xref: /openbmc/linux/arch/sh/lib/checksum.S (revision 11a163f2) 
1/* SPDX-License-Identifier: GPL-2.0+
2 *
3 * $Id: checksum.S,v 1.10 2001/07/06 13:11:32 gniibe Exp $
4 *
5 * INET		An implementation of the TCP/IP protocol suite for the LINUX
6 *		operating system.  INET is implemented using the  BSD Socket
7 *		interface as the means of communication with the user level.
8 *
9 *		IP/TCP/UDP checksumming routines
10 *
11 * Authors:	Jorge Cwik, <jorge@laser.satlink.net>
12 *		Arnt Gulbrandsen, <agulbra@nvg.unit.no>
13 *		Tom May, <ftom@netcom.com>
14 *              Pentium Pro/II routines:
15 *              Alexander Kjeldaas <astor@guardian.no>
16 *              Finn Arne Gangstad <finnag@guardian.no>
17 *		Lots of code moved from tcp.c and ip.c; see those files
18 *		for more names.
19 *
20 * Changes:     Ingo Molnar, converted csum_partial_copy() to 2.1 exception
21 *			     handling.
22 *		Andi Kleen,  add zeroing on error
23 *                   converted to pure assembler
24 *
25 * SuperH version:  Copyright (C) 1999  Niibe Yutaka
26 */
27
28#include <asm/errno.h>
29#include <linux/linkage.h>
30
31/*
32 * computes a partial checksum, e.g. for TCP/UDP fragments
33 */
34
35/*
36 * asmlinkage __wsum csum_partial(const void *buf, int len, __wsum sum);
37 */
38
39.text
40ENTRY(csum_partial)
41	  /*
42	   * Experiments with Ethernet and SLIP connections show that buff
43	   * is aligned on either a 2-byte or 4-byte boundary.  We get at
44	   * least a twofold speedup on 486 and Pentium if it is 4-byte aligned.
45	   * Fortunately, it is easy to convert 2-byte alignment to 4-byte
46	   * alignment for the unrolled loop.
47	   */
48	mov	r4, r0
49	tst	#3, r0		! Check alignment.
50	bt/s	2f		! Jump if alignment is ok.
51	 mov	r4, r7		! Keep a copy to check for alignment
52	!
53	tst	#1, r0		! Check alignment.
54	bt	21f		! Jump if alignment is boundary of 2bytes.
55
56	! buf is odd
57	tst	r5, r5
58	add	#-1, r5
59	bt	9f
60	mov.b	@r4+, r0
61	extu.b	r0, r0
62	addc	r0, r6		! t=0 from previous tst
63	mov	r6, r0
64	shll8	r6
65	shlr16	r0
66	shlr8	r0
67	or	r0, r6
68	mov	r4, r0
69	tst	#2, r0
70	bt	2f
7121:
72	! buf is 2 byte aligned (len could be 0)
73	add	#-2, r5		! Alignment uses up two bytes.
74	cmp/pz	r5		!
75	bt/s	1f		! Jump if we had at least two bytes.
76	 clrt
77	bra	6f
78	 add	#2, r5		! r5 was < 2.  Deal with it.
