summaryrefslogtreecommitdiffstats
path: root/arch/xtensa/lib/usercopy.S
blob: 4172b73b036485b83a67acb3bde7650f01b515cb (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
/*
 *  arch/xtensa/lib/usercopy.S
 *
 *  Copy to/from user space (derived from arch/xtensa/lib/hal/memcopy.S)
 *
 *  DO NOT COMBINE this function with <arch/xtensa/lib/hal/memcopy.S>.
 *  It needs to remain separate and distinct.  The hal files are part
 *  of the Xtensa link-time HAL, and those files may differ per
 *  processor configuration.  Patching the kernel for another
 *  processor configuration includes replacing the hal files, and we
 *  could lose the special functionality for accessing user-space
 *  memory during such a patch.  We sacrifice a little code space here
 *  in favor to simplify code maintenance.
 *
 *  This file is subject to the terms and conditions of the GNU General
 *  Public License.  See the file "COPYING" in the main directory of
 *  this archive for more details.
 *
 *  Copyright (C) 2002 Tensilica Inc.
 */


/*
 * size_t __xtensa_copy_user (void *dst, const void *src, size_t len);
 *
 * The returned value is the number of bytes not copied.  Implies zero
 * is success.
 *
 * The general case algorithm is as follows:
 *   If the destination and source are both aligned,
 *     do 16B chunks with a loop, and then finish up with
 *     8B, 4B, 2B, and 1B copies conditional on the length.
 *   If destination is aligned and source unaligned,
 *     do the same, but use SRC to align the source data.
 *   If destination is unaligned, align it by conditionally
 *     copying 1B and 2B and then retest.
 *   This code tries to use fall-through braches for the common
 *     case of aligned destinations (except for the branches to
 *     the alignment label).
 *
 * Register use:
 *	a0/ return address
 *	a1/ stack pointer
 *	a2/ return value
 *	a3/ src
 *	a4/ length
 *	a5/ dst
 *	a6/ tmp
 *	a7/ tmp
 *	a8/ tmp
 *	a9/ tmp
 *	a10/ tmp
 *	a11/ original length
 */

#include <variant/core.h>
#include <asm/asmmacro.h>

#ifdef __XTENSA_EB__
#define ALIGN(R, W0, W1) src	R, W0, W1
#define SSA8(R)	ssa8b R
#else
#define ALIGN(R, W0, W1) src	R, W1, W0
#define SSA8(R)	ssa8l R
#endif

	.text
	.align	4
	.global	__xtensa_copy_user
	.type	__xtensa_copy_user,@function
__xtensa_copy_user:
	entry	sp, 16		# minimal stack frame
	# a2/ dst, a3/ src, a4/ len
	mov	a5, a2		# copy dst so that a2 is return value
	mov	a11, a4		# preserve original len for error case
.Lcommon:
	bbsi.l	a2, 0, .Ldst1mod2 # if dst is 1 mod 2
	bbsi.l	a2, 1, .Ldst2mod4 # if dst is 2 mod 4
.Ldstaligned:	# return here from .Ldstunaligned when dst is aligned
	srli	a7, a4, 4	# number of loop iterations with 16B
				# per iteration
	movi	a8, 3		  # if source is also aligned,
	bnone	a3, a8, .Laligned # then use word copy
	SSA8(	a3)		# set shift amount from byte offset
	bnez	a4, .Lsrcunaligned
	movi	a2, 0		# return success for len==0
	retw

/*
 * Destination is unaligned
 */

.Ldst1mod2:	# dst is only byte aligned
	bltui	a4, 7, .Lbytecopy	# do short copies byte by byte

	# copy 1 byte
EX(10f)	l8ui	a6, a3, 0
	addi	a3, a3,  1
EX(10f)	s8i	a6, a5,  0
	addi	a5, a5,  1
	addi	a4, a4, -1
	bbci.l	a5, 1, .Ldstaligned	# if dst is now aligned, then
					# return to main algorithm
.Ldst2mod4:	# dst 16-bit aligned
	# copy 2 bytes
	bltui	a4, 6, .Lbytecopy	# do short copies byte by byte
EX(10f)	l8ui	a6, a3, 0
EX(10f)	l8ui	a7, a3, 1
	addi	a3, a3,  2
EX(10f)	s8i	a6, a5,  0
EX(10f)	s8i	a7, a5,  1
	addi	a5, a5,  2
	addi	a4, a4, -2
	j	.Ldstaligned	# dst is now aligned, return to main algorithm

/*
 * Byte by byte copy
 */
	.align	4
	.byte	0		# 1 mod 4 alignment for LOOPNEZ
				# (0 mod 4 alignment for LBEG)
.Lbytecopy:
#if XCHAL_HAVE_LOOPS
	loopnez	a4, .Lbytecopydone
#else /* !XCHAL_HAVE_LOOPS */
	beqz	a4, .Lbytecopydone
	add	a7, a3, a4	# a7 = end address for source
#endif /* !XCHAL_HAVE_LOOPS */
.Lnextbyte:
EX(10f)	l8ui	a6, a3, 0
	addi	a3, a3, 1
EX(10f)	s8i	a6, a5, 0
	addi	a5, a5, 1
#if !XCHAL_HAVE_LOOPS
	blt	a3, a7, .Lnextbyte
#endif /* !XCHAL_HAVE_LOOPS */
.Lbytecopydone:
	movi	a2, 0		# return success for len bytes copied
	retw

