1 /* $NetBSD: in_cksum_arm.S,v 1.2 2003/09/23 10:01:36 scw Exp $ */
2
3 /*-
4 * Copyright 2003 Wasabi Systems, Inc.
5 * All rights reserved.
6 *
7 * Written by Steve C. Woodford for Wasabi Systems, Inc.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 * 3. All advertising materials mentioning features or use of this software
18 * must display the following acknowledgement:
19 * This product includes software developed for the NetBSD Project by
20 * Wasabi Systems, Inc.
21 * 4. The name of Wasabi Systems, Inc. may not be used to endorse
22 * or promote products derived from this software without specific prior
23 * written permission.
24 *
25 * THIS SOFTWARE IS PROVIDED BY WASABI SYSTEMS, INC. ``AS IS'' AND
26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
27 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
28 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL WASABI SYSTEMS, INC
29 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
30 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
31 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
32 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
33 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
34 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
35 * POSSIBILITY OF SUCH DAMAGE.
36 *
37 */
38
39 /*
40 * Hand-optimised in_cksum() and in4_cksum() implementations for ARM/armv5e
41 */
42
43 #include "opt_inet.h"
44
45 #include <machine/asm.h>
46 #include "assym.s"
47 __FBSDID("$FreeBSD: src/sys/arm/arm/in_cksum_arm.S,v 1.7 2007/10/13 12:05:03 cognet Exp $");
48
49 /*
50 * int in_cksum(struct mbuf *m, int len)
51 *
52 * Entry:
53 * r0 m
54 * r1 len
55 *
56 * NOTE: Assumes 'm' is *never* NULL.
57 */
58 /* LINTSTUB: Func: int in_cksum(struct mbuf *, int) */
59 ENTRY(in_cksum)
60 stmfd sp!, {r4-r11,lr}
61 mov r8, #0x00
62 mov r9, r1
63 mov r10, #0x00
64 mov ip, r0
65
66 .Lin_cksum_loop:
67 ldr r1, [ip, #(M_LEN)]
68 ldr r0, [ip, #(M_DATA)]
69 ldr ip, [ip, #(M_NEXT)]
70 .Lin_cksum_entry4:
71 cmp r9, r1
72 movlt r1, r9
73 sub r9, r9, r1
74 eor r11, r10, r0
75 add r10, r10, r1
76 adds r2, r1, #0x00
77 blne _ASM_LABEL(L_cksumdata)
78 tst r11, #0x01
79 movne r2, r2, ror #8
80 adds r8, r8, r2
81 adc r8, r8, #0x00
82 cmp ip, #0x00
83 bne .Lin_cksum_loop
84
85 mov r1, #0xff
86 orr r1, r1, #0xff00
87 and r0, r8, r1
88 add r0, r0, r8, lsr #16
89 add r0, r0, r0, lsr #16
90 and r0, r0, r1
91 eor r0, r0, r1
92 ldmfd sp!, {r4-r11,pc}
93
94
95 ENTRY(do_cksum)
96 stmfd sp!, {r4-r7, lr}
97 bl L_cksumdata
98 mov r0, r2
99 ldmfd sp!, {r4-r7, pc}
100 /*
101 * The main in*_cksum() workhorse...