791:
80	mov.w	@r4+, r0
81	extu.w	r0, r0
82	addc	r0, r6
83	bf	2f
84	add	#1, r6
852:
86	! buf is 4 byte aligned (len could be 0)
87	mov	r5, r1
88	mov	#-5, r0
89	shld	r0, r1
90	tst	r1, r1
91	bt/s	4f		! if it's =0, go to 4f
92	 clrt
93	.align	2
943:
95	mov.l	@r4+, r0
96	mov.l	@r4+, r2
97	mov.l	@r4+, r3
98	addc	r0, r6
99	mov.l	@r4+, r0
100	addc	r2, r6
101	mov.l	@r4+, r2
102	addc	r3, r6
103	mov.l	@r4+, r3
104	addc	r0, r6
105	mov.l	@r4+, r0
106	addc	r2, r6
107	mov.l	@r4+, r2
108	addc	r3, r6
109	addc	r0, r6
110	addc	r2, r6
111	movt	r0
112	dt	r1
113	bf/s	3b
114	 cmp/eq	#1, r0
115	! here, we know r1==0
116	addc	r1, r6			! add carry to r6
1174:
118	mov	r5, r0
119	and	#0x1c, r0
120	tst	r0, r0
121	bt	6f
122	! 4 bytes or more remaining
123	mov	r0, r1
124	shlr2	r1
125	mov	#0, r2
1265:
127	addc	r2, r6
128	mov.l	@r4+, r2
129	movt	r0
130	dt	r1
131	bf/s	5b
132	 cmp/eq	#1, r0
133	addc	r2, r6
134	addc	r1, r6		! r1==0 here, so it means add carry-bit
1356:
136	! 3 bytes or less remaining
137	mov	#3, r0
138	and	r0, r5
139	tst	r5, r5
140	bt	9f		! if it's =0 go to 9f
141	mov	#2, r1
142	cmp/hs  r1, r5
143	bf	7f
144	mov.w	@r4+, r0
145	extu.w	r0, r0
146	cmp/eq	r1, r5
147	bt/s	8f
148	 clrt
149	shll16	r0
150	addc	r0, r6
1517:
152	mov.b	@r4+, r0
153	extu.b	r0, r0
154#ifndef	__LITTLE_ENDIAN__
155	shll8	r0
156#endif
1578:
158	addc	r0, r6
159	mov	#0, r0
160	addc	r0, r6
1619:
162	! Check if the buffer was misaligned, if so realign sum
163	mov	r7, r0
164	tst	#1, r0
165	bt	10f
166	mov	r6, r0
167	shll8	r6
168	shlr16	r0
169	shlr8	r0
170	or	r0, r6
17110:
172	rts
173	 mov	r6, r0
174
175/*
176unsigned int csum_partial_copy_generic (const char *src, char *dst, int len)
177 */
178
179/*
180 * Copy from ds while checksumming, otherwise like csum_partial with initial
181 * sum being ~0U
182 */
183
184#define EXC(...)			\
185	9999: __VA_ARGS__ ;		\
186	.section __ex_table, "a";	\
187	.long 9999b, 6001f	;	\
188	.previous
189
190!
191! r4:	const char *SRC
192! r5:	char *DST
193! r6:	int LEN
194!
195ENTRY(csum_partial_copy_generic)
196	mov	#-1,r7
197	mov	#3,r0		! Check src and dest are equally aligned
198	mov	r4,r1
199	and	r0,r1
200	and	r5,r0
201	cmp/eq	r1,r0
202	bf	3f		! Different alignments, use slow version
203	tst	#1,r0		! Check dest word aligned
204	bf	3f		! If not, do it the slow way
205
206	mov	#2,r0
207	tst	r0,r5		! Check dest alignment.
208	bt	2f		! Jump if alignment is ok.
209	add	#-2,r6		! Alignment uses up two bytes.
210	cmp/pz	r6		! Jump if we had at least two bytes.
211	bt/s	1f
212	 clrt
213	add	#2,r6		! r6 was < 2.	Deal with it.
214	bra	4f
215	 mov	r6,r2
216
2173:	! Handle different src and dest alignments.
218	! This is not common, so simple byte by byte copy will do.
