summaryrefslogtreecommitdiffstats
path: root/arch/parisc/lib/lusercopy.S
blob: b428d29e45fb939ca3d8688b5acfd3b52d9d224d (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
/* SPDX-License-Identifier: GPL-2.0-or-later */
/*
 *    User Space Access Routines
 *
 *    Copyright (C) 2000-2002 Hewlett-Packard (John Marvin)
 *    Copyright (C) 2000 Richard Hirst <rhirst with parisc-linux.org>
 *    Copyright (C) 2001 Matthieu Delahaye <delahaym at esiee.fr>
 *    Copyright (C) 2003 Randolph Chung <tausq with parisc-linux.org>
 *    Copyright (C) 2017 Helge Deller <deller@gmx.de>
 *    Copyright (C) 2017 John David Anglin <dave.anglin@bell.net>
 */

/*
 * These routines still have plenty of room for optimization
 * (word & doubleword load/store, dual issue, store hints, etc.).
 */

/*
 * The following routines assume that space register 3 (sr3) contains
 * the space id associated with the current users address space.
 */


	.text
	
#include <asm/assembly.h>
#include <asm/errno.h>
#include <linux/linkage.h>

	/*
	 * unsigned long lclear_user(void *to, unsigned long n)
	 *
	 * Returns 0 for success.
	 * otherwise, returns number of bytes not transferred.
	 */

ENTRY_CFI(lclear_user)
	comib,=,n   0,%r25,$lclu_done
$lclu_loop:
	addib,<>    -1,%r25,$lclu_loop
1:	stbs,ma     %r0,1(%sr3,%r26)

$lclu_done:
	bv          %r0(%r2)
	copy        %r25,%r28

2:	b           $lclu_done
	ldo         1(%r25),%r25

	ASM_EXCEPTIONTABLE_ENTRY(1b,2b)
ENDPROC_CFI(lclear_user)


/*
 * unsigned long pa_memcpy(void *dstp, const void *srcp, unsigned long len)
 *
 * Inputs:
 * - sr1 already contains space of source region
 * - sr2 already contains space of destination region
 *
 * Returns:
 * - number of bytes that could not be copied.
 *   On success, this will be zero.
 *
 * This code is based on a C-implementation of a copy routine written by
 * Randolph Chung, which in turn was derived from the glibc.
 *
 * Several strategies are tried to try to get the best performance for various
 * conditions. In the optimal case, we copy by loops that copy 32- or 16-bytes
 * at a time using general registers.  Unaligned copies are handled either by
 * aligning the destination and then using shift-and-write method, or in a few
 * cases by falling back to a byte-at-a-time copy.
 *
 * Testing with various alignments and buffer sizes shows that this code is
 * often >10x faster than a simple byte-at-a-time copy, even for strangely
 * aligned operands. It is interesting to note that the glibc version of memcpy
 * (written in C) is actually quite fast already. This routine is able to beat
 * it by 30-40% for aligned copies because of the loop unrolling, but in some
 * cases the glibc version is still slightly faster. This lends more
 * credibility that gcc can generate very good code as long as we are careful.
 *
 * Possible optimizations:
 * - add cache prefetching
 * - try not to use the post-increment address modifiers; they may create
 *   additional interlocks. Assumption is that those were only efficient on old
 *   machines (pre PA8000 processors)
 */

	dst = arg0
	src = arg1
	len = arg2
	end = arg3
	t1  = r19
	t2  = r20
	t3  = r21
	t4  = r22
	srcspc = sr1
	dstspc = sr2

	t0 = r1
	a1 = t1
	a2 = t2
	a3 = t3
	a0 = t4

	save_src = ret0
	save_dst = ret1
	save_len = r31

ENTRY_CFI(pa_memcpy)
	/* Last destination address */
	add	dst,len,end

	/* short copy with less than 16 bytes? */
	cmpib,COND(>>=),n 15,len,.Lbyte_loop

	/* same alignment? */
	xor	src,dst,t0
	extru	t0,31,2,t1
	cmpib,<>,n  0,t1,.Lunaligned_copy

#ifdef CONFIG_64BIT
	/* only do 64-bit copies if we can get aligned. */
	extru	t0,31,3,t1
	cmpib,<>,n  0,t1,.Lalign_loop32

	/* loop until we are 64-bit aligned */
.Lalign_loop64:
	extru	dst,31,3,t1
	cmpib,=,n	0,t1,.Lcopy_loop_16_start
20:	ldb,ma	1(srcspc,src),t1
21:	stb,ma	t1,1(dstspc,dst)
	b	.Lalign_loop64
	ldo	-1(len),len

	ASM_EXCEPTIONTABLE_ENTRY(20b,.Lcopy_done)
	ASM_EXCEPTIONTABLE_ENTRY(21b,.Lcopy_done)

.Lcopy_loop_16_start:
	ldi	31,t0
.Lcopy_loop_16:
	cmpb,COND(>>=),n t0,len,.Lword_loop

