summaryrefslogtreecommitdiff
path: root/arch/xtensa/lib/checksum.S
blob: c6e73b12e51971cb7c28913abc7a5e9ec83f673b (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
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
/* SPDX-License-Identifier: GPL-2.0-or-later */
/*
 * INET		An implementation of the TCP/IP protocol suite for the LINUX
 *		operating system.  INET is implemented using the  BSD Socket
 *		interface as the means of communication with the user level.
 *
 *		IP/TCP/UDP checksumming routines
 *
 * Xtensa version:  Copyright (C) 2001 Tensilica, Inc. by Kevin Chea
 *                  Optimized by Joe Taylor
 */

#include <linux/errno.h>
#include <linux/linkage.h>
#include <asm/asmmacro.h>
#include <asm/core.h>

/*
 * computes a partial checksum, e.g. for TCP/UDP fragments
 */

/*
 * unsigned int csum_partial(const unsigned char *buf, int len,
 *                           unsigned int sum);
 *    a2 = buf
 *    a3 = len
 *    a4 = sum
 *
 * This function assumes 2- or 4-byte alignment.  Other alignments will fail!
 */

/* ONES_ADD converts twos-complement math to ones-complement. */
#define ONES_ADD(sum, val)	  \
	add	sum, sum, val	; \
	bgeu	sum, val, 99f	; \
	addi	sum, sum, 1	; \
99:				;

.text
ENTRY(csum_partial)

	/*
	 * Experiments with Ethernet and SLIP connections show that buf
	 * is aligned on either a 2-byte or 4-byte boundary.
	 */
	entry	sp, 32
	extui	a5, a2, 0, 2
	bnez	a5, 8f		/* branch if 2-byte aligned */
	/* Fall-through on common case, 4-byte alignment */
1:
	srli	a5, a3, 5	/* 32-byte chunks */
#if XCHAL_HAVE_LOOPS
	loopgtz	a5, 2f
#else
	beqz	a5, 2f
	slli	a5, a5, 5
	add	a5, a5, a2	/* a5 = end of last 32-byte chunk */
.Loop1:
#endif
	l32i	a6, a2, 0
	l32i	a7, a2, 4
	ONES_ADD(a4, a6)
	ONES_ADD(a4, a7)
	l32i	a6, a2, 8
	l32i	a7, a2, 12
	ONES_ADD(a4, a6)
	ONES_ADD(a4, a7)
	l32i	a6, a2, 16
	l32i	a7, a2, 20
	ONES_ADD(a4, a6)
	ONES_ADD(a4, a7)
	l32i	a6, a2, 24
	l32i	a7, a2, 28
	ONES_ADD(a4, a6)
	ONES_ADD(a4, a7)
	addi	a2, a2, 4*8
#if !XCHAL_HAVE_LOOPS
	blt	a2, a5, .Loop1
#endif
2:
	extui	a5, a3, 2, 3	/* remaining 4-byte chunks */
#if XCHAL_HAVE_LOOPS
	loopgtz	a5, 3f
#else
	beqz	a5, 3f
	slli	a5, a5, 2
	add	a5, a5, a2	/* a5 = end of last 4-byte chunk */
.Loop2:
#endif
	l32i	a6, a2, 0
	ONES_ADD(a4, a6)
	addi	a2, a2, 4
#if !XCHAL_HAVE_LOOPS
	blt	a2, a5, .Loop2
#endif
3:
	_bbci.l	a3, 1, 5f	/* remaining 2-byte chunk */
	l16ui	a6, a2, 0
	ONES_ADD(a4, a6)
	addi	a2, a2, 2
5:
	_bbci.l	a3, 0, 7f	/* remaining 1-byte chunk */
6:	l8ui	a6, a2, 0
#ifdef __XTENSA_EB__
	slli	a6, a6, 8	/* load byte into bits 8..15 */
#endif
	ONES_ADD(a4, a6)
7:
	mov	a2, a4
	retw

	/* uncommon case, buf is 2-byte aligned */
8:
	beqz	a3, 7b		/* branch if len == 0 */
	beqi	a3, 1, 6b	/* branch if len == 1 */

	extui	a5, a2, 0, 1
	bnez	a5, 8f		/* branch if 1-byte aligned */

	l16ui	a6, a2, 0	/* common case, len >= 2 */
	ONES_ADD(a4, a6)
	addi	a2, a2, 2	/* adjust buf */
	addi	a3, a3, -2	/* adjust len */
	j	1b		/* now buf is 4-byte aligned */

