startprog64.S revision 1.3.16.2
1 1.3.16.2 jdolecek /* $NetBSD: startprog64.S,v 1.3.16.2 2017/12/03 11:36:19 jdolecek Exp $ */ 2 1.3.16.2 jdolecek /* NetBSD: startprog.S,v 1.3 2003/02/01 14:48:18 dsl Exp */ 3 1.3.16.2 jdolecek 4 1.3.16.2 jdolecek /* starts program in protected mode / flat space 5 1.3.16.2 jdolecek with given stackframe 6 1.3.16.2 jdolecek needs global variables flatcodeseg and flatdataseg 7 1.3.16.2 jdolecek (gdt offsets) 8 1.3.16.2 jdolecek derived from: NetBSD:sys/arch/i386/boot/asm.S 9 1.3.16.2 jdolecek */ 10 1.3.16.2 jdolecek 11 1.3.16.2 jdolecek /* 12 1.3.16.2 jdolecek * Ported to boot 386BSD by Julian Elischer (julian (at) tfs.com) Sept 1992 13 1.3.16.2 jdolecek * 14 1.3.16.2 jdolecek * Mach Operating System 15 1.3.16.2 jdolecek * Copyright (c) 1992, 1991 Carnegie Mellon University 16 1.3.16.2 jdolecek * All Rights Reserved. 17 1.3.16.2 jdolecek * 18 1.3.16.2 jdolecek * Permission to use, copy, modify and distribute this software and its 19 1.3.16.2 jdolecek * documentation is hereby granted, provided that both the copyright 20 1.3.16.2 jdolecek * notice and this permission notice appear in all copies of the 21 1.3.16.2 jdolecek * software, derivative works or modified versions, and any portions 22 1.3.16.2 jdolecek * thereof, and that both notices appear in supporting documentation. 23 1.3.16.2 jdolecek * 24 1.3.16.2 jdolecek * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 25 1.3.16.2 jdolecek * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND FOR 26 1.3.16.2 jdolecek * ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 27 1.3.16.2 jdolecek * 28 1.3.16.2 jdolecek * Carnegie Mellon requests users of this software to return to 29 1.3.16.2 jdolecek * 30 1.3.16.2 jdolecek * Software Distribution Coordinator or Software.Distribution (at) CS.CMU.EDU 31 1.3.16.2 jdolecek * School of Computer Science 32 1.3.16.2 jdolecek * Carnegie Mellon University 33 1.3.16.2 jdolecek * Pittsburgh PA 15213-3890 34 1.3.16.2 jdolecek * 35 1.3.16.2 jdolecek * any improvements or extensions that they make and grant Carnegie Mellon 36 1.3.16.2 jdolecek * the rights to redistribute these changes. 37 1.3.16.2 jdolecek */ 38 1.3.16.2 jdolecek 39 1.3.16.2 jdolecek /* 40 1.3.16.2 jdolecek Copyright 1988, 1989, 1990, 1991, 1992 41 1.3.16.2 jdolecek by Intel Corporation, Santa Clara, California. 42 1.3.16.2 jdolecek 43 1.3.16.2 jdolecek All Rights Reserved 44 1.3.16.2 jdolecek 45 1.3.16.2 jdolecek Permission to use, copy, modify, and distribute this software and 46 1.3.16.2 jdolecek its documentation for any purpose and without fee is hereby 47 1.3.16.2 jdolecek granted, provided that the above copyright notice appears in all 48 1.3.16.2 jdolecek copies and that both the copyright notice and this permission notice 49 1.3.16.2 jdolecek appear in supporting documentation, and that the name of Intel 50 1.3.16.2 jdolecek not be used in advertising or publicity pertaining to distribution 51 1.3.16.2 jdolecek of the software without specific, written prior permission. 