219	mov	r6,r2
220	shlr	r6
221	tst	r6,r6
222	bt	4f
223	clrt
224	.align	2
2255:
226EXC(	mov.b	@r4+,r1 	)
227EXC(	mov.b	@r4+,r0		)
228	extu.b	r1,r1
229EXC(	mov.b	r1,@r5		)
230EXC(	mov.b	r0,@(1,r5)	)
231	extu.b	r0,r0
232	add	#2,r5
233
234#ifdef	__LITTLE_ENDIAN__
235	shll8	r0
236#else
237	shll8	r1
238#endif
239	or	r1,r0
240
241	addc	r0,r7
242	movt	r0
243	dt	r6
244	bf/s	5b
245	 cmp/eq	#1,r0
246	mov	#0,r0
247	addc	r0, r7
248
249	mov	r2, r0
250	tst	#1, r0
251	bt	7f
252	bra	5f
253	 clrt
254
255	! src and dest equally aligned, but to a two byte boundary.
256	! Handle first two bytes as a special case
257	.align	2
2581:
259EXC(	mov.w	@r4+,r0		)
260EXC(	mov.w	r0,@r5		)
261	add	#2,r5
262	extu.w	r0,r0
263	addc	r0,r7
264	mov	#0,r0
265	addc	r0,r7
2662:
267	mov	r6,r2
268	mov	#-5,r0
269	shld	r0,r6
270	tst	r6,r6
271	bt/s	2f
272	 clrt
273	.align	2
2741:
275EXC(	mov.l	@r4+,r0		)
276EXC(	mov.l	@r4+,r1		)
277	addc	r0,r7
278EXC(	mov.l	r0,@r5		)
279EXC(	mov.l	r1,@(4,r5)	)
280	addc	r1,r7
281
282EXC(	mov.l	@r4+,r0		)
283EXC(	mov.l	@r4+,r1		)
284	addc	r0,r7
285EXC(	mov.l	r0,@(8,r5)	)
286EXC(	mov.l	r1,@(12,r5)	)
287	addc	r1,r7
288
289EXC(	mov.l	@r4+,r0 	)
290EXC(	mov.l	@r4+,r1		)
291	addc	r0,r7
292EXC(	mov.l	r0,@(16,r5)	)
293EXC(	mov.l	r1,@(20,r5)	)
294	addc	r1,r7
295
296EXC(	mov.l	@r4+,r0		)
297EXC(	mov.l	@r4+,r1		)
298	addc	r0,r7
299EXC(	mov.l	r0,@(24,r5)	)
300EXC(	mov.l	r1,@(28,r5)	)
301	addc	r1,r7
302	add	#32,r5
303	movt	r0
304	dt	r6
305	bf/s	1b
306	 cmp/eq	#1,r0
307	mov	#0,r0
308	addc	r0,r7
309
3102:	mov	r2,r6
311	mov	#0x1c,r0
312	and	r0,r6
313	cmp/pl	r6
314	bf/s	4f
315	 clrt
316	shlr2	r6
3173:
318EXC(	mov.l	@r4+,r0	)
319	addc	r0,r7
320EXC(	mov.l	r0,@r5	)
321	add	#4,r5
322	movt	r0
323	dt	r6
324	bf/s	3b
325	 cmp/eq	#1,r0
326	mov	#0,r0
327	addc	r0,r7
3284:	mov	r2,r6
329	mov	#3,r0
330	and	r0,r6
331	cmp/pl	r6
332	bf	7f
333	mov	#2,r1
334	cmp/hs	r1,r6
335	bf	5f
336EXC(	mov.w	@r4+,r0	)
337EXC(	mov.w	r0,@r5	)
338	extu.w	r0,r0
339	add	#2,r5
340	cmp/eq	r1,r6
341	bt/s	6f
342	 clrt
343	shll16	r0
344	addc	r0,r7
3455:
346EXC(	mov.b	@r4+,r0	)
347EXC(	mov.b	r0,@r5	)
348	extu.b	r0,r0
349#ifndef	__LITTLE_ENDIAN__
350	shll8	r0
351#endif
3526:	addc	r0,r7
353	mov	#0,r0
354	addc	r0,r7
3557:
356
357# Exception handler:
358.section .fixup, "ax"
359
3606001:
361	rts
362	 mov	#0,r0
363.previous
364	rts
365	 mov	r7,r0
366