27877c478bd9Sstevel@tonic-gate /* 27887c478bd9Sstevel@tonic-gate * This is called just after configure() in startup(). 27897c478bd9Sstevel@tonic-gate * 27907c478bd9Sstevel@tonic-gate * The ISALIST concept is a bit hopeless on Intel, because 27917c478bd9Sstevel@tonic-gate * there's no guarantee of an ever-more-capable processor 27927c478bd9Sstevel@tonic-gate * given that various parts of the instruction set may appear 27937c478bd9Sstevel@tonic-gate * and disappear between different implementations. 27947c478bd9Sstevel@tonic-gate * 27957c478bd9Sstevel@tonic-gate * While it would be possible to correct it and even enhance 27967c478bd9Sstevel@tonic-gate * it somewhat, the explicit hardware capability bitmask allows 27977c478bd9Sstevel@tonic-gate * more flexibility. 27987c478bd9Sstevel@tonic-gate * 27997c478bd9Sstevel@tonic-gate * So, we just leave this alone. 28007c478bd9Sstevel@tonic-gate */ 28017c478bd9Sstevel@tonic-gate void 28027c478bd9Sstevel@tonic-gate setx86isalist(void) 28037c478bd9Sstevel@tonic-gate { 28047c478bd9Sstevel@tonic-gate char *tp; 28057c478bd9Sstevel@tonic-gate size_t len; 28067c478bd9Sstevel@tonic-gate extern char *isa_list; 28077c478bd9Sstevel@tonic-gate 28087c478bd9Sstevel@tonic-gate #define TBUFSIZE 1024 28097c478bd9Sstevel@tonic-gate 28107c478bd9Sstevel@tonic-gate tp = kmem_alloc(TBUFSIZE, KM_SLEEP); 28117c478bd9Sstevel@tonic-gate *tp = '\0'; 28127c478bd9Sstevel@tonic-gate 28137c478bd9Sstevel@tonic-gate #if defined(__amd64) 28147c478bd9Sstevel@tonic-gate (void) strcpy(tp, "amd64 "); 28157c478bd9Sstevel@tonic-gate #endif 28167c478bd9Sstevel@tonic-gate 28177c478bd9Sstevel@tonic-gate switch (x86_vendor) { 28187c478bd9Sstevel@tonic-gate case X86_VENDOR_Intel: 28197c478bd9Sstevel@tonic-gate case X86_VENDOR_AMD: 28207c478bd9Sstevel@tonic-gate case X86_VENDOR_TM: 28217c478bd9Sstevel@tonic-gate if (x86_feature & X86_CMOV) { 28227c478bd9Sstevel@tonic-gate /* 28237c478bd9Sstevel@tonic-gate * Pentium Pro or later 28247c478bd9Sstevel@tonic-gate */ 28257c478bd9Sstevel@tonic-gate (void) strcat(tp, "pentium_pro"); 28267c478bd9Sstevel@tonic-gate (void) strcat(tp, x86_feature & X86_MMX ? 28277c478bd9Sstevel@tonic-gate "+mmx pentium_pro " : " "); 28287c478bd9Sstevel@tonic-gate } 28297c478bd9Sstevel@tonic-gate /*FALLTHROUGH*/ 28307c478bd9Sstevel@tonic-gate case X86_VENDOR_Cyrix: 28317c478bd9Sstevel@tonic-gate /* 28327c478bd9Sstevel@tonic-gate * The Cyrix 6x86 does not have any Pentium features 28337c478bd9Sstevel@tonic-gate * accessible while not at privilege level 0. 28347c478bd9Sstevel@tonic-gate */ 28357c478bd9Sstevel@tonic-gate if (x86_feature & X86_CPUID) { 28367c478bd9Sstevel@tonic-gate (void) strcat(tp, "pentium"); 28377c478bd9Sstevel@tonic-gate (void) strcat(tp, x86_feature & X86_MMX ? 