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 /* 22a563a037Sbholler * Copyright 2008 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> 11719397407SSherry Moore #include <sys/multiboot.h> 118ae115bc7Smrj 119843e1988Sjohnlev #ifdef __xpv 1201d03c31eSjohnlev 121843e1988Sjohnlev #include <sys/hypervisor.h> 122843e1988Sjohnlev #include <sys/xen_mmu.h> 123843e1988Sjohnlev #include <sys/evtchn_impl.h> 124843e1988Sjohnlev #include <sys/gnttab.h> 125843e1988Sjohnlev #include <sys/xpv_panic.h> 126843e1988Sjohnlev #include <xen/sys/xenbus_comms.h> 127843e1988Sjohnlev #include <xen/public/physdev.h> 1281d03c31eSjohnlev 129843e1988Sjohnlev extern void xen_late_startup(void); 130ae115bc7Smrj 1311d03c31eSjohnlev struct xen_evt_data cpu0_evt_data; 1321d03c31eSjohnlev 1331d03c31eSjohnlev #endif /* __xpv */ 1347c478bd9Sstevel@tonic-gate 1357c478bd9Sstevel@tonic-gate extern void progressbar_init(void); 1367c478bd9Sstevel@tonic-gate extern void progressbar_start(void); 1379acbbeafSnn extern void brand_init(void); 13806fb6a36Sdv extern void pcf_init(void); 1397c478bd9Sstevel@tonic-gate 140d7d93655Sblakej extern int size_pse_array(pgcnt_t, int); 141d7d93655Sblakej 142ed5289f9SKen Erickson #if defined(_SOFT_HOSTID) 143ed5289f9SKen Erickson 144ed5289f9SKen Erickson #include <sys/rtc.h> 145ed5289f9SKen Erickson 146ed5289f9SKen Erickson static int32_t set_soft_hostid(void); 147ed5289f9SKen Erickson extern char hw_serial[]; 148ed5289f9SKen Erickson static char hostid_file[] = "/etc/hostid"; 149ed5289f9SKen Erickson 150ed5289f9SKen Erickson #endif 151ed5289f9SKen Erickson 152*94f1124eSVikram Hegde void *gfx_devinfo_list; 153*94f1124eSVikram Hegde int startup_amd_iommu_disable; 154*94f1124eSVikram Hegde char *startup_amd_iommu_disable_list; 155ed5289f9SKen Erickson 1567c478bd9Sstevel@tonic-gate /* 1577c478bd9Sstevel@tonic-gate * XXX make declaration below "static" when drivers no longer use this 1587c478bd9Sstevel@tonic-gate * interface. 1597c478bd9Sstevel@tonic-gate */ 1607c478bd9Sstevel@tonic-gate extern caddr_t p0_va; /* Virtual address for accessing physical page 0 */ 1617c478bd9Sstevel@tonic-gate 1627c478bd9Sstevel@tonic-gate /* 1637c478bd9Sstevel@tonic-gate * segkp 1647c478bd9Sstevel@tonic-gate */ 1657c478bd9Sstevel@tonic-gate extern int segkp_fromheap; 1667c478bd9Sstevel@tonic-gate 1677c478bd9Sstevel@tonic-gate static void kvm_init(void); 1687c478bd9Sstevel@tonic-gate static void startup_init(void); 1697c478bd9Sstevel@tonic-gate static void startup_memlist(void); 170ae115bc7Smrj static void startup_kmem(void); 1717c478bd9Sstevel@tonic-gate static void startup_modules(void); 1727c478bd9Sstevel@tonic-gate static void startup_vm(void); 1737c478bd9Sstevel@tonic-gate static void startup_end(void); 17435b1ab99Sjosephb static void layout_kernel_va(void); 1757c478bd9Sstevel@tonic-gate 1767c478bd9Sstevel@tonic-gate /* 1777c478bd9Sstevel@tonic-gate * Declare these as initialized data so we can patch them. 1787c478bd9Sstevel@tonic-gate */ 179f53ad214Skchow #ifdef __i386 18035b1ab99Sjosephb 181f53ad214Skchow /* 182f53ad214Skchow * Due to virtual address space limitations running in 32 bit mode, restrict 18335b1ab99Sjosephb * the amount of physical memory configured to a max of PHYSMEM pages (16g). 184f53ad214Skchow * 185f53ad214Skchow * If the physical max memory size of 64g were allowed to be configured, the 186f53ad214Skchow * size of user virtual address space will be less than 1g. A limited user 187f53ad214Skchow * address space greatly reduces the range of applications that can run. 188f53ad214Skchow * 18935b1ab99Sjosephb * If more physical memory than PHYSMEM is required, users should preferably 19035b1ab99Sjosephb * run in 64 bit mode which has far looser virtual address space limitations. 191f53ad214Skchow * 192f53ad214Skchow * If 64 bit mode is not available (as in IA32) and/or more physical memory 19335b1ab99Sjosephb * than PHYSMEM is required in 32 bit mode, physmem can be set to the desired 194f53ad214Skchow * value or to 0 (to configure all available memory) via eeprom(1M). kernelbase 195f53ad214Skchow * should also be carefully tuned to balance out the need of the user 196f53ad214Skchow * application while minimizing the risk of kernel heap exhaustion due to 197f53ad214Skchow * kernelbase being set too high. 198f53ad214Skchow */ 19935b1ab99Sjosephb #define PHYSMEM 0x400000 200f53ad214Skchow 20135b1ab99Sjosephb #else /* __amd64 */ 20235b1ab99Sjosephb 20335b1ab99Sjosephb /* 20435b1ab99Sjosephb * For now we can handle memory with physical addresses up to about 20535b1ab99Sjosephb * 64 Terabytes. This keeps the kernel above the VA hole, leaving roughly 20635b1ab99Sjosephb * half the VA space for seg_kpm. When systems get bigger than 64TB this 20735b1ab99Sjosephb * code will need revisiting. There is an implicit assumption that there 20835b1ab99Sjosephb * are no *huge* holes in the physical address space too. 20935b1ab99Sjosephb */ 21035b1ab99Sjosephb #define TERABYTE (1ul << 40) 21135b1ab99Sjosephb #define PHYSMEM_MAX64 mmu_btop(64 * TERABYTE) 21235b1ab99Sjosephb #define PHYSMEM PHYSMEM_MAX64 21335b1ab99Sjosephb #define AMD64_VA_HOLE_END 0xFFFF800000000000ul 21435b1ab99Sjosephb 21535b1ab99Sjosephb #endif /* __amd64 */ 21635b1ab99Sjosephb 21735b1ab99Sjosephb pgcnt_t physmem = PHYSMEM; 2187c478bd9Sstevel@tonic-gate pgcnt_t obp_pages; /* Memory used by PROM for its text and data */ 2197c478bd9Sstevel@tonic-gate 2207c478bd9Sstevel@tonic-gate char *kobj_file_buf; 2217c478bd9Sstevel@tonic-gate int kobj_file_bufsize; /* set in /etc/system */ 2227c478bd9Sstevel@tonic-gate 2237c478bd9Sstevel@tonic-gate /* Global variables for MP support. Used in mp_startup */ 22419397407SSherry Moore caddr_t rm_platter_va = 0; 2257c478bd9Sstevel@tonic-gate uint32_t rm_platter_pa; 2267c478bd9Sstevel@tonic-gate 22783f9b804Skchow int auto_lpg_disable = 1; 22883f9b804Skchow 2297c478bd9Sstevel@tonic-gate /* 2307c478bd9Sstevel@tonic-gate * Some CPUs have holes in the middle of the 64-bit virtual address range. 2317c478bd9Sstevel@tonic-gate */ 2327c478bd9Sstevel@tonic-gate uintptr_t hole_start, hole_end; 2337c478bd9Sstevel@tonic-gate 2347c478bd9Sstevel@tonic-gate /* 2357c478bd9Sstevel@tonic-gate * kpm mapping window 2367c478bd9Sstevel@tonic-gate */ 2377c478bd9Sstevel@tonic-gate caddr_t kpm_vbase; 2387c478bd9Sstevel@tonic-gate size_t kpm_size; 23935b1ab99Sjosephb static int kpm_desired; 24035b1ab99Sjosephb #ifdef __amd64 24135b1ab99Sjosephb static uintptr_t segkpm_base = (uintptr_t)SEGKPM_BASE; 24235b1ab99Sjosephb #endif 2437c478bd9Sstevel@tonic-gate 2447c478bd9Sstevel@tonic-gate /* 2457c478bd9Sstevel@tonic-gate * Configuration parameters set at boot time. 2467c478bd9Sstevel@tonic-gate */ 2477c478bd9Sstevel@tonic-gate 2487c478bd9Sstevel@tonic-gate caddr_t econtig; /* end of first block of contiguous kernel */ 2497c478bd9Sstevel@tonic-gate 2507c478bd9Sstevel@tonic-gate struct bootops *bootops = 0; /* passed in from boot */ 2517c478bd9Sstevel@tonic-gate struct bootops **bootopsp; 2527c478bd9Sstevel@tonic-gate struct boot_syscalls *sysp; /* passed in from boot */ 2537c478bd9Sstevel@tonic-gate 2547c478bd9Sstevel@tonic-gate char bootblock_fstype[16]; 2557c478bd9Sstevel@tonic-gate 2567c478bd9Sstevel@tonic-gate char kern_bootargs[OBP_MAXPATHLEN]; 25719397407SSherry Moore char kern_bootfile[OBP_MAXPATHLEN]; 2587c478bd9Sstevel@tonic-gate 259ad23a2dbSjohansen /* 260ad23a2dbSjohansen * ZFS zio segment. This allows us to exclude large portions of ZFS data that 261ad23a2dbSjohansen * gets cached in kmem caches on the heap. If this is set to zero, we allocate 262ad23a2dbSjohansen * zio buffers from their own segment, otherwise they are allocated from the 263ad23a2dbSjohansen * heap. The optimization of allocating zio buffers from their own segment is 264ad23a2dbSjohansen * only valid on 64-bit kernels. 265ad23a2dbSjohansen */ 266ad23a2dbSjohansen #if defined(__amd64) 267ad23a2dbSjohansen int segzio_fromheap = 0; 268ad23a2dbSjohansen #else 269ad23a2dbSjohansen int segzio_fromheap = 1; 270ad23a2dbSjohansen #endif 271ad23a2dbSjohansen 2727c478bd9Sstevel@tonic-gate /* 2737c478bd9Sstevel@tonic-gate * new memory fragmentations are possible in startup() due to BOP_ALLOCs. this 2747c478bd9Sstevel@tonic-gate * depends on number of BOP_ALLOC calls made and requested size, memory size 2757c478bd9Sstevel@tonic-gate * combination and whether boot.bin memory needs to be freed. 2767c478bd9Sstevel@tonic-gate */ 2777c478bd9Sstevel@tonic-gate #define POSS_NEW_FRAGMENTS 12 2787c478bd9Sstevel@tonic-gate 2797c478bd9Sstevel@tonic-gate /* 2807c478bd9Sstevel@tonic-gate * VM data structures 2817c478bd9Sstevel@tonic-gate */ 2827c478bd9Sstevel@tonic-gate long page_hashsz; /* Size of page hash table (power of two) */ 2837c478bd9Sstevel@tonic-gate struct page *pp_base; /* Base of initial system page struct array */ 2847c478bd9Sstevel@tonic-gate struct page **page_hash; /* Page hash table */ 285d7d93655Sblakej pad_mutex_t *pse_mutex; /* Locks protecting pp->p_selock */ 286d7d93655Sblakej size_t pse_table_size; /* Number of mutexes in pse_mutex[] */ 287d7d93655Sblakej int pse_shift; /* log2(pse_table_size) */ 2887c478bd9Sstevel@tonic-gate struct seg ktextseg; /* Segment used for kernel executable image */ 2897c478bd9Sstevel@tonic-gate struct seg kvalloc; /* Segment used for "valloc" mapping */ 2907c478bd9Sstevel@tonic-gate struct seg kpseg; /* Segment used for pageable kernel virt mem */ 2917c478bd9Sstevel@tonic-gate struct seg kmapseg; /* Segment used for generic kernel mappings */ 2927c478bd9Sstevel@tonic-gate struct seg kdebugseg; /* Segment used for the kernel debugger */ 2937c478bd9Sstevel@tonic-gate 2947c478bd9Sstevel@tonic-gate struct seg *segkmap = &kmapseg; /* Kernel generic mapping segment */ 295ae115bc7Smrj static struct seg *segmap = &kmapseg; /* easier to use name for in here */ 296ae115bc7Smrj 2977c478bd9Sstevel@tonic-gate struct seg *segkp = &kpseg; /* Pageable kernel virtual memory segment */ 2987c478bd9Sstevel@tonic-gate 2997c478bd9Sstevel@tonic-gate #if defined(__amd64) 3007c478bd9Sstevel@tonic-gate struct seg kvseg_core; /* Segment used for the core heap */ 3017c478bd9Sstevel@tonic-gate struct seg kpmseg; /* Segment used for physical mapping */ 3027c478bd9Sstevel@tonic-gate struct seg *segkpm = &kpmseg; /* 64bit kernel physical mapping segment */ 3037c478bd9Sstevel@tonic-gate #else 3047c478bd9Sstevel@tonic-gate struct seg *segkpm = NULL; /* Unused on IA32 */ 3057c478bd9Sstevel@tonic-gate #endif 3067c478bd9Sstevel@tonic-gate 3077c478bd9Sstevel@tonic-gate caddr_t segkp_base; /* Base address of segkp */ 308ad23a2dbSjohansen caddr_t segzio_base; /* Base address of segzio */ 3097c478bd9Sstevel@tonic-gate #if defined(__amd64) 3107c478bd9Sstevel@tonic-gate pgcnt_t segkpsize = btop(SEGKPDEFSIZE); /* size of segkp segment in pages */ 3117c478bd9Sstevel@tonic-gate #else 3127c478bd9Sstevel@tonic-gate pgcnt_t segkpsize = 0; 3137c478bd9Sstevel@tonic-gate #endif 314ad23a2dbSjohansen pgcnt_t segziosize = 0; /* size of zio segment in pages */ 3157c478bd9Sstevel@tonic-gate 316ae115bc7Smrj /* 317ae115bc7Smrj * VA range available to the debugger 318ae115bc7Smrj */ 319ae115bc7Smrj const caddr_t kdi_segdebugbase = (const caddr_t)SEGDEBUGBASE; 320ae115bc7Smrj const size_t kdi_segdebugsize = SEGDEBUGSIZE; 321ae115bc7Smrj 3227c478bd9Sstevel@tonic-gate struct memseg *memseg_base; 3237c478bd9Sstevel@tonic-gate struct vnode unused_pages_vp; 3247c478bd9Sstevel@tonic-gate 3257c478bd9Sstevel@tonic-gate #define FOURGB 0x100000000LL 3267c478bd9Sstevel@tonic-gate 3277c478bd9Sstevel@tonic-gate struct memlist *memlist; 3287c478bd9Sstevel@tonic-gate 3297c478bd9Sstevel@tonic-gate caddr_t s_text; /* start of kernel text segment */ 3307c478bd9Sstevel@tonic-gate caddr_t e_text; /* end of kernel text segment */ 3317c478bd9Sstevel@tonic-gate caddr_t s_data; /* start of kernel data segment */ 3327c478bd9Sstevel@tonic-gate caddr_t e_data; /* end of kernel data segment */ 3337c478bd9Sstevel@tonic-gate caddr_t modtext; /* start of loadable module text reserved */ 3347c478bd9Sstevel@tonic-gate caddr_t e_modtext; /* end of loadable module text reserved */ 3357c478bd9Sstevel@tonic-gate caddr_t moddata; /* start of loadable module data reserved */ 3367c478bd9Sstevel@tonic-gate caddr_t e_moddata; /* end of loadable module data reserved */ 3377c478bd9Sstevel@tonic-gate 3387c478bd9Sstevel@tonic-gate struct memlist *phys_install; /* Total installed physical memory */ 3397c478bd9Sstevel@tonic-gate struct memlist *phys_avail; /* Total available physical memory */ 3407c478bd9Sstevel@tonic-gate 3417c478bd9Sstevel@tonic-gate /* 3427c478bd9Sstevel@tonic-gate * kphysm_init returns the number of pages that were processed 3437c478bd9Sstevel@tonic-gate */ 344ae115bc7Smrj static pgcnt_t kphysm_init(page_t *, pgcnt_t); 3457c478bd9Sstevel@tonic-gate 3467c478bd9Sstevel@tonic-gate #define IO_PROP_SIZE 64 /* device property size */ 3477c478bd9Sstevel@tonic-gate 3487c478bd9Sstevel@tonic-gate /* 3497c478bd9Sstevel@tonic-gate * a couple useful roundup macros 3507c478bd9Sstevel@tonic-gate */ 3517c478bd9Sstevel@tonic-gate #define ROUND_UP_PAGE(x) \ 3527c478bd9Sstevel@tonic-gate ((uintptr_t)P2ROUNDUP((uintptr_t)(x), (uintptr_t)MMU_PAGESIZE)) 3537c478bd9Sstevel@tonic-gate #define ROUND_UP_LPAGE(x) \ 3547c478bd9Sstevel@tonic-gate ((uintptr_t)P2ROUNDUP((uintptr_t)(x), mmu.level_size[1])) 3557c478bd9Sstevel@tonic-gate #define ROUND_UP_4MEG(x) \ 356ae115bc7Smrj ((uintptr_t)P2ROUNDUP((uintptr_t)(x), (uintptr_t)FOUR_MEG)) 3577c478bd9Sstevel@tonic-gate #define ROUND_UP_TOPLEVEL(x) \ 3587c478bd9Sstevel@tonic-gate ((uintptr_t)P2ROUNDUP((uintptr_t)(x), mmu.level_size[mmu.max_level])) 3597c478bd9Sstevel@tonic-gate 3607c478bd9Sstevel@tonic-gate /* 3617c478bd9Sstevel@tonic-gate * 32-bit Kernel's Virtual memory layout. 3627c478bd9Sstevel@tonic-gate * +-----------------------+ 363ae115bc7Smrj * | | 3647c478bd9Sstevel@tonic-gate * 0xFFC00000 -|-----------------------|- ARGSBASE 3657c478bd9Sstevel@tonic-gate * | debugger | 3667c478bd9Sstevel@tonic-gate * 0xFF800000 -|-----------------------|- SEGDEBUGBASE 3677c478bd9Sstevel@tonic-gate * | Kernel Data | 3687c478bd9Sstevel@tonic-gate * 0xFEC00000 -|-----------------------| 3697c478bd9Sstevel@tonic-gate * | Kernel Text | 370843e1988Sjohnlev * 0xFE800000 -|-----------------------|- KERNEL_TEXT (0xFB400000 on Xen) 371ae115bc7Smrj * |--- GDT ---|- GDT page (GDT_VA) 372ae115bc7Smrj * |--- debug info ---|- debug info (DEBUG_INFO_VA) 373ae115bc7Smrj * | | 374ae115bc7Smrj * | page_t structures | 3757c478bd9Sstevel@tonic-gate * | memsegs, memlists, | 3767c478bd9Sstevel@tonic-gate * | page hash, etc. | 377ae115bc7Smrj * --- -|-----------------------|- ekernelheap, valloc_base (floating) 378ae115bc7Smrj * | | (segkp is just an arena in the heap) 3797c478bd9Sstevel@tonic-gate * | | 3807c478bd9Sstevel@tonic-gate * | kvseg | 3817c478bd9Sstevel@tonic-gate * | | 3827c478bd9Sstevel@tonic-gate * | | 3837c478bd9Sstevel@tonic-gate * --- -|-----------------------|- kernelheap (floating) 3847c478bd9Sstevel@tonic-gate * | Segkmap | 385ae115bc7Smrj * 0xC3002000 -|-----------------------|- segmap_start (floating) 3867c478bd9Sstevel@tonic-gate * | Red Zone | 3877c478bd9Sstevel@tonic-gate * 0xC3000000 -|-----------------------|- kernelbase / userlimit (floating) 3887c478bd9Sstevel@tonic-gate * | | || 3897c478bd9Sstevel@tonic-gate * | Shared objects | \/ 3907c478bd9Sstevel@tonic-gate * | | 3917c478bd9Sstevel@tonic-gate * : : 3927c478bd9Sstevel@tonic-gate * | user data | 3937c478bd9Sstevel@tonic-gate * |-----------------------| 3947c478bd9Sstevel@tonic-gate * | user text | 3957c478bd9Sstevel@tonic-gate * 0x08048000 -|-----------------------| 3967c478bd9Sstevel@tonic-gate * | user stack | 3977c478bd9Sstevel@tonic-gate * : : 3987c478bd9Sstevel@tonic-gate * | invalid | 3997c478bd9Sstevel@tonic-gate * 0x00000000 +-----------------------+ 4007c478bd9Sstevel@tonic-gate * 4017c478bd9Sstevel@tonic-gate * 4027c478bd9Sstevel@tonic-gate * 64-bit Kernel's Virtual memory layout. (assuming 64 bit app) 4037c478bd9Sstevel@tonic-gate * +-----------------------+ 404ae115bc7Smrj * | | 4057c478bd9Sstevel@tonic-gate * 0xFFFFFFFF.FFC00000 |-----------------------|- ARGSBASE 4067c478bd9Sstevel@tonic-gate * | debugger (?) | 4077c478bd9Sstevel@tonic-gate * 0xFFFFFFFF.FF800000 |-----------------------|- SEGDEBUGBASE 4087c478bd9Sstevel@tonic-gate * | unused | 4097c478bd9Sstevel@tonic-gate * +-----------------------+ 4107c478bd9Sstevel@tonic-gate * | Kernel Data | 4117c478bd9Sstevel@tonic-gate * 0xFFFFFFFF.FBC00000 |-----------------------| 4127c478bd9Sstevel@tonic-gate * | Kernel Text | 4137c478bd9Sstevel@tonic-gate * 0xFFFFFFFF.FB800000 |-----------------------|- KERNEL_TEXT 414ae115bc7Smrj * |--- GDT ---|- GDT page (GDT_VA) 415ae115bc7Smrj * |--- debug info ---|- debug info (DEBUG_INFO_VA) 416ae115bc7Smrj * | | 4177c478bd9Sstevel@tonic-gate * | Core heap | (used for loadable modules) 4187c478bd9Sstevel@tonic-gate * 0xFFFFFFFF.C0000000 |-----------------------|- core_base / ekernelheap 4197c478bd9Sstevel@tonic-gate * | Kernel | 4207c478bd9Sstevel@tonic-gate * | heap | 4217c478bd9Sstevel@tonic-gate * 0xFFFFFXXX.XXX00000 |-----------------------|- kernelheap (floating) 422ae115bc7Smrj * | segmap | 423ae115bc7Smrj * 0xFFFFFXXX.XXX00000 |-----------------------|- segmap_start (floating) 4247c478bd9Sstevel@tonic-gate * | device mappings | 4257c478bd9Sstevel@tonic-gate * 0xFFFFFXXX.XXX00000 |-----------------------|- toxic_addr (floating) 426ad23a2dbSjohansen * | segzio | 427ad23a2dbSjohansen * 0xFFFFFXXX.XXX00000 |-----------------------|- segzio_base (floating) 4287c478bd9Sstevel@tonic-gate * | segkp | 429ae115bc7Smrj * --- |-----------------------|- segkp_base (floating) 430ae115bc7Smrj * | page_t structures | valloc_base + valloc_sz 431ae115bc7Smrj * | memsegs, memlists, | 432ae115bc7Smrj * | page hash, etc. | 43335b1ab99Sjosephb * 0xFFFFFF00.00000000 |-----------------------|- valloc_base (lower if > 1TB) 4347c478bd9Sstevel@tonic-gate * | segkpm | 4357c478bd9Sstevel@tonic-gate * 0xFFFFFE00.00000000 |-----------------------| 4367c478bd9Sstevel@tonic-gate * | Red Zone | 43735b1ab99Sjosephb * 0xFFFFFD80.00000000 |-----------------------|- KERNELBASE (lower if > 1TB) 4387c478bd9Sstevel@tonic-gate * | User stack |- User space memory 4397c478bd9Sstevel@tonic-gate * | | 4407c478bd9Sstevel@tonic-gate * | shared objects, etc | (grows downwards) 4417c478bd9Sstevel@tonic-gate * : : 4427c478bd9Sstevel@tonic-gate * | | 4437c478bd9Sstevel@tonic-gate * 0xFFFF8000.00000000 |-----------------------| 4447c478bd9Sstevel@tonic-gate * | | 4457c478bd9Sstevel@tonic-gate * | VA Hole / unused | 4467c478bd9Sstevel@tonic-gate * | | 4477c478bd9Sstevel@tonic-gate * 0x00008000.00000000 |-----------------------| 4487c478bd9Sstevel@tonic-gate * | | 4497c478bd9Sstevel@tonic-gate * | | 4507c478bd9Sstevel@tonic-gate * : : 4517c478bd9Sstevel@tonic-gate * | user heap | (grows upwards) 4527c478bd9Sstevel@tonic-gate * | | 4537c478bd9Sstevel@tonic-gate * | user data | 4547c478bd9Sstevel@tonic-gate * |-----------------------| 4557c478bd9Sstevel@tonic-gate * | user text | 4567c478bd9Sstevel@tonic-gate * 0x00000000.04000000 |-----------------------| 4577c478bd9Sstevel@tonic-gate * | invalid | 4587c478bd9Sstevel@tonic-gate * 0x00000000.00000000 +-----------------------+ 4597c478bd9Sstevel@tonic-gate * 4607c478bd9Sstevel@tonic-gate * A 32 bit app on the 64 bit kernel sees the same layout as on the 32 bit 4617c478bd9Sstevel@tonic-gate * kernel, except that userlimit is raised to 0xfe000000 4627c478bd9Sstevel@tonic-gate * 4637c478bd9Sstevel@tonic-gate * Floating values: 4647c478bd9Sstevel@tonic-gate * 4657c478bd9Sstevel@tonic-gate * valloc_base: start of the kernel's memory management/tracking data 466ae115bc7Smrj * structures. This region contains page_t structures for 467ae115bc7Smrj * physical memory, memsegs, memlists, and the page hash. 4687c478bd9Sstevel@tonic-gate * 4697c478bd9Sstevel@tonic-gate * core_base: start of the kernel's "core" heap area on 64-bit systems. 