	/* case: odd-byte aligned, len > 1
	 * This case is dog slow, so don't give us an odd address.
	 * (I don't think this ever happens, but just in case.)
	 */
8:
	srli	a5, a3, 2	/* 4-byte chunks */
#if XCHAL_HAVE_LOOPS
	loopgtz	a5, 2f
#else
	beqz	a5, 2f
	slli	a5, a5, 2
	add	a5, a5, a2	/* a5 = end of last 4-byte chunk */
.Loop3:
#endif
	l8ui	a6, a2, 0	/* bits 24..31 */
	l16ui	a7, a2, 1	/* bits  8..23 */
	l8ui	a8, a2, 3	/* bits  0.. 8 */
#ifdef	__XTENSA_EB__
	slli	a6, a6, 24
#else
	slli	a8, a8, 24
#endif
	slli	a7, a7, 8
	or	a7, a7, a6
	or	a7, a7, a8
	ONES_ADD(a4, a7)
	addi	a2, a2, 4
#if !XCHAL_HAVE_LOOPS
	blt	a2, a5, .Loop3
#endif
2:
	_bbci.l	a3, 1, 3f	/* remaining 2-byte chunk, still odd addr */
	l8ui	a6, a2, 0
	l8ui	a7, a2, 1
#ifdef	__XTENSA_EB__
	slli	a6, a6, 8
#else
	slli	a7, a7, 8
#endif
	or	a7, a7, a6
	ONES_ADD(a4, a7)
	addi	a2, a2, 2
3:
	j	5b		/* branch to handle the remaining byte */

ENDPROC(csum_partial)

/*
 * Copy from ds while checksumming, otherwise like csum_partial
 */

/*
unsigned int csum_partial_copy_generic (const char *src, char *dst, int len,
					int sum, int *src_err_ptr, int *dst_err_ptr)
	a2  = src
	a3  = dst
	a4  = len
	a5  = sum
	a6  = src_err_ptr
	a7  = dst_err_ptr
	a8  = temp
	a9  = temp
	a10 = temp
	a11 = original len for exception handling
	a12 = original dst for exception handling

    This function is optimized for 4-byte aligned addresses.  Other
    alignments work, but not nearly as efficiently.
 */

ENTRY(csum_partial_copy_generic)

	entry	sp, 32
	mov	a12, a3
	mov	a11, a4
	or	a10, a2, a3

	/* We optimize the following alignment tests for the 4-byte
	aligned case.  Two bbsi.l instructions might seem more optimal
	(commented out below).  However, both labels 5: and 3: are out
	of the imm8 range, so the assembler relaxes them into
	equivalent bbci.l, j combinations, which is actually
	slower. */

	extui	a9, a10, 0, 2
	beqz	a9, 1f		/* branch if both are 4-byte aligned */
	bbsi.l	a10, 0, 5f	/* branch if one address is odd */
	j	3f		/* one address is 2-byte aligned */

/*	_bbsi.l	a10, 0, 5f */	/* branch if odd address */
/*	_bbsi.l	a10, 1, 3f */	/* branch if 2-byte-aligned address */

1:
	/* src and dst are both 4-byte aligned */
	srli	a10, a4, 5	/* 32-byte chunks */
#if XCHAL_HAVE_LOOPS
	loopgtz	a10, 2f
#else
	beqz	a10, 2f
	slli	a10, a10, 5
	add	a10, a10, a2	/* a10 = end of last 32-byte src chunk */
.Loop5:
#endif
EX(10f)	l32i	a9, a2, 0
EX(10f)	l32i	a8, a2, 4
EX(11f)	s32i	a9, a3, 0
EX(11f)	s32i	a8, a3, 4
	ONES_ADD(a5, a9)
	ONES_ADD(a5, a8)
EX(10f)	l32i	a9, a2, 8
EX(10f)	l32i	a8, a2, 12
EX(11f)	s32i	a9, a3, 8
EX(11f)	s32i	a8, a3, 12
	ONES_ADD(a5, a9)
	ONES_ADD(a5, a8)
EX(10f)	l32i	a9, a2, 16
EX(10f)	l32i	a8, a2, 20
EX(11f)	s32i	a9, a3, 16
EX(11f)	s32i	a8, a3, 20
	ONES_ADD(a5, a9)
	ONES_ADD(a5, a8)
EX(10f)	l32i	a9, a2, 24
EX(10f)	l32i	a8, a2, 28
EX(11f)	s32i	a9, a3, 24
EX(11f)	s32i	a8, a3, 28
	ONES_ADD(a5, a9)
	ONES_ADD(a5, a8)
	addi	a2, a2, 32
	addi	a3, a3, 32
#if !XCHAL_HAVE_LOOPS
	blt	a2, a10, .Loop5
#endif
2:
	extui	a10, a4, 2, 3	/* remaining 4-byte chunks */
	extui	a4, a4, 0, 2	/* reset len for general-case, 2-byte chunks */
#if XCHAL_HAVE_LOOPS
	loopgtz	a10, 3f
#else
	beqz	a10, 3f
	slli	a10, a10, 2
	add	a10, a10, a2	/* a10 = end of last 4-byte src chunk */
.Loop6:
#endif
EX(10f)	l32i	a9, a2, 0
EX(11f)	s32i	a9, a3, 0
	ONES_ADD(a5, a9)
	addi	a2, a2, 4
	addi	a3, a3, 4
#if !XCHAL_HAVE_LOOPS
	blt	a2, a10, .Loop6
#endif
3:
	/*
	Control comes to here in two cases: (1) It may fall through
	to here from the 4-byte alignment case to process, at most,
	one 2-byte chunk.  (2) It branches to here from above if
	either src or dst is 2-byte aligned, and we process all bytes
	here, except for perhaps a trailing odd byte.  It's
	inefficient, so align your addresses to 4-byte boundaries.

