1 /* SPDX-License-Identifier: GPL-2.0 */
2 /* checksum.S: Sparc optimized checksum code.
3 *
4 * Copyright(C) 1995 Linus Torvalds
5 * Copyright(C) 1995 Miguel de Icaza
6 * Copyright(C) 1996 David S. Miller
7 * Copyright(C) 1997 Jakub Jelinek
8 *
9 * derived from:
10 * Linux/Alpha checksum c-code
11 * Linux/ix86 inline checksum assembly
12 * RFC1071 Computing the Internet Checksu
13 * David Mosberger-Tang for optimized ref
14 * BSD4.4 portable checksum routine
15 */
16
17 #include <linux/export.h>
18 #include <asm/errno.h>
19
20 #define CSUM_BIGCHUNK(buf, offset, sum, t0, t1
21 ldd [buf + offset + 0x00], t0;
22 ldd [buf + offset + 0x08], t2;
23 addxcc t0, sum, sum;
24 addxcc t1, sum, sum;
25 ldd [buf + offset + 0x10], t4;
26 addxcc t2, sum, sum;
27 addxcc t3, sum, sum;
28 ldd [buf + offset + 0x18], t0;
29 addxcc t4, sum, sum;
30 addxcc t5, sum, sum;
31 addxcc t0, sum, sum;
32 addxcc t1, sum, sum;
33
34 #define CSUM_LASTCHUNK(buf, offset, sum, t0, t
35 ldd [buf - offset - 0x08], t0;
36 ldd [buf - offset - 0x00], t2;
37 addxcc t0, sum, sum;
38 addxcc t1, sum, sum;
39 addxcc t2, sum, sum;
40 addxcc t3, sum, sum;
41
42 /* Do end cruft out of band to get bet
43 csum_partial_end_cruft:
44 be 1f
45 andcc %o1, 4, %g0
46 ldd [%o0], %g2
47 addcc %g2, %o2, %o2
48 addxcc %g3, %o2, %o2
49 add %o0, 8, %o0
50 addx %g0, %o2, %o2
51 andcc %o1, 4, %g0
52 1: be 1f
53 andcc %o1, 3, %o1
54 ld [%o0], %g2
55 addcc %g2, %o2, %o2
56 add %o0, 4, %o0
57 addx %g0, %o2, %o2
58 andcc %o1, 3, %g0
59 1: be 1f
60 addcc %o1, -1, %g0
61 bne 2f
62 subcc %o1, 2, %o1
63 b 4f
64 or %g0, %g0, %o4
65 2: lduh [%o0], %o4
66 be 6f
67 add %o0, 2, %o0
68 sll %o4, 16, %o4
69 4: ldub [%o0], %o5
70 sll %o5, 8, %o5
71 or %o5, %o4, %o4
72 6: addcc %o4, %o2, %o2
73 1: retl
74 addx %g0, %o2, %o0
75
76 /* Also do alignment out of band to ge
77 csum_partial_fix_alignment:
78 cmp %o1, 6
79 bl cpte - 0x4
80 andcc %o0, 0x2, %g0
81 be 1f
82 andcc %o0, 0x4, %g0
83 lduh [%o0 + 0x00], %g2
84 sub %o1, 2, %o1
85 add %o0, 2, %o0
86 sll %g2, 16, %g2
87 addcc %g2, %o2, %o2
88 srl %o2, 16, %g3
89 addx %g0, %g3, %g2
90 sll %o2, 16, %o2
91 sll %g2, 16, %g3
92 srl %o2, 16, %o2
93 andcc %o0, 0x4, %g0
94 or %g3, %o2, %o2
95 1: be cpa
96 andcc %o1, 0xffffff80, %o3
97 ld [%o0 + 0x00], %g2
98 sub %o1, 4, %o1
99 addcc %g2, %o2, %o2
100 add %o0, 4, %o0
101 addx %g0, %o2, %o2
102 b cpa
103 andcc %o1, 0xffffff80, %o3
104
105 /* The common case is to get called wi
106 * buffer of size 0x20. Follow the co
107 */
108 .globl csum_partial
109 EXPORT_SYMBOL(csum_partial)
110 csum_partial: /* %o0=buf, %o
111 andcc %o0, 0x7, %g0