4707c478bd9Sstevel@tonic-gate * This area is intended to be used for global data as well as for module 4717c478bd9Sstevel@tonic-gate * text/data that does not fit into the nucleus pages. The core heap is 4727c478bd9Sstevel@tonic-gate * restricted to a 2GB range, allowing every address within it to be 4737c478bd9Sstevel@tonic-gate * accessed using rip-relative addressing 4747c478bd9Sstevel@tonic-gate * 4757c478bd9Sstevel@tonic-gate * ekernelheap: end of kernelheap and start of segmap. 4767c478bd9Sstevel@tonic-gate * 4777c478bd9Sstevel@tonic-gate * kernelheap: start of kernel heap. On 32-bit systems, this starts right 4787c478bd9Sstevel@tonic-gate * above a red zone that separates the user's address space from the 4797c478bd9Sstevel@tonic-gate * kernel's. On 64-bit systems, it sits above segkp and segkpm. 4807c478bd9Sstevel@tonic-gate * 481ae115bc7Smrj * segmap_start: start of segmap. The length of segmap can be modified 48252eef812SAmrita Sadhukhan * through eeprom. The default length is 16MB on 32-bit systems and 64MB 48352eef812SAmrita Sadhukhan * on 64-bit systems. 4847c478bd9Sstevel@tonic-gate * 4857c478bd9Sstevel@tonic-gate * kernelbase: On a 32-bit kernel the default value of 0xd4000000 will be 4867c478bd9Sstevel@tonic-gate * decreased by 2X the size required for page_t. This allows the kernel 4877c478bd9Sstevel@tonic-gate * heap to grow in size with physical memory. With sizeof(page_t) == 80 4887c478bd9Sstevel@tonic-gate * bytes, the following shows the values of kernelbase and kernel heap 4897c478bd9Sstevel@tonic-gate * sizes for different memory configurations (assuming default segmap and 4907c478bd9Sstevel@tonic-gate * segkp sizes). 4917c478bd9Sstevel@tonic-gate * 4927c478bd9Sstevel@tonic-gate * mem size for kernelbase kernel heap 4937c478bd9Sstevel@tonic-gate * size page_t's size 4947c478bd9Sstevel@tonic-gate * ---- --------- ---------- ----------- 4957c478bd9Sstevel@tonic-gate * 1gb 0x01400000 0xd1800000 684MB 4967c478bd9Sstevel@tonic-gate * 2gb 0x02800000 0xcf000000 704MB 4977c478bd9Sstevel@tonic-gate * 4gb 0x05000000 0xca000000 744MB 4987c478bd9Sstevel@tonic-gate * 6gb 0x07800000 0xc5000000 784MB 4997c478bd9Sstevel@tonic-gate * 8gb 0x0a000000 0xc0000000 824MB 5007c478bd9Sstevel@tonic-gate * 16gb 0x14000000 0xac000000 984MB 5017c478bd9Sstevel@tonic-gate * 32gb 0x28000000 0x84000000 1304MB 5027c478bd9Sstevel@tonic-gate * 64gb 0x50000000 0x34000000 1944MB (*) 5037c478bd9Sstevel@tonic-gate * 5047c478bd9Sstevel@tonic-gate * kernelbase is less than the abi minimum of 0xc0000000 for memory 5057c478bd9Sstevel@tonic-gate * configurations above 8gb. 5067c478bd9Sstevel@tonic-gate * 5077c478bd9Sstevel@tonic-gate * (*) support for memory configurations above 32gb will require manual tuning 5087c478bd9Sstevel@tonic-gate * of kernelbase to balance out the need of user applications. 5097c478bd9Sstevel@tonic-gate */ 5107c478bd9Sstevel@tonic-gate 5117c478bd9Sstevel@tonic-gate /* real-time-clock initialization parameters */ 512ae115bc7Smrj extern time_t process_rtc_config_file(void); 5137c478bd9Sstevel@tonic-gate 5147c478bd9Sstevel@tonic-gate uintptr_t kernelbase; 515ae115bc7Smrj uintptr_t postbootkernelbase; /* not set till boot loader is gone */ 5167c478bd9Sstevel@tonic-gate uintptr_t eprom_kernelbase; 5177c478bd9Sstevel@tonic-gate size_t segmapsize; 518ae115bc7Smrj uintptr_t segmap_start; 5197c478bd9Sstevel@tonic-gate int segmapfreelists; 5207c478bd9Sstevel@tonic-gate pgcnt_t npages; 521ae115bc7Smrj pgcnt_t orig_npages; 5227c478bd9Sstevel@tonic-gate size_t core_size; /* size of "core" heap */ 5237c478bd9Sstevel@tonic-gate uintptr_t core_base; /* base address of "core" heap */ 5247c478bd9Sstevel@tonic-gate 5257c478bd9Sstevel@tonic-gate /* 5267c478bd9Sstevel@tonic-gate * List of bootstrap pages. We mark these as allocated in startup. 5277c478bd9Sstevel@tonic-gate * release_bootstrap() will free them when we're completely done with 5287c478bd9Sstevel@tonic-gate * the bootstrap. 5297c478bd9Sstevel@tonic-gate */ 530ae115bc7Smrj static page_t *bootpages; 531ae115bc7Smrj 532ae115bc7Smrj /* 533ae115bc7Smrj * boot time pages that have a vnode from the ramdisk will keep that forever. 534ae115bc7Smrj */ 535ae115bc7Smrj static page_t *rd_pages; 5367c478bd9Sstevel@tonic-gate 53719397407SSherry Moore /* 53819397407SSherry Moore * Lower 64K 53919397407SSherry Moore */ 54019397407SSherry Moore static page_t *lower_pages = NULL; 54119397407SSherry Moore static int lower_pages_count = 0; 54219397407SSherry Moore 5437c478bd9Sstevel@tonic-gate struct system_hardware system_hardware; 5447c478bd9Sstevel@tonic-gate 545551bc2a6Smrj /* 546551bc2a6Smrj * Is this Solaris instance running in a fully virtualized xVM domain? 547551bc2a6Smrj */ 548551bc2a6Smrj int xpv_is_hvm = 0; 549551bc2a6Smrj 5507c478bd9Sstevel@tonic-gate /* 5517c478bd9Sstevel@tonic-gate * Enable some debugging messages concerning memory usage... 5527c478bd9Sstevel@tonic-gate */ 5537c478bd9Sstevel@tonic-gate static void 554ae115bc7Smrj print_memlist(char *title, struct memlist *mp) 5557c478bd9Sstevel@tonic-gate { 5567c478bd9Sstevel@tonic-gate prom_printf("MEMLIST: %s:\n", title); 5577c478bd9Sstevel@tonic-gate while (mp != NULL) { 5587c478bd9Sstevel@tonic-gate prom_printf("\tAddress 0x%" PRIx64 ", size 0x%" PRIx64 "\n", 5597c478bd9Sstevel@tonic-gate mp->address, mp->size); 5607c478bd9Sstevel@tonic-gate mp = mp->next; 5617c478bd9Sstevel@tonic-gate } 5627c478bd9Sstevel@tonic-gate } 5637c478bd9Sstevel@tonic-gate 5647c478bd9Sstevel@tonic-gate /* 5657c478bd9Sstevel@tonic-gate * XX64 need a comment here.. are these just default values, surely 5667c478bd9Sstevel@tonic-gate * we read the "cpuid" type information to figure this out. 5677c478bd9Sstevel@tonic-gate */ 5687c478bd9Sstevel@tonic-gate int l2cache_sz = 0x80000; 5697c478bd9Sstevel@tonic-gate int l2cache_linesz = 0x40; 5707c478bd9Sstevel@tonic-gate int l2cache_assoc = 1; 5717c478bd9Sstevel@tonic-gate 5722cb27123Saguzovsk static size_t textrepl_min_gb = 10; 5732cb27123Saguzovsk 5747c478bd9Sstevel@tonic-gate /* 5757c478bd9Sstevel@tonic-gate * on 64 bit we use a predifined VA range for mapping devices in the kernel 5767c478bd9Sstevel@tonic-gate * on 32 bit the mappings are intermixed in the heap, so we use a bit map 5777c478bd9Sstevel@tonic-gate */ 5787c478bd9Sstevel@tonic-gate #ifdef __amd64 5797c478bd9Sstevel@tonic-gate 5807c478bd9Sstevel@tonic-gate vmem_t *device_arena; 5817c478bd9Sstevel@tonic-gate uintptr_t toxic_addr = (uintptr_t)NULL; 582ae115bc7Smrj size_t toxic_size = 1024 * 1024 * 1024; /* Sparc uses 1 gig too */ 5837c478bd9Sstevel@tonic-gate 5847c478bd9Sstevel@tonic-gate #else /* __i386 */ 5857c478bd9Sstevel@tonic-gate 5867c478bd9Sstevel@tonic-gate ulong_t *toxic_bit_map; /* one bit for each 4k of VA in heap_arena */ 5877c478bd9Sstevel@tonic-gate size_t toxic_bit_map_len = 0; /* in bits */ 5887c478bd9Sstevel@tonic-gate 5897c478bd9Sstevel@tonic-gate #endif /* __i386 */ 5907c478bd9Sstevel@tonic-gate 5917c478bd9Sstevel@tonic-gate /* 5927c478bd9Sstevel@tonic-gate * Simple boot time debug facilities 5937c478bd9Sstevel@tonic-gate */ 5947c478bd9Sstevel@tonic-gate static char *prm_dbg_str[] = { 5957c478bd9Sstevel@tonic-gate "%s:%d: '%s' is 0x%x\n", 5967c478bd9Sstevel@tonic-gate "%s:%d: '%s' is 0x%llx\n" 5977c478bd9Sstevel@tonic-gate }; 5987c478bd9Sstevel@tonic-gate 5997c478bd9Sstevel@tonic-gate int prom_debug; 6007c478bd9Sstevel@tonic-gate 6017c478bd9Sstevel@tonic-gate #define PRM_DEBUG(q) if (prom_debug) \ 6027c478bd9Sstevel@tonic-gate prom_printf(prm_dbg_str[sizeof (q) >> 3], "startup.c", __LINE__, #q, q); 6037c478bd9Sstevel@tonic-gate #define PRM_POINT(q) if (prom_debug) \ 6047c478bd9Sstevel@tonic-gate prom_printf("%s:%d: %s\n", "startup.c", __LINE__, q); 6057c478bd9Sstevel@tonic-gate 6067c478bd9Sstevel@tonic-gate /* 6077c478bd9Sstevel@tonic-gate * This structure is used to keep track of the intial allocations 6087c478bd9Sstevel@tonic-gate * done in startup_memlist(). The value of NUM_ALLOCATIONS needs to 6097c478bd9Sstevel@tonic-gate * be >= the number of ADD_TO_ALLOCATIONS() executed in the code. 6107c478bd9Sstevel@tonic-gate */ 6117c478bd9Sstevel@tonic-gate #define NUM_ALLOCATIONS 7 6127c478bd9Sstevel@tonic-gate int num_allocations = 0; 6137c478bd9Sstevel@tonic-gate struct { 6147c478bd9Sstevel@tonic-gate void **al_ptr; 6157c478bd9Sstevel@tonic-gate size_t al_size; 6167c478bd9Sstevel@tonic-gate } allocations[NUM_ALLOCATIONS]; 6177c478bd9Sstevel@tonic-gate size_t valloc_sz = 0; 6187c478bd9Sstevel@tonic-gate uintptr_t valloc_base; 6197c478bd9Sstevel@tonic-gate 6207c478bd9Sstevel@tonic-gate #define ADD_TO_ALLOCATIONS(ptr, size) { \ 6217c478bd9Sstevel@tonic-gate size = ROUND_UP_PAGE(size); \ 6227c478bd9Sstevel@tonic-gate if (num_allocations == NUM_ALLOCATIONS) \ 6237c478bd9Sstevel@tonic-gate panic("too many ADD_TO_ALLOCATIONS()"); \ 6247c478bd9Sstevel@tonic-gate allocations[num_allocations].al_ptr = (void**)&ptr; \ 6257c478bd9Sstevel@tonic-gate allocations[num_allocations].al_size = size; \ 6267c478bd9Sstevel@tonic-gate valloc_sz += size; \ 6277c478bd9Sstevel@tonic-gate ++num_allocations; \ 6287c478bd9Sstevel@tonic-gate } 6297c478bd9Sstevel@tonic-gate 630ae115bc7Smrj /* 631ae115bc7Smrj * Allocate all the initial memory needed by the page allocator. 632ae115bc7Smrj */ 6337c478bd9Sstevel@tonic-gate static void 6347c478bd9Sstevel@tonic-gate perform_allocations(void) 6357c478bd9Sstevel@tonic-gate { 6367c478bd9Sstevel@tonic-gate caddr_t mem; 6377c478bd9Sstevel@tonic-gate int i; 638ae115bc7Smrj int valloc_align; 6397c478bd9Sstevel@tonic-gate 640ae115bc7Smrj PRM_DEBUG(valloc_base); 641ae115bc7Smrj PRM_DEBUG(valloc_sz); 642ae115bc7Smrj valloc_align = mmu.level_size[mmu.max_page_level > 0]; 643ae115bc7Smrj mem = BOP_ALLOC(bootops, (caddr_t)valloc_base, valloc_sz, valloc_align); 6447c478bd9Sstevel@tonic-gate if (mem != (caddr_t)valloc_base) 6457c478bd9Sstevel@tonic-gate panic("BOP_ALLOC() failed"); 6467c478bd9Sstevel@tonic-gate bzero(mem, valloc_sz); 6477c478bd9Sstevel@tonic-gate for (i = 0; i < num_allocations; ++i) { 6487c478bd9Sstevel@tonic-gate *allocations[i].al_ptr = (void *)mem; 6497c478bd9Sstevel@tonic-gate mem += allocations[i].al_size; 6507c478bd9Sstevel@tonic-gate } 6517c478bd9Sstevel@tonic-gate } 6527c478bd9Sstevel@tonic-gate 6537c478bd9Sstevel@tonic-gate /* 6547c478bd9Sstevel@tonic-gate * Our world looks like this at startup time. 6557c478bd9Sstevel@tonic-gate * 6567c478bd9Sstevel@tonic-gate * In a 32-bit OS, boot loads the kernel text at 0xfe800000 and kernel data 6577c478bd9Sstevel@tonic-gate * at 0xfec00000. On a 64-bit OS, kernel text and data are loaded at 6587c478bd9Sstevel@tonic-gate * 0xffffffff.fe800000 and 0xffffffff.fec00000 respectively. Those 6597c478bd9Sstevel@tonic-gate * addresses are fixed in the binary at link time. 6607c478bd9Sstevel@tonic-gate * 6617c478bd9Sstevel@tonic-gate * On the text page: 6627c478bd9Sstevel@tonic-gate * unix/genunix/krtld/module text loads. 6637c478bd9Sstevel@tonic-gate * 6647c478bd9Sstevel@tonic-gate * On the data page: 665ae115bc7Smrj * unix/genunix/krtld/module data loads. 666ae115bc7Smrj * 6677c478bd9Sstevel@tonic-gate * Machine-dependent startup code 6687c478bd9Sstevel@tonic-gate */ 6697c478bd9Sstevel@tonic-gate void 6707c478bd9Sstevel@tonic-gate startup(void) 6717c478bd9Sstevel@tonic-gate { 672843e1988Sjohnlev #if !defined(__xpv) 67375bcd456Sjg extern void startup_bios_disk(void); 67475bcd456Sjg extern void startup_pci_bios(void); 67519397407SSherry Moore extern int post_fastreboot; 676843e1988Sjohnlev #endif 677a563a037Sbholler extern cpuset_t cpu_ready_set; 678a563a037Sbholler 679843e1988Sjohnlev /* 680843e1988Sjohnlev * Make sure that nobody tries to use sekpm until we have 681843e1988Sjohnlev * initialized it properly. 682843e1988Sjohnlev */ 6837c478bd9Sstevel@tonic-gate #if defined(__amd64) 68435b1ab99Sjosephb kpm_desired = 1; 6857c478bd9Sstevel@tonic-gate #endif 6867c478bd9Sstevel@tonic-gate kpm_enable = 0; 687a563a037Sbholler CPUSET_ONLY(cpu_ready_set, 0); /* cpu 0 is boot cpu */ 6887c478bd9Sstevel@tonic-gate 689843e1988Sjohnlev #if defined(__xpv) /* XXPV fix me! */ 690843e1988Sjohnlev { 691843e1988Sjohnlev extern int segvn_use_regions; 692843e1988Sjohnlev segvn_use_regions = 0; 693843e1988Sjohnlev } 694843e1988Sjohnlev #endif 6957c478bd9Sstevel@tonic-gate progressbar_init(); 6967c478bd9Sstevel@tonic-gate startup_init(); 697ab4a9bebSjohnlev #if defined(__xpv) 698ab4a9bebSjohnlev startup_xen_version(); 699ab4a9bebSjohnlev #endif 7007c478bd9Sstevel@tonic-gate startup_memlist(); 701ae115bc7Smrj startup_kmem(); 70235b1ab99Sjosephb startup_vm(); 703843e1988Sjohnlev #if !defined(__xpv) 70419397407SSherry Moore if (!post_fastreboot) 70519397407SSherry Moore startup_pci_bios(); 706e4b86885SCheng Sean Ye #endif 707e4b86885SCheng Sean Ye #if defined(__xpv) 708e4b86885SCheng Sean Ye startup_xen_mca(); 709843e1988Sjohnlev #endif 7107c478bd9Sstevel@tonic-gate startup_modules(); 711843e1988Sjohnlev #if !defined(__xpv) 71219397407SSherry Moore if (!post_fastreboot) 71319397407SSherry Moore startup_bios_disk(); 714843e1988Sjohnlev #endif 7157c478bd9Sstevel@tonic-gate startup_end(); 7167c478bd9Sstevel@tonic-gate progressbar_start(); 7177c478bd9Sstevel@tonic-gate } 7187c478bd9Sstevel@tonic-gate 7197c478bd9Sstevel@tonic-gate static void 7207c478bd9Sstevel@tonic-gate startup_init() 7217c478bd9Sstevel@tonic-gate { 7227c478bd9Sstevel@tonic-gate PRM_POINT("startup_init() starting..."); 7237c478bd9Sstevel@tonic-gate 7247c478bd9Sstevel@tonic-gate /* 7257c478bd9Sstevel@tonic-gate * Complete the extraction of cpuid data 7267c478bd9Sstevel@tonic-gate */ 7277c478bd9Sstevel@tonic-gate cpuid_pass2(CPU); 7287c478bd9Sstevel@tonic-gate 7297c478bd9Sstevel@tonic-gate (void) check_boot_version(BOP_GETVERSION(bootops)); 7307c478bd9Sstevel@tonic-gate 7317c478bd9Sstevel@tonic-gate /* 7327c478bd9Sstevel@tonic-gate * Check for prom_debug in boot environment 7337c478bd9Sstevel@tonic-gate */ 7347c478bd9Sstevel@tonic-gate if (BOP_GETPROPLEN(bootops, "prom_debug") >= 0) { 7357c478bd9Sstevel@tonic-gate ++prom_debug; 7367c478bd9Sstevel@tonic-gate PRM_POINT("prom_debug found in boot enviroment"); 7377c478bd9Sstevel@tonic-gate } 7387c478bd9Sstevel@tonic-gate 7397c478bd9Sstevel@tonic-gate /* 7407c478bd9Sstevel@tonic-gate * Collect node, cpu and memory configuration information. 7417c478bd9Sstevel@tonic-gate */ 7427c478bd9Sstevel@tonic-gate get_system_configuration(); 7437c478bd9Sstevel@tonic-gate 7447c478bd9Sstevel@tonic-gate /* 7457c478bd9Sstevel@tonic-gate * Halt if this is an unsupported processor. 7467c478bd9Sstevel@tonic-gate */ 7477c478bd9Sstevel@tonic-gate if (x86_type == X86_TYPE_486 || x86_type == X86_TYPE_CYRIX_486) { 7487c478bd9Sstevel@tonic-gate printf("\n486 processor (\"%s\") detected.\n", 7497c478bd9Sstevel@tonic-gate CPU->cpu_brandstr); 7507c478bd9Sstevel@tonic-gate halt("This processor is not supported by this release " 7517c478bd9Sstevel@tonic-gate "of Solaris."); 7527c478bd9Sstevel@tonic-gate } 7537c478bd9Sstevel@tonic-gate 7547c478bd9Sstevel@tonic-gate PRM_POINT("startup_init() done"); 7557c478bd9Sstevel@tonic-gate } 7567c478bd9Sstevel@tonic-gate 7577c478bd9Sstevel@tonic-gate /* 7587c478bd9Sstevel@tonic-gate * Callback for copy_memlist_filter() to filter nucleus, kadb/kmdb, (ie. 7597c478bd9Sstevel@tonic-gate * everything mapped above KERNEL_TEXT) pages from phys_avail. Note it 7607c478bd9Sstevel@tonic-gate * also filters out physical page zero. There is some reliance on the 7617c478bd9Sstevel@tonic-gate * boot loader allocating only a few contiguous physical memory chunks. 