1 /* SPDX-License-Identifier: GPL-2.0 */ <<
2 /* linux/arch/sparc/lib/memset.S: Sparc optimi 1 /* linux/arch/sparc/lib/memset.S: Sparc optimized memset, bzero and clear_user code
3 * Copyright (C) 1991,1996 Free Software Found 2 * Copyright (C) 1991,1996 Free Software Foundation
4 * Copyright (C) 1996,1997 Jakub Jelinek (jj@s 3 * Copyright (C) 1996,1997 Jakub Jelinek (jj@sunsite.mff.cuni.cz)
5 * Copyright (C) 1996 David S. Miller (davem@c 4 * Copyright (C) 1996 David S. Miller (davem@caip.rutgers.edu)
6 * 5 *
7 * Calls to memset returns initial %o0. Calls !! 6 * Returns 0, if ok, and number of bytes not yet set if exception
8 * number of bytes not yet set if exception oc !! 7 * occurs and we were called as clear_user.
9 * clear_user. <<
10 */ 8 */
11 9
12 #include <linux/export.h> <<
13 #include <asm/ptrace.h> 10 #include <asm/ptrace.h>
14 11
15 /* Work around cpp -rob */ 12 /* Work around cpp -rob */
16 #define ALLOC #alloc 13 #define ALLOC #alloc
17 #define EXECINSTR #execinstr 14 #define EXECINSTR #execinstr
18 #define EX(x,y,a,b) 15 #define EX(x,y,a,b) \
19 98: x,y; 16 98: x,y; \
20 .section .fixup,ALLOC,EXECINSTR; 17 .section .fixup,ALLOC,EXECINSTR; \
21 .align 4; 18 .align 4; \
22 99: retl; !! 19 99: ba 30f; \
23 a, b, %o0; 20 a, b, %o0; \
24 .section __ex_table,ALLOC; 21 .section __ex_table,ALLOC; \
25 .align 4; 22 .align 4; \
26 .word 98b, 99b; 23 .word 98b, 99b; \
27 .text; 24 .text; \
28 .align 4 25 .align 4
29 26
30 #define STORE(source, base, offset, n) !! 27 #define EXT(start,end,handler) \
31 98: std source, [base + offset + n]; <<
32 .section .fixup,ALLOC,EXECINSTR; <<
33 .align 4; <<
34 99: ba 30f; <<
35 sub %o3, n - offset, %o3; <<
36 .section __ex_table,ALLOC; 28 .section __ex_table,ALLOC; \
37 .align 4; 29 .align 4; \
38 .word 98b, 99b; !! 30 .word start, 0, end, handler; \
39 .text; 31 .text; \
40 .align 4; !! 32 .align 4
41 <<
42 #define STORE_LAST(source, base, offset, n) <<
43 EX(std source, [base - offset - n], <<
44 add %o1, offset + n); <<
45 33
46 /* Please don't change these macros, unless yo 34 /* Please don't change these macros, unless you change the logic
47 * in the .fixup section below as well. 35 * in the .fixup section below as well.
48 * Store 64 bytes at (BASE + OFFSET) using val 36 * Store 64 bytes at (BASE + OFFSET) using value SOURCE. */
49 #define ZERO_BIG_BLOCK(base, offset, source) !! 37 #define ZERO_BIG_BLOCK(base, offset, source) \
50 STORE(source, base, offset, 0x00); !! 38 std source, [base + offset + 0x00]; \
51 STORE(source, base, offset, 0x08); !! 39 std source, [base + offset + 0x08]; \
52 STORE(source, base, offset, 0x10); !! 40 std source, [base + offset + 0x10]; \
53 STORE(source, base, offset, 0x18); !! 41 std source, [base + offset + 0x18]; \
54 STORE(source, base, offset, 0x20); !! 42 std source, [base + offset + 0x20]; \
