1 /* SPDX-License-Identifier: GPL-2.0 */
2 /*
3 * arch/alpha/lib/stxncpy.S
4 * Contributed by Richard Henderson (rth@tamu.
5 *
6 * Copy no more than COUNT bytes of the null-t
7 * SRC to DST.
8 *
9 * This is an internal routine used by strncpy
10 * As such, it uses special linkage convention
11 * of these public functions more efficient.
12 *
13 * On input:
14 * t9 = return address
15 * a0 = DST
16 * a1 = SRC
17 * a2 = COUNT
18 *
19 * Furthermore, COUNT may not be zero.
20 *
21 * On output:
22 * t0 = last word written
23 * t10 = bitmask (with one bit set) indic
24 * the end of the range specified b
25 * t12 = bitmask (with one bit set) indic
26 * a0 = unaligned address of the last *w
27 * a2 = the number of full words left in
28 *
29 * Furthermore, v0, a3-a5, t11, and $at are un
30 */
31
32 #include <asm/regdef.h>
33
34 .set noat
35 .set noreorder
36
37 .text
38
39 /* There is a problem with either gdb (as of 4
40 doesn't like putting the entry point for a
41 middle of the procedure descriptor. Work a
42 aligned copy in its own procedure descripto
43
44 .ent stxncpy_aligned
45 .align 3
46 stxncpy_aligned:
47 .frame sp, 0, t9, 0
48 .prologue 0
49
50 /* On entry to this basic block:
51 t0 == the first destination word fo
52 t1 == the first source word. */
53
54 /* Create the 1st output word and dete
55 lda t2, -1 # e1 : buil
56 mskqh t2, a1, t2 # e0 : de
57 mskqh t1, a1, t3 # e0 :
58 ornot t1, t2, t2 # .. e1 :
59 mskql t0, a1, t0 # e0 : asse
60 cmpbge zero, t2, t8 # .. e1 : bits
61 or t0, t3, t0 # e0 :
62 beq a2, $a_eoc # .. e1 :
63 bne t8, $a_eos # .. e1 :
64
65 /* On entry to this basic block:
66 t0 == a source word not containing
67
68 $a_loop:
69 stq_u t0, 0(a0) # e0 :
70 addq a0, 8, a0 # .. e1 :
71 ldq_u t0, 0(a1) # e0 :
72 addq a1, 8, a1 # .. e1 :
73 subq a2, 1, a2 # e0 :
74 cmpbge zero, t0, t8 # .. e1 (stall
75 beq a2, $a_eoc # e1 :
76 beq t8, $a_loop # e1 :
77
78 /* Take care of the final (partial) wo
79 the end-of-count bit is set in t8 i
80
81 On entry to this basic block we hav
82 t0 == the source word containing th
83 t8 == the cmpbge mask that found it
84
85 $a_eos:
86 negq t8, t12 # e0 : find
87 and t8, t12, t12 # e1 (stall)
88
89 /* For the sake of the cache, don't re
90 if we're not going to need it. */
91 and t12, 0x80, t6 # e0 :
92 bne t6, 1f # .. e1 (zdb)
93
94 /* We're doing a partial word store an
95 our source and original destination
96 ldq_u t1, 0(a0) # e0 :
97 subq t12, 1, t6 # .. e1 :
98 or t12, t6, t8 # e0 :
99 unop #
100 zapnot t0, t8, t0 # e0 : clea
101 zap t1, t8, t1 # .. e1 : clea
102 or t0, t1, t0 # e1 :
103
104 1: stq_u t0, 0(a0) # e0 :
105 ret (t9) # e1 :
106
107 /* Add the end-of-count bit to the eos
108 $a_eoc:
109 or t10, t8, t8
110 br $a_eos
111
112 .end stxncpy_aligned
113
114 .align 3
115 .ent __stxncpy
116 .globl __stxncpy
117 __stxncpy:
118 .frame sp, 0, t9, 0
119 .prologue 0
120
121 /* Are source and destination co-align
122 xor a0, a1, t1 # e0 :