7627c478bd9Sstevel@tonic-gate */ 7637c478bd9Sstevel@tonic-gate static void 7647c478bd9Sstevel@tonic-gate avail_filter(uint64_t *addr, uint64_t *size) 7657c478bd9Sstevel@tonic-gate { 7667c478bd9Sstevel@tonic-gate uintptr_t va; 7677c478bd9Sstevel@tonic-gate uintptr_t next_va; 7687c478bd9Sstevel@tonic-gate pfn_t pfn; 7697c478bd9Sstevel@tonic-gate uint64_t pfn_addr; 7707c478bd9Sstevel@tonic-gate uint64_t pfn_eaddr; 7717c478bd9Sstevel@tonic-gate uint_t prot; 7727c478bd9Sstevel@tonic-gate size_t len; 7737c478bd9Sstevel@tonic-gate uint_t change; 7747c478bd9Sstevel@tonic-gate 7757c478bd9Sstevel@tonic-gate if (prom_debug) 7767c478bd9Sstevel@tonic-gate prom_printf("\tFilter: in: a=%" PRIx64 ", s=%" PRIx64 "\n", 7777c478bd9Sstevel@tonic-gate *addr, *size); 7787c478bd9Sstevel@tonic-gate 7797c478bd9Sstevel@tonic-gate /* 7807c478bd9Sstevel@tonic-gate * page zero is required for BIOS.. never make it available 7817c478bd9Sstevel@tonic-gate */ 7827c478bd9Sstevel@tonic-gate if (*addr == 0) { 7837c478bd9Sstevel@tonic-gate *addr += MMU_PAGESIZE; 7847c478bd9Sstevel@tonic-gate *size -= MMU_PAGESIZE; 7857c478bd9Sstevel@tonic-gate } 7867c478bd9Sstevel@tonic-gate 7877c478bd9Sstevel@tonic-gate /* 788ae115bc7Smrj * First we trim from the front of the range. Since kbm_probe() 7897c478bd9Sstevel@tonic-gate * walks ranges in virtual order, but addr/size are physical, we need 7907c478bd9Sstevel@tonic-gate * to the list until no changes are seen. This deals with the case 7917c478bd9Sstevel@tonic-gate * where page "p" is mapped at v, page "p + PAGESIZE" is mapped at w 7927c478bd9Sstevel@tonic-gate * but w < v. 7937c478bd9Sstevel@tonic-gate */ 7947c478bd9Sstevel@tonic-gate do { 7957c478bd9Sstevel@tonic-gate change = 0; 7967c478bd9Sstevel@tonic-gate for (va = KERNEL_TEXT; 797ae115bc7Smrj *size > 0 && kbm_probe(&va, &len, &pfn, &prot) != 0; 7987c478bd9Sstevel@tonic-gate va = next_va) { 7997c478bd9Sstevel@tonic-gate 8007c478bd9Sstevel@tonic-gate next_va = va + len; 801ae115bc7Smrj pfn_addr = pfn_to_pa(pfn); 8027c478bd9Sstevel@tonic-gate pfn_eaddr = pfn_addr + len; 8037c478bd9Sstevel@tonic-gate 8047c478bd9Sstevel@tonic-gate if (pfn_addr <= *addr && pfn_eaddr > *addr) { 8057c478bd9Sstevel@tonic-gate change = 1; 8067c478bd9Sstevel@tonic-gate while (*size > 0 && len > 0) { 8077c478bd9Sstevel@tonic-gate *addr += MMU_PAGESIZE; 8087c478bd9Sstevel@tonic-gate *size -= MMU_PAGESIZE; 8097c478bd9Sstevel@tonic-gate len -= MMU_PAGESIZE; 8107c478bd9Sstevel@tonic-gate } 8117c478bd9Sstevel@tonic-gate } 8127c478bd9Sstevel@tonic-gate } 8137c478bd9Sstevel@tonic-gate if (change && prom_debug) 8147c478bd9Sstevel@tonic-gate prom_printf("\t\ttrim: a=%" PRIx64 ", s=%" PRIx64 "\n", 8157c478bd9Sstevel@tonic-gate *addr, *size); 8167c478bd9Sstevel@tonic-gate } while (change); 8177c478bd9Sstevel@tonic-gate 8187c478bd9Sstevel@tonic-gate /* 8197c478bd9Sstevel@tonic-gate * Trim pages from the end of the range. 8207c478bd9Sstevel@tonic-gate */ 8217c478bd9Sstevel@tonic-gate for (va = KERNEL_TEXT; 822ae115bc7Smrj *size > 0 && kbm_probe(&va, &len, &pfn, &prot) != 0; 8237c478bd9Sstevel@tonic-gate va = next_va) { 8247c478bd9Sstevel@tonic-gate 8257c478bd9Sstevel@tonic-gate next_va = va + len; 826ae115bc7Smrj pfn_addr = pfn_to_pa(pfn); 8277c478bd9Sstevel@tonic-gate 8287c478bd9Sstevel@tonic-gate if (pfn_addr >= *addr && pfn_addr < *addr + *size) 8297c478bd9Sstevel@tonic-gate *size = pfn_addr - *addr; 8307c478bd9Sstevel@tonic-gate } 8317c478bd9Sstevel@tonic-gate 8327c478bd9Sstevel@tonic-gate if (prom_debug) 8337c478bd9Sstevel@tonic-gate prom_printf("\tFilter out: a=%" PRIx64 ", s=%" PRIx64 "\n", 8347c478bd9Sstevel@tonic-gate *addr, *size); 8357c478bd9Sstevel@tonic-gate } 8367c478bd9Sstevel@tonic-gate 8377c478bd9Sstevel@tonic-gate static void 8387c478bd9Sstevel@tonic-gate kpm_init() 8397c478bd9Sstevel@tonic-gate { 8407c478bd9Sstevel@tonic-gate struct segkpm_crargs b; 8417c478bd9Sstevel@tonic-gate 8427c478bd9Sstevel@tonic-gate /* 8437c478bd9Sstevel@tonic-gate * These variables were all designed for sfmmu in which segkpm is 8447c478bd9Sstevel@tonic-gate * mapped using a single pagesize - either 8KB or 4MB. On x86, we 8457c478bd9Sstevel@tonic-gate * might use 2+ page sizes on a single machine, so none of these 8467c478bd9Sstevel@tonic-gate * variables have a single correct value. They are set up as if we 8477c478bd9Sstevel@tonic-gate * always use a 4KB pagesize, which should do no harm. In the long 8487c478bd9Sstevel@tonic-gate * run, we should get rid of KPM's assumption that only a single 8497c478bd9Sstevel@tonic-gate * pagesize is used. 8507c478bd9Sstevel@tonic-gate */ 8517c478bd9Sstevel@tonic-gate kpm_pgshft = MMU_PAGESHIFT; 8527c478bd9Sstevel@tonic-gate kpm_pgsz = MMU_PAGESIZE; 8537c478bd9Sstevel@tonic-gate kpm_pgoff = MMU_PAGEOFFSET; 8547c478bd9Sstevel@tonic-gate kpmp2pshft = 0; 8557c478bd9Sstevel@tonic-gate kpmpnpgs = 1; 8567c478bd9Sstevel@tonic-gate ASSERT(((uintptr_t)kpm_vbase & (kpm_pgsz - 1)) == 0); 8577c478bd9Sstevel@tonic-gate 8587c478bd9Sstevel@tonic-gate PRM_POINT("about to create segkpm"); 8597c478bd9Sstevel@tonic-gate rw_enter(&kas.a_lock, RW_WRITER); 8607c478bd9Sstevel@tonic-gate 8617c478bd9Sstevel@tonic-gate if (seg_attach(&kas, kpm_vbase, kpm_size, segkpm) < 0) 8627c478bd9Sstevel@tonic-gate panic("cannot attach segkpm"); 8637c478bd9Sstevel@tonic-gate 8647c478bd9Sstevel@tonic-gate b.prot = PROT_READ | PROT_WRITE; 8657c478bd9Sstevel@tonic-gate b.nvcolors = 1; 8667c478bd9Sstevel@tonic-gate 8677c478bd9Sstevel@tonic-gate if (segkpm_create(segkpm, (caddr_t)&b) != 0) 8687c478bd9Sstevel@tonic-gate panic("segkpm_create segkpm"); 8697c478bd9Sstevel@tonic-gate 8707c478bd9Sstevel@tonic-gate rw_exit(&kas.a_lock); 871ae115bc7Smrj } 8727c478bd9Sstevel@tonic-gate 873ae115bc7Smrj /* 874ae115bc7Smrj * The debug info page provides enough information to allow external 875ae115bc7Smrj * inspectors (e.g. when running under a hypervisor) to bootstrap 876ae115bc7Smrj * themselves into allowing full-blown kernel debugging. 877ae115bc7Smrj */ 878ae115bc7Smrj static void 879ae115bc7Smrj init_debug_info(void) 880ae115bc7Smrj { 881ae115bc7Smrj caddr_t mem; 882ae115bc7Smrj debug_info_t *di; 883ae115bc7Smrj 884ae115bc7Smrj #ifndef __lint 885ae115bc7Smrj ASSERT(sizeof (debug_info_t) < MMU_PAGESIZE); 886ae115bc7Smrj #endif 887ae115bc7Smrj 888ae115bc7Smrj mem = BOP_ALLOC(bootops, (caddr_t)DEBUG_INFO_VA, MMU_PAGESIZE, 889ae115bc7Smrj MMU_PAGESIZE); 890ae115bc7Smrj 891ae115bc7Smrj if (mem != (caddr_t)DEBUG_INFO_VA) 892ae115bc7Smrj panic("BOP_ALLOC() failed"); 893ae115bc7Smrj bzero(mem, MMU_PAGESIZE); 894ae115bc7Smrj 895ae115bc7Smrj di = (debug_info_t *)mem; 896ae115bc7Smrj 897ae115bc7Smrj di->di_magic = DEBUG_INFO_MAGIC; 898ae115bc7Smrj di->di_version = DEBUG_INFO_VERSION; 899843e1988Sjohnlev di->di_modules = (uintptr_t)&modules; 900843e1988Sjohnlev di->di_s_text = (uintptr_t)s_text; 901843e1988Sjohnlev di->di_e_text = (uintptr_t)e_text; 902843e1988Sjohnlev di->di_s_data = (uintptr_t)s_data; 903843e1988Sjohnlev di->di_e_data = (uintptr_t)e_data; 904843e1988Sjohnlev di->di_hat_htable_off = offsetof(hat_t, hat_htable); 905843e1988Sjohnlev di->di_ht_pfn_off = offsetof(htable_t, ht_pfn); 9067c478bd9Sstevel@tonic-gate } 9077c478bd9Sstevel@tonic-gate 9087c478bd9Sstevel@tonic-gate /* 909ae115bc7Smrj * Build the memlists and other kernel essential memory system data structures. 910ae115bc7Smrj * This is everything at valloc_base. 9117c478bd9Sstevel@tonic-gate */ 9127c478bd9Sstevel@tonic-gate static void 9137c478bd9Sstevel@tonic-gate startup_memlist(void) 9147c478bd9Sstevel@tonic-gate { 9157c478bd9Sstevel@tonic-gate size_t memlist_sz; 9167c478bd9Sstevel@tonic-gate size_t memseg_sz; 9177c478bd9Sstevel@tonic-gate size_t pagehash_sz; 9187c478bd9Sstevel@tonic-gate size_t pp_sz; 9197c478bd9Sstevel@tonic-gate uintptr_t va; 9207c478bd9Sstevel@tonic-gate size_t len; 9217c478bd9Sstevel@tonic-gate uint_t prot; 9227c478bd9Sstevel@tonic-gate pfn_t pfn; 9237c478bd9Sstevel@tonic-gate int memblocks; 9247c478bd9Sstevel@tonic-gate caddr_t pagecolor_mem; 9257c478bd9Sstevel@tonic-gate size_t pagecolor_memsz; 9267c478bd9Sstevel@tonic-gate caddr_t page_ctrs_mem; 9277c478bd9Sstevel@tonic-gate size_t page_ctrs_size; 928d7d93655Sblakej size_t pse_table_alloc_size; 9297c478bd9Sstevel@tonic-gate struct memlist *current; 9307c478bd9Sstevel@tonic-gate extern void startup_build_mem_nodes(struct memlist *); 9317c478bd9Sstevel@tonic-gate 9327c478bd9Sstevel@tonic-gate /* XX64 fix these - they should be in include files */ 9337c478bd9Sstevel@tonic-gate extern size_t page_coloring_init(uint_t, int, int); 9347c478bd9Sstevel@tonic-gate extern void page_coloring_setup(caddr_t); 9357c478bd9Sstevel@tonic-gate 9367c478bd9Sstevel@tonic-gate PRM_POINT("startup_memlist() starting..."); 9377c478bd9Sstevel@tonic-gate 9387c478bd9Sstevel@tonic-gate /* 9397c478bd9Sstevel@tonic-gate * Use leftover large page nucleus text/data space for loadable modules. 9407c478bd9Sstevel@tonic-gate * Use at most MODTEXT/MODDATA. 9417c478bd9Sstevel@tonic-gate */ 942ae115bc7Smrj len = kbm_nucleus_size; 943ae115bc7Smrj ASSERT(len > MMU_PAGESIZE); 9447c478bd9Sstevel@tonic-gate 945ae115bc7Smrj moddata = (caddr_t)ROUND_UP_PAGE(e_data); 946ae115bc7Smrj e_moddata = (caddr_t)P2ROUNDUP((uintptr_t)e_data, (uintptr_t)len); 947ae115bc7Smrj if (e_moddata - moddata > MODDATA) 948ae115bc7Smrj e_moddata = moddata + MODDATA; 9497c478bd9Sstevel@tonic-gate 950ae115bc7Smrj modtext = (caddr_t)ROUND_UP_PAGE(e_text); 951ae115bc7Smrj e_modtext = (caddr_t)P2ROUNDUP((uintptr_t)e_text, (uintptr_t)len); 952ae115bc7Smrj if (e_modtext - modtext > MODTEXT) 953ae115bc7Smrj e_modtext = modtext + MODTEXT; 9547c478bd9Sstevel@tonic-gate 9557c478bd9Sstevel@tonic-gate econtig = e_moddata; 9567c478bd9Sstevel@tonic-gate 9577c478bd9Sstevel@tonic-gate PRM_DEBUG(modtext); 9587c478bd9Sstevel@tonic-gate PRM_DEBUG(e_modtext); 9597c478bd9Sstevel@tonic-gate PRM_DEBUG(moddata); 9607c478bd9Sstevel@tonic-gate PRM_DEBUG(e_moddata); 9617c478bd9Sstevel@tonic-gate PRM_DEBUG(econtig); 9627c478bd9Sstevel@tonic-gate 9637c478bd9Sstevel@tonic-gate /* 964ae115bc7Smrj * Examine the boot loader physical memory map to find out: 9657c478bd9Sstevel@tonic-gate * - total memory in system - physinstalled 9667c478bd9Sstevel@tonic-gate * - the max physical address - physmax 967ae115bc7Smrj * - the number of discontiguous segments of memory. 9687c478bd9Sstevel@tonic-gate */ 9697c478bd9Sstevel@tonic-gate if (prom_debug) 970ae115bc7Smrj print_memlist("boot physinstalled", 9717c478bd9Sstevel@tonic-gate bootops->boot_mem->physinstalled); 9727c478bd9Sstevel@tonic-gate installed_top_size(bootops->boot_mem->physinstalled, &physmax, 9737c478bd9Sstevel@tonic-gate &physinstalled, &memblocks); 9747c478bd9Sstevel@tonic-gate PRM_DEBUG(physmax); 9757c478bd9Sstevel@tonic-gate PRM_DEBUG(physinstalled); 9767c478bd9Sstevel@tonic-gate PRM_DEBUG(memblocks); 9777c478bd9Sstevel@tonic-gate 9787c478bd9Sstevel@tonic-gate /* 9797c478bd9Sstevel@tonic-gate * Initialize hat's mmu parameters. 9807c478bd9Sstevel@tonic-gate * Check for enforce-prot-exec in boot environment. It's used to 9817c478bd9Sstevel@tonic-gate * enable/disable support for the page table entry NX bit. 9827c478bd9Sstevel@tonic-gate * The default is to enforce PROT_EXEC on processors that support NX. 9837c478bd9Sstevel@tonic-gate * Boot seems to round up the "len", but 8 seems to be big enough. 9847c478bd9Sstevel@tonic-gate */ 9857c478bd9Sstevel@tonic-gate mmu_init(); 9867c478bd9Sstevel@tonic-gate 9877c478bd9Sstevel@tonic-gate #ifdef __i386 9887c478bd9Sstevel@tonic-gate /* 9897c478bd9Sstevel@tonic-gate * physmax is lowered if there is more memory than can be 9907c478bd9Sstevel@tonic-gate * physically addressed in 32 bit (PAE/non-PAE) modes. 9917c478bd9Sstevel@tonic-gate */ 9927c478bd9Sstevel@tonic-gate if (mmu.pae_hat) { 9937c478bd9Sstevel@tonic-gate if (PFN_ABOVE64G(physmax)) { 9947c478bd9Sstevel@tonic-gate physinstalled -= (physmax - (PFN_64G - 1)); 9957c478bd9Sstevel@tonic-gate physmax = PFN_64G - 1; 9967c478bd9Sstevel@tonic-gate } 9977c478bd9Sstevel@tonic-gate } else { 9987c478bd9Sstevel@tonic-gate if (PFN_ABOVE4G(physmax)) { 9997c478bd9Sstevel@tonic-gate physinstalled -= (physmax - (PFN_4G - 1)); 10007c478bd9Sstevel@tonic-gate physmax = PFN_4G - 1; 10017c478bd9Sstevel@tonic-gate } 10027c478bd9Sstevel@tonic-gate } 10037c478bd9Sstevel@tonic-gate #endif 10047c478bd9Sstevel@tonic-gate 10057c478bd9Sstevel@tonic-gate startup_build_mem_nodes(bootops->boot_mem->physinstalled); 10067c478bd9Sstevel@tonic-gate 10077c478bd9Sstevel@tonic-gate if (BOP_GETPROPLEN(bootops, "enforce-prot-exec") >= 0) { 10087c478bd9Sstevel@tonic-gate int len = BOP_GETPROPLEN(bootops, "enforce-prot-exec"); 10097c478bd9Sstevel@tonic-gate char value[8]; 10107c478bd9Sstevel@tonic-gate 10117c478bd9Sstevel@tonic-gate if (len < 8) 10127c478bd9Sstevel@tonic-gate (void) BOP_GETPROP(bootops, "enforce-prot-exec", value); 10137c478bd9Sstevel@tonic-gate else 10147c478bd9Sstevel@tonic-gate (void) strcpy(value, ""); 10157c478bd9Sstevel@tonic-gate if (strcmp(value, "off") == 0) 10167c478bd9Sstevel@tonic-gate mmu.pt_nx = 0; 10177c478bd9Sstevel@tonic-gate } 10187c478bd9Sstevel@tonic-gate PRM_DEBUG(mmu.pt_nx); 10197c478bd9Sstevel@tonic-gate 10207c478bd9Sstevel@tonic-gate /* 10217c478bd9Sstevel@tonic-gate * We will need page_t's for every page in the system, except for 10227c478bd9Sstevel@tonic-gate * memory mapped at or above above the start of the kernel text segment. 10237c478bd9Sstevel@tonic-gate * 10247c478bd9Sstevel@tonic-gate * pages above e_modtext are attributed to kernel debugger (obp_pages) 10257c478bd9Sstevel@tonic-gate */ 10267c478bd9Sstevel@tonic-gate npages = physinstalled - 1; /* avail_filter() skips page 0, so "- 1" */ 10277c478bd9Sstevel@tonic-gate obp_pages = 0; 10287c478bd9Sstevel@tonic-gate va = KERNEL_TEXT; 1029ae115bc7Smrj while (kbm_probe(&va, &len, &pfn, &prot) != 0) { 10307c478bd9Sstevel@tonic-gate npages -= len >> MMU_PAGESHIFT; 10317c478bd9Sstevel@tonic-gate if (va >= (uintptr_t)e_moddata) 10327c478bd9Sstevel@tonic-gate obp_pages += len >> MMU_PAGESHIFT; 10337c478bd9Sstevel@tonic-gate va += len; 10347c478bd9Sstevel@tonic-gate } 10357c478bd9Sstevel@tonic-gate PRM_DEBUG(npages); 10367c478bd9Sstevel@tonic-gate PRM_DEBUG(obp_pages); 10377c478bd9Sstevel@tonic-gate 10387c478bd9Sstevel@tonic-gate /* 103935b1ab99Sjosephb * If physmem is patched to be non-zero, use it instead of the computed 104035b1ab99Sjosephb * value unless it is larger than the actual amount of memory on hand. 10417c478bd9Sstevel@tonic-gate */ 10424944b02eSkchow if (physmem == 0 || physmem > npages) { 10437c478bd9Sstevel@tonic-gate physmem = npages; 10444944b02eSkchow } else if (physmem < npages) { 1045f53ad214Skchow orig_npages = npages; 10467c478bd9Sstevel@tonic-gate npages = physmem; 10474944b02eSkchow } 10487c478bd9Sstevel@tonic-gate PRM_DEBUG(physmem); 10497c478bd9Sstevel@tonic-gate 10507c478bd9Sstevel@tonic-gate /* 10517c478bd9Sstevel@tonic-gate * We now compute the sizes of all the initial allocations for 10527c478bd9Sstevel@tonic-gate * structures the kernel needs in order do kmem_alloc(). These 10537c478bd9Sstevel@tonic-gate * include: 10547c478bd9Sstevel@tonic-gate * memsegs 10557c478bd9Sstevel@tonic-gate * memlists 10567c478bd9Sstevel@tonic-gate * page hash table 10577c478bd9Sstevel@tonic-gate * page_t's 10587c478bd9Sstevel@tonic-gate * page coloring data structs 10597c478bd9Sstevel@tonic-gate */ 10607c478bd9Sstevel@tonic-gate memseg_sz = sizeof (struct memseg) * (memblocks + POSS_NEW_FRAGMENTS); 10617c478bd9Sstevel@tonic-gate ADD_TO_ALLOCATIONS(memseg_base, memseg_sz); 10627c478bd9Sstevel@tonic-gate PRM_DEBUG(memseg_sz); 10637c478bd9Sstevel@tonic-gate 10647c478bd9Sstevel@tonic-gate /* 1065ae115bc7Smrj * Reserve space for memlists. There's no real good way to know exactly 1066ae115bc7Smrj * how much room we'll need, but this should be a good upper bound. 10677c478bd9Sstevel@tonic-gate */ 10687c478bd9Sstevel@tonic-gate memlist_sz = ROUND_UP_PAGE(2 * sizeof (struct memlist) * 10697c478bd9Sstevel@tonic-gate (memblocks + POSS_NEW_FRAGMENTS)); 10707c478bd9Sstevel@tonic-gate ADD_TO_ALLOCATIONS(memlist, memlist_sz); 10717c478bd9Sstevel@tonic-gate PRM_DEBUG(memlist_sz); 10727c478bd9Sstevel@tonic-gate 10737c478bd9Sstevel@tonic-gate /* 10747c478bd9Sstevel@tonic-gate * The page structure hash table size is a power of 2 10757c478bd9Sstevel@tonic-gate * such that the average hash chain length is PAGE_HASHAVELEN. 10767c478bd9Sstevel@tonic-gate */ 10777c478bd9Sstevel@tonic-gate page_hashsz = npages / PAGE_HASHAVELEN; 10787c478bd9Sstevel@tonic-gate page_hashsz = 1 << highbit(page_hashsz); 10797c478bd9Sstevel@tonic-gate pagehash_sz = sizeof (struct page *) * page_hashsz; 10807c478bd9Sstevel@tonic-gate ADD_TO_ALLOCATIONS(page_hash, pagehash_sz); 10817c478bd9Sstevel@tonic-gate PRM_DEBUG(pagehash_sz); 10827c478bd9Sstevel@tonic-gate 10837c478bd9Sstevel@tonic-gate /* 1084ae115bc7Smrj * Set aside room for the page structures themselves. 10857c478bd9Sstevel@tonic-gate */ 1086ae115bc7Smrj PRM_DEBUG(npages); 1087ae115bc7Smrj pp_sz = sizeof (struct page) * npages; 10887c478bd9Sstevel@tonic-gate ADD_TO_ALLOCATIONS(pp_base, pp_sz); 10897c478bd9Sstevel@tonic-gate PRM_DEBUG(pp_sz); 10907c478bd9Sstevel@tonic-gate 10917c478bd9Sstevel@tonic-gate /* 10927c478bd9Sstevel@tonic-gate * determine l2 cache info and memory size for page coloring 10937c478bd9Sstevel@tonic-gate */ 10947c478bd9Sstevel@tonic-gate (void) getl2cacheinfo(CPU, 10957c478bd9Sstevel@tonic-gate &l2cache_sz, &l2cache_linesz, &l2cache_assoc); 10967c478bd9Sstevel@tonic-gate pagecolor_memsz = 10977c478bd9Sstevel@tonic-gate page_coloring_init(l2cache_sz, l2cache_linesz, l2cache_assoc); 10987c478bd9Sstevel@tonic-gate ADD_TO_ALLOCATIONS(pagecolor_mem, pagecolor_memsz); 10997c478bd9Sstevel@tonic-gate PRM_DEBUG(pagecolor_memsz); 11007c478bd9Sstevel@tonic-gate 11017c478bd9Sstevel@tonic-gate page_ctrs_size = page_ctrs_sz(); 11027c478bd9Sstevel@tonic-gate ADD_TO_ALLOCATIONS(page_ctrs_mem, page_ctrs_size); 11037c478bd9Sstevel@tonic-gate PRM_DEBUG(page_ctrs_size); 11047c478bd9Sstevel@tonic-gate 1105d7d93655Sblakej /* 1106d7d93655Sblakej * Allocate the array that protects pp->p_selock. 