Thu Sep 18 21:35:17 2008 UTC ()
Annotate some of the assembler.
Remove 2 redundant register-register moves.


(dsl)
diff -r1.11 -r1.12 src/sys/arch/amd64/amd64/copy.S
cvs diff -r1.11 -r1.12 src/sys/arch/amd64/amd64/copy.S (expand / switch to unified diff)

--- src/sys/arch/amd64/amd64/copy.S 2008/05/30 11:03:29 1.11
+++ src/sys/arch/amd64/amd64/copy.S 2008/09/18 21:35:17 1.12
@@ -1,14 +1,14 @@ @@ -1,14 +1,14 @@
1/* $NetBSD: copy.S,v 1.11 2008/05/30 11:03:29 ad Exp $ */ 1/* $NetBSD: copy.S,v 1.12 2008/09/18 21:35:17 dsl Exp $ */
2 2
3/* 3/*
4 * Copyright (c) 2001 Wasabi Systems, Inc. 4 * Copyright (c) 2001 Wasabi Systems, Inc.
5 * All rights reserved. 5 * All rights reserved.
6 * 6 *
7 * Written by Frank van der Linden for Wasabi Systems, Inc. 7 * Written by Frank van der Linden for Wasabi Systems, Inc.
8 * 8 *
9 * Redistribution and use in source and binary forms, with or without 9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions 10 * modification, are permitted provided that the following conditions
11 * are met: 11 * are met:
12 * 1. Redistributions of source code must retain the above copyright 12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer. 13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright 14 * 2. Redistributions in binary form must reproduce the above copyright
@@ -136,95 +136,98 @@ ENTRY(kcopy) @@ -136,95 +136,98 @@ ENTRY(kcopy)
136 rep 136 rep
137 movsq 137 movsq
138 138
139 movq %rdx,%rcx 139 movq %rdx,%rcx
140 andl $7,%ecx # any bytes left? 140 andl $7,%ecx # any bytes left?
141 rep 141 rep
142 movsb 142 movsb
143 143
144 GET_CURPCB(%rdx) 144 GET_CURPCB(%rdx)
145 popq PCB_ONFAULT(%rdx) 145 popq PCB_ONFAULT(%rdx)
146 xorq %rax,%rax 146 xorq %rax,%rax
147 ret 147 ret
148 148
 149# Using 'rep movs' to copy backwards is not as fast as for forwards copies
 150# and ought not be done when the copy doesn't acually overlap.
 151# However kcopy() isn't used any that looks even vaguely used often.
 152# I'm also not sure it is ever asked to do overlapping copies!
 153
1491: addq %rcx,%rdi # copy backward 1541: addq %rcx,%rdi # copy backward
150 addq %rcx,%rsi 155 addq %rcx,%rsi
151 std 156 std
152 andq $7,%rcx # any fractional bytes? 157 andq $7,%rcx # any fractional bytes?