10:	ldd	0(srcspc,src),t1
11:	ldd	8(srcspc,src),t2
	ldo	16(src),src
12:	std,ma	t1,8(dstspc,dst)
13:	std,ma	t2,8(dstspc,dst)
14:	ldd	0(srcspc,src),t1
15:	ldd	8(srcspc,src),t2
	ldo	16(src),src
16:	std,ma	t1,8(dstspc,dst)
17:	std,ma	t2,8(dstspc,dst)

	ASM_EXCEPTIONTABLE_ENTRY(10b,.Lcopy_done)
	ASM_EXCEPTIONTABLE_ENTRY(11b,.Lcopy16_fault)
	ASM_EXCEPTIONTABLE_ENTRY(12b,.Lcopy_done)
	ASM_EXCEPTIONTABLE_ENTRY(13b,.Lcopy_done)
	ASM_EXCEPTIONTABLE_ENTRY(14b,.Lcopy_done)
	ASM_EXCEPTIONTABLE_ENTRY(15b,.Lcopy16_fault)
	ASM_EXCEPTIONTABLE_ENTRY(16b,.Lcopy_done)
	ASM_EXCEPTIONTABLE_ENTRY(17b,.Lcopy_done)

	b	.Lcopy_loop_16
	ldo	-32(len),len

.Lword_loop:
	cmpib,COND(>>=),n 3,len,.Lbyte_loop
20:	ldw,ma	4(srcspc,src),t1
21:	stw,ma	t1,4(dstspc,dst)
	b	.Lword_loop
	ldo	-4(len),len

	ASM_EXCEPTIONTABLE_ENTRY(20b,.Lcopy_done)
	ASM_EXCEPTIONTABLE_ENTRY(21b,.Lcopy_done)

#endif /* CONFIG_64BIT */

	/* loop until we are 32-bit aligned */
.Lalign_loop32:
	extru	dst,31,2,t1
	cmpib,=,n	0,t1,.Lcopy_loop_8
20:	ldb,ma	1(srcspc,src),t1
21:	stb,ma	t1,1(dstspc,dst)
	b	.Lalign_loop32
	ldo	-1(len),len

	ASM_EXCEPTIONTABLE_ENTRY(20b,.Lcopy_done)
	ASM_EXCEPTIONTABLE_ENTRY(21b,.Lcopy_done)


.Lcopy_loop_8:
	cmpib,COND(>>=),n 15,len,.Lbyte_loop

10:	ldw	0(srcspc,src),t1
11:	ldw	4(srcspc,src),t2
12:	stw,ma	t1,4(dstspc,dst)
13:	stw,ma	t2,4(dstspc,dst)
14:	ldw	8(srcspc,src),t1
15:	ldw	12(srcspc,src),t2
	ldo	16(src),src
16:	stw,ma	t1,4(dstspc,dst)
17:	stw,ma	t2,4(dstspc,dst)

	ASM_EXCEPTIONTABLE_ENTRY(10b,.Lcopy_done)
	ASM_EXCEPTIONTABLE_ENTRY(11b,.Lcopy8_fault)
	ASM_EXCEPTIONTABLE_ENTRY(12b,.Lcopy_done)
	ASM_EXCEPTIONTABLE_ENTRY(13b,.Lcopy_done)
	ASM_EXCEPTIONTABLE_ENTRY(14b,.Lcopy_done)
	ASM_EXCEPTIONTABLE_ENTRY(15b,.Lcopy8_fault)
	ASM_EXCEPTIONTABLE_ENTRY(16b,.Lcopy_done)
	ASM_EXCEPTIONTABLE_ENTRY(17b,.Lcopy_done)

	b	.Lcopy_loop_8
	ldo	-16(len),len

.Lbyte_loop:
	cmpclr,COND(<>) len,%r0,%r0
	b,n	.Lcopy_done
20:	ldb	0(srcspc,src),t1
	ldo	1(src),src
21:	stb,ma	t1,1(dstspc,dst)
	b	.Lbyte_loop
	ldo	-1(len),len

	ASM_EXCEPTIONTABLE_ENTRY(20b,.Lcopy_done)
	ASM_EXCEPTIONTABLE_ENTRY(21b,.Lcopy_done)

.Lcopy_done:
	bv	%r0(%r2)
	sub	end,dst,ret0


	/* src and dst are not aligned the same way. */
	/* need to go the hard way */
.Lunaligned_copy:
	/* align until dst is 32bit-word-aligned */
	extru	dst,31,2,t1
	cmpib,=,n	0,t1,.Lcopy_dstaligned
20:	ldb	0(srcspc,src),t1
	ldo	1(src),src
21:	stb,ma	t1,1(dstspc,dst)
	b	.Lunaligned_copy
	ldo	-1(len),len

	ASM_EXCEPTIONTABLE_ENTRY(20b,.Lcopy_done)
	ASM_EXCEPTIONTABLE_ENTRY(21b,.Lcopy_done)