1107d7d93655Sblakej */ 1108d7d93655Sblakej pse_shift = size_pse_array(physmem, max_ncpus); 1109d7d93655Sblakej pse_table_size = 1 << pse_shift; 1110d7d93655Sblakej pse_table_alloc_size = pse_table_size * sizeof (pad_mutex_t); 1111d7d93655Sblakej ADD_TO_ALLOCATIONS(pse_mutex, pse_table_alloc_size); 1112d7d93655Sblakej 1113ae115bc7Smrj #if defined(__amd64) 11147c478bd9Sstevel@tonic-gate valloc_sz = ROUND_UP_LPAGE(valloc_sz); 1115ae115bc7Smrj valloc_base = VALLOC_BASE; 111635b1ab99Sjosephb 111735b1ab99Sjosephb /* 111835b1ab99Sjosephb * The default values of VALLOC_BASE and SEGKPM_BASE should work 111935b1ab99Sjosephb * for values of physmax up to 1 Terabyte. They need adjusting when 112035b1ab99Sjosephb * memory is at addresses above 1 TB. 112135b1ab99Sjosephb */ 112235b1ab99Sjosephb if (physmax + 1 > mmu_btop(TERABYTE)) { 112335b1ab99Sjosephb uint64_t kpm_resv_amount = mmu_ptob(physmax + 1); 112435b1ab99Sjosephb 112535b1ab99Sjosephb /* Round to largest possible pagesize for now */ 112635b1ab99Sjosephb kpm_resv_amount = P2ROUNDUP(kpm_resv_amount, ONE_GIG); 112735b1ab99Sjosephb 112835b1ab99Sjosephb segkpm_base = -(2 * kpm_resv_amount); /* down from top VA */ 112935b1ab99Sjosephb 113035b1ab99Sjosephb /* make sure we leave some space for user apps above hole */ 113135b1ab99Sjosephb segkpm_base = MAX(segkpm_base, AMD64_VA_HOLE_END + TERABYTE); 113235b1ab99Sjosephb if (segkpm_base > SEGKPM_BASE) 113335b1ab99Sjosephb segkpm_base = SEGKPM_BASE; 113435b1ab99Sjosephb PRM_DEBUG(segkpm_base); 113535b1ab99Sjosephb 113635b1ab99Sjosephb valloc_base = segkpm_base + kpm_resv_amount; 113735b1ab99Sjosephb PRM_DEBUG(valloc_base); 113835b1ab99Sjosephb } 1139ae115bc7Smrj #else /* __i386 */ 1140ae115bc7Smrj valloc_base = (uintptr_t)(MISC_VA_BASE - valloc_sz); 1141ae115bc7Smrj valloc_base = P2ALIGN(valloc_base, mmu.level_size[1]); 11427c478bd9Sstevel@tonic-gate PRM_DEBUG(valloc_base); 114335b1ab99Sjosephb #endif /* __i386 */ 1144ae115bc7Smrj 1145ae115bc7Smrj /* 1146ae115bc7Smrj * do all the initial allocations 1147ae115bc7Smrj */ 1148ae115bc7Smrj perform_allocations(); 1149ae115bc7Smrj 1150ae115bc7Smrj /* 1151ae115bc7Smrj * Build phys_install and phys_avail in kernel memspace. 1152ae115bc7Smrj * - phys_install should be all memory in the system. 1153ae115bc7Smrj * - phys_avail is phys_install minus any memory mapped before this 1154ae115bc7Smrj * point above KERNEL_TEXT. 1155ae115bc7Smrj */ 1156ae115bc7Smrj current = phys_install = memlist; 1157ae115bc7Smrj copy_memlist_filter(bootops->boot_mem->physinstalled, ¤t, NULL); 1158ae115bc7Smrj if ((caddr_t)current > (caddr_t)memlist + memlist_sz) 1159ae115bc7Smrj panic("physinstalled was too big!"); 1160ae115bc7Smrj if (prom_debug) 1161ae115bc7Smrj print_memlist("phys_install", phys_install); 1162ae115bc7Smrj 1163ae115bc7Smrj phys_avail = current; 1164ae115bc7Smrj PRM_POINT("Building phys_avail:\n"); 1165ae115bc7Smrj copy_memlist_filter(bootops->boot_mem->physinstalled, ¤t, 1166ae115bc7Smrj avail_filter); 1167ae115bc7Smrj if ((caddr_t)current > (caddr_t)memlist + memlist_sz) 1168ae115bc7Smrj panic("physavail was too big!"); 1169ae115bc7Smrj if (prom_debug) 1170ae115bc7Smrj print_memlist("phys_avail", phys_avail); 1171ae115bc7Smrj 1172ae115bc7Smrj /* 1173ae115bc7Smrj * setup page coloring 1174ae115bc7Smrj */ 1175ae115bc7Smrj page_coloring_setup(pagecolor_mem); 1176ae115bc7Smrj page_lock_init(); /* currently a no-op */ 1177ae115bc7Smrj 1178ae115bc7Smrj /* 1179ae115bc7Smrj * free page list counters 1180ae115bc7Smrj */ 1181ae115bc7Smrj (void) page_ctrs_alloc(page_ctrs_mem); 1182ae115bc7Smrj 118306fb6a36Sdv /* 118406fb6a36Sdv * Size the pcf array based on the number of cpus in the box at 118506fb6a36Sdv * boot time. 118606fb6a36Sdv */ 118706fb6a36Sdv 118806fb6a36Sdv pcf_init(); 118906fb6a36Sdv 1190ae115bc7Smrj /* 1191ae115bc7Smrj * Initialize the page structures from the memory lists. 1192ae115bc7Smrj */ 1193ae115bc7Smrj availrmem_initial = availrmem = freemem = 0; 1194ae115bc7Smrj PRM_POINT("Calling kphysm_init()..."); 1195ae115bc7Smrj npages = kphysm_init(pp_base, npages); 1196ae115bc7Smrj PRM_POINT("kphysm_init() done"); 1197ae115bc7Smrj PRM_DEBUG(npages); 1198ae115bc7Smrj 1199ae115bc7Smrj init_debug_info(); 1200ae115bc7Smrj 1201ae115bc7Smrj /* 1202ae115bc7Smrj * Now that page_t's have been initialized, remove all the 1203ae115bc7Smrj * initial allocation pages from the kernel free page lists. 1204ae115bc7Smrj */ 1205ae115bc7Smrj boot_mapin((caddr_t)valloc_base, valloc_sz); 12060cfdb603Sjosephb boot_mapin((caddr_t)MISC_VA_BASE, MISC_VA_SIZE); 1207ae115bc7Smrj PRM_POINT("startup_memlist() done"); 1208ae115bc7Smrj 1209ae115bc7Smrj PRM_DEBUG(valloc_sz); 12102cb27123Saguzovsk 1211567d55e1Saguzovsk #if defined(__amd64) 1212567d55e1Saguzovsk if ((availrmem >> (30 - MMU_PAGESHIFT)) >= 1213567d55e1Saguzovsk textrepl_min_gb && l2cache_sz <= 2 << 20) { 1214567d55e1Saguzovsk extern size_t textrepl_size_thresh; 12152cb27123Saguzovsk textrepl_size_thresh = (16 << 20) - 1; 12162cb27123Saguzovsk } 1217567d55e1Saguzovsk #endif 1218ae115bc7Smrj } 1219ae115bc7Smrj 1220ae115bc7Smrj /* 1221ae115bc7Smrj * Layout the kernel's part of address space and initialize kmem allocator. 1222ae115bc7Smrj */ 1223ae115bc7Smrj static void 1224ae115bc7Smrj startup_kmem(void) 1225ae115bc7Smrj { 1226932dc8e5Sdp extern void page_set_colorequiv_arr(void); 1227932dc8e5Sdp 1228ae115bc7Smrj PRM_POINT("startup_kmem() starting..."); 12297c478bd9Sstevel@tonic-gate 12307c478bd9Sstevel@tonic-gate #if defined(__amd64) 12317c478bd9Sstevel@tonic-gate if (eprom_kernelbase && eprom_kernelbase != KERNELBASE) 12327c478bd9Sstevel@tonic-gate cmn_err(CE_NOTE, "!kernelbase cannot be changed on 64-bit " 12337c478bd9Sstevel@tonic-gate "systems."); 123435b1ab99Sjosephb kernelbase = segkpm_base - KERNEL_REDZONE_SIZE; 12357c478bd9Sstevel@tonic-gate core_base = (uintptr_t)COREHEAP_BASE; 1236ae115bc7Smrj core_size = (size_t)MISC_VA_BASE - COREHEAP_BASE; 12377c478bd9Sstevel@tonic-gate #else /* __i386 */ 12387c478bd9Sstevel@tonic-gate /* 12397c478bd9Sstevel@tonic-gate * We configure kernelbase based on: 12407c478bd9Sstevel@tonic-gate * 12417c478bd9Sstevel@tonic-gate * 1. user specified kernelbase via eeprom command. Value cannot exceed 12427c478bd9Sstevel@tonic-gate * KERNELBASE_MAX. we large page align eprom_kernelbase 12437c478bd9Sstevel@tonic-gate * 12447c478bd9Sstevel@tonic-gate * 2. Default to KERNELBASE and adjust to 2X less the size for page_t. 12457c478bd9Sstevel@tonic-gate * On large memory systems we must lower kernelbase to allow 12467c478bd9Sstevel@tonic-gate * enough room for page_t's for all of memory. 12477c478bd9Sstevel@tonic-gate * 12487c478bd9Sstevel@tonic-gate * The value set here, might be changed a little later. 12497c478bd9Sstevel@tonic-gate */ 12507c478bd9Sstevel@tonic-gate if (eprom_kernelbase) { 12517c478bd9Sstevel@tonic-gate kernelbase = eprom_kernelbase & mmu.level_mask[1]; 12527c478bd9Sstevel@tonic-gate if (kernelbase > KERNELBASE_MAX) 12537c478bd9Sstevel@tonic-gate kernelbase = KERNELBASE_MAX; 12547c478bd9Sstevel@tonic-gate } else { 12557c478bd9Sstevel@tonic-gate kernelbase = (uintptr_t)KERNELBASE; 12567c478bd9Sstevel@tonic-gate kernelbase -= ROUND_UP_4MEG(2 * valloc_sz); 12577c478bd9Sstevel@tonic-gate } 12587c478bd9Sstevel@tonic-gate ASSERT((kernelbase & mmu.level_offset[1]) == 0); 1259ae115bc7Smrj core_base = valloc_base; 12607c478bd9Sstevel@tonic-gate core_size = 0; 1261ae115bc7Smrj #endif /* __i386 */ 12627c478bd9Sstevel@tonic-gate 12637c478bd9Sstevel@tonic-gate PRM_DEBUG(core_base); 12647c478bd9Sstevel@tonic-gate PRM_DEBUG(core_size); 1265ae115bc7Smrj PRM_DEBUG(kernelbase); 12667c478bd9Sstevel@tonic-gate 126735b1ab99Sjosephb #if defined(__i386) 12687c478bd9Sstevel@tonic-gate segkp_fromheap = 1; 1269ae115bc7Smrj #endif /* __i386 */ 127035b1ab99Sjosephb 12717c478bd9Sstevel@tonic-gate ekernelheap = (char *)core_base; 1272ae115bc7Smrj PRM_DEBUG(ekernelheap); 12737c478bd9Sstevel@tonic-gate 12747c478bd9Sstevel@tonic-gate /* 12757c478bd9Sstevel@tonic-gate * Now that we know the real value of kernelbase, 12767c478bd9Sstevel@tonic-gate * update variables that were initialized with a value of 12777c478bd9Sstevel@tonic-gate * KERNELBASE (in common/conf/param.c). 12787c478bd9Sstevel@tonic-gate * 12797c478bd9Sstevel@tonic-gate * XXX The problem with this sort of hackery is that the 12807c478bd9Sstevel@tonic-gate * compiler just may feel like putting the const declarations 12817c478bd9Sstevel@tonic-gate * (in param.c) into the .text section. Perhaps they should 12827c478bd9Sstevel@tonic-gate * just be declared as variables there? 12837c478bd9Sstevel@tonic-gate */ 12847c478bd9Sstevel@tonic-gate 12857c478bd9Sstevel@tonic-gate *(uintptr_t *)&_kernelbase = kernelbase; 12867c478bd9Sstevel@tonic-gate *(uintptr_t *)&_userlimit = kernelbase; 1287ccbaea4fSjosephb #if defined(__amd64) 1288ccbaea4fSjosephb *(uintptr_t *)&_userlimit -= KERNELBASE - USERLIMIT; 1289ccbaea4fSjosephb #else 12907c478bd9Sstevel@tonic-gate *(uintptr_t *)&_userlimit32 = _userlimit; 12917c478bd9Sstevel@tonic-gate #endif 12927c478bd9Sstevel@tonic-gate PRM_DEBUG(_kernelbase); 12937c478bd9Sstevel@tonic-gate PRM_DEBUG(_userlimit); 12947c478bd9Sstevel@tonic-gate PRM_DEBUG(_userlimit32); 12957c478bd9Sstevel@tonic-gate 129635b1ab99Sjosephb layout_kernel_va(); 129735b1ab99Sjosephb 129835b1ab99Sjosephb #if defined(__i386) 129935b1ab99Sjosephb /* 130035b1ab99Sjosephb * If segmap is too large we can push the bottom of the kernel heap 130135b1ab99Sjosephb * higher than the base. Or worse, it could exceed the top of the 130235b1ab99Sjosephb * VA space entirely, causing it to wrap around. 130335b1ab99Sjosephb */ 130435b1ab99Sjosephb if (kernelheap >= ekernelheap || (uintptr_t)kernelheap < kernelbase) 130535b1ab99Sjosephb panic("too little address space available for kernelheap," 130635b1ab99Sjosephb " use eeprom for lower kernelbase or smaller segmapsize"); 130735b1ab99Sjosephb #endif /* __i386 */ 130835b1ab99Sjosephb 13097c478bd9Sstevel@tonic-gate /* 13107c478bd9Sstevel@tonic-gate * Initialize the kernel heap. Note 3rd argument must be > 1st. 13117c478bd9Sstevel@tonic-gate */ 131235b1ab99Sjosephb kernelheap_init(kernelheap, ekernelheap, 131335b1ab99Sjosephb kernelheap + MMU_PAGESIZE, 1314ae115bc7Smrj (void *)core_base, (void *)(core_base + core_size)); 13157c478bd9Sstevel@tonic-gate 1316843e1988Sjohnlev #if defined(__xpv) 1317843e1988Sjohnlev /* 1318843e1988Sjohnlev * Link pending events struct into cpu struct 1319843e1988Sjohnlev */ 1320843e1988Sjohnlev CPU->cpu_m.mcpu_evt_pend = &cpu0_evt_data; 1321843e1988Sjohnlev #endif 13227c478bd9Sstevel@tonic-gate /* 13237c478bd9Sstevel@tonic-gate * Initialize kernel memory allocator. 13247c478bd9Sstevel@tonic-gate */ 13257c478bd9Sstevel@tonic-gate kmem_init(); 13267c478bd9Sstevel@tonic-gate 1327932dc8e5Sdp /* 1328932dc8e5Sdp * Factor in colorequiv to check additional 'equivalent' bins 1329932dc8e5Sdp */ 1330932dc8e5Sdp page_set_colorequiv_arr(); 1331932dc8e5Sdp 13327c478bd9Sstevel@tonic-gate /* 13337c478bd9Sstevel@tonic-gate * print this out early so that we know what's going on 13347c478bd9Sstevel@tonic-gate */ 13357c478bd9Sstevel@tonic-gate cmn_err(CE_CONT, "?features: %b\n", x86_feature, FMT_X86_FEATURE); 13367c478bd9Sstevel@tonic-gate 13377c478bd9Sstevel@tonic-gate /* 13387c478bd9Sstevel@tonic-gate * Initialize bp_mapin(). 13397c478bd9Sstevel@tonic-gate */ 13407c478bd9Sstevel@tonic-gate bp_init(MMU_PAGESIZE, HAT_STORECACHING_OK); 13417c478bd9Sstevel@tonic-gate 1342f53ad214Skchow /* 1343f53ad214Skchow * orig_npages is non-zero if physmem has been configured for less 1344f53ad214Skchow * than the available memory. 1345f53ad214Skchow */ 1346f53ad214Skchow if (orig_npages) { 134735b1ab99Sjosephb cmn_err(CE_WARN, "!%slimiting physmem to 0x%lx of 0x%lx pages", 134835b1ab99Sjosephb (npages == PHYSMEM ? "Due to virtual address space " : ""), 134935b1ab99Sjosephb npages, orig_npages); 1350f53ad214Skchow } 13517c478bd9Sstevel@tonic-gate #if defined(__i386) 13527c478bd9Sstevel@tonic-gate if (eprom_kernelbase && (eprom_kernelbase != kernelbase)) 13537c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "kernelbase value, User specified 0x%lx, " 13547c478bd9Sstevel@tonic-gate "System using 0x%lx", 13557c478bd9Sstevel@tonic-gate (uintptr_t)eprom_kernelbase, (uintptr_t)kernelbase); 13567c478bd9Sstevel@tonic-gate #endif 13577c478bd9Sstevel@tonic-gate 13587c478bd9Sstevel@tonic-gate #ifdef KERNELBASE_ABI_MIN 13597c478bd9Sstevel@tonic-gate if (kernelbase < (uintptr_t)KERNELBASE_ABI_MIN) { 13607c478bd9Sstevel@tonic-gate cmn_err(CE_NOTE, "!kernelbase set to 0x%lx, system is not " 13617c478bd9Sstevel@tonic-gate "i386 ABI compliant.", (uintptr_t)kernelbase); 13627c478bd9Sstevel@tonic-gate } 13637c478bd9Sstevel@tonic-gate #endif 13647c478bd9Sstevel@tonic-gate 1365843e1988Sjohnlev #ifdef __xpv 1366843e1988Sjohnlev /* 1367843e1988Sjohnlev * Some of the xen start information has to be relocated up 1368843e1988Sjohnlev * into the kernel's permanent address space. 1369843e1988Sjohnlev */ 1370843e1988Sjohnlev PRM_POINT("calling xen_relocate_start_info()"); 1371843e1988Sjohnlev xen_relocate_start_info(); 1372843e1988Sjohnlev PRM_POINT("xen_relocate_start_info() done"); 1373843e1988Sjohnlev 1374843e1988Sjohnlev /* 1375843e1988Sjohnlev * (Update the vcpu pointer in our cpu structure to point into 1376843e1988Sjohnlev * the relocated shared info.) 1377843e1988Sjohnlev */ 1378843e1988Sjohnlev CPU->cpu_m.mcpu_vcpu_info = 1379843e1988Sjohnlev &HYPERVISOR_shared_info->vcpu_info[CPU->cpu_id]; 1380843e1988Sjohnlev #endif 1381843e1988Sjohnlev 1382ae115bc7Smrj PRM_POINT("startup_kmem() done"); 13837c478bd9Sstevel@tonic-gate } 13847c478bd9Sstevel@tonic-gate 1385551bc2a6Smrj #ifndef __xpv 1386551bc2a6Smrj /* 1387551bc2a6Smrj * If we have detected that we are running in an HVM environment, we need 1388551bc2a6Smrj * to prepend the PV driver directory to the module search path. 1389551bc2a6Smrj */ 1390551bc2a6Smrj #define HVM_MOD_DIR "/platform/i86hvm/kernel" 1391551bc2a6Smrj static void 1392551bc2a6Smrj update_default_path() 1393551bc2a6Smrj { 1394551bc2a6Smrj char *current, *newpath; 1395551bc2a6Smrj int newlen; 1396551bc2a6Smrj 1397551bc2a6Smrj /* 1398551bc2a6Smrj * We are about to resync with krtld. krtld will reset its 1399551bc2a6Smrj * internal module search path iff Solaris has set default_path. 1400551bc2a6Smrj * We want to be sure we're prepending this new directory to the 1401551bc2a6Smrj * right search path. 1402551bc2a6Smrj */ 1403551bc2a6Smrj current = (default_path == NULL) ? kobj_module_path : default_path; 1404551bc2a6Smrj 1405551bc2a6Smrj newlen = strlen(HVM_MOD_DIR) + strlen(current) + 1; 1406551bc2a6Smrj newpath = kmem_alloc(newlen, KM_SLEEP); 1407551bc2a6Smrj (void) strcpy(newpath, HVM_MOD_DIR); 1408551bc2a6Smrj (void) strcat(newpath, " "); 1409551bc2a6Smrj (void) strcat(newpath, current); 1410551bc2a6Smrj 1411551bc2a6Smrj default_path = newpath; 1412551bc2a6Smrj } 1413551bc2a6Smrj #endif 1414551bc2a6Smrj 14157c478bd9Sstevel@tonic-gate static void 14167c478bd9Sstevel@tonic-gate startup_modules(void) 14177c478bd9Sstevel@tonic-gate { 1418ed5289f9SKen Erickson int cnt; 14197c478bd9Sstevel@tonic-gate extern void prom_setup(void); 1420ed5289f9SKen Erickson int32_t v, h; 1421ed5289f9SKen Erickson char d[11]; 1422ed5289f9SKen Erickson char *cp; 1423e4b86885SCheng Sean Ye cmi_hdl_t hdl; 14247c478bd9Sstevel@tonic-gate 14257c478bd9Sstevel@tonic-gate PRM_POINT("startup_modules() starting..."); 1426843e1988Sjohnlev 1427843e1988Sjohnlev #ifndef __xpv 14287c478bd9Sstevel@tonic-gate /* 14297c478bd9Sstevel@tonic-gate * Initialize ten-micro second timer so that drivers will 14307c478bd9Sstevel@tonic-gate * not get short changed in their init phase. This was 14317c478bd9Sstevel@tonic-gate * not getting called until clkinit which, on fast cpu's 14327c478bd9Sstevel@tonic-gate * caused the drv_usecwait to be way too short. 14337c478bd9Sstevel@tonic-gate */ 14347c478bd9Sstevel@tonic-gate microfind(); 1435551bc2a6Smrj 1436551bc2a6Smrj if (xpv_is_hvm) 1437551bc2a6Smrj update_default_path(); 1438843e1988Sjohnlev #endif 14397c478bd9Sstevel@tonic-gate 14407c478bd9Sstevel@tonic-gate /* 14417c478bd9Sstevel@tonic-gate * Read the GMT lag from /etc/rtc_config. 14427c478bd9Sstevel@tonic-gate */ 1443ae115bc7Smrj sgmtl(process_rtc_config_file()); 14447c478bd9Sstevel@tonic-gate 14457c478bd9Sstevel@tonic-gate /* 14467c478bd9Sstevel@tonic-gate * Calculate default settings of system parameters based upon 14477c478bd9Sstevel@tonic-gate * maxusers, yet allow to be overridden via the /etc/system file. 14487c478bd9Sstevel@tonic-gate */ 14497c478bd9Sstevel@tonic-gate param_calc(0); 14507c478bd9Sstevel@tonic-gate 14517c478bd9Sstevel@tonic-gate mod_setup(); 14527c478bd9Sstevel@tonic-gate 14537c478bd9Sstevel@tonic-gate /* 14547c478bd9Sstevel@tonic-gate * Initialize system parameters. 14557c478bd9Sstevel@tonic-gate */ 14567c478bd9Sstevel@tonic-gate param_init(); 14577c478bd9Sstevel@tonic-gate 14589acbbeafSnn /* 14599acbbeafSnn * Initialize the default brands 14609acbbeafSnn */ 14619acbbeafSnn brand_init(); 14629acbbeafSnn 14637c478bd9Sstevel@tonic-gate /* 14647c478bd9Sstevel@tonic-gate * maxmem is the amount of physical memory we're playing with. 14657c478bd9Sstevel@tonic-gate */ 14667c478bd9Sstevel@tonic-gate maxmem = physmem; 14677c478bd9Sstevel@tonic-gate 14687c478bd9Sstevel@tonic-gate /* 14697c478bd9Sstevel@tonic-gate * Initialize segment management stuff. 14707c478bd9Sstevel@tonic-gate */ 14717c478bd9Sstevel@tonic-gate seg_init(); 14727c478bd9Sstevel@tonic-gate 14737c478bd9Sstevel@tonic-gate if (modload("fs", "specfs") == -1) 14747c478bd9Sstevel@tonic-gate halt("Can't load specfs"); 14757c478bd9Sstevel@tonic-gate 14767c478bd9Sstevel@tonic-gate if (modload("fs", "devfs") == -1) 14777c478bd9Sstevel@tonic-gate halt("Can't load devfs"); 14787c478bd9Sstevel@tonic-gate 1479facf4a8dSllai if (modload("fs", "dev") == -1) 1480facf4a8dSllai halt("Can't load dev"); 1481facf4a8dSllai 148245916cd2Sjpk (void) modloadonly("sys", "lbl_edition"); 148345916cd2Sjpk 14847c478bd9Sstevel@tonic-gate dispinit(); 14857c478bd9Sstevel@tonic-gate 14867c478bd9Sstevel@tonic-gate /* 14877c478bd9Sstevel@tonic-gate * This is needed here to initialize hw_serial[] for cluster booting. 14887c478bd9Sstevel@tonic-gate */ 1489ed5289f9SKen Erickson if ((h = set_soft_hostid()) == -1) 1490ed5289f9SKen Erickson cmn_err(CE_WARN, "Unable to set hostid"); 1491ed5289f9SKen Erickson else { 1492ed5289f9SKen Erickson for (v = h, cnt = 0; cnt < 10; cnt++) { 1493ed5289f9SKen Erickson d[cnt] = v % 10; 1494ed5289f9SKen Erickson v /= 10; 1495ed5289f9SKen Erickson if (v == 0) 1496ed5289f9SKen Erickson break; 1497ed5289f9SKen Erickson } 1498ed5289f9SKen Erickson for (cp = hw_serial; cnt >= 0; cnt--) 1499ed5289f9SKen Erickson *cp++ = d[cnt] + '0'; 1500ed5289f9SKen Erickson *cp = 0; 1501ed5289f9SKen Erickson } 15027c478bd9Sstevel@tonic-gate 15037c478bd9Sstevel@tonic-gate /* Read cluster configuration data. */ 15047c478bd9Sstevel@tonic-gate clconf_init(); 15057c478bd9Sstevel@tonic-gate 1506843e1988Sjohnlev #if defined(__xpv) 1507843e1988Sjohnlev ec_init(); 1508843e1988Sjohnlev gnttab_init(); 1509843e1988Sjohnlev (void) xs_early_init(); 1510843e1988Sjohnlev #endif /* __xpv */ 1511843e1988Sjohnlev 15127c478bd9Sstevel@tonic-gate /* 15137c478bd9Sstevel@tonic-gate * Create a kernel device tree. First, create rootnex and 15147c478bd9Sstevel@tonic-gate * then invoke bus specific code to probe devices. 