102 *
103 * Entry parameters:
104 * r0 Pointer to buffer
105 * r1 Buffer length
106 * lr Return address
107 *
108 * Returns:
109 * r2 Accumulated 32-bit sum
110 *
111 * Clobbers:
112 * r0-r7
113 */
114 /* LINTSTUB: Ignore */
115 ASENTRY_NP(L_cksumdata)
116 #ifdef _ARM_ARCH_5E
117 pld [r0] /* Pre-fetch the start of the buffer */
118 #endif
119 mov r2, #0
120
121 /* We first have to word-align the buffer. */
122 ands r7, r0, #0x03
123 beq .Lcksumdata_wordaligned
124 rsb r7, r7, #0x04
125 cmp r1, r7 /* Enough bytes left to make it? */
126 blt .Lcksumdata_endgame
127 cmp r7, #0x02
128 ldrb r4, [r0], #0x01 /* Fetch 1st byte */
129 ldrgeb r5, [r0], #0x01 /* Fetch 2nd byte */
130 movlt r5, #0x00
131 ldrgtb r6, [r0], #0x01 /* Fetch 3rd byte */
132 movle r6, #0x00
133 /* Combine the three bytes depending on endianness and alignment */
134 #ifdef __ARMEB__
135 orreq r2, r5, r4, lsl #8
136 orreq r2, r2, r6, lsl #24
137 orrne r2, r4, r5, lsl #8
138 orrne r2, r2, r6, lsl #16
139 #else
140 orreq r2, r4, r5, lsl #8
141 orreq r2, r2, r6, lsl #16
142 orrne r2, r5, r4, lsl #8
143 orrne r2, r2, r6, lsl #24
144 #endif
145 subs r1, r1, r7 /* Update length */
146 RETeq /* All done? */
147
148 /* Buffer is now word aligned */
149 .Lcksumdata_wordaligned:
150 #ifdef _ARM_ARCH_5E
151 cmp r1, #0x04 /* Less than 4 bytes left? */
152 blt .Lcksumdata_endgame /* Yup */
153
154 /* Now quad-align, if necessary */
155 ands r7, r0, #0x04
156 ldrne r7, [r0], #0x04
157 subne r1, r1, #0x04
158 subs r1, r1, #0x40
159 blt .Lcksumdata_bigloop_end /* Note: C flag clear if branch taken */
160
161 /*
162 * Buffer is now quad aligned. Sum 64 bytes at a time.
163 * Note: First ldrd is hoisted above the loop, together with
164 * setting r6 to zero to avoid stalling for results in the
165 * loop. (r7 is live, from above).
166 */
167 ldrd r4, [r0], #0x08
168 mov r6, #0x00
169 .Lcksumdata_bigloop:
170 pld [r0, #0x18]
171 adds r2, r2, r6
172 adcs r2, r2, r7
173 ldrd r6, [r0], #0x08
174 adcs r2, r2, r4
175 adcs r2, r2, r5
176 ldrd r4, [r0], #0x08
177 adcs r2, r2, r6
178 adcs r2, r2, r7
179 ldrd r6, [r0], #0x08
180 adcs r2, r2, r4
181 adcs r2, r2, r5
182 ldrd r4, [r0], #0x08
183 adcs r2, r2, r6
184 adcs r2, r2, r7
185 pld [r0, #0x18]
186 ldrd r6, [r0], #0x08
187 adcs r2, r2, r4
188 adcs r2, r2, r5
189 ldrd r4, [r0], #0x08
190 adcs r2, r2, r6
191 adcs r2, r2, r7
192 ldrd r6, [r0], #0x08
193 adcs r2, r2, r4
194 adcs r2, r2, r5
195 adc r2, r2, #0x00
196 subs r1, r1, #0x40
197 ldrged r4, [r0], #0x08
198 bge .Lcksumdata_bigloop
199
200 adds r2, r2, r6 /* r6/r7 still need summing */
201 .Lcksumdata_bigloop_end:
202 adcs r2, r2, r7
203 adc r2, r2, #0x00
204
205 #else /* !_ARM_ARCH_5E */
206
207 subs r1, r1, #0x40
208 blt .Lcksumdata_bigloop_end
209
210 .Lcksumdata_bigloop:
211 ldmia r0!, {r3, r4, r5, r6}
212 adds r2, r2, r3
213 adcs r2, r2, r4
214 adcs r2, r2, r5
215 ldmia r0!, {r3, r4, r5, r7}
216 adcs r2, r2, r6
217 adcs r2, r2, r3
218 adcs r2, r2, r4
219 adcs r2, r2, r5