112 bne csum_partial_fix_alignment
113 sethi %hi(cpte - 8), %g7
114 andcc %o1, 0xffffff80, %o3
115 cpa: be 3f
116 andcc %o1, 0x70, %g1
117 5: CSUM_BIGCHUNK(%o0, 0x00, %o2, %o4, %o5
118 CSUM_BIGCHUNK(%o0, 0x20, %o2, %o4, %o5
119 CSUM_BIGCHUNK(%o0, 0x40, %o2, %o4, %o5
120 CSUM_BIGCHUNK(%o0, 0x60, %o2, %o4, %o5
121 addx %g0, %o2, %o2
122 subcc %o3, 128, %o3
123 bne 5b
124 add %o0, 128, %o0
125 andcc %o1, 0x70, %g1
126 3: be cpte
127 andcc %o1, 0xf, %g0
128 srl %g1, 1, %o4
129 sub %g7, %g1, %g7
130 sub %g7, %o4, %g7
131 jmp %g7 + %lo(cpte - 8)
132 add %o0, %g1, %o0
133 cptbl: CSUM_LASTCHUNK(%o0, 0x68, %o2, %g2, %g
134 CSUM_LASTCHUNK(%o0, 0x58, %o2, %g2, %g
135 CSUM_LASTCHUNK(%o0, 0x48, %o2, %g2, %g
136 CSUM_LASTCHUNK(%o0, 0x38, %o2, %g2, %g
137 CSUM_LASTCHUNK(%o0, 0x28, %o2, %g2, %g
138 CSUM_LASTCHUNK(%o0, 0x18, %o2, %g2, %g
139 CSUM_LASTCHUNK(%o0, 0x08, %o2, %g2, %g
140 addx %g0, %o2, %o2
141 andcc %o1, 0xf, %g0
142 cpte: bne csum_partial_end_cruft
143 andcc %o1, 8, %g0
144 cpout: retl
145 mov %o2, %o0
146
147 /* Work around cpp -rob */
148 #define ALLOC #alloc
149 #define EXECINSTR #execinstr
150 #define EX(x,y)
151 98: x,y;
152 .section __ex_table,ALLOC;
153 .align 4;
154 .word 98b, cc_fault;
155 .text;
156 .align 4
157
158 /* This aligned version executes typic
159 * is the best I can do. I say 8.5 be
160 * the next ldd in the main unrolled l
161 * If you change these macros (includi
162 * please check the fixup code below a
163 */
164 #define CSUMCOPY_BIGCHUNK_ALIGNED(src, dst, su
165 EX(ldd [src + off + 0x00], t0);
166 EX(ldd [src + off + 0x08], t2);
167 addxcc t0, sum, sum;
168 EX(ldd [src + off + 0x10], t4);
169 addxcc t1, sum, sum;
170 EX(ldd [src + off + 0x18], t6);
171 addxcc t2, sum, sum;
172 EX(std t0, [dst + off + 0x00]);
173 addxcc t3, sum, sum;
174 EX(std t2, [dst + off + 0x08]);
175 addxcc t4, sum, sum;
176 EX(std t4, [dst + off + 0x10]);
177 addxcc t5, sum, sum;
178 EX(std t6, [dst + off + 0x18]);
179 addxcc t6, sum, sum;
180 addxcc t7, sum, sum;
181
182 /* 12 superscalar cycles seems to be t
183 * because of this we thus do all the
184 * Viking MXCC into streaming mode. H
185 */
186 #define CSUMCOPY_BIGCHUNK(src, dst, sum, off,
187 EX(ldd [src + off + 0x00], t0);
188 EX(ldd [src + off + 0x08], t2);
189 EX(ldd [src + off + 0x10], t4);
190 EX(ldd [src + off + 0x18], t6);
191 EX(st t0, [dst + off + 0x00]);
192 addxcc t0, sum, sum;
193 EX(st t1, [dst + off + 0x04]);
194 addxcc t1, sum, sum;
195 EX(st t2, [dst + off + 0x08]);
196 addxcc t2, sum, sum;
197 EX(st t3, [dst + off + 0x0c]);
198 addxcc t3, sum, sum;
199 EX(st t4, [dst + off + 0x10]);