15157c478bd9Sstevel@tonic-gate */ 15167c478bd9Sstevel@tonic-gate setup_ddi(); 15177aec1d6eScindi 1518e4b86885SCheng Sean Ye /* 1519e4b86885SCheng Sean Ye * Set up the CPU module subsystem for the boot cpu in the native 1520e4b86885SCheng Sean Ye * case, and all physical cpu resource in the xpv dom0 case. 1521e4b86885SCheng Sean Ye * Modifies the device tree, so this must be done after 1522e4b86885SCheng Sean Ye * setup_ddi(). 1523e4b86885SCheng Sean Ye */ 1524e4b86885SCheng Sean Ye #ifdef __xpv 1525e4b86885SCheng Sean Ye /* 1526e4b86885SCheng Sean Ye * If paravirtualized and on dom0 then we initialize all physical 1527e4b86885SCheng Sean Ye * cpu handles now; if paravirtualized on a domU then do not 1528e4b86885SCheng Sean Ye * initialize. 1529e4b86885SCheng Sean Ye */ 1530e4b86885SCheng Sean Ye if (DOMAIN_IS_INITDOMAIN(xen_info)) { 1531e4b86885SCheng Sean Ye xen_mc_lcpu_cookie_t cpi; 1532e4b86885SCheng Sean Ye 1533e4b86885SCheng Sean Ye for (cpi = xen_physcpu_next(NULL); cpi != NULL; 1534e4b86885SCheng Sean Ye cpi = xen_physcpu_next(cpi)) { 1535e4b86885SCheng Sean Ye if ((hdl = cmi_init(CMI_HDL_SOLARIS_xVM_MCA, 1536e4b86885SCheng Sean Ye xen_physcpu_chipid(cpi), xen_physcpu_coreid(cpi), 1537e4b86885SCheng Sean Ye xen_physcpu_strandid(cpi))) != NULL && 1538e4b86885SCheng Sean Ye (x86_feature & X86_MCA)) 153920c794b3Sgavinm cmi_mca_init(hdl); 154020c794b3Sgavinm } 154120c794b3Sgavinm } 1542e4b86885SCheng Sean Ye #else 1543e4b86885SCheng Sean Ye /* 1544e4b86885SCheng Sean Ye * Initialize a handle for the boot cpu - others will initialize 1545e4b86885SCheng Sean Ye * as they startup. Do not do this if we know we are in an HVM domU. 1546e4b86885SCheng Sean Ye */ 1547e4b86885SCheng Sean Ye if (!xpv_is_hvm && 1548e4b86885SCheng Sean Ye (hdl = cmi_init(CMI_HDL_NATIVE, cmi_ntv_hwchipid(CPU), 1549e4b86885SCheng Sean Ye cmi_ntv_hwcoreid(CPU), cmi_ntv_hwstrandid(CPU))) != NULL && 1550e4b86885SCheng Sean Ye (x86_feature & X86_MCA)) 1551e4b86885SCheng Sean Ye cmi_mca_init(hdl); 155220c794b3Sgavinm #endif /* __xpv */ 15537aec1d6eScindi 15547c478bd9Sstevel@tonic-gate /* 15557c478bd9Sstevel@tonic-gate * Fake a prom tree such that /dev/openprom continues to work 15567c478bd9Sstevel@tonic-gate */ 1557ae115bc7Smrj PRM_POINT("startup_modules: calling prom_setup..."); 15587c478bd9Sstevel@tonic-gate prom_setup(); 1559ae115bc7Smrj PRM_POINT("startup_modules: done"); 15607c478bd9Sstevel@tonic-gate 15617c478bd9Sstevel@tonic-gate /* 15627c478bd9Sstevel@tonic-gate * Load all platform specific modules 15637c478bd9Sstevel@tonic-gate */ 1564ae115bc7Smrj PRM_POINT("startup_modules: calling psm_modload..."); 15657c478bd9Sstevel@tonic-gate psm_modload(); 15667c478bd9Sstevel@tonic-gate 15677c478bd9Sstevel@tonic-gate PRM_POINT("startup_modules() done"); 15687c478bd9Sstevel@tonic-gate } 15697c478bd9Sstevel@tonic-gate 1570ae115bc7Smrj /* 1571ae115bc7Smrj * claim a "setaside" boot page for use in the kernel 1572ae115bc7Smrj */ 1573ae115bc7Smrj page_t * 1574ae115bc7Smrj boot_claim_page(pfn_t pfn) 15757c478bd9Sstevel@tonic-gate { 1576ae115bc7Smrj page_t *pp; 15777c478bd9Sstevel@tonic-gate 1578ae115bc7Smrj pp = page_numtopp_nolock(pfn); 1579ae115bc7Smrj ASSERT(pp != NULL); 15807c478bd9Sstevel@tonic-gate 1581ae115bc7Smrj if (PP_ISBOOTPAGES(pp)) { 1582ae115bc7Smrj if (pp->p_next != NULL) 1583ae115bc7Smrj pp->p_next->p_prev = pp->p_prev; 1584ae115bc7Smrj if (pp->p_prev == NULL) 1585ae115bc7Smrj bootpages = pp->p_next; 1586ae115bc7Smrj else 1587ae115bc7Smrj pp->p_prev->p_next = pp->p_next; 1588ae115bc7Smrj } else { 1589ae115bc7Smrj /* 1590ae115bc7Smrj * htable_attach() expects a base pagesize page 1591ae115bc7Smrj */ 1592ae115bc7Smrj if (pp->p_szc != 0) 1593ae115bc7Smrj page_boot_demote(pp); 1594ae115bc7Smrj pp = page_numtopp(pfn, SE_EXCL); 1595ae115bc7Smrj } 1596ae115bc7Smrj return (pp); 15977c478bd9Sstevel@tonic-gate } 15987c478bd9Sstevel@tonic-gate 15997c478bd9Sstevel@tonic-gate /* 16007c478bd9Sstevel@tonic-gate * Walk through the pagetables looking for pages mapped in by boot. If the 16017c478bd9Sstevel@tonic-gate * setaside flag is set the pages are expected to be returned to the 16027c478bd9Sstevel@tonic-gate * kernel later in boot, so we add them to the bootpages list. 16037c478bd9Sstevel@tonic-gate */ 16047c478bd9Sstevel@tonic-gate static void 16057c478bd9Sstevel@tonic-gate protect_boot_range(uintptr_t low, uintptr_t high, int setaside) 16067c478bd9Sstevel@tonic-gate { 16077c478bd9Sstevel@tonic-gate uintptr_t va = low; 16087c478bd9Sstevel@tonic-gate size_t len; 16097c478bd9Sstevel@tonic-gate uint_t prot; 16107c478bd9Sstevel@tonic-gate pfn_t pfn; 16117c478bd9Sstevel@tonic-gate page_t *pp; 16127c478bd9Sstevel@tonic-gate pgcnt_t boot_protect_cnt = 0; 16137c478bd9Sstevel@tonic-gate 1614ae115bc7Smrj while (kbm_probe(&va, &len, &pfn, &prot) != 0 && va < high) { 16157c478bd9Sstevel@tonic-gate if (va + len >= high) 16167c478bd9Sstevel@tonic-gate panic("0x%lx byte mapping at 0x%p exceeds boot's " 16177c478bd9Sstevel@tonic-gate "legal range.", len, (void *)va); 16187c478bd9Sstevel@tonic-gate 16197c478bd9Sstevel@tonic-gate while (len > 0) { 16207c478bd9Sstevel@tonic-gate pp = page_numtopp_alloc(pfn); 16217c478bd9Sstevel@tonic-gate if (pp != NULL) { 16227c478bd9Sstevel@tonic-gate if (setaside == 0) 16237c478bd9Sstevel@tonic-gate panic("Unexpected mapping by boot. " 16247c478bd9Sstevel@tonic-gate "addr=%p pfn=%lx\n", 16257c478bd9Sstevel@tonic-gate (void *)va, pfn); 16267c478bd9Sstevel@tonic-gate 16277c478bd9Sstevel@tonic-gate pp->p_next = bootpages; 1628ae115bc7Smrj pp->p_prev = NULL; 1629ae115bc7Smrj PP_SETBOOTPAGES(pp); 1630ae115bc7Smrj if (bootpages != NULL) { 1631ae115bc7Smrj bootpages->p_prev = pp; 1632ae115bc7Smrj } 16337c478bd9Sstevel@tonic-gate bootpages = pp; 16347c478bd9Sstevel@tonic-gate ++boot_protect_cnt; 16357c478bd9Sstevel@tonic-gate } 16367c478bd9Sstevel@tonic-gate 16377c478bd9Sstevel@tonic-gate ++pfn; 16387c478bd9Sstevel@tonic-gate len -= MMU_PAGESIZE; 16397c478bd9Sstevel@tonic-gate va += MMU_PAGESIZE; 16407c478bd9Sstevel@tonic-gate } 16417c478bd9Sstevel@tonic-gate } 16427c478bd9Sstevel@tonic-gate PRM_DEBUG(boot_protect_cnt); 16437c478bd9Sstevel@tonic-gate } 16447c478bd9Sstevel@tonic-gate 1645ae115bc7Smrj /* 164635b1ab99Sjosephb * 1647ae115bc7Smrj */ 16487c478bd9Sstevel@tonic-gate static void 164935b1ab99Sjosephb layout_kernel_va(void) 16507c478bd9Sstevel@tonic-gate { 165135b1ab99Sjosephb PRM_POINT("layout_kernel_va() starting..."); 16527c478bd9Sstevel@tonic-gate /* 1653ae115bc7Smrj * Establish the final size of the kernel's heap, size of segmap, 1654ae115bc7Smrj * segkp, etc. 16557c478bd9Sstevel@tonic-gate */ 16567c478bd9Sstevel@tonic-gate 16577c478bd9Sstevel@tonic-gate #if defined(__amd64) 16587c478bd9Sstevel@tonic-gate 165935b1ab99Sjosephb kpm_vbase = (caddr_t)segkpm_base; 166035b1ab99Sjosephb kpm_size = ROUND_UP_LPAGE(mmu_ptob(physmax + 1)); 166135b1ab99Sjosephb if ((uintptr_t)kpm_vbase + kpm_size > (uintptr_t)valloc_base) 166235b1ab99Sjosephb panic("not enough room for kpm!"); 1663ae115bc7Smrj PRM_DEBUG(kpm_size); 1664ae115bc7Smrj PRM_DEBUG(kpm_vbase); 16657c478bd9Sstevel@tonic-gate 16667c478bd9Sstevel@tonic-gate /* 1667ae115bc7Smrj * By default we create a seg_kp in 64 bit kernels, it's a little 1668ae115bc7Smrj * faster to access than embedding it in the heap. 16697c478bd9Sstevel@tonic-gate */ 1670ae115bc7Smrj segkp_base = (caddr_t)valloc_base + valloc_sz; 16717c478bd9Sstevel@tonic-gate if (!segkp_fromheap) { 16727c478bd9Sstevel@tonic-gate size_t sz = mmu_ptob(segkpsize); 16737c478bd9Sstevel@tonic-gate 16747c478bd9Sstevel@tonic-gate /* 1675ae115bc7Smrj * determine size of segkp 16767c478bd9Sstevel@tonic-gate */ 16777c478bd9Sstevel@tonic-gate if (sz < SEGKPMINSIZE || sz > SEGKPMAXSIZE) { 16787c478bd9Sstevel@tonic-gate sz = SEGKPDEFSIZE; 16797c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "!Illegal value for segkpsize. " 16807c478bd9Sstevel@tonic-gate "segkpsize has been reset to %ld pages", 16817c478bd9Sstevel@tonic-gate mmu_btop(sz)); 16827c478bd9Sstevel@tonic-gate } 16837c478bd9Sstevel@tonic-gate sz = MIN(sz, MAX(SEGKPMINSIZE, mmu_ptob(physmem))); 16847c478bd9Sstevel@tonic-gate 16857c478bd9Sstevel@tonic-gate segkpsize = mmu_btop(ROUND_UP_LPAGE(sz)); 16867c478bd9Sstevel@tonic-gate } 1687ae115bc7Smrj PRM_DEBUG(segkp_base); 1688ae115bc7Smrj PRM_DEBUG(segkpsize); 16897c478bd9Sstevel@tonic-gate 169035b1ab99Sjosephb /* 169135b1ab99Sjosephb * segzio is used for ZFS cached data. It uses a distinct VA 169235b1ab99Sjosephb * segment (from kernel heap) so that we can easily tell not to 169335b1ab99Sjosephb * include it in kernel crash dumps on 64 bit kernels. The trick is 169435b1ab99Sjosephb * to give it lots of VA, but not constrain the kernel heap. 169535b1ab99Sjosephb * We scale the size of segzio linearly with physmem up to 169635b1ab99Sjosephb * SEGZIOMAXSIZE. Above that amount it scales at 50% of physmem. 169735b1ab99Sjosephb */ 1698ae115bc7Smrj segzio_base = segkp_base + mmu_ptob(segkpsize); 1699ae115bc7Smrj if (segzio_fromheap) { 1700ae115bc7Smrj segziosize = 0; 1701ae115bc7Smrj } else { 170235b1ab99Sjosephb size_t physmem_size = mmu_ptob(physmem); 170335b1ab99Sjosephb size_t size = (segziosize == 0) ? 170435b1ab99Sjosephb physmem_size : mmu_ptob(segziosize); 1705ad23a2dbSjohansen 170635b1ab99Sjosephb if (size < SEGZIOMINSIZE) 1707ad23a2dbSjohansen size = SEGZIOMINSIZE; 170835b1ab99Sjosephb if (size > SEGZIOMAXSIZE) { 1709a778305aSjohansen size = SEGZIOMAXSIZE; 171035b1ab99Sjosephb if (physmem_size > size) 171135b1ab99Sjosephb size += (physmem_size - size) / 2; 1712ad23a2dbSjohansen } 1713ad23a2dbSjohansen segziosize = mmu_btop(ROUND_UP_LPAGE(size)); 1714ad23a2dbSjohansen } 1715ae115bc7Smrj PRM_DEBUG(segziosize); 1716ae115bc7Smrj PRM_DEBUG(segzio_base); 1717ad23a2dbSjohansen 17187c478bd9Sstevel@tonic-gate /* 1719ae115bc7Smrj * Put the range of VA for device mappings next, kmdb knows to not 1720ae115bc7Smrj * grep in this range of addresses. 17217c478bd9Sstevel@tonic-gate */ 1722ae115bc7Smrj toxic_addr = 1723ae115bc7Smrj ROUND_UP_LPAGE((uintptr_t)segzio_base + mmu_ptob(segziosize)); 17247c478bd9Sstevel@tonic-gate PRM_DEBUG(toxic_addr); 1725ae115bc7Smrj segmap_start = ROUND_UP_LPAGE(toxic_addr + toxic_size); 1726ae115bc7Smrj #else /* __i386 */ 1727ae115bc7Smrj segmap_start = ROUND_UP_LPAGE(kernelbase); 1728ae115bc7Smrj #endif /* __i386 */ 1729ae115bc7Smrj PRM_DEBUG(segmap_start); 17307c478bd9Sstevel@tonic-gate 17317c478bd9Sstevel@tonic-gate /* 173252eef812SAmrita Sadhukhan * Users can change segmapsize through eeprom. If the variable 173352eef812SAmrita Sadhukhan * is tuned through eeprom, there is no upper bound on the 1734ed5289f9SKen Erickson * size of segmap. 17357c478bd9Sstevel@tonic-gate */ 17367c478bd9Sstevel@tonic-gate segmapsize = MAX(ROUND_UP_LPAGE(segmapsize), SEGMAPDEFAULT); 17377c478bd9Sstevel@tonic-gate 17387c478bd9Sstevel@tonic-gate #if defined(__i386) 17397c478bd9Sstevel@tonic-gate /* 17407c478bd9Sstevel@tonic-gate * 32-bit systems don't have segkpm or segkp, so segmap appears at 17417c478bd9Sstevel@tonic-gate * the bottom of the kernel's address range. Set aside space for a 1742ae115bc7Smrj * small red zone just below the start of segmap. 17437c478bd9Sstevel@tonic-gate */ 1744ae115bc7Smrj segmap_start += KERNEL_REDZONE_SIZE; 17457c478bd9Sstevel@tonic-gate segmapsize -= KERNEL_REDZONE_SIZE; 17467c478bd9Sstevel@tonic-gate #endif 17477c478bd9Sstevel@tonic-gate 1748ae115bc7Smrj PRM_DEBUG(segmap_start); 17497c478bd9Sstevel@tonic-gate PRM_DEBUG(segmapsize); 175035b1ab99Sjosephb kernelheap = (caddr_t)ROUND_UP_LPAGE(segmap_start + segmapsize); 175135b1ab99Sjosephb PRM_DEBUG(kernelheap); 175235b1ab99Sjosephb PRM_POINT("layout_kernel_va() done..."); 175335b1ab99Sjosephb } 175435b1ab99Sjosephb 175535b1ab99Sjosephb /* 175635b1ab99Sjosephb * Finish initializing the VM system, now that we are no longer 175735b1ab99Sjosephb * relying on the boot time memory allocators. 175835b1ab99Sjosephb */ 175935b1ab99Sjosephb static void 176035b1ab99Sjosephb startup_vm(void) 176135b1ab99Sjosephb { 176235b1ab99Sjosephb struct segmap_crargs a; 176335b1ab99Sjosephb 176435b1ab99Sjosephb extern int use_brk_lpg, use_stk_lpg; 176535b1ab99Sjosephb 176635b1ab99Sjosephb PRM_POINT("startup_vm() starting..."); 176735b1ab99Sjosephb 176835b1ab99Sjosephb /* 176935b1ab99Sjosephb * Initialize the hat layer. 177035b1ab99Sjosephb */ 177135b1ab99Sjosephb hat_init(); 17727c478bd9Sstevel@tonic-gate 1773ae115bc7Smrj /* 1774ae115bc7Smrj * Do final allocations of HAT data structures that need to 1775ae115bc7Smrj * be allocated before quiescing the boot loader. 1776ae115bc7Smrj */ 1777ae115bc7Smrj PRM_POINT("Calling hat_kern_alloc()..."); 1778ae115bc7Smrj hat_kern_alloc((caddr_t)segmap_start, segmapsize, ekernelheap); 1779ae115bc7Smrj PRM_POINT("hat_kern_alloc() done"); 1780ae115bc7Smrj 1781843e1988Sjohnlev #ifndef __xpv 1782ae115bc7Smrj /* 17831d03c31eSjohnlev * Setup Page Attribute Table 1784ae115bc7Smrj */ 17851d03c31eSjohnlev pat_sync(); 1786843e1988Sjohnlev #endif 1787ae115bc7Smrj 1788ae115bc7Smrj /* 1789ae115bc7Smrj * The next two loops are done in distinct steps in order 1790ae115bc7Smrj * to be sure that any page that is doubly mapped (both above 1791ae115bc7Smrj * KERNEL_TEXT and below kernelbase) is dealt with correctly. 1792ae115bc7Smrj * Note this may never happen, but it might someday. 1793ae115bc7Smrj */ 1794ae115bc7Smrj bootpages = NULL; 1795ae115bc7Smrj PRM_POINT("Protecting boot pages"); 1796ae115bc7Smrj 1797ae115bc7Smrj /* 1798ae115bc7Smrj * Protect any pages mapped above KERNEL_TEXT that somehow have 1799ae115bc7Smrj * page_t's. This can only happen if something weird allocated 1800ae115bc7Smrj * in this range (like kadb/kmdb). 1801ae115bc7Smrj */ 1802ae115bc7Smrj protect_boot_range(KERNEL_TEXT, (uintptr_t)-1, 0); 1803ae115bc7Smrj 1804ae115bc7Smrj /* 1805ae115bc7Smrj * Before we can take over memory allocation/mapping from the boot 1806ae115bc7Smrj * loader we must remove from our free page lists any boot allocated 1807ae115bc7Smrj * pages that stay mapped until release_bootstrap(). 1808ae115bc7Smrj */ 1809ae115bc7Smrj protect_boot_range(0, kernelbase, 1); 1810ae115bc7Smrj 1811843e1988Sjohnlev 1812ae115bc7Smrj /* 1813ae115bc7Smrj * Switch to running on regular HAT (not boot_mmu) 1814ae115bc7Smrj */ 1815ae115bc7Smrj PRM_POINT("Calling hat_kern_setup()..."); 1816ae115bc7Smrj hat_kern_setup(); 1817ae115bc7Smrj 1818ae115bc7Smrj /* 1819ae115bc7Smrj * It is no longer safe to call BOP_ALLOC(), so make sure we don't. 1820ae115bc7Smrj */ 1821ae115bc7Smrj bop_no_more_mem(); 1822ae115bc7Smrj 1823ae115bc7Smrj PRM_POINT("hat_kern_setup() done"); 1824ae115bc7Smrj 1825ae115bc7Smrj hat_cpu_online(CPU); 1826ae115bc7Smrj 18277c478bd9Sstevel@tonic-gate /* 18287c478bd9Sstevel@tonic-gate * Initialize VM system 18297c478bd9Sstevel@tonic-gate */ 18307c478bd9Sstevel@tonic-gate PRM_POINT("Calling kvm_init()..."); 18317c478bd9Sstevel@tonic-gate kvm_init(); 18327c478bd9Sstevel@tonic-gate PRM_POINT("kvm_init() done"); 18337c478bd9Sstevel@tonic-gate 18347c478bd9Sstevel@tonic-gate /* 18357c478bd9Sstevel@tonic-gate * Tell kmdb that the VM system is now working 18367c478bd9Sstevel@tonic-gate */ 18377c478bd9Sstevel@tonic-gate if (boothowto & RB_DEBUG) 18387c478bd9Sstevel@tonic-gate kdi_dvec_vmready(); 18397c478bd9Sstevel@tonic-gate 1840843e1988Sjohnlev #if defined(__xpv) 1841843e1988Sjohnlev /* 1842843e1988Sjohnlev * Populate the I/O pool on domain 0 1843843e1988Sjohnlev */ 1844843e1988Sjohnlev if (DOMAIN_IS_INITDOMAIN(xen_info)) { 1845843e1988Sjohnlev extern long populate_io_pool(void); 1846843e1988Sjohnlev long init_io_pool_cnt; 1847843e1988Sjohnlev 1848843e1988Sjohnlev PRM_POINT("Populating reserve I/O page pool"); 1849843e1988Sjohnlev init_io_pool_cnt = populate_io_pool(); 1850843e1988Sjohnlev PRM_DEBUG(init_io_pool_cnt); 1851843e1988Sjohnlev } 1852843e1988Sjohnlev #endif 18537c478bd9Sstevel@tonic-gate /* 18547c478bd9Sstevel@tonic-gate * Mangle the brand string etc. 18557c478bd9Sstevel@tonic-gate */ 18567c478bd9Sstevel@tonic-gate cpuid_pass3(CPU); 18577c478bd9Sstevel@tonic-gate 18587c478bd9Sstevel@tonic-gate #if defined(__amd64) 18597c478bd9Sstevel@tonic-gate 18607c478bd9Sstevel@tonic-gate /* 18617c478bd9Sstevel@tonic-gate * Create the device arena for toxic (to dtrace/kmdb) mappings. 18627c478bd9Sstevel@tonic-gate */ 18637c478bd9Sstevel@tonic-gate device_arena = vmem_create("device", (void *)toxic_addr, 18647c478bd9Sstevel@tonic-gate toxic_size, MMU_PAGESIZE, NULL, NULL, NULL, 0, VM_SLEEP); 18657c478bd9Sstevel@tonic-gate 18667c478bd9Sstevel@tonic-gate #else /* __i386 */ 18677c478bd9Sstevel@tonic-gate 18687c478bd9Sstevel@tonic-gate /* 18697c478bd9Sstevel@tonic-gate * allocate the bit map that tracks toxic pages 18707c478bd9Sstevel@tonic-gate */ 1871ae115bc7Smrj toxic_bit_map_len = btop((ulong_t)(valloc_base - kernelbase)); 18727c478bd9Sstevel@tonic-gate PRM_DEBUG(toxic_bit_map_len); 18737c478bd9Sstevel@tonic-gate toxic_bit_map = 18747c478bd9Sstevel@tonic-gate kmem_zalloc(BT_SIZEOFMAP(toxic_bit_map_len), KM_NOSLEEP); 18757c478bd9Sstevel@tonic-gate ASSERT(toxic_bit_map != NULL); 18767c478bd9Sstevel@tonic-gate PRM_DEBUG(toxic_bit_map); 18777c478bd9Sstevel@tonic-gate 18787c478bd9Sstevel@tonic-gate #endif /* __i386 */ 18797c478bd9Sstevel@tonic-gate 18807c478bd9Sstevel@tonic-gate 18817c478bd9Sstevel@tonic-gate /* 18827c478bd9Sstevel@tonic-gate * Now that we've got more VA, as well as the ability to allocate from 18837c478bd9Sstevel@tonic-gate * it, tell the debugger. 