28387c478bd9Sstevel@tonic-gate "+mmx pentium " : " "); 28397c478bd9Sstevel@tonic-gate } 28407c478bd9Sstevel@tonic-gate break; 28417c478bd9Sstevel@tonic-gate default: 28427c478bd9Sstevel@tonic-gate break; 28437c478bd9Sstevel@tonic-gate } 28447c478bd9Sstevel@tonic-gate (void) strcat(tp, "i486 i386 i86"); 28457c478bd9Sstevel@tonic-gate len = strlen(tp) + 1; /* account for NULL at end of string */ 28467c478bd9Sstevel@tonic-gate isa_list = strcpy(kmem_alloc(len, KM_SLEEP), tp); 28477c478bd9Sstevel@tonic-gate kmem_free(tp, TBUFSIZE); 28487c478bd9Sstevel@tonic-gate 28497c478bd9Sstevel@tonic-gate #undef TBUFSIZE 28507c478bd9Sstevel@tonic-gate } 28517c478bd9Sstevel@tonic-gate 28527c478bd9Sstevel@tonic-gate 28537c478bd9Sstevel@tonic-gate #ifdef __amd64 28547c478bd9Sstevel@tonic-gate 28557c478bd9Sstevel@tonic-gate void * 28567c478bd9Sstevel@tonic-gate device_arena_alloc(size_t size, int vm_flag) 28577c478bd9Sstevel@tonic-gate { 28587c478bd9Sstevel@tonic-gate return (vmem_alloc(device_arena, size, vm_flag)); 28597c478bd9Sstevel@tonic-gate } 28607c478bd9Sstevel@tonic-gate 28617c478bd9Sstevel@tonic-gate void 28627c478bd9Sstevel@tonic-gate device_arena_free(void *vaddr, size_t size) 28637c478bd9Sstevel@tonic-gate { 28647c478bd9Sstevel@tonic-gate vmem_free(device_arena, vaddr, size); 28657c478bd9Sstevel@tonic-gate } 28667c478bd9Sstevel@tonic-gate 2867ae115bc7Smrj #else /* __i386 */ 28687c478bd9Sstevel@tonic-gate 28697c478bd9Sstevel@tonic-gate void * 28707c478bd9Sstevel@tonic-gate device_arena_alloc(size_t size, int vm_flag) 28717c478bd9Sstevel@tonic-gate { 28727c478bd9Sstevel@tonic-gate caddr_t vaddr; 28737c478bd9Sstevel@tonic-gate uintptr_t v; 28747c478bd9Sstevel@tonic-gate size_t start; 28757c478bd9Sstevel@tonic-gate size_t end; 28767c478bd9Sstevel@tonic-gate 28777c478bd9Sstevel@tonic-gate vaddr = vmem_alloc(heap_arena, size, vm_flag); 28787c478bd9Sstevel@tonic-gate if (vaddr == NULL) 28797c478bd9Sstevel@tonic-gate return (NULL); 28807c478bd9Sstevel@tonic-gate 28817c478bd9Sstevel@tonic-gate v = (uintptr_t)vaddr; 28827c478bd9Sstevel@tonic-gate ASSERT(v >= kernelbase); 2883ae115bc7Smrj ASSERT(v + size <= valloc_base); 28847c478bd9Sstevel@tonic-gate 28857c478bd9Sstevel@tonic-gate start = btop(v - kernelbase); 28867c478bd9Sstevel@tonic-gate end = btop(v + size - 1 - kernelbase); 28877c478bd9Sstevel@tonic-gate ASSERT(start < toxic_bit_map_len); 28887c478bd9Sstevel@tonic-gate ASSERT(end < toxic_bit_map_len); 28897c478bd9Sstevel@tonic-gate 28907c478bd9Sstevel@tonic-gate while (start <= end) { 28917c478bd9Sstevel@tonic-gate BT_ATOMIC_SET(toxic_bit_map, start); 28927c478bd9Sstevel@tonic-gate ++start; 28937c478bd9Sstevel@tonic-gate } 28947c478bd9Sstevel@tonic-gate return (vaddr); 28957c478bd9Sstevel@tonic-gate } 28967c478bd9Sstevel@tonic-gate 28977c478bd9Sstevel@tonic-gate void 28987c478bd9Sstevel@tonic-gate device_arena_free(void *vaddr, size_t size) 28997c478bd9Sstevel@tonic-gate { 29007c478bd9Sstevel@tonic-gate uintptr_t v = (uintptr_t)vaddr; 29017c478bd9Sstevel@tonic-gate size_t start; 29027c478bd9Sstevel@tonic-gate size_t end; 29037c478bd9Sstevel@tonic-gate 29047c478bd9Sstevel@tonic-gate ASSERT(v >= kernelbase); 2905ae115bc7Smrj ASSERT(v + size <= valloc_base); 