Loading...
1/* SPDX-License-Identifier: GPL-2.0 */
2/* linux/arch/sparc/lib/memset.S: Sparc optimized memset, bzero and clear_user code
3 * Copyright (C) 1991,1996 Free Software Foundation
4 * Copyright (C) 1996,1997 Jakub Jelinek (jj@sunsite.mff.cuni.cz)
5 * Copyright (C) 1996 David S. Miller (davem@caip.rutgers.edu)
6 *
7 * Calls to memset returns initial %o0. Calls to bzero returns 0, if ok, and
8 * number of bytes not yet set if exception occurs and we were called as
9 * clear_user.
10 */
11
12#include <asm/ptrace.h>
13#include <asm/export.h>
14
15/* Work around cpp -rob */
16#define ALLOC #alloc
17#define EXECINSTR #execinstr
18#define EX(x,y,a,b) \
1998: x,y; \
20 .section .fixup,ALLOC,EXECINSTR; \
21 .align 4; \
2299: retl; \
23 a, b, %o0; \
24 .section __ex_table,ALLOC; \
25 .align 4; \
26 .word 98b, 99b; \
27 .text; \
28 .align 4
29
30#define STORE(source, base, offset, n) \
3198: std source, [base + offset + n]; \
32 .section .fixup,ALLOC,EXECINSTR; \
33 .align 4; \
3499: ba 30f; \
35 sub %o3, n - offset, %o3; \
36 .section __ex_table,ALLOC; \
37 .align 4; \
38 .word 98b, 99b; \
39 .text; \
40 .align 4;
41
42#define STORE_LAST(source, base, offset, n) \
43 EX(std source, [base - offset - n], \
44 add %o1, offset + n);
45
46/* Please don't change these macros, unless you change the logic
47 * in the .fixup section below as well.
48 * Store 64 bytes at (BASE + OFFSET) using value SOURCE. */
49#define ZERO_BIG_BLOCK(base, offset, source) \
50 STORE(source, base, offset, 0x00); \
51 STORE(source, base, offset, 0x08); \
52 STORE(source, base, offset, 0x10); \
53 STORE(source, base, offset, 0x18); \
54 STORE(source, base, offset, 0x20); \
55 STORE(source, base, offset, 0x28); \
56 STORE(source, base, offset, 0x30); \
57 STORE(source, base, offset, 0x38);
58
59#define ZERO_LAST_BLOCKS(base, offset, source) \
60 STORE_LAST(source, base, offset, 0x38); \
61 STORE_LAST(source, base, offset, 0x30); \
62 STORE_LAST(source, base, offset, 0x28); \
63 STORE_LAST(source, base, offset, 0x20); \
64 STORE_LAST(source, base, offset, 0x18); \
65 STORE_LAST(source, base, offset, 0x10); \
66 STORE_LAST(source, base, offset, 0x08); \
67 STORE_LAST(source, base, offset, 0x00);
68
69 .text
70 .align 4
71
72 .globl __bzero_begin
73__bzero_begin:
74
75 .globl __bzero
76 .type __bzero,#function
77 .globl memset
78 EXPORT_SYMBOL(__bzero)
79 EXPORT_SYMBOL(memset)
80memset:
81 mov %o0, %g1
82 mov 1, %g4
83 and %o1, 0xff, %g3
84 sll %g3, 8, %g2
85 or %g3, %g2, %g3
86 sll %g3, 16, %g2
87 or %g3, %g2, %g3
88 b 1f
89 mov %o2, %o1
903:
91 cmp %o2, 3
92 be 2f
93 EX(stb %g3, [%o0], sub %o1, 0)
94
95 cmp %o2, 2
96 be 2f
97 EX(stb %g3, [%o0 + 0x01], sub %o1, 1)