18847c478bd9Sstevel@tonic-gate */ 18857c478bd9Sstevel@tonic-gate if (boothowto & RB_DEBUG) 18867c478bd9Sstevel@tonic-gate kdi_dvec_memavail(); 18877c478bd9Sstevel@tonic-gate 18887c478bd9Sstevel@tonic-gate /* 18897c478bd9Sstevel@tonic-gate * The following code installs a special page fault handler (#pf) 18907c478bd9Sstevel@tonic-gate * to work around a pentium bug. 18917c478bd9Sstevel@tonic-gate */ 1892843e1988Sjohnlev #if !defined(__amd64) && !defined(__xpv) 18937c478bd9Sstevel@tonic-gate if (x86_type == X86_TYPE_P5) { 1894ae115bc7Smrj desctbr_t idtr; 18957c478bd9Sstevel@tonic-gate gate_desc_t *newidt; 18967c478bd9Sstevel@tonic-gate 18977c478bd9Sstevel@tonic-gate if ((newidt = kmem_zalloc(MMU_PAGESIZE, KM_NOSLEEP)) == NULL) 18987c478bd9Sstevel@tonic-gate panic("failed to install pentium_pftrap"); 18997c478bd9Sstevel@tonic-gate 19000cfdb603Sjosephb bcopy(idt0, newidt, NIDT * sizeof (*idt0)); 19017c478bd9Sstevel@tonic-gate set_gatesegd(&newidt[T_PGFLT], &pentium_pftrap, 1902843e1988Sjohnlev KCS_SEL, SDT_SYSIGT, TRP_KPL); 19037c478bd9Sstevel@tonic-gate 19047c478bd9Sstevel@tonic-gate (void) as_setprot(&kas, (caddr_t)newidt, MMU_PAGESIZE, 19050cfdb603Sjosephb PROT_READ | PROT_EXEC); 19067c478bd9Sstevel@tonic-gate 19070cfdb603Sjosephb CPU->cpu_idt = newidt; 19080cfdb603Sjosephb idtr.dtr_base = (uintptr_t)CPU->cpu_idt; 19090cfdb603Sjosephb idtr.dtr_limit = (NIDT * sizeof (*idt0)) - 1; 1910ae115bc7Smrj wr_idtr(&idtr); 19117c478bd9Sstevel@tonic-gate } 19127c478bd9Sstevel@tonic-gate #endif /* !__amd64 */ 19137c478bd9Sstevel@tonic-gate 1914843e1988Sjohnlev #if !defined(__xpv) 19157c478bd9Sstevel@tonic-gate /* 19167c478bd9Sstevel@tonic-gate * Map page pfn=0 for drivers, such as kd, that need to pick up 19177c478bd9Sstevel@tonic-gate * parameters left there by controllers/BIOS. 19187c478bd9Sstevel@tonic-gate */ 19197c478bd9Sstevel@tonic-gate PRM_POINT("setup up p0_va"); 19207c478bd9Sstevel@tonic-gate p0_va = i86devmap(0, 1, PROT_READ); 19217c478bd9Sstevel@tonic-gate PRM_DEBUG(p0_va); 1922843e1988Sjohnlev #endif 19237c478bd9Sstevel@tonic-gate 19247c478bd9Sstevel@tonic-gate cmn_err(CE_CONT, "?mem = %luK (0x%lx)\n", 19257c478bd9Sstevel@tonic-gate physinstalled << (MMU_PAGESHIFT - 10), ptob(physinstalled)); 19267c478bd9Sstevel@tonic-gate 192783f9b804Skchow /* 192883f9b804Skchow * disable automatic large pages for small memory systems or 192983f9b804Skchow * when the disable flag is set. 193002bc52beSkchow * 193102bc52beSkchow * Do not yet consider page sizes larger than 2m/4m. 193283f9b804Skchow */ 1933ec25b48fSsusans if (!auto_lpg_disable && mmu.max_page_level > 0) { 1934ec25b48fSsusans max_uheap_lpsize = LEVEL_SIZE(1); 1935ec25b48fSsusans max_ustack_lpsize = LEVEL_SIZE(1); 1936ec25b48fSsusans max_privmap_lpsize = LEVEL_SIZE(1); 1937ec25b48fSsusans max_uidata_lpsize = LEVEL_SIZE(1); 1938ec25b48fSsusans max_utext_lpsize = LEVEL_SIZE(1); 1939ec25b48fSsusans max_shm_lpsize = LEVEL_SIZE(1); 1940ec25b48fSsusans } 1941ec25b48fSsusans if (physmem < privm_lpg_min_physmem || mmu.max_page_level == 0 || 1942ec25b48fSsusans auto_lpg_disable) { 1943beb1bda0Sdavemq use_brk_lpg = 0; 1944beb1bda0Sdavemq use_stk_lpg = 0; 1945ec25b48fSsusans } 194602bc52beSkchow mcntl0_lpsize = LEVEL_SIZE(mmu.umax_page_level); 1947beb1bda0Sdavemq 19487c478bd9Sstevel@tonic-gate PRM_POINT("Calling hat_init_finish()..."); 19497c478bd9Sstevel@tonic-gate hat_init_finish(); 19507c478bd9Sstevel@tonic-gate PRM_POINT("hat_init_finish() done"); 19517c478bd9Sstevel@tonic-gate 19527c478bd9Sstevel@tonic-gate /* 19537c478bd9Sstevel@tonic-gate * Initialize the segkp segment type. 19547c478bd9Sstevel@tonic-gate */ 19557c478bd9Sstevel@tonic-gate rw_enter(&kas.a_lock, RW_WRITER); 1956ae115bc7Smrj PRM_POINT("Attaching segkp"); 1957ae115bc7Smrj if (segkp_fromheap) { 19587c478bd9Sstevel@tonic-gate segkp->s_as = &kas; 1959ae115bc7Smrj } else if (seg_attach(&kas, (caddr_t)segkp_base, mmu_ptob(segkpsize), 1960ae115bc7Smrj segkp) < 0) { 1961ae115bc7Smrj panic("startup: cannot attach segkp"); 1962ae115bc7Smrj /*NOTREACHED*/ 19637c478bd9Sstevel@tonic-gate } 1964ae115bc7Smrj PRM_POINT("Doing segkp_create()"); 19657c478bd9Sstevel@tonic-gate if (segkp_create(segkp) != 0) { 19667c478bd9Sstevel@tonic-gate panic("startup: segkp_create failed"); 19677c478bd9Sstevel@tonic-gate /*NOTREACHED*/ 19687c478bd9Sstevel@tonic-gate } 19697c478bd9Sstevel@tonic-gate PRM_DEBUG(segkp); 19707c478bd9Sstevel@tonic-gate rw_exit(&kas.a_lock); 19717c478bd9Sstevel@tonic-gate 19727c478bd9Sstevel@tonic-gate /* 19737c478bd9Sstevel@tonic-gate * kpm segment 19747c478bd9Sstevel@tonic-gate */ 19757c478bd9Sstevel@tonic-gate segmap_kpm = 0; 19767c478bd9Sstevel@tonic-gate if (kpm_desired) { 19777c478bd9Sstevel@tonic-gate kpm_init(); 19787c478bd9Sstevel@tonic-gate kpm_enable = 1; 1979a5652762Spraks vpm_enable = 1; 19807c478bd9Sstevel@tonic-gate } 19817c478bd9Sstevel@tonic-gate 19827c478bd9Sstevel@tonic-gate /* 19837c478bd9Sstevel@tonic-gate * Now create segmap segment. 19847c478bd9Sstevel@tonic-gate */ 19857c478bd9Sstevel@tonic-gate rw_enter(&kas.a_lock, RW_WRITER); 1986ae115bc7Smrj if (seg_attach(&kas, (caddr_t)segmap_start, segmapsize, segmap) < 0) { 1987ae115bc7Smrj panic("cannot attach segmap"); 19887c478bd9Sstevel@tonic-gate /*NOTREACHED*/ 19897c478bd9Sstevel@tonic-gate } 1990ae115bc7Smrj PRM_DEBUG(segmap); 19917c478bd9Sstevel@tonic-gate 19927c478bd9Sstevel@tonic-gate a.prot = PROT_READ | PROT_WRITE; 19937c478bd9Sstevel@tonic-gate a.shmsize = 0; 19947c478bd9Sstevel@tonic-gate a.nfreelist = segmapfreelists; 19957c478bd9Sstevel@tonic-gate 1996ae115bc7Smrj if (segmap_create(segmap, (caddr_t)&a) != 0) 1997ae115bc7Smrj panic("segmap_create segmap"); 19987c478bd9Sstevel@tonic-gate rw_exit(&kas.a_lock); 19997c478bd9Sstevel@tonic-gate 20007c478bd9Sstevel@tonic-gate setup_vaddr_for_ppcopy(CPU); 20017c478bd9Sstevel@tonic-gate 20027c478bd9Sstevel@tonic-gate segdev_init(); 2003843e1988Sjohnlev #if defined(__xpv) 2004843e1988Sjohnlev if (DOMAIN_IS_INITDOMAIN(xen_info)) 2005843e1988Sjohnlev #endif 2006843e1988Sjohnlev pmem_init(); 2007ae115bc7Smrj 20087c478bd9Sstevel@tonic-gate PRM_POINT("startup_vm() done"); 20097c478bd9Sstevel@tonic-gate } 20107c478bd9Sstevel@tonic-gate 2011ae115bc7Smrj /* 2012ae115bc7Smrj * Load a tod module for the non-standard tod part found on this system. 2013ae115bc7Smrj */ 2014ae115bc7Smrj static void 2015ae115bc7Smrj load_tod_module(char *todmod) 2016ae115bc7Smrj { 2017ae115bc7Smrj if (modload("tod", todmod) == -1) 2018ae115bc7Smrj halt("Can't load TOD module"); 2019ae115bc7Smrj } 2020ae115bc7Smrj 2021*94f1124eSVikram Hegde #if !defined(__xpv) 2022*94f1124eSVikram Hegde static void 2023*94f1124eSVikram Hegde get_amd_iommu_boot_props(void) 2024*94f1124eSVikram Hegde { 2025*94f1124eSVikram Hegde int len; 2026*94f1124eSVikram Hegde char *disable; 2027*94f1124eSVikram Hegde 2028*94f1124eSVikram Hegde if ((len = do_bsys_getproplen(NULL, "amd-iommu")) > 0) { 2029*94f1124eSVikram Hegde disable = kmem_zalloc(len, KM_SLEEP); 2030*94f1124eSVikram Hegde (void) do_bsys_getprop(NULL, "amd-iommu", disable); 2031*94f1124eSVikram Hegde if (strcmp(disable, "no") == 0) { 2032*94f1124eSVikram Hegde startup_amd_iommu_disable = 1; 2033*94f1124eSVikram Hegde } 2034*94f1124eSVikram Hegde kmem_free(disable, len); 2035*94f1124eSVikram Hegde } 2036*94f1124eSVikram Hegde 2037*94f1124eSVikram Hegde if ((len = do_bsys_getproplen(NULL, "amd-iommu-disable-list")) > 0) { 2038*94f1124eSVikram Hegde startup_amd_iommu_disable_list = kmem_zalloc(len, KM_SLEEP); 2039*94f1124eSVikram Hegde (void) do_bsys_getprop(NULL, "amd-iommu-disable-list", 2040*94f1124eSVikram Hegde startup_amd_iommu_disable_list); 2041*94f1124eSVikram Hegde } 2042*94f1124eSVikram Hegde } 2043*94f1124eSVikram Hegde #endif 2044*94f1124eSVikram Hegde 20457c478bd9Sstevel@tonic-gate static void 20467c478bd9Sstevel@tonic-gate startup_end(void) 20477c478bd9Sstevel@tonic-gate { 2048dd4eeefdSeota int i; 20497c478bd9Sstevel@tonic-gate extern void setx86isalist(void); 20507c478bd9Sstevel@tonic-gate 20517c478bd9Sstevel@tonic-gate PRM_POINT("startup_end() starting..."); 20527c478bd9Sstevel@tonic-gate 20537c478bd9Sstevel@tonic-gate /* 20547c478bd9Sstevel@tonic-gate * Perform tasks that get done after most of the VM 20557c478bd9Sstevel@tonic-gate * initialization has been done but before the clock 20567c478bd9Sstevel@tonic-gate * and other devices get started. 20577c478bd9Sstevel@tonic-gate */ 20587c478bd9Sstevel@tonic-gate kern_setup1(); 20597c478bd9Sstevel@tonic-gate 20607c478bd9Sstevel@tonic-gate /* 20617c478bd9Sstevel@tonic-gate * Perform CPC initialization for this CPU. 20627c478bd9Sstevel@tonic-gate */ 20637c478bd9Sstevel@tonic-gate kcpc_hw_init(CPU); 20647c478bd9Sstevel@tonic-gate 2065ee88d2b9Skchow #if defined(OPTERON_WORKAROUND_6323525) 2066ee88d2b9Skchow if (opteron_workaround_6323525) 2067ee88d2b9Skchow patch_workaround_6323525(); 2068ee88d2b9Skchow #endif 2069ae115bc7Smrj /* 2070ae115bc7Smrj * If needed, load TOD module now so that ddi_get_time(9F) etc. work 2071ae115bc7Smrj * (For now, "needed" is defined as set tod_module_name in /etc/system) 2072ae115bc7Smrj */ 2073ae115bc7Smrj if (tod_module_name != NULL) { 2074ae115bc7Smrj PRM_POINT("load_tod_module()"); 2075ae115bc7Smrj load_tod_module(tod_module_name); 2076ae115bc7Smrj } 2077ae115bc7Smrj 2078843e1988Sjohnlev #if defined(__xpv) 2079843e1988Sjohnlev /* 2080843e1988Sjohnlev * Forceload interposing TOD module for the hypervisor. 2081843e1988Sjohnlev */ 2082843e1988Sjohnlev PRM_POINT("load_tod_module()"); 2083843e1988Sjohnlev load_tod_module("xpvtod"); 2084843e1988Sjohnlev #endif 2085843e1988Sjohnlev 20867c478bd9Sstevel@tonic-gate /* 20877c478bd9Sstevel@tonic-gate * Configure the system. 20887c478bd9Sstevel@tonic-gate */ 20897c478bd9Sstevel@tonic-gate PRM_POINT("Calling configure()..."); 20907c478bd9Sstevel@tonic-gate configure(); /* set up devices */ 20917c478bd9Sstevel@tonic-gate PRM_POINT("configure() done"); 20927c478bd9Sstevel@tonic-gate 20937c478bd9Sstevel@tonic-gate /* 20947c478bd9Sstevel@tonic-gate * Set the isa_list string to the defined instruction sets we 20957c478bd9Sstevel@tonic-gate * support. 20967c478bd9Sstevel@tonic-gate */ 20977c478bd9Sstevel@tonic-gate setx86isalist(); 2098100b72f4Sandrei cpu_intr_alloc(CPU, NINTR_THREADS); 20997c478bd9Sstevel@tonic-gate psm_install(); 21007c478bd9Sstevel@tonic-gate 21017c478bd9Sstevel@tonic-gate /* 21027c478bd9Sstevel@tonic-gate * We're done with bootops. We don't unmap the bootstrap yet because 21037c478bd9Sstevel@tonic-gate * we're still using bootsvcs. 21047c478bd9Sstevel@tonic-gate */ 2105ae115bc7Smrj PRM_POINT("NULLing out bootops"); 2106ae115bc7Smrj *bootopsp = (struct bootops *)NULL; 21077c478bd9Sstevel@tonic-gate bootops = (struct bootops *)NULL; 21087c478bd9Sstevel@tonic-gate 2109843e1988Sjohnlev #if defined(__xpv) 2110843e1988Sjohnlev ec_init_debug_irq(); 2111843e1988Sjohnlev xs_domu_init(); 2112843e1988Sjohnlev #endif 21137c478bd9Sstevel@tonic-gate PRM_POINT("Enabling interrupts"); 21147c478bd9Sstevel@tonic-gate (*picinitf)(); 21157c478bd9Sstevel@tonic-gate sti(); 2116843e1988Sjohnlev #if defined(__xpv) 2117843e1988Sjohnlev ASSERT(CPU->cpu_m.mcpu_vcpu_info->evtchn_upcall_mask == 0); 2118843e1988Sjohnlev xen_late_startup(); 2119843e1988Sjohnlev #endif 21207c478bd9Sstevel@tonic-gate 21217c478bd9Sstevel@tonic-gate (void) add_avsoftintr((void *)&softlevel1_hdl, 1, softlevel1, 212235b1ab99Sjosephb "softlevel1", NULL, NULL); /* XXX to be moved later */ 21237c478bd9Sstevel@tonic-gate 2124dd4eeefdSeota /* 2125dd4eeefdSeota * Register these software interrupts for ddi timer. 2126dd4eeefdSeota * Software interrupts up to the level 10 are supported. 2127dd4eeefdSeota */ 2128dd4eeefdSeota for (i = DDI_IPL_1; i <= DDI_IPL_10; i++) { 2129dd4eeefdSeota char name[sizeof ("timer_softintr") + 2]; 2130dd4eeefdSeota (void) sprintf(name, "timer_softintr%02d", i); 2131dd4eeefdSeota (void) add_avsoftintr((void *)&softlevel_hdl[i-1], i, 2132dd4eeefdSeota (avfunc)timer_softintr, name, (caddr_t)(uintptr_t)i, NULL); 2133dd4eeefdSeota } 2134dd4eeefdSeota 21357125fcbdSVikram Hegde #if !defined(__xpv) 2136*94f1124eSVikram Hegde get_amd_iommu_boot_props(); 2137*94f1124eSVikram Hegde 21387125fcbdSVikram Hegde if (modload("drv", "amd_iommu") < 0) { 21397125fcbdSVikram Hegde PRM_POINT("No AMD IOMMU present\n"); 2140*94f1124eSVikram Hegde } else if (ddi_hold_installed_driver(ddi_name_to_major( 2141*94f1124eSVikram Hegde "amd_iommu")) == NULL) { 21427125fcbdSVikram Hegde prom_printf("ERROR: failed to attach AMD IOMMU\n"); 21437125fcbdSVikram Hegde } 21447125fcbdSVikram Hegde #endif 21457125fcbdSVikram Hegde 21467c478bd9Sstevel@tonic-gate PRM_POINT("startup_end() done"); 21477c478bd9Sstevel@tonic-gate } 21487c478bd9Sstevel@tonic-gate 21497c478bd9Sstevel@tonic-gate extern char hw_serial[]; 215096d008c2SKen Erickson /* 215196d008c2SKen Erickson * Don't remove the following 2 variables. They are necessary 215296d008c2SKen Erickson * for reading the hostid from the legacy file (/kernel/misc/sysinit). 215396d008c2SKen Erickson */ 21547c478bd9Sstevel@tonic-gate char *_hs1107 = hw_serial; 21557c478bd9Sstevel@tonic-gate ulong_t _bdhs34; 21567c478bd9Sstevel@tonic-gate 21577c478bd9Sstevel@tonic-gate void 21587c478bd9Sstevel@tonic-gate post_startup(void) 21597c478bd9Sstevel@tonic-gate { 21607c478bd9Sstevel@tonic-gate /* 21617c478bd9Sstevel@tonic-gate * Set the system wide, processor-specific flags to be passed 21627c478bd9Sstevel@tonic-gate * to userland via the aux vector for performance hints and 21637c478bd9Sstevel@tonic-gate * instruction set extensions. 21647c478bd9Sstevel@tonic-gate */ 21657c478bd9Sstevel@tonic-gate bind_hwcap(); 21667c478bd9Sstevel@tonic-gate 2167843e1988Sjohnlev #ifdef __xpv 2168843e1988Sjohnlev if (DOMAIN_IS_INITDOMAIN(xen_info)) 2169843e1988Sjohnlev #endif 2170843e1988Sjohnlev { 2171843e1988Sjohnlev /* 2172843e1988Sjohnlev * Load the System Management BIOS into the global ksmbios 2173843e1988Sjohnlev * handle, if an SMBIOS is present on this system. 2174843e1988Sjohnlev */ 2175843e1988Sjohnlev ksmbios = smbios_open(NULL, SMB_VERSION, ksmbios_flags, NULL); 21767c478bd9Sstevel@tonic-gate 2177843e1988Sjohnlev #if defined(__xpv) 2178843e1988Sjohnlev xpv_panic_init(); 2179843e1988Sjohnlev #else 2180843e1988Sjohnlev /* 2181843e1988Sjohnlev * Startup the memory scrubber. 2182843e1988Sjohnlev * XXPV This should be running somewhere .. 2183843e1988Sjohnlev */ 2184e4b86885SCheng Sean Ye if (!xpv_is_hvm) 2185e4b86885SCheng Sean Ye memscrub_init(); 2186843e1988Sjohnlev #endif 2187843e1988Sjohnlev } 21887c478bd9Sstevel@tonic-gate 21897aec1d6eScindi /* 21907aec1d6eScindi * Complete CPU module initialization 21917aec1d6eScindi */ 219220c794b3Sgavinm cmi_post_startup(); 21937aec1d6eScindi 21947c478bd9Sstevel@tonic-gate /* 219584ab085aSmws * Perform forceloading tasks for /etc/system. 21967c478bd9Sstevel@tonic-gate */ 219784ab085aSmws (void) mod_sysctl(SYS_FORCELOAD, NULL); 21987c478bd9Sstevel@tonic-gate 21997c478bd9Sstevel@tonic-gate /* 22007c478bd9Sstevel@tonic-gate * ON4.0: Force /proc module in until clock interrupt handle fixed 22017c478bd9Sstevel@tonic-gate * ON4.0: This must be fixed or restated in /etc/systems. 22027c478bd9Sstevel@tonic-gate */ 22037c478bd9Sstevel@tonic-gate (void) modload("fs", "procfs"); 22047c478bd9Sstevel@tonic-gate 2205c35aa225Smarx (void) i_ddi_attach_hw_nodes("pit_beep"); 2206c35aa225Smarx 22077c478bd9Sstevel@tonic-gate #if defined(__i386) 22087c478bd9Sstevel@tonic-gate /* 22097c478bd9Sstevel@tonic-gate * Check for required functional Floating Point hardware, 22107c478bd9Sstevel@tonic-gate * unless FP hardware explicitly disabled. 22117c478bd9Sstevel@tonic-gate */ 22127c478bd9Sstevel@tonic-gate if (fpu_exists && (fpu_pentium_fdivbug || fp_kind == FP_NO)) 22137c478bd9Sstevel@tonic-gate halt("No working FP hardware found"); 22147c478bd9Sstevel@tonic-gate #endif 22157c478bd9Sstevel@tonic-gate 22167c478bd9Sstevel@tonic-gate maxmem = freemem; 22177c478bd9Sstevel@tonic-gate 22187c478bd9Sstevel@tonic-gate add_cpunode2devtree(CPU->cpu_id, CPU->cpu_m.mcpu_cpi); 22197c478bd9Sstevel@tonic-gate } 22207c478bd9Sstevel@tonic-gate 22217c478bd9Sstevel@tonic-gate static int 222219397407SSherry Moore pp_in_range(page_t *pp, uint64_t low_addr, uint64_t high_addr) 22237c478bd9Sstevel@tonic-gate { 222419397407SSherry Moore return ((pp->p_pagenum >= btop(low_addr)) && 222519397407SSherry Moore (pp->p_pagenum < btopr(high_addr))); 22267c478bd9Sstevel@tonic-gate } 22277c478bd9Sstevel@tonic-gate 22287c478bd9Sstevel@tonic-gate void 22297c478bd9Sstevel@tonic-gate release_bootstrap(void) 22307c478bd9Sstevel@tonic-gate { 22317c478bd9Sstevel@tonic-gate int root_is_ramdisk; 22327c478bd9Sstevel@tonic-gate page_t *pp; 22337c478bd9Sstevel@tonic-gate extern void kobj_boot_unmountroot(void); 22347c478bd9Sstevel@tonic-gate extern dev_t rootdev; 223555d507a9SSeth Goldberg #if !defined(__xpv) 223655d507a9SSeth Goldberg pfn_t pfn; 223755d507a9SSeth Goldberg #endif 22387c478bd9Sstevel@tonic-gate 22397c478bd9Sstevel@tonic-gate /* unmount boot ramdisk and release kmem usage */ 22407c478bd9Sstevel@tonic-gate kobj_boot_unmountroot(); 22417c478bd9Sstevel@tonic-gate 22427c478bd9Sstevel@tonic-gate /* 22437c478bd9Sstevel@tonic-gate * We're finished using the boot loader so free its pages. 22447c478bd9Sstevel@tonic-gate */ 22457c478bd9Sstevel@tonic-gate PRM_POINT("Unmapping lower boot pages"); 224619397407SSherry Moore 2247ae115bc7Smrj clear_boot_mappings(0, _userlimit); 224819397407SSherry Moore 2249ae115bc7Smrj postbootkernelbase = kernelbase; 22507c478bd9Sstevel@tonic-gate 22517c478bd9Sstevel@tonic-gate /* 22527c478bd9Sstevel@tonic-gate * If root isn't on ramdisk, destroy the hardcoded 22537c478bd9Sstevel@tonic-gate * ramdisk node now and release the memory. Else, 22547c478bd9Sstevel@tonic-gate * ramdisk memory is kept in rd_pages. 