	a2 = src
	a3 = dst
	a4 = len
	a5 = sum
	*/
	srli	a10, a4, 1	/* 2-byte chunks */
#if XCHAL_HAVE_LOOPS
	loopgtz	a10, 4f
#else
	beqz	a10, 4f
	slli	a10, a10, 1
	add	a10, a10, a2	/* a10 = end of last 2-byte src chunk */
.Loop7:
#endif
EX(10f)	l16ui	a9, a2, 0
EX(11f)	s16i	a9, a3, 0
	ONES_ADD(a5, a9)
	addi	a2, a2, 2
	addi	a3, a3, 2
#if !XCHAL_HAVE_LOOPS
	blt	a2, a10, .Loop7
#endif
4:
	/* This section processes a possible trailing odd byte. */
	_bbci.l	a4, 0, 8f	/* 1-byte chunk */
EX(10f)	l8ui	a9, a2, 0
EX(11f)	s8i	a9, a3, 0
#ifdef __XTENSA_EB__
	slli	a9, a9, 8	/* shift byte to bits 8..15 */
#endif
	ONES_ADD(a5, a9)
8:
	mov	a2, a5
	retw

5:
	/* Control branch to here when either src or dst is odd.  We
	process all bytes using 8-bit accesses.  Grossly inefficient,
	so don't feed us an odd address. */

	srli	a10, a4, 1	/* handle in pairs for 16-bit csum */
#if XCHAL_HAVE_LOOPS
	loopgtz	a10, 6f
#else
	beqz	a10, 6f
	slli	a10, a10, 1
	add	a10, a10, a2	/* a10 = end of last odd-aligned, 2-byte src chunk */
.Loop8:
#endif
EX(10f)	l8ui	a9, a2, 0
EX(10f)	l8ui	a8, a2, 1
EX(11f)	s8i	a9, a3, 0
EX(11f)	s8i	a8, a3, 1
#ifdef __XTENSA_EB__
	slli	a9, a9, 8	/* combine into a single 16-bit value */
#else				/* for checksum computation */
	slli	a8, a8, 8
#endif
	or	a9, a9, a8
	ONES_ADD(a5, a9)
	addi	a2, a2, 2
	addi	a3, a3, 2
#if !XCHAL_HAVE_LOOPS
	blt	a2, a10, .Loop8
#endif
6:
	j	4b		/* process the possible trailing odd byte */

ENDPROC(csum_partial_copy_generic)


# Exception handler:
.section .fixup, "ax"
/*
	a6  = src_err_ptr
	a7  = dst_err_ptr
	a11 = original len for exception handling
	a12 = original dst for exception handling
*/

10:
	_movi	a2, -EFAULT
	s32i	a2, a6, 0	/* src_err_ptr */

	# clear the complete destination - computing the rest
	# is too much work
	movi	a2, 0
#if XCHAL_HAVE_LOOPS
	loopgtz	a11, 2f
#else
	beqz	a11, 2f
	add	a11, a11, a12	/* a11 = ending address */
.Leloop:
#endif
	s8i	a2, a12, 0
	addi	a12, a12, 1
#if !XCHAL_HAVE_LOOPS
	blt	a12, a11, .Leloop
#endif
2:
	retw

11:
	movi	a2, -EFAULT
	s32i	a2, a7, 0	/* dst_err_ptr */
	movi	a2, 0
	retw

.previous