123 and a0, 7, t0 # .. e1 : find
124 and t1, 7, t1 # e0 :
125 addq a2, t0, a2 # .. e1 : bias
126 subq a2, 1, a2 # e0 :
127 and a2, 7, t2 # e1 :
128 srl a2, 3, a2 # e0 : a2 =
129 addq zero, 1, t10 # .. e1 :
130 sll t10, t2, t10 # e0 : t10
131 bne t1, $unaligned # .. e1 :
132
133 /* We are co-aligned; take care of a p
134
135 ldq_u t1, 0(a1) # e0 : load
136 addq a1, 8, a1 # .. e1 :
137
138 beq t0, stxncpy_aligned # avoi
139 ldq_u t0, 0(a0) # e0 :
140 br stxncpy_aligned # .. e1 :
141
142
143 /* The source and destination are not co-align
144 and cope. We have to be very careful about
145 causing a SEGV. */
146
147 .align 3
148 $u_head:
149 /* We know just enough now to be able
150 full source word. We can still fin
151 that prevents us from outputting th
152
153 On entry to this basic block:
154 t0 == the first dest word, unmasked
155 t1 == the shifted low bits of the f
156 t6 == bytemask that is -1 in dest w
157
158 ldq_u t2, 8(a1) # e0 : load
159 addq a1, 8, a1 # .. e1 :
160 mskql t0, a0, t0 # e0 : mask
161 extqh t2, a1, t4 # e0 :
162 or t1, t4, t1 # e1 : firs
163 mskqh t1, a0, t1 # e0 : mask
164 or t0, t1, t0 # e0 : firs
165 or t0, t6, t6 # e1 : mask
166 cmpbge zero, t6, t8 # e0 :
167 beq a2, $u_eocfin # .. e1 :
168 lda t6, -1 # e0 :
169 bne t8, $u_final # .. e1 :
170
171 mskql t6, a1, t6 # e0 : mask
172 nop # .. e1 :
173 stq_u t0, 0(a0) # e0 : stor
174 or t6, t2, t2 # .. e1 :
175 cmpbge zero, t2, t8 # e0 : find
176 addq a0, 8, a0 # .. e1 :
177 subq a2, 1, a2 # e0 :
178 bne t8, $u_late_head_exit # .. e
179
180 /* Finally, we've got all the stupid l
181 of and we can set up to enter the m
182
183 extql t2, a1, t1 # e0 : posi
184 beq a2, $u_eoc # .. e1 :
185 ldq_u t2, 8(a1) # e0 : read
186 addq a1, 8, a1 # .. e1 :
187 extqh t2, a1, t0 # e0 : posi
188 cmpbge zero, t2, t8 # .. e1 :
189 nop # e0 :
190 bne t8, $u_eos # .. e1 :
191
192 /* Unaligned copy main loop. In order
193 the loop is structured to detect ze
194 This has, unfortunately, effectivel
195 iteration out into the head and hal
196 prevent nastiness from accumulating
197 to run as fast as possible.
198
199 On entry to this basic block:
200 t0 == the shifted low-order bits fr
201 t1 == the shifted high-order bits f
202 t2 == the unshifted current source
203
204 We further know that t2 does not co
205
206 .align 3
207 $u_loop:
208 or t0, t1, t0 # e0 : curr
209 subq a2, 1, a2 # .. e1 : decr
210 stq_u t0, 0(a0) # e0 : save
211 addq a0, 8, a0 # .. e1 :
212 extql t2, a1, t1 # e0 : extr
213 beq a2, $u_eoc # .. e1 :
214 ldq_u t2, 8(a1) # e0 : load
215 addq a1, 8, a1 # .. e1 :
216 nop # e0 :
217 cmpbge zero, t2, t8 # e1 : test
218 extqh t2, a1, t0 # e0 : extr
219 beq t8, $u_loop # .. e1 :
220
221 /* We've found a zero somewhere in the
222 If it resides in the lower half, we
223 word to write out, and if it reside
224 have one full and one partial word
225
226 On entry to this basic block:
227 t0 == the shifted low-order bits fr
228 t1 == the shifted high-order bits f