22557c478bd9Sstevel@tonic-gate */ 22567c478bd9Sstevel@tonic-gate root_is_ramdisk = (getmajor(rootdev) == ddi_name_to_major("ramdisk")); 22577c478bd9Sstevel@tonic-gate if (!root_is_ramdisk) { 22587c478bd9Sstevel@tonic-gate dev_info_t *dip = ddi_find_devinfo("ramdisk", -1, 0); 22597c478bd9Sstevel@tonic-gate ASSERT(dip && ddi_get_parent(dip) == ddi_root_node()); 22607c478bd9Sstevel@tonic-gate ndi_rele_devi(dip); /* held from ddi_find_devinfo */ 22617c478bd9Sstevel@tonic-gate (void) ddi_remove_child(dip, 0); 22627c478bd9Sstevel@tonic-gate } 22637c478bd9Sstevel@tonic-gate 22647c478bd9Sstevel@tonic-gate PRM_POINT("Releasing boot pages"); 22657c478bd9Sstevel@tonic-gate while (bootpages) { 226619397407SSherry Moore extern uint64_t ramdisk_start, ramdisk_end; 22677c478bd9Sstevel@tonic-gate pp = bootpages; 22687c478bd9Sstevel@tonic-gate bootpages = pp->p_next; 226919397407SSherry Moore 227019397407SSherry Moore 227119397407SSherry Moore /* Keep pages for the lower 64K */ 227219397407SSherry Moore if (pp_in_range(pp, 0, 0x40000)) { 227319397407SSherry Moore pp->p_next = lower_pages; 227419397407SSherry Moore lower_pages = pp; 227519397407SSherry Moore lower_pages_count++; 227619397407SSherry Moore continue; 227719397407SSherry Moore } 227819397407SSherry Moore 227919397407SSherry Moore 228019397407SSherry Moore if (root_is_ramdisk && pp_in_range(pp, ramdisk_start, 228119397407SSherry Moore ramdisk_end)) { 22827c478bd9Sstevel@tonic-gate pp->p_next = rd_pages; 22837c478bd9Sstevel@tonic-gate rd_pages = pp; 22847c478bd9Sstevel@tonic-gate continue; 22857c478bd9Sstevel@tonic-gate } 22867c478bd9Sstevel@tonic-gate pp->p_next = (struct page *)0; 2287ae115bc7Smrj pp->p_prev = (struct page *)0; 2288ae115bc7Smrj PP_CLRBOOTPAGES(pp); 22897c478bd9Sstevel@tonic-gate page_free(pp, 1); 22907c478bd9Sstevel@tonic-gate } 2291ae115bc7Smrj PRM_POINT("Boot pages released"); 22927c478bd9Sstevel@tonic-gate 2293843e1988Sjohnlev #if !defined(__xpv) 2294843e1988Sjohnlev /* XXPV -- note this following bunch of code needs to be revisited in Xen 3.0 */ 22957c478bd9Sstevel@tonic-gate /* 229655d507a9SSeth Goldberg * Find 1 page below 1 MB so that other processors can boot up or 229755d507a9SSeth Goldberg * so that any processor can resume. 22987c478bd9Sstevel@tonic-gate * Make sure it has a kernel VA as well as a 1:1 mapping. 22997c478bd9Sstevel@tonic-gate * We should have just free'd one up. 23007c478bd9Sstevel@tonic-gate */ 230119397407SSherry Moore 230219397407SSherry Moore /* 230319397407SSherry Moore * 0x10 pages is 64K. Leave the bottom 64K alone 230419397407SSherry Moore * for BIOS. 230519397407SSherry Moore */ 230619397407SSherry Moore for (pfn = 0x10; pfn < btop(1*1024*1024); pfn++) { 230755d507a9SSeth Goldberg if (page_numtopp_alloc(pfn) == NULL) 230855d507a9SSeth Goldberg continue; 230955d507a9SSeth Goldberg rm_platter_va = i86devmap(pfn, 1, 231055d507a9SSeth Goldberg PROT_READ | PROT_WRITE | PROT_EXEC); 231155d507a9SSeth Goldberg rm_platter_pa = ptob(pfn); 231255d507a9SSeth Goldberg hat_devload(kas.a_hat, 231355d507a9SSeth Goldberg (caddr_t)(uintptr_t)rm_platter_pa, MMU_PAGESIZE, 231455d507a9SSeth Goldberg pfn, PROT_READ | PROT_WRITE | PROT_EXEC, 231555d507a9SSeth Goldberg HAT_LOAD_NOCONSIST); 231655d507a9SSeth Goldberg break; 23177c478bd9Sstevel@tonic-gate } 231855d507a9SSeth Goldberg if (pfn == btop(1*1024*1024) && use_mp) 231955d507a9SSeth Goldberg panic("No page below 1M available for starting " 232055d507a9SSeth Goldberg "other processors or for resuming from system-suspend"); 2321843e1988Sjohnlev #endif /* !__xpv */ 23227c478bd9Sstevel@tonic-gate } 23237c478bd9Sstevel@tonic-gate 23247c478bd9Sstevel@tonic-gate /* 23257c478bd9Sstevel@tonic-gate * Initialize the platform-specific parts of a page_t. 23267c478bd9Sstevel@tonic-gate */ 23277c478bd9Sstevel@tonic-gate void 23287c478bd9Sstevel@tonic-gate add_physmem_cb(page_t *pp, pfn_t pnum) 23297c478bd9Sstevel@tonic-gate { 23307c478bd9Sstevel@tonic-gate pp->p_pagenum = pnum; 23317c478bd9Sstevel@tonic-gate pp->p_mapping = NULL; 23327c478bd9Sstevel@tonic-gate pp->p_embed = 0; 23337c478bd9Sstevel@tonic-gate pp->p_share = 0; 23347c478bd9Sstevel@tonic-gate pp->p_mlentry = 0; 23357c478bd9Sstevel@tonic-gate } 23367c478bd9Sstevel@tonic-gate 23377c478bd9Sstevel@tonic-gate /* 23387c478bd9Sstevel@tonic-gate * kphysm_init() initializes physical memory. 23397c478bd9Sstevel@tonic-gate */ 23407c478bd9Sstevel@tonic-gate static pgcnt_t 23417c478bd9Sstevel@tonic-gate kphysm_init( 2342ae115bc7Smrj page_t *pp, 23437c478bd9Sstevel@tonic-gate pgcnt_t npages) 23447c478bd9Sstevel@tonic-gate { 23457c478bd9Sstevel@tonic-gate struct memlist *pmem; 23467c478bd9Sstevel@tonic-gate struct memseg *cur_memseg; 23477c478bd9Sstevel@tonic-gate pfn_t base_pfn; 23487c478bd9Sstevel@tonic-gate pgcnt_t num; 23497c478bd9Sstevel@tonic-gate pgcnt_t pages_done = 0; 23507c478bd9Sstevel@tonic-gate uint64_t addr; 23517c478bd9Sstevel@tonic-gate uint64_t size; 23527c478bd9Sstevel@tonic-gate extern pfn_t ddiphysmin; 23537c478bd9Sstevel@tonic-gate 23547c478bd9Sstevel@tonic-gate ASSERT(page_hash != NULL && page_hashsz != 0); 23557c478bd9Sstevel@tonic-gate 2356ae115bc7Smrj cur_memseg = memseg_base; 23577c478bd9Sstevel@tonic-gate for (pmem = phys_avail; pmem && npages; pmem = pmem->next) { 23587c478bd9Sstevel@tonic-gate /* 23597c478bd9Sstevel@tonic-gate * In a 32 bit kernel can't use higher memory if we're 23607c478bd9Sstevel@tonic-gate * not booting in PAE mode. This check takes care of that. 23617c478bd9Sstevel@tonic-gate */ 23627c478bd9Sstevel@tonic-gate addr = pmem->address; 23637c478bd9Sstevel@tonic-gate size = pmem->size; 23647c478bd9Sstevel@tonic-gate if (btop(addr) > physmax) 23657c478bd9Sstevel@tonic-gate continue; 23667c478bd9Sstevel@tonic-gate 23677c478bd9Sstevel@tonic-gate /* 23687c478bd9Sstevel@tonic-gate * align addr and size - they may not be at page boundaries 23697c478bd9Sstevel@tonic-gate */ 23707c478bd9Sstevel@tonic-gate if ((addr & MMU_PAGEOFFSET) != 0) { 23717c478bd9Sstevel@tonic-gate addr += MMU_PAGEOFFSET; 23727c478bd9Sstevel@tonic-gate addr &= ~(uint64_t)MMU_PAGEOFFSET; 23737c478bd9Sstevel@tonic-gate size -= addr - pmem->address; 23747c478bd9Sstevel@tonic-gate } 23757c478bd9Sstevel@tonic-gate 23766bb54764Skchow /* only process pages below or equal to physmax */ 23776bb54764Skchow if ((btop(addr + size) - 1) > physmax) 23786bb54764Skchow size = ptob(physmax - btop(addr) + 1); 23797c478bd9Sstevel@tonic-gate 23807c478bd9Sstevel@tonic-gate num = btop(size); 23817c478bd9Sstevel@tonic-gate if (num == 0) 23827c478bd9Sstevel@tonic-gate continue; 23837c478bd9Sstevel@tonic-gate 23847c478bd9Sstevel@tonic-gate if (num > npages) 23857c478bd9Sstevel@tonic-gate num = npages; 23867c478bd9Sstevel@tonic-gate 23877c478bd9Sstevel@tonic-gate npages -= num; 23887c478bd9Sstevel@tonic-gate pages_done += num; 23897c478bd9Sstevel@tonic-gate base_pfn = btop(addr); 23907c478bd9Sstevel@tonic-gate 23917c478bd9Sstevel@tonic-gate if (prom_debug) 23927c478bd9Sstevel@tonic-gate prom_printf("MEMSEG addr=0x%" PRIx64 23937c478bd9Sstevel@tonic-gate " pgs=0x%lx pfn 0x%lx-0x%lx\n", 23947c478bd9Sstevel@tonic-gate addr, num, base_pfn, base_pfn + num); 23957c478bd9Sstevel@tonic-gate 23967c478bd9Sstevel@tonic-gate /* 2397ae115bc7Smrj * Ignore pages below ddiphysmin to simplify ddi memory 23987c478bd9Sstevel@tonic-gate * allocation with non-zero addr_lo requests. 23997c478bd9Sstevel@tonic-gate */ 24007c478bd9Sstevel@tonic-gate if (base_pfn < ddiphysmin) { 2401ae115bc7Smrj if (base_pfn + num <= ddiphysmin) 24027c478bd9Sstevel@tonic-gate continue; 24037c478bd9Sstevel@tonic-gate pp += (ddiphysmin - base_pfn); 24047c478bd9Sstevel@tonic-gate num -= (ddiphysmin - base_pfn); 24057c478bd9Sstevel@tonic-gate base_pfn = ddiphysmin; 24067c478bd9Sstevel@tonic-gate } 2407ae115bc7Smrj 24087c478bd9Sstevel@tonic-gate /* 24097c478bd9Sstevel@tonic-gate * Build the memsegs entry 24107c478bd9Sstevel@tonic-gate */ 24117c478bd9Sstevel@tonic-gate cur_memseg->pages = pp; 24127c478bd9Sstevel@tonic-gate cur_memseg->epages = pp + num; 24137c478bd9Sstevel@tonic-gate cur_memseg->pages_base = base_pfn; 24147c478bd9Sstevel@tonic-gate cur_memseg->pages_end = base_pfn + num; 24157c478bd9Sstevel@tonic-gate 24167c478bd9Sstevel@tonic-gate /* 2417ae115bc7Smrj * Insert into memseg list in decreasing pfn range order. 24187c478bd9Sstevel@tonic-gate * Low memory is typically more fragmented such that this 24197c478bd9Sstevel@tonic-gate * ordering keeps the larger ranges at the front of the list 24207c478bd9Sstevel@tonic-gate * for code that searches memseg. 2421ae115bc7Smrj * This ASSERTS that the memsegs coming in from boot are in 2422ae115bc7Smrj * increasing physical address order and not contiguous. 24237c478bd9Sstevel@tonic-gate */ 2424ae115bc7Smrj if (memsegs != NULL) { 2425ae115bc7Smrj ASSERT(cur_memseg->pages_base >= memsegs->pages_end); 2426ae115bc7Smrj cur_memseg->next = memsegs; 24277c478bd9Sstevel@tonic-gate } 2428ae115bc7Smrj memsegs = cur_memseg; 24297c478bd9Sstevel@tonic-gate 24307c478bd9Sstevel@tonic-gate /* 24317c478bd9Sstevel@tonic-gate * add_physmem() initializes the PSM part of the page 24327c478bd9Sstevel@tonic-gate * struct by calling the PSM back with add_physmem_cb(). 24337c478bd9Sstevel@tonic-gate * In addition it coalesces pages into larger pages as 24347c478bd9Sstevel@tonic-gate * it initializes them. 24357c478bd9Sstevel@tonic-gate */ 24367c478bd9Sstevel@tonic-gate add_physmem(pp, num, base_pfn); 24377c478bd9Sstevel@tonic-gate cur_memseg++; 24387c478bd9Sstevel@tonic-gate availrmem_initial += num; 24397c478bd9Sstevel@tonic-gate availrmem += num; 24407c478bd9Sstevel@tonic-gate 2441ae115bc7Smrj pp += num; 24427c478bd9Sstevel@tonic-gate } 24437c478bd9Sstevel@tonic-gate 24447c478bd9Sstevel@tonic-gate PRM_DEBUG(availrmem_initial); 24457c478bd9Sstevel@tonic-gate PRM_DEBUG(availrmem); 24467c478bd9Sstevel@tonic-gate PRM_DEBUG(freemem); 24477c478bd9Sstevel@tonic-gate build_pfn_hash(); 24487c478bd9Sstevel@tonic-gate return (pages_done); 24497c478bd9Sstevel@tonic-gate } 24507c478bd9Sstevel@tonic-gate 24517c478bd9Sstevel@tonic-gate /* 24527c478bd9Sstevel@tonic-gate * Kernel VM initialization. 24537c478bd9Sstevel@tonic-gate */ 24547c478bd9Sstevel@tonic-gate static void 24557c478bd9Sstevel@tonic-gate kvm_init(void) 24567c478bd9Sstevel@tonic-gate { 24577c478bd9Sstevel@tonic-gate ASSERT((((uintptr_t)s_text) & MMU_PAGEOFFSET) == 0); 24587c478bd9Sstevel@tonic-gate 24597c478bd9Sstevel@tonic-gate /* 24607c478bd9Sstevel@tonic-gate * Put the kernel segments in kernel address space. 24617c478bd9Sstevel@tonic-gate */ 24627c478bd9Sstevel@tonic-gate rw_enter(&kas.a_lock, RW_WRITER); 24637c478bd9Sstevel@tonic-gate as_avlinit(&kas); 24647c478bd9Sstevel@tonic-gate 24657c478bd9Sstevel@tonic-gate (void) seg_attach(&kas, s_text, e_moddata - s_text, &ktextseg); 24667c478bd9Sstevel@tonic-gate (void) segkmem_create(&ktextseg); 24677c478bd9Sstevel@tonic-gate 24687c478bd9Sstevel@tonic-gate (void) seg_attach(&kas, (caddr_t)valloc_base, valloc_sz, &kvalloc); 24697c478bd9Sstevel@tonic-gate (void) segkmem_create(&kvalloc); 24707c478bd9Sstevel@tonic-gate 247135b1ab99Sjosephb (void) seg_attach(&kas, kernelheap, 247235b1ab99Sjosephb ekernelheap - kernelheap, &kvseg); 24737c478bd9Sstevel@tonic-gate (void) segkmem_create(&kvseg); 24747c478bd9Sstevel@tonic-gate 2475ae115bc7Smrj if (core_size > 0) { 2476ae115bc7Smrj PRM_POINT("attaching kvseg_core"); 2477ae115bc7Smrj (void) seg_attach(&kas, (caddr_t)core_base, core_size, 2478ae115bc7Smrj &kvseg_core); 2479ae115bc7Smrj (void) segkmem_create(&kvseg_core); 2480ae115bc7Smrj } 2481ad23a2dbSjohansen 2482ae115bc7Smrj if (segziosize > 0) { 2483ae115bc7Smrj PRM_POINT("attaching segzio"); 2484ad23a2dbSjohansen (void) seg_attach(&kas, segzio_base, mmu_ptob(segziosize), 2485ad23a2dbSjohansen &kzioseg); 2486ad23a2dbSjohansen (void) segkmem_zio_create(&kzioseg); 2487ad23a2dbSjohansen 2488ad23a2dbSjohansen /* create zio area covering new segment */ 2489ad23a2dbSjohansen segkmem_zio_init(segzio_base, mmu_ptob(segziosize)); 2490ad23a2dbSjohansen } 24917c478bd9Sstevel@tonic-gate 2492ae115bc7Smrj (void) seg_attach(&kas, kdi_segdebugbase, kdi_segdebugsize, &kdebugseg); 24937c478bd9Sstevel@tonic-gate (void) segkmem_create(&kdebugseg); 24947c478bd9Sstevel@tonic-gate 24957c478bd9Sstevel@tonic-gate rw_exit(&kas.a_lock); 24967c478bd9Sstevel@tonic-gate 24977c478bd9Sstevel@tonic-gate /* 24987c478bd9Sstevel@tonic-gate * Ensure that the red zone at kernelbase is never accessible. 24997c478bd9Sstevel@tonic-gate */ 2500ae115bc7Smrj PRM_POINT("protecting redzone"); 25017c478bd9Sstevel@tonic-gate (void) as_setprot(&kas, (caddr_t)kernelbase, KERNEL_REDZONE_SIZE, 0); 25027c478bd9Sstevel@tonic-gate 25037c478bd9Sstevel@tonic-gate /* 25047c478bd9Sstevel@tonic-gate * Make the text writable so that it can be hot patched by DTrace. 25057c478bd9Sstevel@tonic-gate */ 25067c478bd9Sstevel@tonic-gate (void) as_setprot(&kas, s_text, e_modtext - s_text, 25077c478bd9Sstevel@tonic-gate PROT_READ | PROT_WRITE | PROT_EXEC); 25087c478bd9Sstevel@tonic-gate 25097c478bd9Sstevel@tonic-gate /* 25107c478bd9Sstevel@tonic-gate * Make data writable until end. 25117c478bd9Sstevel@tonic-gate */ 25127c478bd9Sstevel@tonic-gate (void) as_setprot(&kas, s_data, e_moddata - s_data, 25137c478bd9Sstevel@tonic-gate PROT_READ | PROT_WRITE | PROT_EXEC); 25147c478bd9Sstevel@tonic-gate } 25157c478bd9Sstevel@tonic-gate 2516843e1988Sjohnlev #ifndef __xpv 25177c478bd9Sstevel@tonic-gate /* 25181d03c31eSjohnlev * Solaris adds an entry for Write Combining caching to the PAT 25197c478bd9Sstevel@tonic-gate */ 25201d03c31eSjohnlev static uint64_t pat_attr_reg = PAT_DEFAULT_ATTRIBUTE; 25217c478bd9Sstevel@tonic-gate 25227c478bd9Sstevel@tonic-gate void 25231d03c31eSjohnlev pat_sync(void) 25247c478bd9Sstevel@tonic-gate { 25251d03c31eSjohnlev ulong_t cr0, cr0_orig, cr4; 25267c478bd9Sstevel@tonic-gate 25271d03c31eSjohnlev if (!(x86_feature & X86_PAT)) 25287c478bd9Sstevel@tonic-gate return; 25291d03c31eSjohnlev cr0_orig = cr0 = getcr0(); 25301d03c31eSjohnlev cr4 = getcr4(); 25317c478bd9Sstevel@tonic-gate 25321d03c31eSjohnlev /* disable caching and flush all caches and TLBs */ 25331d03c31eSjohnlev cr0 |= CR0_CD; 25341d03c31eSjohnlev cr0 &= ~CR0_NW; 25351d03c31eSjohnlev setcr0(cr0); 25361d03c31eSjohnlev invalidate_cache(); 25371d03c31eSjohnlev if (cr4 & CR4_PGE) { 25381d03c31eSjohnlev setcr4(cr4 & ~(ulong_t)CR4_PGE); 25391d03c31eSjohnlev setcr4(cr4); 25401d03c31eSjohnlev } else { 25411d03c31eSjohnlev reload_cr3(); 25427c478bd9Sstevel@tonic-gate } 25437c478bd9Sstevel@tonic-gate 25441d03c31eSjohnlev /* add our entry to the PAT */ 25451d03c31eSjohnlev wrmsr(REG_PAT, pat_attr_reg); 25467c478bd9Sstevel@tonic-gate 25471d03c31eSjohnlev /* flush TLBs and cache again, then reenable cr0 caching */ 25481d03c31eSjohnlev if (cr4 & CR4_PGE) { 25491d03c31eSjohnlev setcr4(cr4 & ~(ulong_t)CR4_PGE); 25501d03c31eSjohnlev setcr4(cr4); 25511d03c31eSjohnlev } else { 25521d03c31eSjohnlev reload_cr3(); 25537c478bd9Sstevel@tonic-gate } 25547c478bd9Sstevel@tonic-gate invalidate_cache(); 25557c478bd9Sstevel@tonic-gate setcr0(cr0_orig); 25567c478bd9Sstevel@tonic-gate } 25577c478bd9Sstevel@tonic-gate 25581d03c31eSjohnlev #endif /* !__xpv */ 25597c478bd9Sstevel@tonic-gate 2560ed5289f9SKen Erickson #if defined(_SOFT_HOSTID) 2561ed5289f9SKen Erickson /* 2562ed5289f9SKen Erickson * On platforms that do not have a hardware serial number, attempt 2563ed5289f9SKen Erickson * to set one based on the contents of /etc/hostid. If this file does 2564ed5289f9SKen Erickson * not exist, assume that we are to generate a new hostid and set 2565ed5289f9SKen Erickson * it in the kernel, for subsequent saving by a userland process 2566ed5289f9SKen Erickson * once the system is up and the root filesystem is mounted r/w. 2567ed5289f9SKen Erickson * 256896d008c2SKen Erickson * In order to gracefully support upgrade on OpenSolaris, if 256996d008c2SKen Erickson * /etc/hostid does not exist, we will attempt to get a serial number 257096d008c2SKen Erickson * using the legacy method (/kernel/misc/sysinit). 257196d008c2SKen Erickson * 2572ed5289f9SKen Erickson * In an attempt to make the hostid less prone to abuse 2573ed5289f9SKen Erickson * (for license circumvention, etc), we store it in /etc/hostid 2574ed5289f9SKen Erickson * in rot47 format. 2575ed5289f9SKen Erickson */ 2576ed5289f9SKen Erickson extern volatile unsigned long tenmicrodata; 257796d008c2SKen Erickson static int atoi(char *); 2578ed5289f9SKen Erickson 2579ed5289f9SKen Erickson static int32_t 2580ed5289f9SKen Erickson set_soft_hostid(void) 2581ed5289f9SKen Erickson { 2582ed5289f9SKen Erickson struct _buf *file; 2583ed5289f9SKen Erickson char tokbuf[MAXNAMELEN]; 2584ed5289f9SKen Erickson token_t token; 2585ed5289f9SKen Erickson int done = 0; 2586ed5289f9SKen Erickson u_longlong_t tmp; 258796d008c2SKen Erickson int i; 2588ed5289f9SKen Erickson int32_t hostid = -1; 2589ed5289f9SKen Erickson unsigned char *c; 2590ed5289f9SKen Erickson hrtime_t tsc; 2591ed5289f9SKen Erickson 2592ed5289f9SKen Erickson /* 2593ed5289f9SKen Erickson * If /etc/hostid file not found, we'd like to get a pseudo 2594ed5289f9SKen Erickson * random number to use at the hostid. A nice way to do this 2595ed5289f9SKen Erickson * is to read the real time clock. To remain xen-compatible, 2596ed5289f9SKen Erickson * we can't poke the real hardware, so we use tsc_read() to 2597ed5289f9SKen Erickson * read the real time clock. However, there is an ominous 2598ed5289f9SKen Erickson * warning in tsc_read that says it can return zero, so we 2599ed5289f9SKen Erickson * deal with that possibility by falling back to using the 2600ed5289f9SKen Erickson * (hopefully random enough) value in tenmicrodata. 2601ed5289f9SKen Erickson */ 2602ed5289f9SKen Erickson 2603ed5289f9SKen Erickson if ((file = kobj_open_file(hostid_file)) == (struct _buf *)-1) { 260496d008c2SKen Erickson /* 260596d008c2SKen Erickson * hostid file not found - try to load sysinit module 260696d008c2SKen Erickson * and see if it has a nonzero hostid value...use that 260796d008c2SKen Erickson * instead of generating a new hostid here if so. 260896d008c2SKen Erickson */ 260996d008c2SKen Erickson if ((i = modload("misc", "sysinit")) != -1) { 261096d008c2SKen Erickson if (strlen(hw_serial) > 0) 261196d008c2SKen Erickson hostid = (int32_t)atoi(hw_serial); 261296d008c2SKen Erickson (void) modunload(i); 261396d008c2SKen Erickson } 261496d008c2SKen Erickson if (hostid == -1) { 261596d008c2SKen Erickson tsc = tsc_read(); 261696d008c2SKen Erickson if (tsc == 0) /* tsc_read can return zero sometimes */ 261796d008c2SKen Erickson hostid = (int32_t)tenmicrodata & 0x0CFFFFF; 261896d008c2SKen Erickson else 261996d008c2SKen Erickson hostid = (int32_t)tsc & 0x0CFFFFF; 262096d008c2SKen Erickson } 2621ed5289f9SKen Erickson } else { 2622ed5289f9SKen Erickson /* hostid file found */ 2623ed5289f9SKen Erickson while (!done) { 2624ed5289f9SKen Erickson token = kobj_lex(file, tokbuf, sizeof (tokbuf)); 2625ed5289f9SKen Erickson 2626ed5289f9SKen Erickson switch (token) { 2627ed5289f9SKen Erickson case POUND: 2628ed5289f9SKen Erickson /* 2629ed5289f9SKen Erickson * skip comments 2630ed5289f9SKen Erickson */ 2631ed5289f9SKen Erickson kobj_find_eol(file); 2632ed5289f9SKen Erickson break; 2633ed5289f9SKen Erickson case STRING: 2634ed5289f9SKen Erickson /* 2635ed5289f9SKen Erickson * un-rot47 - obviously this 2636ed5289f9SKen Erickson * nonsense is ascii-specific 2637ed5289f9SKen Erickson */ 2638ed5289f9SKen Erickson for (c = (unsigned char *)tokbuf; 2639ed5289f9SKen Erickson *c != '\0'; c++) { 2640ed5289f9SKen Erickson *c += 47; 2641ed5289f9SKen Erickson if (*c > '~') 2642ed5289f9SKen Erickson *c -= 94; 2643ed5289f9SKen Erickson else if (*c < '!') 