200 addxcc t4, sum, sum;
201 EX(st t5, [dst + off + 0x14]);
202 addxcc t5, sum, sum;
203 EX(st t6, [dst + off + 0x18]);
204 addxcc t6, sum, sum;
205 EX(st t7, [dst + off + 0x1c]);
206 addxcc t7, sum, sum;
207
208 /* Yuck, 6 superscalar cycles... */
209 #define CSUMCOPY_LASTCHUNK(src, dst, sum, off,
210 EX(ldd [src - off - 0x08], t0);
211 EX(ldd [src - off - 0x00], t2);
212 addxcc t0, sum, sum;
213 EX(st t0, [dst - off - 0x08]);
214 addxcc t1, sum, sum;
215 EX(st t1, [dst - off - 0x04]);
216 addxcc t2, sum, sum;
217 EX(st t2, [dst - off - 0x00]);
218 addxcc t3, sum, sum;
219 EX(st t3, [dst - off + 0x04]);
220
221 /* Handle the end cruft code out of ba
222 cc_end_cruft:
223 be 1f
224 andcc %o3, 4, %g0
225 EX(ldd [%o0 + 0x00], %g2)
226 add %o1, 8, %o1
227 addcc %g2, %g7, %g7
228 add %o0, 8, %o0
229 addxcc %g3, %g7, %g7
230 EX(st %g2, [%o1 - 0x08])
231 addx %g0, %g7, %g7
232 andcc %o3, 4, %g0
233 EX(st %g3, [%o1 - 0x04])
234 1: be 1f
235 andcc %o3, 3, %o3
236 EX(ld [%o0 + 0x00], %g2)
237 add %o1, 4, %o1
238 addcc %g2, %g7, %g7
239 EX(st %g2, [%o1 - 0x04])
240 addx %g0, %g7, %g7
241 andcc %o3, 3, %g0
242 add %o0, 4, %o0
243 1: be 1f
244 addcc %o3, -1, %g0
245 bne 2f
246 subcc %o3, 2, %o3
247 b 4f
248 or %g0, %g0, %o4
249 2: EX(lduh [%o0 + 0x00], %o4)
250 add %o0, 2, %o0
251 EX(sth %o4, [%o1 + 0x00])
252 be 6f
253 add %o1, 2, %o1
254 sll %o4, 16, %o4
255 4: EX(ldub [%o0 + 0x00], %o5)
256 EX(stb %o5, [%o1 + 0x00])
257 sll %o5, 8, %o5
258 or %o5, %o4, %o4
259 6: addcc %o4, %g7, %g7
260 1: retl
261 addx %g0, %g7, %o0
262
263 /* Also, handle the alignment code out
264 cc_dword_align:
265 cmp %g1, 16
266 bge 1f
267 srl %g1, 1, %o3
268 2: cmp %o3, 0
269 be,a ccte
270 andcc %g1, 0xf, %o3
271 andcc %o3, %o0, %g0 ! Check %o0 on
272 be,a 2b
273 srl %o3, 1, %o3
274 1: andcc %o0, 0x1, %g0
275 bne ccslow
276 andcc %o0, 0x2, %g0
277 be 1f
278 andcc %o0, 0x4, %g0
279 EX(lduh [%o0 + 0x00], %g4)
280 sub %g1, 2, %g1
281 EX(sth %g4, [%o1 + 0x00])
282 add %o0, 2, %o0
283 sll %g4, 16, %g4
284 addcc %g4, %g7, %g7
285 add %o1, 2, %o1
286 srl %g7, 16, %g3
287 addx %g0, %g3, %g4
288 sll %g7, 16, %g7
289 sll %g4, 16, %g3
290 srl %g7, 16, %g7
291 andcc %o0, 0x4, %g0
292 or %g3, %g7, %g7
293 1: be 3f
294 andcc %g1, 0xffffff80, %g0
295 EX(ld [%o0 + 0x00], %g4)
296 sub %g1, 4, %g1
297 EX(st %g4, [%o1 + 0x00])
298 add %o0, 4, %o0
299 addcc %g4, %g7, %g7
300 add %o1, 4, %o1
301 addx %g0, %g7, %g7
302 b 3f
303 andcc %g1, 0xffffff80, %g0
304
305 /* Sun, you just can't beat me, you ju
306 * give up. I'm serious, I am going t
307 * out of you, game over, lights out.