55 STORE(source, base, offset, 0x28); !! 43 std source, [base + offset + 0x28]; \
56 STORE(source, base, offset, 0x30); !! 44 std source, [base + offset + 0x30]; \
57 STORE(source, base, offset, 0x38); !! 45 std source, [base + offset + 0x38];
58 46
59 #define ZERO_LAST_BLOCKS(base, offset, source) 47 #define ZERO_LAST_BLOCKS(base, offset, source) \
60 STORE_LAST(source, base, offset, 0x38) !! 48 std source, [base - offset - 0x38]; \
61 STORE_LAST(source, base, offset, 0x30) !! 49 std source, [base - offset - 0x30]; \
62 STORE_LAST(source, base, offset, 0x28) !! 50 std source, [base - offset - 0x28]; \
63 STORE_LAST(source, base, offset, 0x20) !! 51 std source, [base - offset - 0x20]; \
64 STORE_LAST(source, base, offset, 0x18) !! 52 std source, [base - offset - 0x18]; \
65 STORE_LAST(source, base, offset, 0x10) !! 53 std source, [base - offset - 0x10]; \
66 STORE_LAST(source, base, offset, 0x08) !! 54 std source, [base - offset - 0x08]; \
67 STORE_LAST(source, base, offset, 0x00) !! 55 std source, [base - offset - 0x00];
68 56
69 .text 57 .text
70 .align 4 58 .align 4
71 59
72 .globl __bzero_begin 60 .globl __bzero_begin
73 __bzero_begin: 61 __bzero_begin:
74 62
75 .globl __bzero !! 63 .globl __bzero, __memset,
76 .type __bzero,#function <<
77 .globl memset 64 .globl memset
78 EXPORT_SYMBOL(__bzero) !! 65 .globl __memset_start, __memset_end
79 EXPORT_SYMBOL(memset) !! 66 __memset_start:
>> 67 __memset:
80 memset: 68 memset:
81 mov %o0, %g1 <<
82 mov 1, %g4 <<
83 and %o1, 0xff, %g3 69 and %o1, 0xff, %g3
84 sll %g3, 8, %g2 70 sll %g3, 8, %g2
85 or %g3, %g2, %g3 71 or %g3, %g2, %g3
86 sll %g3, 16, %g2 72 sll %g3, 16, %g2
87 or %g3, %g2, %g3 73 or %g3, %g2, %g3
88 b 1f 74 b 1f
89 mov %o2, %o1 75 mov %o2, %o1
90 3: 76 3:
91 cmp %o2, 3 77 cmp %o2, 3
92 be 2f 78 be 2f
93 EX(stb %g3, [%o0], sub %o1, 0) 79 EX(stb %g3, [%o0], sub %o1, 0)
94 80
95 cmp %o2, 2 81 cmp %o2, 2
96 be 2f 82 be 2f
97 EX(stb %g3, [%o0 + 0x01], sub %o1, 1) 83 EX(stb %g3, [%o0 + 0x01], sub %o1, 1)
98 84
99 EX(stb %g3, [%o0 + 0x02], sub %o1, 2) 85 EX(stb %g3, [%o0 + 0x02], sub %o1, 2)
100 2: 86 2:
101 sub %o2, 4, %o2 87 sub %o2, 4, %o2
102 add %o1, %o2, %o1 88 add %o1, %o2, %o1
103 b 4f 89 b 4f
104 sub %o0, %o2, %o0 90 sub %o0, %o2, %o0
105 91
106 __bzero: 92 __bzero:
107 clr %g4 <<
108 mov %g0, %g3 93 mov %g0, %g3
109 1: 94 1:
110 cmp %o1, 7 95 cmp %o1, 7
111 bleu 7f 96 bleu 7f
112 andcc %o0, 3, %o2 97 andcc %o0, 3, %o2
113 98
114 bne 3b 99 bne 3b
115 4: 100 4:
116 andcc %o0, 4, %g0 101 andcc %o0, 4, %g0
117 102
118 be 2f 103 be 2f
119 mov %g3, %g2 104 mov %g3, %g2
120 105
121 EX(st %g3, [%o0], sub %o1, 0) 106 EX(st %g3, [%o0], sub %o1, 0)
122 sub %o1, 4, %o1 107 sub %o1, 4, %o1
123 add %o0, 4, %o0 108 add %o0, 4, %o0
124 2: 109 2:
125 andcc %o1, 0xffffff80, %o3 ! Now 110 andcc %o1, 0xffffff80, %o3 ! Now everything is 8 aligned and o1 is len to run