29067c478bd9Sstevel@tonic-gate 29077c478bd9Sstevel@tonic-gate start = btop(v - kernelbase); 29087c478bd9Sstevel@tonic-gate end = btop(v + size - 1 - kernelbase); 29097c478bd9Sstevel@tonic-gate ASSERT(start < toxic_bit_map_len); 29107c478bd9Sstevel@tonic-gate ASSERT(end < toxic_bit_map_len); 29117c478bd9Sstevel@tonic-gate 29127c478bd9Sstevel@tonic-gate while (start <= end) { 29137c478bd9Sstevel@tonic-gate ASSERT(BT_TEST(toxic_bit_map, start) != 0); 29147c478bd9Sstevel@tonic-gate BT_ATOMIC_CLEAR(toxic_bit_map, start); 29157c478bd9Sstevel@tonic-gate ++start; 29167c478bd9Sstevel@tonic-gate } 29177c478bd9Sstevel@tonic-gate vmem_free(heap_arena, vaddr, size); 29187c478bd9Sstevel@tonic-gate } 29197c478bd9Sstevel@tonic-gate 29207c478bd9Sstevel@tonic-gate /* 29217c478bd9Sstevel@tonic-gate * returns 1st address in range that is in device arena, or NULL 29227c478bd9Sstevel@tonic-gate * if len is not NULL it returns the length of the toxic range 29237c478bd9Sstevel@tonic-gate */ 29247c478bd9Sstevel@tonic-gate void * 29257c478bd9Sstevel@tonic-gate device_arena_contains(void *vaddr, size_t size, size_t *len) 29267c478bd9Sstevel@tonic-gate { 29277c478bd9Sstevel@tonic-gate uintptr_t v = (uintptr_t)vaddr; 29287c478bd9Sstevel@tonic-gate uintptr_t eaddr = v + size; 29297c478bd9Sstevel@tonic-gate size_t start; 29307c478bd9Sstevel@tonic-gate size_t end; 29317c478bd9Sstevel@tonic-gate 29327c478bd9Sstevel@tonic-gate /* 29337c478bd9Sstevel@tonic-gate * if called very early by kmdb, just return NULL 29347c478bd9Sstevel@tonic-gate */ 29357c478bd9Sstevel@tonic-gate if (toxic_bit_map == NULL) 29367c478bd9Sstevel@tonic-gate return (NULL); 29377c478bd9Sstevel@tonic-gate 29387c478bd9Sstevel@tonic-gate /* 29397c478bd9Sstevel@tonic-gate * First check if we're completely outside the bitmap range. 29407c478bd9Sstevel@tonic-gate */ 2941ae115bc7Smrj if (v >= valloc_base || eaddr < kernelbase) 29427c478bd9Sstevel@tonic-gate return (NULL); 29437c478bd9Sstevel@tonic-gate 29447c478bd9Sstevel@tonic-gate /* 29457c478bd9Sstevel@tonic-gate * Trim ends of search to look at only what the bitmap covers. 29467c478bd9Sstevel@tonic-gate */ 29477c478bd9Sstevel@tonic-gate if (v < kernelbase) 29487c478bd9Sstevel@tonic-gate v = kernelbase; 29497c478bd9Sstevel@tonic-gate start = btop(v - kernelbase); 29507c478bd9Sstevel@tonic-gate end = btop(eaddr - kernelbase); 29517c478bd9Sstevel@tonic-gate if (end >= toxic_bit_map_len) 29527c478bd9Sstevel@tonic-gate end = toxic_bit_map_len; 29537c478bd9Sstevel@tonic-gate 29547c478bd9Sstevel@tonic-gate if (bt_range(toxic_bit_map, &start, &end, end) == 0) 29557c478bd9Sstevel@tonic-gate return (NULL); 29567c478bd9Sstevel@tonic-gate 29577c478bd9Sstevel@tonic-gate v = kernelbase + ptob(start); 29587c478bd9Sstevel@tonic-gate if (len != NULL) 29597c478bd9Sstevel@tonic-gate *len = ptob(end - start); 29607c478bd9Sstevel@tonic-gate return ((void *)v); 29617c478bd9Sstevel@tonic-gate } 29627c478bd9Sstevel@tonic-gate 2963ae115bc7Smrj #endif /* __i386 */ 2964