52 1.3.16.2 jdolecek 53 1.3.16.2 jdolecek INTEL DISCLAIMS ALL WARRANTIES WITH REGARD TO THIS SOFTWARE 54 1.3.16.2 jdolecek INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS, 55 1.3.16.2 jdolecek IN NO EVENT SHALL INTEL BE LIABLE FOR ANY SPECIAL, INDIRECT, OR 56 1.3.16.2 jdolecek CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM 57 1.3.16.2 jdolecek LOSS OF USE, DATA OR PROFITS, WHETHER IN ACTION OF CONTRACT, 58 1.3.16.2 jdolecek NEGLIGENCE, OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION 59 1.3.16.2 jdolecek WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. 60 1.3.16.2 jdolecek */ 61 1.3.16.2 jdolecek 62 1.3.16.2 jdolecek #include <machine/asm.h> 63 1.3.16.2 jdolecek #include <machine/specialreg.h> 64 1.3.16.2 jdolecek 65 1.3.16.2 jdolecek #define CODE_SEGMENT 0x08 66 1.3.16.2 jdolecek #define DATA_SEGMENT 0x10 67 1.3.16.2 jdolecek 68 1.3.16.2 jdolecek .align 16 69 1.3.16.2 jdolecek .globl _C_LABEL(startprog64) 70 1.3.16.2 jdolecek _C_LABEL(startprog64): 71 1.3.16.2 jdolecek .quad 0 72 1.3.16.2 jdolecek 73 1.3.16.2 jdolecek .globl _C_LABEL(startprog64_size) 74 1.3.16.2 jdolecek _C_LABEL(startprog64_size): 75 1.3.16.2 jdolecek .long startprog64_end - _C_LABEL(startprog64_start) 76 1.3.16.2 jdolecek 77 1.3.16.2 jdolecek .text 78 1.3.16.2 jdolecek .p2align 4,,15 79 1.3.16.2 jdolecek 80 1.3.16.2 jdolecek /* 81 1.3.16.2 jdolecek * startprog64(loadddr,entry,stack,kern_load,kern_start,kern_size) 82 1.3.16.2 jdolecek */ 83 1.3.16.2 jdolecek ENTRY(startprog64_start) 84 1.3.16.2 jdolecek start: 85 1.3.16.2 jdolecek /* 86 1.3.16.2 jdolecek * This function is to call the loaded kernel's start() with 87 1.3.16.2 jdolecek * 32bit segment mode from x64 mode. 88 1.3.16.2 jdolecek * %rdi: kernel start address 89 1.3.16.2 jdolecek * %rsi: loaded kernel address 90 1.3.16.2 jdolecek * %rdx: stack address 91 1.3.16.2 jdolecek * %rcx: loaded kernel size 92 1.3.16.2 jdolecek * %r8 : loaded start address 93 1.3.16.2 jdolecek * %r9 : kernel entry address 94 1.3.16.2 jdolecek */ 95 1.3.16.2 jdolecek 96 1.3.16.2 jdolecek cld /* LynxOS depends on it */ 97 1.3.16.2 jdolecek 98 1.3.16.2 jdolecek cli 99 1.3.16.2 jdolecek 100 1.3.16.2 jdolecek /* Copy kernel */ 101 1.3.16.2 jdolecek mov %rcx, %r12 /* original kernel size */ 102 1.3.16.2 jdolecek movq %rdi, %r11 /* for misaligned check */ 103 1.3.16.2 jdolecek 104 1.3.16.2 jdolecek #if !defined(NO_OVERLAP) 105 1.3.16.2 jdolecek movq %rdi, %r13 106 1.3.16.2 jdolecek subq %rsi, %r13 107 1.3.16.2 jdolecek #endif 108 1.3.16.2 jdolecek 109 1.3.16.2 jdolecek shrq $3, %rcx /* count for copy by words */ 110 1.3.16.2 jdolecek jz 8f /* j if less than 8 bytes */ 111 1.3.16.2 jdolecek 112 1.3.16.2 jdolecek lea -8(%rdi, %r12), %r14 /* target address of last 8 */ 113 1.3.16.2 jdolecek mov -8(%rsi, %r12), %r15 /* get last word */ 114 1.3.16.2 jdolecek #if !defined(NO_OVERLAP) 115 1.3.16.2 jdolecek cmpq %r12, %r13 /* overlapping? */ 116 1.3.16.2 jdolecek jb 10f 117 1.3.16.2 jdolecek #endif 118 1.3.16.2 jdolecek 119 1.3.16.2 jdolecek /* 120 1.3.16.2 jdolecek * Non-overlaping, copy forwards. 