126 be 9f 111 be 9f
127 andcc %o1, 0x78, %o2 112 andcc %o1, 0x78, %o2
128 10: 113 10:
129 ZERO_BIG_BLOCK(%o0, 0x00, %g2) 114 ZERO_BIG_BLOCK(%o0, 0x00, %g2)
130 subcc %o3, 128, %o3 115 subcc %o3, 128, %o3
131 ZERO_BIG_BLOCK(%o0, 0x40, %g2) 116 ZERO_BIG_BLOCK(%o0, 0x40, %g2)
>> 117 11:
>> 118 EXT(10b, 11b, 20f)
132 bne 10b 119 bne 10b
133 add %o0, 128, %o0 120 add %o0, 128, %o0
134 121
135 orcc %o2, %g0, %g0 122 orcc %o2, %g0, %g0
136 9: 123 9:
137 be 13f 124 be 13f
138 andcc %o1, 7, %o1 125 andcc %o1, 7, %o1
139 126
140 srl %o2, 1, %o3 127 srl %o2, 1, %o3
141 set 13f, %o4 128 set 13f, %o4
142 sub %o4, %o3, %o4 129 sub %o4, %o3, %o4
143 jmp %o4 130 jmp %o4
144 add %o0, %o2, %o0 131 add %o0, %o2, %o0
145 132
>> 133 12:
146 ZERO_LAST_BLOCKS(%o0, 0x48, %g2) 134 ZERO_LAST_BLOCKS(%o0, 0x48, %g2)
147 ZERO_LAST_BLOCKS(%o0, 0x08, %g2) 135 ZERO_LAST_BLOCKS(%o0, 0x08, %g2)
148 13: 136 13:
149 be 8f 137 be 8f
150 andcc %o1, 4, %g0 138 andcc %o1, 4, %g0
151 139
152 be 1f 140 be 1f
153 andcc %o1, 2, %g0 141 andcc %o1, 2, %g0
154 142
155 EX(st %g3, [%o0], and %o1, 7) 143 EX(st %g3, [%o0], and %o1, 7)
156 add %o0, 4, %o0 144 add %o0, 4, %o0
157 1: 145 1:
158 be 1f 146 be 1f
159 andcc %o1, 1, %g0 147 andcc %o1, 1, %g0
160 148
161 EX(sth %g3, [%o0], and %o1, 3) 149 EX(sth %g3, [%o0], and %o1, 3)
162 add %o0, 2, %o0 150 add %o0, 2, %o0
163 1: 151 1:
164 bne,a 8f 152 bne,a 8f
165 EX(stb %g3, [%o0], and %o1, 1) 153 EX(stb %g3, [%o0], and %o1, 1)
166 8: 154 8:
167 b 0f !! 155 retl
168 nop !! 156 clr %o0
169 7: 157 7:
170 be 13b 158 be 13b
171 orcc %o1, 0, %g0 159 orcc %o1, 0, %g0
172 160
173 be 0f 161 be 0f
174 8: 162 8:
175 add %o0, 1, %o0 163 add %o0, 1, %o0
176 subcc %o1, 1, %o1 164 subcc %o1, 1, %o1
177 bne 8b 165 bne 8b
178 EX(stb %g3, [%o0 - 1], add %o1, 1) 166 EX(stb %g3, [%o0 - 1], add %o1, 1)
179 0: 167 0:
180 andcc %g4, 1, %g0 <<
181 be 5f <<
182 nop <<
183 retl <<
184 mov %g1, %o0 <<
185 5: <<
186 retl 168 retl
187 clr %o0 169 clr %o0
>> 170 __memset_end:
188 171
189 .section .fixup,#alloc,#execinstr 172 .section .fixup,#alloc,#execinstr
190 .align 4 173 .align 4
>> 174 20:
>> 175 cmp %g2, 8
>> 176 bleu 1f
>> 177 and %o1, 0x7f, %o1
>> 178 sub %g2, 9, %g2
>> 179 add %o3, 64, %o3
>> 180 1:
>> 181 sll %g2, 3, %g2
>> 182 add %o3, %o1, %o0
>> 183 b 30f
>> 184 sub %o0, %g2, %o0
>> 185 21:
>> 186 mov 8, %o0
>> 187 and %o1, 7, %o1
>> 188 sub %o0, %g2, %o0
>> 189 sll %o0, 3, %o0
>> 190 b 30f
>> 191 add %o0, %o1, %o0
191 30: 192 30:
192 and %o1, 0x7f, %o1 !! 193 /* %o4 is faulting address, %o5 is %pc where fault occurred */
193 retl !! 194 save %sp, -104, %sp
194 add %o3, %o1, %o0 !! 195 mov %i5, %o0
>> 196 mov %i7, %o1
>> 197 call lookup_fault
>> 198 mov %i4, %o2
>> 199 ret
>> 200 restore
195 201
196 .globl __bzero_end 202 .globl __bzero_end
197 __bzero_end: 203 __bzero_end:
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.