.Lcopy_dstaligned:

	/* store src, dst and len in safe place */
	copy	src,save_src
	copy	dst,save_dst
	copy	len,save_len

	/* len now needs give number of words to copy */
	SHRREG	len,2,len

	/*
	 * Copy from a not-aligned src to an aligned dst using shifts.
	 * Handles 4 words per loop.
	 */

	depw,z src,28,2,t0
	subi 32,t0,t0
	mtsar t0
	extru len,31,2,t0
	cmpib,= 2,t0,.Lcase2
	/* Make src aligned by rounding it down.  */
	depi 0,31,2,src

	cmpiclr,<> 3,t0,%r0
	b,n .Lcase3
	cmpiclr,<> 1,t0,%r0
	b,n .Lcase1
.Lcase0:
	cmpb,COND(=) %r0,len,.Lcda_finish
	nop

1:	ldw,ma 4(srcspc,src), a3
	ASM_EXCEPTIONTABLE_ENTRY(1b,.Lcda_rdfault)
1:	ldw,ma 4(srcspc,src), a0
	ASM_EXCEPTIONTABLE_ENTRY(1b,.Lcda_rdfault)
	b,n .Ldo3
.Lcase1:
1:	ldw,ma 4(srcspc,src), a2
	ASM_EXCEPTIONTABLE_ENTRY(1b,.Lcda_rdfault)
1:	ldw,ma 4(srcspc,src), a3
	ASM_EXCEPTIONTABLE_ENTRY(1b,.Lcda_rdfault)
	ldo -1(len),len
	cmpb,COND(=),n %r0,len,.Ldo0
.Ldo4:
1:	ldw,ma 4(srcspc,src), a0
	ASM_EXCEPTIONTABLE_ENTRY(1b,.Lcda_rdfault)
	shrpw a2, a3, %sar, t0
1:	stw,ma t0, 4(dstspc,dst)
	ASM_EXCEPTIONTABLE_ENTRY(1b,.Lcopy_done)
.Ldo3:
1:	ldw,ma 4(srcspc,src), a1
	ASM_EXCEPTIONTABLE_ENTRY(1b,.Lcda_rdfault)
	shrpw a3, a0, %sar, t0
1:	stw,ma t0, 4(dstspc,dst)
	ASM_EXCEPTIONTABLE_ENTRY(1b,.Lcopy_done)
.Ldo2:
1:	ldw,ma 4(srcspc,src), a2
	ASM_EXCEPTIONTABLE_ENTRY(1b,.Lcda_rdfault)
	shrpw a0, a1, %sar, t0
1:	stw,ma t0, 4(dstspc,dst)
	ASM_EXCEPTIONTABLE_ENTRY(1b,.Lcopy_done)
.Ldo1:
1:	ldw,ma 4(srcspc,src), a3
	ASM_EXCEPTIONTABLE_ENTRY(1b,.Lcda_rdfault)
	shrpw a1, a2, %sar, t0
1:	stw,ma t0, 4(dstspc,dst)
	ASM_EXCEPTIONTABLE_ENTRY(1b,.Lcopy_done)
	ldo -4(len),len
	cmpb,COND(<>) %r0,len,.Ldo4
	nop
.Ldo0:
	shrpw a2, a3, %sar, t0
1:	stw,ma t0, 4(dstspc,dst)
	ASM_EXCEPTIONTABLE_ENTRY(1b,.Lcopy_done)

.Lcda_rdfault:
.Lcda_finish:
	/* calculate new src, dst and len and jump to byte-copy loop */
	sub	dst,save_dst,t0
	add	save_src,t0,src
	b	.Lbyte_loop
	sub	save_len,t0,len

.Lcase3:
1:	ldw,ma 4(srcspc,src), a0
	ASM_EXCEPTIONTABLE_ENTRY(1b,.Lcda_rdfault)
1:	ldw,ma 4(srcspc,src), a1
	ASM_EXCEPTIONTABLE_ENTRY(1b,.Lcda_rdfault)
	b .Ldo2
	ldo 1(len),len
.Lcase2:
1:	ldw,ma 4(srcspc,src), a1
	ASM_EXCEPTIONTABLE_ENTRY(1b,.Lcda_rdfault)
1:	ldw,ma 4(srcspc,src), a2
	ASM_EXCEPTIONTABLE_ENTRY(1b,.Lcda_rdfault)
	b .Ldo1
	ldo 2(len),len


	/* fault exception fixup handlers: */
#ifdef CONFIG_64BIT
.Lcopy16_fault:
	b	.Lcopy_done
10:	std,ma	t1,8(dstspc,dst)
	ASM_EXCEPTIONTABLE_ENTRY(10b,.Lcopy_done)
#endif

.Lcopy8_fault:
	b	.Lcopy_done
10:	stw,ma	t1,4(dstspc,dst)
	ASM_EXCEPTIONTABLE_ENTRY(10b,.Lcopy_done)
ENDPROC_CFI(pa_memcpy)

	.end