220 ldmia r0!, {r3, r4, r5, r6}
221 adcs r2, r2, r7
222 adcs r2, r2, r3
223 adcs r2, r2, r4
224 adcs r2, r2, r5
225 ldmia r0!, {r3, r4, r5, r7}
226 adcs r2, r2, r6
227 adcs r2, r2, r3
228 adcs r2, r2, r4
229 adcs r2, r2, r5
230 adcs r2, r2, r7
231 adc r2, r2, #0x00
232 subs r1, r1, #0x40
233 bge .Lcksumdata_bigloop
234 .Lcksumdata_bigloop_end:
235 #endif
236
237 adds r1, r1, #0x40
238 RETeq
239 cmp r1, #0x20
240
241 #ifdef _ARM_ARCH_5E
242 ldrged r4, [r0], #0x08 /* Avoid stalling pld and result */
243 blt .Lcksumdata_less_than_32
244 pld [r0, #0x18]
245 ldrd r6, [r0], #0x08
246 adds r2, r2, r4
247 adcs r2, r2, r5
248 ldrd r4, [r0], #0x08
249 adcs r2, r2, r6
250 adcs r2, r2, r7
251 ldrd r6, [r0], #0x08
252 adcs r2, r2, r4
253 adcs r2, r2, r5
254 adcs r2, r2, r6 /* XXX: Unavoidable result stall */
255 adcs r2, r2, r7
256 #else
257 blt .Lcksumdata_less_than_32
258 ldmia r0!, {r3, r4, r5, r6}
259 adds r2, r2, r3
260 adcs r2, r2, r4
261 adcs r2, r2, r5
262 ldmia r0!, {r3, r4, r5, r7}
263 adcs r2, r2, r6
264 adcs r2, r2, r3
265 adcs r2, r2, r4
266 adcs r2, r2, r5
267 adcs r2, r2, r7
268 #endif
269 adc r2, r2, #0x00
270 subs r1, r1, #0x20
271 RETeq
272
273 .Lcksumdata_less_than_32:
274 /* There are less than 32 bytes left */
275 and r3, r1, #0x18
276 rsb r4, r3, #0x18
277 sub r1, r1, r3
278 adds r4, r4, r4, lsr #1 /* Side effect: Clear carry flag */
279 addne pc, pc, r4
280 nop
281
282 /*
283 * Note: We use ldm here, even on armv5e, since the combined issue/result
284 * latencies for ldm and ldrd are the same. Using ldm avoids needless #ifdefs.
285 */
286 /* At least 24 bytes remaining... */
287 ldmia r0!, {r4, r5}
288 adcs r2, r2, r4
289 adcs r2, r2, r5
290
291 /* At least 16 bytes remaining... */
292 ldmia r0!, {r4, r5}
293 adcs r2, r2, r4
294 adcs r2, r2, r5
295
296 /* At least 8 bytes remaining... */
297 ldmia r0!, {r4, r5}
298 adcs r2, r2, r4
299 adcs r2, r2, r5
300
301 /* Less than 8 bytes remaining... */
302 adc r2, r2, #0x00
303 subs r1, r1, #0x04
304 blt .Lcksumdata_lessthan4
305
306 ldr r4, [r0], #0x04
307 sub r1, r1, #0x04
308 adds r2, r2, r4
309 adc r2, r2, #0x00
310
311 /* Deal with < 4 bytes remaining */
312 .Lcksumdata_lessthan4:
313 adds r1, r1, #0x04
314 RETeq
315
316 /* Deal with 1 to 3 remaining bytes, possibly misaligned */
317 .Lcksumdata_endgame:
318 ldrb r3, [r0] /* Fetch first byte */
319 cmp r1, #0x02
320 ldrgeb r4, [r0, #0x01] /* Fetch 2nd and 3rd as necessary */
321 movlt r4, #0x00
322 ldrgtb r5, [r0, #0x02]
323 movle r5, #0x00
324 /* Combine the three bytes depending on endianness and alignment */
325 tst r0, #0x01
326 #ifdef __ARMEB__
327 orreq r3, r4, r3, lsl #8
328 orreq r3, r3, r5, lsl #24
329 orrne r3, r3, r4, lsl #8
330 orrne r3, r3, r5, lsl #16
331 #else
332 orreq r3, r3, r4, lsl #8
333 orreq r3, r3, r5, lsl #16
334 orrne r3, r4, r3, lsl #8
335 orrne r3, r3, r5, lsl #24
336 #endif
337 adds r2, r2, r3
338 adc r2, r2, #0x00
339 RET
Cache object: f61485d3a3e46d65e003a6735cc1e845
|