121 1.3.16.2 jdolecek * Newer Intel cpus (Nehalem) will do 16byte read/write transfers 122 1.3.16.2 jdolecek * if %ecx is more than 76. 123 1.3.16.2 jdolecek * AMD might do something similar some day. 124 1.3.16.2 jdolecek */ 125 1.3.16.2 jdolecek and $7, %r11 /* destination misaligned ? */ 126 1.3.16.2 jdolecek jnz 2f 127 1.3.16.2 jdolecek rep 128 1.3.16.2 jdolecek movsq 129 1.3.16.2 jdolecek mov %r15, (%r14) /* write last word */ 130 1.3.16.2 jdolecek jmp .Lcopy_done 131 1.3.16.2 jdolecek 132 1.3.16.2 jdolecek /* 133 1.3.16.2 jdolecek * Destination misaligned 134 1.3.16.2 jdolecek * AMD say it is better to align the destination (not the source). 135 1.3.16.2 jdolecek * This will also re-align copies if the source and dest are both 136 1.3.16.2 jdolecek * misaligned by the same amount) 137 1.3.16.2 jdolecek * (I think Nehalem will use its accelerated copy if the source 138 1.3.16.2 jdolecek * and destination have the same alignment.) 139 1.3.16.2 jdolecek */ 140 1.3.16.2 jdolecek 2: 141 1.3.16.2 jdolecek lea -9(%r11, %r12), %rcx /* post re-alignment count */ 142 1.3.16.2 jdolecek neg %r11 /* now -1 .. -7 */ 143 1.3.16.2 jdolecek mov (%rsi), %r12 /* get first word */ 144 1.3.16.2 jdolecek mov %rdi, %r13 /* target for first word */ 145 1.3.16.2 jdolecek lea 8(%rsi, %r11), %rsi 146 1.3.16.2 jdolecek lea 8(%rdi, %r11), %rdi 147 1.3.16.2 jdolecek shr $3, %rcx 148 1.3.16.2 jdolecek rep 149 1.3.16.2 jdolecek movsq 150 1.3.16.2 jdolecek mov %r12, (%r13) /* write first word */ 151 1.3.16.2 jdolecek mov %r15, (%r14) /* write last word */ 152 1.3.16.2 jdolecek jmp .Lcopy_done 153 1.3.16.2 jdolecek 154 1.3.16.2 jdolecek #if !defined(NO_OVERLAP) 155 1.3.16.2 jdolecek /* Must copy backwards. 156 1.3.16.2 jdolecek * Reverse copy is probably easy to code faster than 'rep movds' 157 1.3.16.2 jdolecek * since that requires (IIRC) an extra clock every 3 iterations (AMD). 158 1.3.16.2 jdolecek * However I don't suppose anything cares that much! 159 1.3.16.2 jdolecek * The big cost is the std/cld pair - reputedly 50+ cycles on Netburst P4. 160 1.3.16.2 jdolecek * The copy is aligned with the buffer start (more likely to 161 1.3.16.2 jdolecek * be a multiple of 8 than the end). 162 1.3.16.2 jdolecek */ 163 1.3.16.2 jdolecek 10: 164 1.3.16.2 jdolecek lea -8(%rsi, %rcx, 8), %rsi 165 1.3.16.2 jdolecek lea -8(%rdi, %rcx, 8), %rdi 166 1.3.16.2 jdolecek std 167 1.3.16.2 jdolecek rep 168 1.3.16.2 jdolecek movsq 169 1.3.16.2 jdolecek cld 170 1.3.16.2 jdolecek mov %r15, (%r14) /* write last bytes */ 171 1.3.16.2 jdolecek jmp .Lcopy_done 172 1.3.16.2 jdolecek #endif 173 1.3.16.2 jdolecek 174 1.3.16.2 jdolecek /* Less than 8 bytes to copy, copy by bytes */ 175 1.3.16.2 jdolecek /* Intel Nehalem optimise 'rep movsb' for <= 7 bytes (9-15 clocks). 176 1.3.16.2 jdolecek * For longer transfers it is 50+ ! 