98
99 EX(stb %g3, [%o0 + 0x02], sub %o1, 2)
1002:
101 sub %o2, 4, %o2
102 add %o1, %o2, %o1
103 b 4f
104 sub %o0, %o2, %o0
105
106__bzero:
107 clr %g4
108 mov %g0, %g3
1091:
110 cmp %o1, 7
111 bleu 7f
112 andcc %o0, 3, %o2
113
114 bne 3b
1154:
116 andcc %o0, 4, %g0
117
118 be 2f
119 mov %g3, %g2
120
121 EX(st %g3, [%o0], sub %o1, 0)
122 sub %o1, 4, %o1
123 add %o0, 4, %o0
1242:
125 andcc %o1, 0xffffff80, %o3 ! Now everything is 8 aligned and o1 is len to run
126 be 9f
127 andcc %o1, 0x78, %o2
12810:
129 ZERO_BIG_BLOCK(%o0, 0x00, %g2)
130 subcc %o3, 128, %o3
131 ZERO_BIG_BLOCK(%o0, 0x40, %g2)
132 bne 10b
133 add %o0, 128, %o0
134
135 orcc %o2, %g0, %g0
1369:
137 be 13f
138 andcc %o1, 7, %o1
139
140 srl %o2, 1, %o3
141 set 13f, %o4
142 sub %o4, %o3, %o4
143 jmp %o4
144 add %o0, %o2, %o0
145
146 ZERO_LAST_BLOCKS(%o0, 0x48, %g2)
147 ZERO_LAST_BLOCKS(%o0, 0x08, %g2)
14813:
149 be 8f
150 andcc %o1, 4, %g0
151
152 be 1f
153 andcc %o1, 2, %g0
154
155 EX(st %g3, [%o0], and %o1, 7)
156 add %o0, 4, %o0
1571:
158 be 1f
159 andcc %o1, 1, %g0
160
161 EX(sth %g3, [%o0], and %o1, 3)
162 add %o0, 2, %o0
1631:
164 bne,a 8f
165 EX(stb %g3, [%o0], and %o1, 1)
1668:
167 b 0f
168 nop
1697:
170 be 13b
171 orcc %o1, 0, %g0
172
173 be 0f
1748:
175 add %o0, 1, %o0
176 subcc %o1, 1, %o1
177 bne 8b
178 EX(stb %g3, [%o0 - 1], add %o1, 1)
1790:
180 andcc %g4, 1, %g0
181 be 5f
182 nop
183 retl
184 mov %g1, %o0
1855:
186 retl
187 clr %o0
188
189 .section .fixup,#alloc,#execinstr
190 .align 4
19130:
192 and %o1, 0x7f, %o1
193 retl
194 add %o3, %o1, %o0
195
196 .globl __bzero_end
197__bzero_end:
1/* linux/arch/sparc/lib/memset.S: Sparc optimized memset, bzero and clear_user code
2 * Copyright (C) 1991,1996 Free Software Foundation
3 * Copyright (C) 1996,1997 Jakub Jelinek (jj@sunsite.mff.cuni.cz)
4 * Copyright (C) 1996 David S. Miller (davem@caip.rutgers.edu)
5 *
6 * Calls to memset returns initial %o0. Calls to bzero returns 0, if ok, and
7 * number of bytes not yet set if exception occurs and we were called as
8 * clear_user.
9 */
10
11#include <asm/ptrace.h>
12#include <asm/export.h>
13
14/* Work around cpp -rob */
15#define ALLOC #alloc
16#define EXECINSTR #execinstr
17#define EX(x,y,a,b) \
1898: x,y; \
19 .section .fixup,ALLOC,EXECINSTR; \
20 .align 4; \
2199: ba 30f; \
22 a, b, %o0; \
23 .section __ex_table,ALLOC; \
24 .align 4; \
25 .word 98b, 99b; \
26 .text; \
27 .align 4
28
29#define EXT(start,end,handler) \
30 .section __ex_table,ALLOC; \
31 .align 4; \
32 .word start, 0, end, handler; \
33 .text; \
34 .align 4
35
36/* Please don't change these macros, unless you change the logic
37 * in the .fixup section below as well.