308 */
309 .align 8
310 .globl __csum_partial_copy_sparc_gene
311 EXPORT_SYMBOL(__csum_partial_copy_spar
312 __csum_partial_copy_sparc_generic:
313 /* %o0
314 xor %o0, %o1, %o4 ! get
315 andcc %o4, 3, %g0 ! chec
316 bne ccslow ! bett
317 andcc %o0, 7, %g0 ! need
318 bne cc_dword_align ! yes,
319 andcc %g1, 0xffffff80, %g0 ! can
320 3: be 3f ! nope
321 andcc %o1, 4, %g0 ! dest
322 be ccdbl + 4 ! 8 by
323 5: CSUMCOPY_BIGCHUNK(%o0,%o1,%g7,0x00,%o4
324 CSUMCOPY_BIGCHUNK(%o0,%o1,%g7,0x20,%o4
325 CSUMCOPY_BIGCHUNK(%o0,%o1,%g7,0x40,%o4
326 CSUMCOPY_BIGCHUNK(%o0,%o1,%g7,0x60,%o4
327 sub %g1, 128, %g1 ! detr
328 addx %g0, %g7, %g7 ! add
329 andcc %g1, 0xffffff80, %g0 ! more
330 add %o0, 128, %o0 ! adva
331 bne 5b ! we d
332 add %o1, 128, %o1 ! adva
333 3: andcc %g1, 0x70, %o2 ! can
334 ccmerge:be ccte ! nope
335 andcc %g1, 0xf, %o3 ! get
336 srl %o2, 1, %o4 ! begi
337 sethi %hi(12f), %o5 ! set
338 add %o0, %o2, %o0 ! adva
339 sub %o5, %o4, %o5 ! cont
340 sll %o2, 1, %g2 ! cons
341 sub %o5, %g2, %o5 ! some
342 jmp %o5 + %lo(12f) ! jump
343 add %o1, %o2, %o1 ! adva
344 cctbl: CSUMCOPY_LASTCHUNK(%o0,%o1,%g7,0x68,%g
345 CSUMCOPY_LASTCHUNK(%o0,%o1,%g7,0x58,%g
346 CSUMCOPY_LASTCHUNK(%o0,%o1,%g7,0x48,%g
347 CSUMCOPY_LASTCHUNK(%o0,%o1,%g7,0x38,%g
348 CSUMCOPY_LASTCHUNK(%o0,%o1,%g7,0x28,%g
349 CSUMCOPY_LASTCHUNK(%o0,%o1,%g7,0x18,%g
350 CSUMCOPY_LASTCHUNK(%o0,%o1,%g7,0x08,%g
351 12: addx %g0, %g7, %g7
352 andcc %o3, 0xf, %g0 ! chec
353 ccte: bne cc_end_cruft ! some
354 andcc %o3, 8, %g0 ! begi
355 retl ! retu
356 mov %g7, %o0 ! give
357 ccdbl: CSUMCOPY_BIGCHUNK_ALIGNED(%o0,%o1,%g7,
358 CSUMCOPY_BIGCHUNK_ALIGNED(%o0,%o1,%g7,
359 CSUMCOPY_BIGCHUNK_ALIGNED(%o0,%o1,%g7,
360 CSUMCOPY_BIGCHUNK_ALIGNED(%o0,%o1,%g7,
361 sub %g1, 128, %g1 ! detr
362 addx %g0, %g7, %g7 ! add
363 andcc %g1, 0xffffff80, %g0 ! more
364 add %o0, 128, %o0 ! adva
365 bne ccdbl ! we d
366 add %o1, 128, %o1 ! adva
367 b ccmerge ! fini