177 1.3.16.2 jdolecek */ 178 1.3.16.2 jdolecek 8: mov %r12, %rcx 179 1.3.16.2 jdolecek 180 1.3.16.2 jdolecek #if !defined(NO_OVERLAP) 181 1.3.16.2 jdolecek cmpq %r12, %r13 /* overlapping? */ 182 1.3.16.2 jdolecek jb 81f 183 1.3.16.2 jdolecek #endif 184 1.3.16.2 jdolecek 185 1.3.16.2 jdolecek /* nope, copy forwards. */ 186 1.3.16.2 jdolecek rep 187 1.3.16.2 jdolecek movsb 188 1.3.16.2 jdolecek jmp .Lcopy_done 189 1.3.16.2 jdolecek 190 1.3.16.2 jdolecek #if !defined(NO_OVERLAP) 191 1.3.16.2 jdolecek /* Must copy backwards */ 192 1.3.16.2 jdolecek 81: 193 1.3.16.2 jdolecek lea -1(%rsi, %rcx), %rsi 194 1.3.16.2 jdolecek lea -1(%rdi, %rcx), %rdi 195 1.3.16.2 jdolecek std 196 1.3.16.2 jdolecek rep 197 1.3.16.2 jdolecek movsb 198 1.3.16.2 jdolecek cld 199 1.3.16.2 jdolecek #endif 200 1.3.16.2 jdolecek /* End of copy kernel */ 201 1.3.16.2 jdolecek .Lcopy_done: 202 1.3.16.2 jdolecek 203 1.3.16.2 jdolecek mov %r8, %rdi /* %rdi: loaded start address */ 204 1.3.16.2 jdolecek mov %r9, %rsi /* %rsi: kernel entry address */ 205 1.3.16.2 jdolecek 206 1.3.16.2 jdolecek /* Prepare jump address */ 207 1.3.16.2 jdolecek lea (start32a - start)(%rdi), %rax 208 1.3.16.2 jdolecek movl %eax, (start32r - start)(%rdi) 209 1.3.16.2 jdolecek 210 1.3.16.2 jdolecek /* Setup GDT */ 211 1.3.16.2 jdolecek lea (gdt - start)(%rdi), %rax 212 1.3.16.2 jdolecek mov %rax, (gdtrr - start)(%rdi) 213 1.3.16.2 jdolecek lgdt (gdtr - start)(%rdi) 214 1.3.16.2 jdolecek 215 1.3.16.2 jdolecek /* Jump to set %cs */ 216 1.3.16.2 jdolecek ljmp *(start32r - start)(%rdi) 217 1.3.16.2 jdolecek 218 1.3.16.2 jdolecek .align 4 219 1.3.16.2 jdolecek .code32 220 1.3.16.2 jdolecek start32a: 221 1.3.16.2 jdolecek movl $DATA_SEGMENT, %eax 222 1.3.16.2 jdolecek movw %ax, %ds 223 1.3.16.2 jdolecek movw %ax, %es 224 1.3.16.2 jdolecek movw %ax, %fs 225 1.3.16.2 jdolecek movw %ax, %gs 226 1.3.16.2 jdolecek movw %ax, %ss 227 1.3.16.2 jdolecek 228 1.3.16.2 jdolecek movl %edx, %esp 229 1.3.16.2 jdolecek 230 1.3.16.2 jdolecek /* Disable Paging in CR0 */ 231 1.3.16.2 jdolecek movl %cr0, %eax 232 1.3.16.2 jdolecek andl $(~CR0_PG), %eax 233 1.3.16.2 jdolecek movl %eax, %cr0 234 1.3.16.2 jdolecek 235 1.3.16.2 jdolecek /* Disable PAE in CR4 */ 236 1.3.16.2 jdolecek movl %cr4, %eax 237 1.3.16.2 jdolecek andl $(~CR4_PAE), %eax 238 1.3.16.2 jdolecek movl %eax, %cr4 239 1.3.16.2 jdolecek 240 1.3.16.2 jdolecek jmp start32b 241 1.3.16.2 jdolecek 242 1.3.16.2 jdolecek .align 4 243 1.3.16.2 jdolecek start32b: 244 1.3.16.2 jdolecek xor %eax, %eax 245 1.3.16.2 jdolecek call *%esi 246 1.3.16.2 jdolecek 247 1.3.16.2 jdolecek .align 16 248 1.3.16.2 jdolecek start32r: 249 1.3.16.2 jdolecek .long 0 250 1.3.16.2 jdolecek .long CODE_SEGMENT 251 1.3.16.2 jdolecek .align 16 252 1.3.16.2 jdolecek gdt: 253 1.3.16.2 jdolecek .long 0, 0 254 1.3.16.2 jdolecek .byte 0xff, 0xff, 0x00, 0x00, 0x00, 0x9f, 0xcf, 0x00 255 1.3.16.2 jdolecek .byte 0xff, 0xff, 0x00, 0x00, 0x00, 0x93, 0xcf, 0x00 256 1.3.16.2 jdolecek gdtr: 257 1.3.16.2 jdolecek .word gdtr - gdt 258 1.3.16.2 jdolecek gdtrr: 259 1.3.16.2 jdolecek .quad 260 1.3.16.2 jdolecek start32end: 261 1.3.16.2 jdolecek /* Space for the stack */ 262 1.3.16.2 jdolecek .align 16 263 1.3.16.2 jdolecek .space 8192 264 1.3.16.2 jdolecek startprog64_end: 265
Indexes created Sat Oct 25 16:10:12 GMT 2025