38 * Store 64 bytes at (BASE + OFFSET) using value SOURCE. */
39#define ZERO_BIG_BLOCK(base, offset, source) \
40 std source, [base + offset + 0x00]; \
41 std source, [base + offset + 0x08]; \
42 std source, [base + offset + 0x10]; \
43 std source, [base + offset + 0x18]; \
44 std source, [base + offset + 0x20]; \
45 std source, [base + offset + 0x28]; \
46 std source, [base + offset + 0x30]; \
47 std source, [base + offset + 0x38];
48
49#define ZERO_LAST_BLOCKS(base, offset, source) \
50 std source, [base - offset - 0x38]; \
51 std source, [base - offset - 0x30]; \
52 std source, [base - offset - 0x28]; \
53 std source, [base - offset - 0x20]; \
54 std source, [base - offset - 0x18]; \
55 std source, [base - offset - 0x10]; \
56 std source, [base - offset - 0x08]; \
57 std source, [base - offset - 0x00];
58
59 .text
60 .align 4
61
62 .globl __bzero_begin
63__bzero_begin:
64
65 .globl __bzero
66 .globl memset
67 EXPORT_SYMBOL(__bzero)
68 EXPORT_SYMBOL(memset)
69 .globl __memset_start, __memset_end
70__memset_start:
71memset:
72 mov %o0, %g1
73 mov 1, %g4
74 and %o1, 0xff, %g3
75 sll %g3, 8, %g2
76 or %g3, %g2, %g3
77 sll %g3, 16, %g2
78 or %g3, %g2, %g3
79 b 1f
80 mov %o2, %o1
813:
82 cmp %o2, 3
83 be 2f
84 EX(stb %g3, [%o0], sub %o1, 0)
85
86 cmp %o2, 2
87 be 2f
88 EX(stb %g3, [%o0 + 0x01], sub %o1, 1)
89
90 EX(stb %g3, [%o0 + 0x02], sub %o1, 2)
912:
92 sub %o2, 4, %o2
93 add %o1, %o2, %o1
94 b 4f
95 sub %o0, %o2, %o0
96
97__bzero:
98 clr %g4
99 mov %g0, %g3
1001:
101 cmp %o1, 7
102 bleu 7f
103 andcc %o0, 3, %o2
104
105 bne 3b
1064:
107 andcc %o0, 4, %g0
108
109 be 2f
110 mov %g3, %g2
111
112 EX(st %g3, [%o0], sub %o1, 0)
113 sub %o1, 4, %o1
114 add %o0, 4, %o0
1152:
116 andcc %o1, 0xffffff80, %o3 ! Now everything is 8 aligned and o1 is len to run
117 be 9f
118 andcc %o1, 0x78, %o2
11910:
120 ZERO_BIG_BLOCK(%o0, 0x00, %g2)
121 subcc %o3, 128, %o3
122 ZERO_BIG_BLOCK(%o0, 0x40, %g2)
12311:
124 EXT(10b, 11b, 20f)
125 bne 10b
126 add %o0, 128, %o0
127
128 orcc %o2, %g0, %g0
1299:
130 be 13f
131 andcc %o1, 7, %o1
132
133 srl %o2, 1, %o3
134 set 13f, %o4
135 sub %o4, %o3, %o4
136 jmp %o4
137 add %o0, %o2, %o0
138
13912:
140 ZERO_LAST_BLOCKS(%o0, 0x48, %g2)
141 ZERO_LAST_BLOCKS(%o0, 0x08, %g2)
14213:
143 be 8f
144 andcc %o1, 4, %g0
145
146 be 1f
147 andcc %o1, 2, %g0
148
149 EX(st %g3, [%o0], and %o1, 7)
150 add %o0, 4, %o0
1511:
152 be 1f
153 andcc %o1, 1, %g0
154
155 EX(sth %g3, [%o0], and %o1, 3)
156 add %o0, 2, %o0
1571:
158 bne,a 8f
159 EX(stb %g3, [%o0], and %o1, 1)
1608:
161 b 0f
162 nop
1637:
164 be 13b
165 orcc %o1, 0, %g0
166
167 be 0f
1688:
169 add %o0, 1, %o0
170 subcc %o1, 1, %o1
171 bne 8b
172 EX(stb %g3, [%o0 - 1], add %o1, 1)
1730:
174 andcc %g4, 1, %g0
175 be 5f
176 nop
177 retl
178 mov %g1, %o0
1795:
180 retl
181 clr %o0
182__memset_end:
183
184 .section .fixup,#alloc,#execinstr
185 .align 4
18620:
187 cmp %g2, 8
188 bleu 1f
189 and %o1, 0x7f, %o1
190 sub %g2, 9, %g2
191 add %o3, 64, %o3
1921:
193 sll %g2, 3, %g2
194 add %o3, %o1, %o0
195 b 30f
196 sub %o0, %g2, %o0
19721:
198 mov 8, %o0
199 and %o1, 7, %o1
200 sub %o0, %g2, %o0
201 sll %o0, 3, %o0
202 b 30f
203 add %o0, %o1, %o0
20430:
205/* %o4 is faulting address, %o5 is %pc where fault occurred */
206 save %sp, -104, %sp
207 mov %i5, %o0
208 mov %i7, %o1
209 call lookup_fault
210 mov %i4, %o2
211 ret
212 restore
213
214 .globl __bzero_end
215__bzero_end: