1 /* SPDX-License-Identifier: GPL-2.0 */
2 /*
3 * arch/alpha/lib/stxcpy.S
4 * Contributed by Richard Henderson (rth@tamu.
5 *
6 * Copy a null-terminated string from SRC to D
7 *
8 * This is an internal routine used by strcpy,
9 * As such, it uses special linkage convention
10 * of these public functions more efficient.
11 *
12 * On input:
13 * t9 = return address
14 * a0 = DST
15 * a1 = SRC
16 *
17 * On output:
18 * t12 = bitmask (with one bit set) indic
19 * a0 = unaligned address of the last *w
20 *
21 * Furthermore, v0, a3-a5, t11, and t12 are un
22 */
23
24 #include <asm/regdef.h>
25
26 .set noat
27 .set noreorder
28
29 .text
30
31 /* There is a problem with either gdb (as of 4
32 doesn't like putting the entry point for a
33 middle of the procedure descriptor. Work a
34 aligned copy in its own procedure descripto
35
36 .ent stxcpy_aligned
37 .align 3
38 stxcpy_aligned:
39 .frame sp, 0, t9
40 .prologue 0
41
42 /* On entry to this basic block:
43 t0 == the first destination word fo
44 t1 == the first source word. */
45
46 /* Create the 1st output word and dete
47 lda t2, -1 # e1 : buil
48 mskqh t2, a1, t2 # e0 : de
49 mskqh t1, a1, t3 # e0 :
50 ornot t1, t2, t2 # .. e1 :
51 mskql t0, a1, t0 # e0 : asse
52 cmpbge zero, t2, t8 # .. e1 : bits
53 or t0, t3, t1 # e0 :
54 bne t8, $a_eos # .. e1 :
55
56 /* On entry to this basic block:
57 t0 == the first destination word fo
58 t1 == a source word not containing
59
60 $a_loop:
61 stq_u t1, 0(a0) # e0 :
62 addq a0, 8, a0 # .. e1 :
63 ldq_u t1, 0(a1) # e0 :
64 addq a1, 8, a1 # .. e1 :
65 cmpbge zero, t1, t8 # e0 (stall)
66 beq t8, $a_loop # .. e1 (zdb)
67
68 /* Take care of the final (partial) wo
69 On entry to this basic block we hav
70 t1 == the source word containing th
71 t8 == the cmpbge mask that found it
72 $a_eos:
73 negq t8, t6 # e0 : find
74 and t8, t6, t12 # e1 (stall)
75
76 /* For the sake of the cache, don't re
77 if we're not going to need it. */
78 and t12, 0x80, t6 # e0 :
79 bne t6, 1f # .. e1 (zdb)
80
81 /* We're doing a partial word store an
82 our source and original destination
83 ldq_u t0, 0(a0) # e0 :
84 subq t12, 1, t6 # .. e1 :
85 zapnot t1, t6, t1 # e0 : clea
86 or t12, t6, t8 # .. e1 :
87 zap t0, t8, t0 # e0 : clea
88 or t0, t1, t1 # e1 :
89
90 1: stq_u t1, 0(a0) # e0 :
91 ret (t9) # .. e1 :
92
93 .end stxcpy_aligned
94
95 .align 3
96 .ent __stxcpy
97 .globl __stxcpy
98 __stxcpy:
99 .frame sp, 0, t9
100 .prologue 0
101
102 /* Are source and destination co-align
103 xor a0, a1, t0 # e0 :
104 unop # :
105 and t0, 7, t0 # e0 :
106 bne t0, $unaligned # .. e1 :
107
108 /* We are co-aligned; take care of a p
109 ldq_u t1, 0(a1) # e0 : load
110 and a0, 7, t0 # .. e1 : take
111 addq a1, 8, a1 # e0
112 beq t0, stxcpy_aligned # .. e
113 ldq_u t0, 0(a0) # e0 :
114 br stxcpy_aligned # .. e1 :
115
116
117 /* The source and destination are not co-align
118 and cope. We have to be very careful about
119 causing a SEGV. */
120
121 .align 3
122 $u_head:
123 /* We know just enough now to be able
124 full source word. We can still fin
125 that prevents us from outputting th
126
127 On entry to this basic block:
128 t0 == the first dest word, for mask
129 t1 == the low bits of the first sou
130 t6 == bytemask that is -1 in dest w
131
132 ldq_u t2, 8(a1) # e0 :
133 addq a1, 8, a1 # .. e1 :
134
135 extql t1, a1, t1 # e0 :
136 extqh t2, a1, t4 # e0 :
137 mskql t0, a0, t0 # e0 :
138 or t1, t4, t1 # .. e1 :
139 mskqh t1, a0, t1 # e0 :
140 or t0, t1, t1 # e1 :
141
142 or t1, t6, t6 # e0 :
143 cmpbge zero, t6, t8 # .. e1 :
144 lda t6, -1 # e0 : for
145 bne t8, $u_final # .. e1 :
146
147 mskql t6, a1, t6 # e0
148 or t6, t2, t2 # e1
149 cmpbge zero, t2, t8 # e0
150 bne t8, $u_late_head_exit # .. e
151
152 /* Finally, we've got all the stupid l
153 of and we can set up to enter the m
154
155 stq_u t1, 0(a0) # e0 : stor
156 addq a0, 8, a0 # .. e1 :
157 extql t2, a1, t0 # e0 : posi
158 ldq_u t2, 8(a1) # .. e1 : read
159 addq a1, 8, a1 # e0 :
160 cmpbge zero, t2, t8 # .. e1 :
161 nop # e0 :
162 bne t8, $u_eos # .. e1 :
163
164 /* Unaligned copy main loop. In order
165 the loop is structured to detect ze
166 This has, unfortunately, effectivel
167 iteration out into the head and hal
168 prevent nastiness from accumulating
169 to run as fast as possible.