368 andcc %g1, 0x70, %o2 ! can
369
370 ccslow: cmp %g1, 0
371 mov 0, %g5
372 bleu 4f
373 andcc %o0, 1, %o5
374 be,a 1f
375 srl %g1, 1, %g4
376 sub %g1, 1, %g1
377 EX(ldub [%o0], %g5)
378 add %o0, 1, %o0
379 EX(stb %g5, [%o1])
380 srl %g1, 1, %g4
381 add %o1, 1, %o1
382 1: cmp %g4, 0
383 be,a 3f
384 andcc %g1, 1, %g0
385 andcc %o0, 2, %g0
386 be,a 1f
387 srl %g4, 1, %g4
388 EX(lduh [%o0], %o4)
389 sub %g1, 2, %g1
390 srl %o4, 8, %g2
391 sub %g4, 1, %g4
392 EX(stb %g2, [%o1])
393 add %o4, %g5, %g5
394 EX(stb %o4, [%o1 + 1])
395 add %o0, 2, %o0
396 srl %g4, 1, %g4
397 add %o1, 2, %o1
398 1: cmp %g4, 0
399 be,a 2f
400 andcc %g1, 2, %g0
401 EX(ld [%o0], %o4)
402 5: srl %o4, 24, %g2
403 srl %o4, 16, %g3
404 EX(stb %g2, [%o1])
405 srl %o4, 8, %g2
406 EX(stb %g3, [%o1 + 1])
407 add %o0, 4, %o0
408 EX(stb %g2, [%o1 + 2])
409 addcc %o4, %g5, %g5
410 EX(stb %o4, [%o1 + 3])
411 addx %g5, %g0, %g5 ! I am now to
412 add %o1, 4, %o1 ! is worthy).
413 subcc %g4, 1, %g4 ! tricks
414 bne,a 5b
415 EX(ld [%o0], %o4)
416 sll %g5, 16, %g2
417 srl %g5, 16, %g5
418 srl %g2, 16, %g2
419 andcc %g1, 2, %g0
420 add %g2, %g5, %g5
421 2: be,a 3f
422 andcc %g1, 1, %g0
423 EX(lduh [%o0], %o4)
424 andcc %g1, 1, %g0
425 srl %o4, 8, %g2
426 add %o0, 2, %o0
427 EX(stb %g2, [%o1])
428 add %g5, %o4, %g5
429 EX(stb %o4, [%o1 + 1])
430 add %o1, 2, %o1
431 3: be,a 1f
432 sll %g5, 16, %o4
433 EX(ldub [%o0], %g2)
434 sll %g2, 8, %o4
435 EX(stb %g2, [%o1])
436 add %g5, %o4, %g5
437 sll %g5, 16, %o4
438 1: addcc %o4, %g5, %g5
439 srl %g5, 16, %o4
440 addx %g0, %o4, %g5
441 orcc %o5, %g0, %g0
442 be 4f
443 srl %g5, 8, %o4
444 and %g5, 0xff, %g2
445 and %o4, 0xff, %o4
446 sll %g2, 8, %g2
447 or %g2, %o4, %g5
448 4: addcc %g7, %g5, %g7
449 retl
450 addx %g0, %g7, %o0
451
452 /* We do these strange calculations for the cs
453 * we only bother with faults on loads... */
454
455 cc_fault:
456 retl
457 clr %o0
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.