/*
 * Destination and source are word-aligned.
 */
	# copy 16 bytes per iteration for word-aligned dst and word-aligned src
	.align	4		# 1 mod 4 alignment for LOOPNEZ
	.byte	0		# (0 mod 4 alignment for LBEG)
.Laligned:
#if XCHAL_HAVE_LOOPS
	loopnez	a7, .Loop1done
#else /* !XCHAL_HAVE_LOOPS */
	beqz	a7, .Loop1done
	slli	a8, a7, 4
	add	a8, a8, a3	# a8 = end of last 16B source chunk
#endif /* !XCHAL_HAVE_LOOPS */
.Loop1:
EX(10f)	l32i	a6, a3,  0
EX(10f)	l32i	a7, a3,  4
EX(10f)	s32i	a6, a5,  0
EX(10f)	l32i	a6, a3,  8
EX(10f)	s32i	a7, a5,  4
EX(10f)	l32i	a7, a3, 12
EX(10f)	s32i	a6, a5,  8
	addi	a3, a3, 16
EX(10f)	s32i	a7, a5, 12
	addi	a5, a5, 16
#if !XCHAL_HAVE_LOOPS
	blt	a3, a8, .Loop1
#endif /* !XCHAL_HAVE_LOOPS */
.Loop1done:
	bbci.l	a4, 3, .L2
	# copy 8 bytes
EX(10f)	l32i	a6, a3,  0
EX(10f)	l32i	a7, a3,  4
	addi	a3, a3,  8
EX(10f)	s32i	a6, a5,  0
EX(10f)	s32i	a7, a5,  4
	addi	a5, a5,  8
.L2:
	bbci.l	a4, 2, .L3
	# copy 4 bytes
EX(10f)	l32i	a6, a3,  0
	addi	a3, a3,  4
EX(10f)	s32i	a6, a5,  0
	addi	a5, a5,  4
.L3:
	bbci.l	a4, 1, .L4
	# copy 2 bytes
EX(10f)	l16ui	a6, a3,  0
	addi	a3, a3,  2
EX(10f)	s16i	a6, a5,  0
	addi	a5, a5,  2
.L4:
	bbci.l	a4, 0, .L5
	# copy 1 byte
EX(10f)	l8ui	a6, a3,  0
EX(10f)	s8i	a6, a5,  0
.L5:
	movi	a2, 0		# return success for len bytes copied
	retw

/*
 * Destination is aligned, Source is unaligned
 */

	.align	4
	.byte	0		# 1 mod 4 alignement for LOOPNEZ
				# (0 mod 4 alignment for LBEG)
.Lsrcunaligned:
	# copy 16 bytes per iteration for word-aligned dst and unaligned src
	and	a10, a3, a8	# save unalignment offset for below
	sub	a3, a3, a10	# align a3 (to avoid sim warnings only; not needed for hardware)
EX(10f)	l32i	a6, a3, 0	# load first word
#if XCHAL_HAVE_LOOPS
	loopnez	a7, .Loop2done
#else /* !XCHAL_HAVE_LOOPS */
	beqz	a7, .Loop2done
	slli	a12, a7, 4
	add	a12, a12, a3	# a12 = end of last 16B source chunk
#endif /* !XCHAL_HAVE_LOOPS */
.Loop2:
EX(10f)	l32i	a7, a3,  4
EX(10f)	l32i	a8, a3,  8
	ALIGN(	a6, a6, a7)
EX(10f)	s32i	a6, a5,  0
EX(10f)	l32i	a9, a3, 12
	ALIGN(	a7, a7, a8)
EX(10f)	s32i	a7, a5,  4
EX(10f)	l32i	a6, a3, 16
	ALIGN(	a8, a8, a9)
EX(10f)	s32i	a8, a5,  8
	addi	a3, a3, 16
	ALIGN(	a9, a9, a6)
EX(10f)	s32i	a9, a5, 12
	addi	a5, a5, 16
#if !XCHAL_HAVE_LOOPS
	blt	a3, a12, .Loop2
#endif /* !XCHAL_HAVE_LOOPS */
.Loop2done:
	bbci.l	a4, 3, .L12
	# copy 8 bytes
EX(10f)	l32i	a7, a3,  4
EX(10f)	l32i	a8, a3,  8
	ALIGN(	a6, a6, a7)
EX(10f)	s32i	a6, a5,  0
	addi	a3, a3,  8
	ALIGN(	a7, a7, a8)
EX(10f)	s32i	a7, a5,  4
	addi	a5, a5,  8
	mov	a6, a8
.L12:
	bbci.l	a4, 2, .L13
	# copy 4 bytes
EX(10f)	l32i	a7, a3,  4
	addi	a3, a3,  4
	ALIGN(	a6, a6, a7)
EX(10f)	s32i	a6, a5,  0
	addi	a5, a5,  4
	mov	a6, a7
.L13:
	add	a3, a3, a10	# readjust a3 with correct misalignment
	bbci.l	a4, 1, .L14
	# copy 2 bytes
EX(10f)	l8ui	a6, a3,  0
EX(10f)	l8ui	a7, a3,  1
	addi	a3, a3,  2
EX(10f)	s8i	a6, a5,  0
EX(10f)	s8i	a7, a5,  1
	addi	a5, a5,  2
.L14:
	bbci.l	a4, 0, .L15
	# copy 1 byte
EX(10f)	l8ui	a6, a3,  0
EX(10f)	s8i	a6, a5,  0
.L15:
	movi	a2, 0		# return success for len bytes copied
	retw


	.section .fixup, "ax"
	.align	4

/* a2 = original dst; a5 = current dst; a11= original len
 * bytes_copied = a5 - a2
 * retval = bytes_not_copied = original len - bytes_copied
 * retval = a11 - (a5 - a2)
 */


10:
	sub	a2, a5, a2	/* a2 <-- bytes copied */
	sub	a2, a11, a2	/* a2 <-- bytes not copied */
	retw