153 decq %rdi 158 decq %rdi
154 decq %rsi 159 decq %rsi
155 rep 160 rep
156 movsb 161 movsb
157 movq %rdx,%rcx # copy remainder by 64-bit words 162 movq %rdx,%rcx # copy remainder by 64-bit words
158 shrq $3,%rcx 163 shrq $3,%rcx
159 subq $7,%rsi 164 subq $7,%rsi
160 subq $7,%rdi 165 subq $7,%rdi
161 rep 166 rep
162 movsq 167 movsq
163 cld 168 cld
164 169
165 GET_CURPCB(%rdx) 170 GET_CURPCB(%rdx)
166 popq PCB_ONFAULT(%rdx) 171 popq PCB_ONFAULT(%rdx)
167 xorq %rax,%rax 172 xorq %rax,%rax
168 ret 173 ret
169 174
170ENTRY(copyout) 175ENTRY(copyout)
171 DEFERRED_SWITCH_CHECK 176 DEFERRED_SWITCH_CHECK
172 pushq $0 177 pushq $0 # Restored to PCB_ONFAULT(%rdx)
173 178
174 xchgq %rdi,%rsi 179 xchgq %rdi,%rsi # kernel address to %rsi, user to %rdi
175 movq %rdx,%rax 180 movq %rdx,%rax # save transfer length (bytes)
176  181
177 movq %rdi,%rdx 182 addq %rdi,%rdx # end address to %rdx
178 addq %rax,%rdx 183 jc _C_LABEL(copy_efault) # jump if wraps
179 jc _C_LABEL(copy_efault) 
180 movq $VM_MAXUSER_ADDRESS,%r8 184 movq $VM_MAXUSER_ADDRESS,%r8
181 cmpq %r8,%rdx 185 cmpq %r8,%rdx
182 ja _C_LABEL(copy_efault) 186 ja _C_LABEL(copy_efault) # jump if end in kernel space
183 187
184 GET_CURPCB(%rdx) 188 GET_CURPCB(%rdx)
185 leaq _C_LABEL(copy_fault)(%rip),%r11 189 leaq _C_LABEL(copy_fault)(%rip),%r11
186 movq %r11,PCB_ONFAULT(%rdx) 190 movq %r11,PCB_ONFAULT(%rdx) # prime fault handler
187 191
188 movq %rax,%rcx 192 movq %rax,%rcx # length
189 shrq $3,%rcx 193 shrq $3,%rcx # count of 8-byte words
190 rep 194 rep
191 movsq 195 movsq # copy from %rsi to %rdi
192 movb %al,%cl 196 movb %al,%cl
193 andb $7,%cl 197 andb $7,%cl # remaining number of bytes
194 rep 198 rep
195 movsb 199 movsb # copy remaining bytes
196 200
197 popq PCB_ONFAULT(%rdx) 201 popq PCB_ONFAULT(%rdx)
198 xorl %eax,%eax 202 xorl %eax,%eax
199 ret 203 ret
200 DEFERRED_SWITCH_CALL 204 DEFERRED_SWITCH_CALL
201 205
202ENTRY(copyin) 206ENTRY(copyin)
203 DEFERRED_SWITCH_CHECK 207 DEFERRED_SWITCH_CHECK
204 GET_CURPCB(%rax) 208 GET_CURPCB(%rax)
205 pushq $0 209 pushq $0
206 leaq _C_LABEL(copy_fault)(%rip),%r11 210 leaq _C_LABEL(copy_fault)(%rip),%r11
207 movq %r11,PCB_ONFAULT(%rax) 211 movq %r11,PCB_ONFAULT(%rax)
208 212
209 xchgq %rdi,%rsi 213 xchgq %rdi,%rsi
210 movq %rdx,%rax 214 movq %rdx,%rax
211 215
212 movq %rsi,%rdx 216 addq %rsi,%rdx # Check source address not wrapped
213 addq %rax,%rdx 
214 jc _C_LABEL(copy_efault) 217 jc _C_LABEL(copy_efault)
215 movq $VM_MAXUSER_ADDRESS,%r8 218 movq $VM_MAXUSER_ADDRESS,%r8
216 cmpq %r8,%rdx 219 cmpq %r8,%rdx
217 ja _C_LABEL(copy_efault) 220 ja _C_LABEL(copy_efault) # j if end in kernel space
218 221
2193: /* bcopy(%rsi, %rdi, %rax); */ 2223: /* bcopy(%rsi, %rdi, %rax); */
220 movq %rax,%rcx 223 movq %rax,%rcx
221 shrq $3,%rcx 224 shrq $3,%rcx
222 rep 225 rep
223 movsq 226 movsq
224 movb %al,%cl 227 movb %al,%cl
225 andb $7,%cl 228 andb $7,%cl
226 rep 229 rep
227 movsb 230 movsb
228 231
229 GET_CURPCB(%rdx) 232 GET_CURPCB(%rdx)
230 popq PCB_ONFAULT(%rdx) 233 popq PCB_ONFAULT(%rdx)