2644ed5289f9SKen Erickson *c += 94; 2645ed5289f9SKen Erickson } 2646ed5289f9SKen Erickson /* 2647ed5289f9SKen Erickson * now we should have a real number 2648ed5289f9SKen Erickson */ 2649ed5289f9SKen Erickson 2650ed5289f9SKen Erickson if (kobj_getvalue(tokbuf, &tmp) != 0) 2651ed5289f9SKen Erickson kobj_file_err(CE_WARN, file, 2652ed5289f9SKen Erickson "Bad value %s for hostid", 2653ed5289f9SKen Erickson tokbuf); 2654ed5289f9SKen Erickson else 2655ed5289f9SKen Erickson hostid = (int32_t)tmp; 2656ed5289f9SKen Erickson 2657ed5289f9SKen Erickson break; 2658ed5289f9SKen Erickson case EOF: 2659ed5289f9SKen Erickson done = 1; 2660ed5289f9SKen Erickson /* FALLTHROUGH */ 2661ed5289f9SKen Erickson case NEWLINE: 2662ed5289f9SKen Erickson kobj_newline(file); 2663ed5289f9SKen Erickson break; 2664ed5289f9SKen Erickson default: 2665ed5289f9SKen Erickson break; 2666ed5289f9SKen Erickson 2667ed5289f9SKen Erickson } 2668ed5289f9SKen Erickson } 2669ed5289f9SKen Erickson if (hostid == -1) /* didn't find a hostid string */ 2670ed5289f9SKen Erickson kobj_file_err(CE_WARN, file, 2671ed5289f9SKen Erickson "hostid missing or corrupt"); 2672ed5289f9SKen Erickson 2673ed5289f9SKen Erickson kobj_close_file(file); 2674ed5289f9SKen Erickson } 2675ed5289f9SKen Erickson /* 2676ed5289f9SKen Erickson * hostid is now the value read from /etc/hostid, or the 2677ed5289f9SKen Erickson * new hostid we generated in this routine or -1 if not set. 2678ed5289f9SKen Erickson */ 2679ed5289f9SKen Erickson return (hostid); 2680ed5289f9SKen Erickson } 268196d008c2SKen Erickson 268296d008c2SKen Erickson static int 268396d008c2SKen Erickson atoi(char *p) 268496d008c2SKen Erickson { 268596d008c2SKen Erickson int i = 0; 268696d008c2SKen Erickson 268796d008c2SKen Erickson while (*p != '\0') 268896d008c2SKen Erickson i = 10 * i + (*p++ - '0'); 268996d008c2SKen Erickson 269096d008c2SKen Erickson return (i); 269196d008c2SKen Erickson } 269296d008c2SKen Erickson 2693ed5289f9SKen Erickson #endif /* _SOFT_HOSTID */ 2694ed5289f9SKen Erickson 26957c478bd9Sstevel@tonic-gate void 269645916cd2Sjpk get_system_configuration(void) 26977c478bd9Sstevel@tonic-gate { 26987c478bd9Sstevel@tonic-gate char prop[32]; 26997c478bd9Sstevel@tonic-gate u_longlong_t nodes_ll, cpus_pernode_ll, lvalue; 27007c478bd9Sstevel@tonic-gate 270135b1ab99Sjosephb if (BOP_GETPROPLEN(bootops, "nodes") > sizeof (prop) || 270235b1ab99Sjosephb BOP_GETPROP(bootops, "nodes", prop) < 0 || 270335b1ab99Sjosephb kobj_getvalue(prop, &nodes_ll) == -1 || 270435b1ab99Sjosephb nodes_ll > MAXNODES || 270535b1ab99Sjosephb BOP_GETPROPLEN(bootops, "cpus_pernode") > sizeof (prop) || 270635b1ab99Sjosephb BOP_GETPROP(bootops, "cpus_pernode", prop) < 0 || 270735b1ab99Sjosephb kobj_getvalue(prop, &cpus_pernode_ll) == -1) { 27087c478bd9Sstevel@tonic-gate system_hardware.hd_nodes = 1; 27097c478bd9Sstevel@tonic-gate system_hardware.hd_cpus_per_node = 0; 27107c478bd9Sstevel@tonic-gate } else { 27117c478bd9Sstevel@tonic-gate system_hardware.hd_nodes = (int)nodes_ll; 27127c478bd9Sstevel@tonic-gate system_hardware.hd_cpus_per_node = (int)cpus_pernode_ll; 27137c478bd9Sstevel@tonic-gate } 271435b1ab99Sjosephb 271535b1ab99Sjosephb if (BOP_GETPROPLEN(bootops, "kernelbase") > sizeof (prop) || 271635b1ab99Sjosephb BOP_GETPROP(bootops, "kernelbase", prop) < 0 || 271735b1ab99Sjosephb kobj_getvalue(prop, &lvalue) == -1) 271835b1ab99Sjosephb eprom_kernelbase = NULL; 27197c478bd9Sstevel@tonic-gate else 272035b1ab99Sjosephb eprom_kernelbase = (uintptr_t)lvalue; 27217c478bd9Sstevel@tonic-gate 272235b1ab99Sjosephb if (BOP_GETPROPLEN(bootops, "segmapsize") > sizeof (prop) || 272335b1ab99Sjosephb BOP_GETPROP(bootops, "segmapsize", prop) < 0 || 272435b1ab99Sjosephb kobj_getvalue(prop, &lvalue) == -1) 27257c478bd9Sstevel@tonic-gate segmapsize = SEGMAPDEFAULT; 272635b1ab99Sjosephb else 27277c478bd9Sstevel@tonic-gate segmapsize = (uintptr_t)lvalue; 27287c478bd9Sstevel@tonic-gate 272935b1ab99Sjosephb if (BOP_GETPROPLEN(bootops, "segmapfreelists") > sizeof (prop) || 273035b1ab99Sjosephb BOP_GETPROP(bootops, "segmapfreelists", prop) < 0 || 273135b1ab99Sjosephb kobj_getvalue(prop, &lvalue) == -1) 27327c478bd9Sstevel@tonic-gate segmapfreelists = 0; /* use segmap driver default */ 273335b1ab99Sjosephb else 27347c478bd9Sstevel@tonic-gate segmapfreelists = (int)lvalue; 27354944b02eSkchow 2736aac11643Sjosephb /* physmem used to be here, but moved much earlier to fakebop.c */ 27377c478bd9Sstevel@tonic-gate } 27387c478bd9Sstevel@tonic-gate 27397c478bd9Sstevel@tonic-gate /* 27407c478bd9Sstevel@tonic-gate * Add to a memory list. 27417c478bd9Sstevel@tonic-gate * start = start of new memory segment 27427c478bd9Sstevel@tonic-gate * len = length of new memory segment in bytes 27437c478bd9Sstevel@tonic-gate * new = pointer to a new struct memlist 27447c478bd9Sstevel@tonic-gate * memlistp = memory list to which to add segment. 27457c478bd9Sstevel@tonic-gate */ 2746ae115bc7Smrj void 27477c478bd9Sstevel@tonic-gate memlist_add( 27487c478bd9Sstevel@tonic-gate uint64_t start, 27497c478bd9Sstevel@tonic-gate uint64_t len, 27507c478bd9Sstevel@tonic-gate struct memlist *new, 27517c478bd9Sstevel@tonic-gate struct memlist **memlistp) 27527c478bd9Sstevel@tonic-gate { 27537c478bd9Sstevel@tonic-gate struct memlist *cur; 27547c478bd9Sstevel@tonic-gate uint64_t end = start + len; 27557c478bd9Sstevel@tonic-gate 27567c478bd9Sstevel@tonic-gate new->address = start; 27577c478bd9Sstevel@tonic-gate new->size = len; 27587c478bd9Sstevel@tonic-gate 27597c478bd9Sstevel@tonic-gate cur = *memlistp; 27607c478bd9Sstevel@tonic-gate 27617c478bd9Sstevel@tonic-gate while (cur) { 27627c478bd9Sstevel@tonic-gate if (cur->address >= end) { 27637c478bd9Sstevel@tonic-gate new->next = cur; 27647c478bd9Sstevel@tonic-gate *memlistp = new; 27657c478bd9Sstevel@tonic-gate new->prev = cur->prev; 27667c478bd9Sstevel@tonic-gate cur->prev = new; 27677c478bd9Sstevel@tonic-gate return; 27687c478bd9Sstevel@tonic-gate } 27697c478bd9Sstevel@tonic-gate ASSERT(cur->address + cur->size <= start); 27707c478bd9Sstevel@tonic-gate if (cur->next == NULL) { 27717c478bd9Sstevel@tonic-gate cur->next = new; 27727c478bd9Sstevel@tonic-gate new->prev = cur; 27737c478bd9Sstevel@tonic-gate new->next = NULL; 27747c478bd9Sstevel@tonic-gate return; 27757c478bd9Sstevel@tonic-gate } 27767c478bd9Sstevel@tonic-gate memlistp = &cur->next; 27777c478bd9Sstevel@tonic-gate cur = cur->next; 27787c478bd9Sstevel@tonic-gate } 27797c478bd9Sstevel@tonic-gate } 27807c478bd9Sstevel@tonic-gate 27817c478bd9Sstevel@tonic-gate void 27827c478bd9Sstevel@tonic-gate kobj_vmem_init(vmem_t **text_arena, vmem_t **data_arena) 27837c478bd9Sstevel@tonic-gate { 27847c478bd9Sstevel@tonic-gate size_t tsize = e_modtext - modtext; 27857c478bd9Sstevel@tonic-gate size_t dsize = e_moddata - moddata; 27867c478bd9Sstevel@tonic-gate 27877c478bd9Sstevel@tonic-gate *text_arena = vmem_create("module_text", tsize ? modtext : NULL, tsize, 27887c478bd9Sstevel@tonic-gate 1, segkmem_alloc, segkmem_free, heaptext_arena, 0, VM_SLEEP); 27897c478bd9Sstevel@tonic-gate *data_arena = vmem_create("module_data", dsize ? moddata : NULL, dsize, 27907c478bd9Sstevel@tonic-gate 1, segkmem_alloc, segkmem_free, heap32_arena, 0, VM_SLEEP); 27917c478bd9Sstevel@tonic-gate } 27927c478bd9Sstevel@tonic-gate 27937c478bd9Sstevel@tonic-gate caddr_t 27947c478bd9Sstevel@tonic-gate kobj_text_alloc(vmem_t *arena, size_t size) 27957c478bd9Sstevel@tonic-gate { 27967c478bd9Sstevel@tonic-gate return (vmem_alloc(arena, size, VM_SLEEP | VM_BESTFIT)); 27977c478bd9Sstevel@tonic-gate } 27987c478bd9Sstevel@tonic-gate 27997c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 28007c478bd9Sstevel@tonic-gate caddr_t 28017c478bd9Sstevel@tonic-gate kobj_texthole_alloc(caddr_t addr, size_t size) 28027c478bd9Sstevel@tonic-gate { 28037c478bd9Sstevel@tonic-gate panic("unexpected call to kobj_texthole_alloc()"); 28047c478bd9Sstevel@tonic-gate /*NOTREACHED*/ 28057c478bd9Sstevel@tonic-gate return (0); 28067c478bd9Sstevel@tonic-gate } 28077c478bd9Sstevel@tonic-gate 28087c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 28097c478bd9Sstevel@tonic-gate void 28107c478bd9Sstevel@tonic-gate kobj_texthole_free(caddr_t addr, size_t size) 28117c478bd9Sstevel@tonic-gate { 28127c478bd9Sstevel@tonic-gate panic("unexpected call to kobj_texthole_free()"); 28137c478bd9Sstevel@tonic-gate } 28147c478bd9Sstevel@tonic-gate 28157c478bd9Sstevel@tonic-gate /* 28167c478bd9Sstevel@tonic-gate * This is called just after configure() in startup(). 28177c478bd9Sstevel@tonic-gate * 28187c478bd9Sstevel@tonic-gate * The ISALIST concept is a bit hopeless on Intel, because 28197c478bd9Sstevel@tonic-gate * there's no guarantee of an ever-more-capable processor 28207c478bd9Sstevel@tonic-gate * given that various parts of the instruction set may appear 28217c478bd9Sstevel@tonic-gate * and disappear between different implementations. 28227c478bd9Sstevel@tonic-gate * 28237c478bd9Sstevel@tonic-gate * While it would be possible to correct it and even enhance 28247c478bd9Sstevel@tonic-gate * it somewhat, the explicit hardware capability bitmask allows 28257c478bd9Sstevel@tonic-gate * more flexibility. 28267c478bd9Sstevel@tonic-gate * 28277c478bd9Sstevel@tonic-gate * So, we just leave this alone. 28287c478bd9Sstevel@tonic-gate */ 28297c478bd9Sstevel@tonic-gate void 28307c478bd9Sstevel@tonic-gate setx86isalist(void) 28317c478bd9Sstevel@tonic-gate { 28327c478bd9Sstevel@tonic-gate char *tp; 28337c478bd9Sstevel@tonic-gate size_t len; 28347c478bd9Sstevel@tonic-gate extern char *isa_list; 28357c478bd9Sstevel@tonic-gate 28367c478bd9Sstevel@tonic-gate #define TBUFSIZE 1024 28377c478bd9Sstevel@tonic-gate 28387c478bd9Sstevel@tonic-gate tp = kmem_alloc(TBUFSIZE, KM_SLEEP); 28397c478bd9Sstevel@tonic-gate *tp = '\0'; 28407c478bd9Sstevel@tonic-gate 28417c478bd9Sstevel@tonic-gate #if defined(__amd64) 28427c478bd9Sstevel@tonic-gate (void) strcpy(tp, "amd64 "); 28437c478bd9Sstevel@tonic-gate #endif 28447c478bd9Sstevel@tonic-gate 28457c478bd9Sstevel@tonic-gate switch (x86_vendor) { 28467c478bd9Sstevel@tonic-gate case X86_VENDOR_Intel: 28477c478bd9Sstevel@tonic-gate case X86_VENDOR_AMD: 28487c478bd9Sstevel@tonic-gate case X86_VENDOR_TM: 28497c478bd9Sstevel@tonic-gate if (x86_feature & X86_CMOV) { 28507c478bd9Sstevel@tonic-gate /* 28517c478bd9Sstevel@tonic-gate * Pentium Pro or later 28527c478bd9Sstevel@tonic-gate */ 28537c478bd9Sstevel@tonic-gate (void) strcat(tp, "pentium_pro"); 28547c478bd9Sstevel@tonic-gate (void) strcat(tp, x86_feature & X86_MMX ? 28557c478bd9Sstevel@tonic-gate "+mmx pentium_pro " : " "); 28567c478bd9Sstevel@tonic-gate } 28577c478bd9Sstevel@tonic-gate /*FALLTHROUGH*/ 28587c478bd9Sstevel@tonic-gate case X86_VENDOR_Cyrix: 28597c478bd9Sstevel@tonic-gate /* 28607c478bd9Sstevel@tonic-gate * The Cyrix 6x86 does not have any Pentium features 28617c478bd9Sstevel@tonic-gate * accessible while not at privilege level 0. 28627c478bd9Sstevel@tonic-gate */ 28637c478bd9Sstevel@tonic-gate if (x86_feature & X86_CPUID) { 28647c478bd9Sstevel@tonic-gate (void) strcat(tp, "pentium"); 28657c478bd9Sstevel@tonic-gate (void) strcat(tp, x86_feature & X86_MMX ? 28667c478bd9Sstevel@tonic-gate "+mmx pentium " : " "); 28677c478bd9Sstevel@tonic-gate } 28687c478bd9Sstevel@tonic-gate break; 28697c478bd9Sstevel@tonic-gate default: 28707c478bd9Sstevel@tonic-gate break; 28717c478bd9Sstevel@tonic-gate } 28727c478bd9Sstevel@tonic-gate (void) strcat(tp, "i486 i386 i86"); 28737c478bd9Sstevel@tonic-gate len = strlen(tp) + 1; /* account for NULL at end of string */ 28747c478bd9Sstevel@tonic-gate isa_list = strcpy(kmem_alloc(len, KM_SLEEP), tp); 28757c478bd9Sstevel@tonic-gate kmem_free(tp, TBUFSIZE); 28767c478bd9Sstevel@tonic-gate 28777c478bd9Sstevel@tonic-gate #undef TBUFSIZE 28787c478bd9Sstevel@tonic-gate } 28797c478bd9Sstevel@tonic-gate 28807c478bd9Sstevel@tonic-gate 28817c478bd9Sstevel@tonic-gate #ifdef __amd64 28827c478bd9Sstevel@tonic-gate 28837c478bd9Sstevel@tonic-gate void * 28847c478bd9Sstevel@tonic-gate device_arena_alloc(size_t size, int vm_flag) 28857c478bd9Sstevel@tonic-gate { 28867c478bd9Sstevel@tonic-gate return (vmem_alloc(device_arena, size, vm_flag)); 28877c478bd9Sstevel@tonic-gate } 28887c478bd9Sstevel@tonic-gate 28897c478bd9Sstevel@tonic-gate void 28907c478bd9Sstevel@tonic-gate device_arena_free(void *vaddr, size_t size) 28917c478bd9Sstevel@tonic-gate { 28927c478bd9Sstevel@tonic-gate vmem_free(device_arena, vaddr, size); 28937c478bd9Sstevel@tonic-gate } 28947c478bd9Sstevel@tonic-gate 2895ae115bc7Smrj #else /* __i386 */ 28967c478bd9Sstevel@tonic-gate 28977c478bd9Sstevel@tonic-gate void * 28987c478bd9Sstevel@tonic-gate device_arena_alloc(size_t size, int vm_flag) 28997c478bd9Sstevel@tonic-gate { 29007c478bd9Sstevel@tonic-gate caddr_t vaddr; 29017c478bd9Sstevel@tonic-gate uintptr_t v; 29027c478bd9Sstevel@tonic-gate size_t start; 29037c478bd9Sstevel@tonic-gate size_t end; 29047c478bd9Sstevel@tonic-gate 29057c478bd9Sstevel@tonic-gate vaddr = vmem_alloc(heap_arena, size, vm_flag); 29067c478bd9Sstevel@tonic-gate if (vaddr == NULL) 29077c478bd9Sstevel@tonic-gate return (NULL); 29087c478bd9Sstevel@tonic-gate 29097c478bd9Sstevel@tonic-gate v = (uintptr_t)vaddr; 29107c478bd9Sstevel@tonic-gate ASSERT(v >= kernelbase); 2911ae115bc7Smrj ASSERT(v + size <= valloc_base); 29127c478bd9Sstevel@tonic-gate 29137c478bd9Sstevel@tonic-gate start = btop(v - kernelbase); 29147c478bd9Sstevel@tonic-gate end = btop(v + size - 1 - kernelbase); 29157c478bd9Sstevel@tonic-gate ASSERT(start < toxic_bit_map_len); 29167c478bd9Sstevel@tonic-gate ASSERT(end < toxic_bit_map_len); 29177c478bd9Sstevel@tonic-gate 29187c478bd9Sstevel@tonic-gate while (start <= end) { 29197c478bd9Sstevel@tonic-gate BT_ATOMIC_SET(toxic_bit_map, start); 29207c478bd9Sstevel@tonic-gate ++start; 29217c478bd9Sstevel@tonic-gate } 29227c478bd9Sstevel@tonic-gate return (vaddr); 29237c478bd9Sstevel@tonic-gate } 29247c478bd9Sstevel@tonic-gate 29257c478bd9Sstevel@tonic-gate void 29267c478bd9Sstevel@tonic-gate device_arena_free(void *vaddr, size_t size) 29277c478bd9Sstevel@tonic-gate { 29287c478bd9Sstevel@tonic-gate uintptr_t v = (uintptr_t)vaddr; 29297c478bd9Sstevel@tonic-gate size_t start; 29307c478bd9Sstevel@tonic-gate size_t end; 29317c478bd9Sstevel@tonic-gate 29327c478bd9Sstevel@tonic-gate ASSERT(v >= kernelbase); 2933ae115bc7Smrj ASSERT(v + size <= valloc_base); 29347c478bd9Sstevel@tonic-gate 29357c478bd9Sstevel@tonic-gate start = btop(v - kernelbase); 29367c478bd9Sstevel@tonic-gate end = btop(v + size - 1 - kernelbase); 29377c478bd9Sstevel@tonic-gate ASSERT(start < toxic_bit_map_len); 29387c478bd9Sstevel@tonic-gate ASSERT(end < toxic_bit_map_len); 29397c478bd9Sstevel@tonic-gate 29407c478bd9Sstevel@tonic-gate while (start <= end) { 29417c478bd9Sstevel@tonic-gate ASSERT(BT_TEST(toxic_bit_map, start) != 0); 29427c478bd9Sstevel@tonic-gate BT_ATOMIC_CLEAR(toxic_bit_map, start); 29437c478bd9Sstevel@tonic-gate ++start; 29447c478bd9Sstevel@tonic-gate } 29457c478bd9Sstevel@tonic-gate vmem_free(heap_arena, vaddr, size); 29467c478bd9Sstevel@tonic-gate } 29477c478bd9Sstevel@tonic-gate 29487c478bd9Sstevel@tonic-gate /* 29497c478bd9Sstevel@tonic-gate * returns 1st address in range that is in device arena, or NULL 29507c478bd9Sstevel@tonic-gate * if len is not NULL it returns the length of the toxic range 29517c478bd9Sstevel@tonic-gate */ 29527c478bd9Sstevel@tonic-gate void * 29537c478bd9Sstevel@tonic-gate device_arena_contains(void *vaddr, size_t size, size_t *len) 29547c478bd9Sstevel@tonic-gate { 29557c478bd9Sstevel@tonic-gate uintptr_t v = (uintptr_t)vaddr; 29567c478bd9Sstevel@tonic-gate uintptr_t eaddr = v + size; 29577c478bd9Sstevel@tonic-gate size_t start; 29587c478bd9Sstevel@tonic-gate size_t end; 29597c478bd9Sstevel@tonic-gate 29607c478bd9Sstevel@tonic-gate /* 29617c478bd9Sstevel@tonic-gate * if called very early by kmdb, just return NULL 29627c478bd9Sstevel@tonic-gate */ 29637c478bd9Sstevel@tonic-gate if (toxic_bit_map == NULL) 29647c478bd9Sstevel@tonic-gate return (NULL); 29657c478bd9Sstevel@tonic-gate 29667c478bd9Sstevel@tonic-gate /* 29677c478bd9Sstevel@tonic-gate * First check if we're completely outside the bitmap range. 29687c478bd9Sstevel@tonic-gate */ 2969ae115bc7Smrj if (v >= valloc_base || eaddr < kernelbase) 29707c478bd9Sstevel@tonic-gate return (NULL); 29717c478bd9Sstevel@tonic-gate 29727c478bd9Sstevel@tonic-gate /* 29737c478bd9Sstevel@tonic-gate * Trim ends of search to look at only what the bitmap covers. 29747c478bd9Sstevel@tonic-gate */ 29757c478bd9Sstevel@tonic-gate if (v < kernelbase) 29767c478bd9Sstevel@tonic-gate v = kernelbase; 29777c478bd9Sstevel@tonic-gate start = btop(v - kernelbase); 29787c478bd9Sstevel@tonic-gate end = btop(eaddr - kernelbase); 29797c478bd9Sstevel@tonic-gate if (end >= toxic_bit_map_len) 29807c478bd9Sstevel@tonic-gate end = toxic_bit_map_len; 29817c478bd9Sstevel@tonic-gate 29827c478bd9Sstevel@tonic-gate if (bt_range(toxic_bit_map, &start, &end, end) == 0) 29837c478bd9Sstevel@tonic-gate return (NULL); 29847c478bd9Sstevel@tonic-gate 29857c478bd9Sstevel@tonic-gate v = kernelbase + ptob(start); 29867c478bd9Sstevel@tonic-gate if (len != NULL) 29877c478bd9Sstevel@tonic-gate *len = ptob(end - start); 29887c478bd9Sstevel@tonic-gate return ((void *)v); 29897c478bd9Sstevel@tonic-gate } 29907c478bd9Sstevel@tonic-gate 2991ae115bc7Smrj #endif /* __i386 */ 2992