229 t2 == the unshifted current source
230 $u_eos:
231 or t0, t1, t0 # e0 : firs
232 nop # .. e1 :
233 cmpbge zero, t0, t8 # e0 : is t
234 bne t8, $u_final # .. e1 (zdb)
235
236 stq_u t0, 0(a0) # e0 : the
237 addq a0, 8, a0 # .. e1 :
238 subq a2, 1, a2 # e1 :
239
240 $u_late_head_exit:
241 extql t2, a1, t0 # .. e0 :
242 cmpbge zero, t0, t8 # e0 :
243 or t8, t10, t6 # e1 :
244 cmoveq a2, t6, t8 # e0 :
245 nop # .. e1 :
246
247 /* Take care of a final (probably part
248 On entry to this basic block:
249 t0 == assembled source word
250 t8 == cmpbge mask that found the nu
251 $u_final:
252 negq t8, t6 # e0 : isol
253 and t6, t8, t12 # e1 :
254
255 and t12, 0x80, t6 # e0 : avoi
256 bne t6, 1f # .. e1 (zdb)
257
258 ldq_u t1, 0(a0) # e0 :
259 subq t12, 1, t6 # .. e1 :
260 or t6, t12, t8 # e0 :
261 zapnot t0, t8, t0 # .. e1 : kill
262 zap t1, t8, t1 # e0 : kill
263 or t0, t1, t0 # e1 :
264
265 1: stq_u t0, 0(a0) # e0 :
266 ret (t9) # .. e1 :
267
268 /* Got to end-of-count before end of s
269 On entry to this basic block:
270 t1 == the shifted high-order bits f
271 $u_eoc:
272 and a1, 7, t6 # e1 :
273 sll t10, t6, t6 # e0 :
274 and t6, 0xff, t6 # e0 :
275 bne t6, 1f # .. e1 :
276
277 ldq_u t2, 8(a1) # e0 : load
278 nop # .. e1 :
279 extqh t2, a1, t0 # e0 : extr
280 or t1, t0, t1 # e1 :
281
282 1: cmpbge zero, t1, t8
283 mov t1, t0
284
285 $u_eocfin: # end-of-count
286 or t10, t8, t8
287 br $u_final
288
289 /* Unaligned copy entry point. */
290 .align 3
291 $unaligned:
292
293 ldq_u t1, 0(a1) # e0 : load
294
295 and a0, 7, t4 # .. e1 : find
296 and a1, 7, t5 # e0 : find
297
298 /* Conditionally load the first destin
299 with 0xff indicating that the desti
300
301 mov zero, t0 # .. e1 :
302 mov zero, t6 # e0 :
303 beq t4, 1f # .. e1 :
304 ldq_u t0, 0(a0) # e0 :
305 lda t6, -1 # .. e1 :
306 mskql t6, a0, t6 # e0 :
307 subq a1, t4, a1 # .. e1 : sub
308
309 /* If source misalignment is larger th
310 extra startup checks to avoid SEGV.
311
312 1: cmplt t4, t5, t12 # e1 :
313 extql t1, a1, t1 # .. e0 : shif
314 lda t2, -1 # e0 : for
315 beq t12, $u_head # .. e1 :
316
317 extql t2, a1, t2 # e0 :
318 cmpbge zero, t1, t8 # .. e1 : is t
319 andnot t2, t6, t2 # e0 : dest
320 or t8, t10, t5 # .. e1 : test
321 cmpbge zero, t2, t3 # e0 :
322 cmoveq a2, t5, t8 # .. e1 :
323 andnot t8, t3, t8 # e0 :
324 beq t8, $u_head # .. e1 (zdb)
325
326 /* At this point we've found a zero in
327 the source. We need to isolate the
328 it into the original destination da
329 that we'll need at least one byte o
330
331 ldq_u t0, 0(a0) # e0 :
332 negq t8, t6 # .. e1 : buil
333 mskqh t1, t4, t1 # e0 :
334 and t6, t8, t12 # .. e1 :
335 subq t12, 1, t6 # e0 :
336 or t6, t12, t8 # e1 :
337
338 zapnot t2, t8, t2 # e0 : prep
339 zapnot t1, t8, t1 # .. e1 : to s
340
341 andnot t0, t2, t0 # e0 : zero
342 or t0, t1, t0 # e1 : and
343 stq_u t0, 0(a0) # e0 :
344 ret (t9) # .. e1 :
345
346 .end __stxncpy
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.