170
171 On entry to this basic block:
172 t0 == the shifted high-order bits f
173 t2 == the unshifted current source
174
175 We further know that t2 does not co
176
177 .align 3
178 $u_loop:
179 extqh t2, a1, t1 # e0 : extr
180 addq a1, 8, a1 # .. e1 :
181 extql t2, a1, t3 # e0 : extr
182 addq a0, 8, a0 # .. e1 :
183 or t0, t1, t1 # e0 : curr
184 ldq_u t2, 0(a1) # .. e1 : load
185 stq_u t1, -8(a0) # e0 : save
186 mov t3, t0 # .. e1 :
187 cmpbge zero, t2, t8 # e0 : test
188 beq t8, $u_loop # .. e1 :
189
190 /* We've found a zero somewhere in the
191 If it resides in the lower half, we
192 word to write out, and if it reside
193 have one full and one partial word
194
195 On entry to this basic block:
196 t0 == the shifted high-order bits f
197 t2 == the unshifted current source
198 $u_eos:
199 extqh t2, a1, t1 # e0 :
200 or t0, t1, t1 # e1 : firs
201
202 cmpbge zero, t1, t8 # e0 : is t
203 bne t8, $u_final # .. e1 (zdb)
204
205 $u_late_head_exit:
206 stq_u t1, 0(a0) # e0 : the
207 addq a0, 8, a0 # .. e1 :
208 extql t2, a1, t1 # e0 :
209 cmpbge zero, t1, t8 # .. e1 :
210
211 /* Take care of a final (probably part
212 On entry to this basic block:
213 t1 == assembled source word
214 t8 == cmpbge mask that found the nu
215 $u_final:
216 negq t8, t6 # e0 : isol
217 and t6, t8, t12 # e1 :
218
219 and t12, 0x80, t6 # e0 : avoi
220 bne t6, 1f # .. e1 (zdb)
221
222 ldq_u t0, 0(a0) # e0 :
223 subq t12, 1, t6 # .. e1 :
224 or t6, t12, t8 # e0 :
225 zapnot t1, t6, t1 # .. e1 : kill
226 zap t0, t8, t0 # e0 : kill
227 or t0, t1, t1 # e1 :
228
229 1: stq_u t1, 0(a0) # e0 :
230 ret (t9) # .. e1 :
231
232 /* Unaligned copy entry point. */
233 .align 3
234 $unaligned:
235
236 ldq_u t1, 0(a1) # e0 : load
237
238 and a0, 7, t4 # .. e1 : find
239 and a1, 7, t5 # e0 : find
240
241 /* Conditionally load the first destin
242 with 0xff indicating that the desti
243
244 mov zero, t0 # .. e1 :
245 mov zero, t6 # e0 :
246 beq t4, 1f # .. e1 :
247 ldq_u t0, 0(a0) # e0 :
248 lda t6, -1 # .. e1 :
249 mskql t6, a0, t6 # e0 :
250 1:
251 subq a1, t4, a1 # .. e1 : sub
252
253 /* If source misalignment is larger th
254 extra startup checks to avoid SEGV.
255
256 cmplt t4, t5, t12 # e0 :
257 beq t12, $u_head # .. e1 (zdb)
258
259 lda t2, -1 # e1 : mask
260 mskqh t2, t5, t2 # e0 :
261 nop # e0 :
262 ornot t1, t2, t3 # .. e1 :
263 cmpbge zero, t3, t8 # e0 : is t
264 beq t8, $u_head # .. e1 (zdb)
265
266 /* At this point we've found a zero in
267 the source. We need to isolate the
268 it into the original destination da
269 that we'll need at least one byte o
270
271 ldq_u t0, 0(a0) # e0 :
272
273 negq t8, t6 # .. e1 : buil
274 and t6, t8, t12 # e0 :
275 and a1, 7, t5 # .. e1 :
276 subq t12, 1, t6 # e0 :
277 or t6, t12, t8 # e1 :
278 srl t12, t5, t12 # e0 : adju
279
280 zapnot t2, t8, t2 # .. e1 : prep
281 and t1, t2, t1 # e1 : to s
282 extql t2, a1, t2 # .. e0 :
283 extql t1, a1, t1 # e0 :
284
285 andnot t0, t2, t0 # .. e1 : zero
286 or t0, t1, t1 # e1 : and
287 stq_u t1, 0(a0) # .. e0 :
288 ret (t9) # e1 :
289
290 .end __stxcpy
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.