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> !! 10 #include <asm/cprefix.h>
13 #include <asm/ptrace.h> 11 #include <asm/ptrace.h>
14 12
15 /* Work around cpp -rob */ 13 /* Work around cpp -rob */
16 #define ALLOC #alloc 14 #define ALLOC #alloc
17 #define EXECINSTR #execinstr 15 #define EXECINSTR #execinstr
18 #define EX(x,y,a,b) 16 #define EX(x,y,a,b) \
19 98: x,y; 17 98: x,y; \
20 .section .fixup,ALLOC,EXECINSTR; 18 .section .fixup,ALLOC,EXECINSTR; \
21 .align 4; 19 .align 4; \
22 99: retl; !! 20 99: ba 30f; \
23 a, b, %o0; 21 a, b, %o0; \
24 .section __ex_table,ALLOC; 22 .section __ex_table,ALLOC; \
25 .align 4; 23 .align 4; \
26 .word 98b, 99b; 24 .word 98b, 99b; \
27 .text; 25 .text; \
28 .align 4 26 .align 4
29 27
30 #define STORE(source, base, offset, n) !! 28 #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; 29 .section __ex_table,ALLOC; \
37 .align 4; 30 .align 4; \
38 .word 98b, 99b; !! 31 .word start, 0, end, handler; \
39 .text; 32 .text; \
40 .align 4; !! 33 .align 4
41 <<
42 #define STORE_LAST(source, base, offset, n) <<
43 EX(std source, [base - offset - n], <<
44 add %o1, offset + n); <<
45 34
46 /* Please don't change these macros, unless yo 35 /* Please don't change these macros, unless you change the logic
47 * in the .fixup section below as well. 36 * in the .fixup section below as well.
48 * Store 64 bytes at (BASE + OFFSET) using val 37 * Store 64 bytes at (BASE + OFFSET) using value SOURCE. */
49 #define ZERO_BIG_BLOCK(base, offset, source) !! 38 #define ZERO_BIG_BLOCK(base, offset, source) \
50 STORE(source, base, offset, 0x00); !! 39 std source, [base + offset + 0x00]; \
51 STORE(source, base, offset, 0x08); !! 40 std source, [base + offset + 0x08]; \
52 STORE(source, base, offset, 0x10); !! 41 std source, [base + offset + 0x10]; \
53 STORE(source, base, offset, 0x18); !! 42 std source, [base + offset + 0x18]; \
54 STORE(source, base, offset, 0x20); !! 43 std source, [base + offset + 0x20]; \
55 STORE(source, base, offset, 0x28); !! 44 std source, [base + offset + 0x28]; \
56 STORE(source, base, offset, 0x30); !! 45 std source, [base + offset + 0x30]; \
57 STORE(source, base, offset, 0x38); !! 46 std source, [base + offset + 0x38];
58 47
59 #define ZERO_LAST_BLOCKS(base, offset, source) 48 #define ZERO_LAST_BLOCKS(base, offset, source) \
60 STORE_LAST(source, base, offset, 0x38) !! 49 std source, [base - offset - 0x38]; \
61 STORE_LAST(source, base, offset, 0x30) !! 50 std source, [base - offset - 0x30]; \
62 STORE_LAST(source, base, offset, 0x28) !! 51 std source, [base - offset - 0x28]; \
63 STORE_LAST(source, base, offset, 0x20) !! 52 std source, [base - offset - 0x20]; \
64 STORE_LAST(source, base, offset, 0x18) !! 53 std source, [base - offset - 0x18]; \
65 STORE_LAST(source, base, offset, 0x10) !! 54 std source, [base - offset - 0x10]; \
66 STORE_LAST(source, base, offset, 0x08) !! 55 std source, [base - offset - 0x08]; \
67 STORE_LAST(source, base, offset, 0x00) !! 56 std source, [base - offset - 0x00];
68 57
69 .text 58 .text
70 .align 4 59 .align 4
71 60
72 .globl __bzero_begin 61 .globl __bzero_begin
73 __bzero_begin: 62 __bzero_begin:
74 63
75 .globl __bzero !! 64 .globl C_LABEL(__bzero), C_LABEL(__memset),
76 .type __bzero,#function !! 65 .globl C_LABEL(memset)
77 .globl memset !! 66 .globl C_LABEL(__memset_start), C_LABEL(__memset_end)
78 EXPORT_SYMBOL(__bzero) !! 67 C_LABEL(__memset_start):
79 EXPORT_SYMBOL(memset) !! 68 C_LABEL(__memset):
80 memset: !! 69 C_LABEL(memset):
81 mov %o0, %g1 <<
82 mov 1, %g4 <<
83 and %o1, 0xff, %g3 70 and %o1, 0xff, %g3
84 sll %g3, 8, %g2 71 sll %g3, 8, %g2
85 or %g3, %g2, %g3 72 or %g3, %g2, %g3
86 sll %g3, 16, %g2 73 sll %g3, 16, %g2
87 or %g3, %g2, %g3 74 or %g3, %g2, %g3
88 b 1f 75 b 1f
89 mov %o2, %o1 76 mov %o2, %o1
90 3: 77 3:
91 cmp %o2, 3 78 cmp %o2, 3
92 be 2f 79 be 2f
93 EX(stb %g3, [%o0], sub %o1, 0) 80 EX(stb %g3, [%o0], sub %o1, 0)
94 81
95 cmp %o2, 2 82 cmp %o2, 2
96 be 2f 83 be 2f
97 EX(stb %g3, [%o0 + 0x01], sub %o1, 1) 84 EX(stb %g3, [%o0 + 0x01], sub %o1, 1)
98 85
99 EX(stb %g3, [%o0 + 0x02], sub %o1, 2) 86 EX(stb %g3, [%o0 + 0x02], sub %o1, 2)
100 2: 87 2:
101 sub %o2, 4, %o2 88 sub %o2, 4, %o2
102 add %o1, %o2, %o1 89 add %o1, %o2, %o1
103 b 4f 90 b 4f
104 sub %o0, %o2, %o0 91 sub %o0, %o2, %o0
105 92
106 __bzero: !! 93 C_LABEL(__bzero):
107 clr %g4 <<
108 mov %g0, %g3 94 mov %g0, %g3
109 1: 95 1:
110 cmp %o1, 7 96 cmp %o1, 7
111 bleu 7f 97 bleu 7f
112 andcc %o0, 3, %o2 98 andcc %o0, 3, %o2
113 99
114 bne 3b 100 bne 3b
115 4: 101 4:
116 andcc %o0, 4, %g0 102 andcc %o0, 4, %g0
117 103
118 be 2f 104 be 2f
119 mov %g3, %g2 105 mov %g3, %g2
120 106
121 EX(st %g3, [%o0], sub %o1, 0) 107 EX(st %g3, [%o0], sub %o1, 0)
122 sub %o1, 4, %o1 108 sub %o1, 4, %o1
123 add %o0, 4, %o0 109 add %o0, 4, %o0
124 2: 110 2:
125 andcc %o1, 0xffffff80, %o3 ! Now 111 andcc %o1, 0xffffff80, %o3 ! Now everything is 8 aligned and o1 is len to run
126 be 9f 112 be 9f
127 andcc %o1, 0x78, %o2 113 andcc %o1, 0x78, %o2
128 10: 114 10:
129 ZERO_BIG_BLOCK(%o0, 0x00, %g2) 115 ZERO_BIG_BLOCK(%o0, 0x00, %g2)
130 subcc %o3, 128, %o3 116 subcc %o3, 128, %o3
131 ZERO_BIG_BLOCK(%o0, 0x40, %g2) 117 ZERO_BIG_BLOCK(%o0, 0x40, %g2)
>> 118 11:
>> 119 EXT(10b, 11b, 20f)
132 bne 10b 120 bne 10b
133 add %o0, 128, %o0 121 add %o0, 128, %o0
134 122
135 orcc %o2, %g0, %g0 123 orcc %o2, %g0, %g0
136 9: 124 9:
137 be 13f 125 be 13f
138 andcc %o1, 7, %o1 126 andcc %o1, 7, %o1
139 127
140 srl %o2, 1, %o3 128 srl %o2, 1, %o3
141 set 13f, %o4 129 set 13f, %o4
142 sub %o4, %o3, %o4 130 sub %o4, %o3, %o4
143 jmp %o4 131 jmp %o4
144 add %o0, %o2, %o0 132 add %o0, %o2, %o0
145 133
>> 134 12:
146 ZERO_LAST_BLOCKS(%o0, 0x48, %g2) 135 ZERO_LAST_BLOCKS(%o0, 0x48, %g2)
147 ZERO_LAST_BLOCKS(%o0, 0x08, %g2) 136 ZERO_LAST_BLOCKS(%o0, 0x08, %g2)
148 13: 137 13:
149 be 8f 138 be 8f
150 andcc %o1, 4, %g0 139 andcc %o1, 4, %g0
151 140
152 be 1f 141 be 1f
153 andcc %o1, 2, %g0 142 andcc %o1, 2, %g0
154 143
155 EX(st %g3, [%o0], and %o1, 7) 144 EX(st %g3, [%o0], and %o1, 7)
156 add %o0, 4, %o0 145 add %o0, 4, %o0
157 1: 146 1:
158 be 1f 147 be 1f
159 andcc %o1, 1, %g0 148 andcc %o1, 1, %g0
160 149
161 EX(sth %g3, [%o0], and %o1, 3) 150 EX(sth %g3, [%o0], and %o1, 3)
162 add %o0, 2, %o0 151 add %o0, 2, %o0
163 1: 152 1:
164 bne,a 8f 153 bne,a 8f
165 EX(stb %g3, [%o0], and %o1, 1) 154 EX(stb %g3, [%o0], and %o1, 1)
166 8: 155 8:
167 b 0f !! 156 retl
168 nop !! 157 clr %o0
169 7: 158 7:
170 be 13b 159 be 13b
171 orcc %o1, 0, %g0 160 orcc %o1, 0, %g0
172 161
173 be 0f 162 be 0f
174 8: 163 8:
175 add %o0, 1, %o0 164 add %o0, 1, %o0
176 subcc %o1, 1, %o1 165 subcc %o1, 1, %o1
177 bne 8b !! 166 bne,a 8b
178 EX(stb %g3, [%o0 - 1], add %o1, 1) 167 EX(stb %g3, [%o0 - 1], add %o1, 1)
179 0: 168 0:
180 andcc %g4, 1, %g0 <<
181 be 5f <<
182 nop <<
183 retl <<
184 mov %g1, %o0 <<
185 5: <<
186 retl 169 retl
187 clr %o0 170 clr %o0
>> 171 C_LABEL(__memset_end):
188 172
189 .section .fixup,#alloc,#execinstr 173 .section .fixup,#alloc,#execinstr
190 .align 4 174 .align 4
>> 175 20:
>> 176 cmp %g2, 8
>> 177 bleu 1f
>> 178 and %o1, 0x7f, %o1
>> 179 sub %g2, 9, %g2
>> 180 add %o3, 64, %o3
>> 181 1:
>> 182 sll %g2, 3, %g2
>> 183 add %o3, %o1, %o0
>> 184 b 30f
>> 185 sub %o0, %g2, %o0
>> 186 21:
>> 187 mov 8, %o0
>> 188 and %o1, 7, %o1
>> 189 sub %o0, %g2, %o0
>> 190 sll %o0, 3, %o0
>> 191 b 30f
>> 192 add %o0, %o1, %o0
191 30: 193 30:
192 and %o1, 0x7f, %o1 !! 194 /* %o4 is faulting address, %o5 is %pc where fault occurred */
193 retl !! 195 save %sp, -104, %sp
194 add %o3, %o1, %o0 !! 196 mov %i5, %o0
>> 197 mov %i7, %o1
>> 198 call C_LABEL(lookup_fault)
>> 199 mov %i4, %o2
>> 200 ret
>> 201 restore
195 202
196 .globl __bzero_end 203 .globl __bzero_end
197 __bzero_end: 204 __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.