1/* checksum.S: Sparc optimized checksum code.
2 *
3 *  Copyright(C) 1995 Linus Torvalds
4 *  Copyright(C) 1995 Miguel de Icaza
5 *  Copyright(C) 1996 David S. Miller
6 *  Copyright(C) 1997 Jakub Jelinek
7 *
8 * derived from:
9 *	Linux/Alpha checksum c-code
10 *      Linux/ix86 inline checksum assembly
11 *      RFC1071 Computing the Internet Checksum (esp. Jacobsons m68k code)
12 *	David Mosberger-Tang for optimized reference c-code
13 *	BSD4.4 portable checksum routine
14 */
15
16#include <asm/cprefix.h>
17#include <asm/errno.h>
18
19#define CSUM_BIGCHUNK(buf, offset, sum, t0, t1, t2, t3, t4, t5)	\
20	ldd	[buf + offset + 0x00], t0;			\
21	ldd	[buf + offset + 0x08], t2;			\
22	addxcc	t0, sum, sum;					\
23	addxcc	t1, sum, sum;					\
24	ldd	[buf + offset + 0x10], t4;			\
25	addxcc	t2, sum, sum;					\
26	addxcc	t3, sum, sum;					\
27	ldd	[buf + offset + 0x18], t0;			\
28	addxcc	t4, sum, sum;					\
29	addxcc	t5, sum, sum;					\
30	addxcc	t0, sum, sum;					\
31	addxcc	t1, sum, sum;
32
33#define CSUM_LASTCHUNK(buf, offset, sum, t0, t1, t2, t3)	\
34	ldd	[buf - offset - 0x08], t0;			\
35	ldd	[buf - offset - 0x00], t2;			\
36	addxcc	t0, sum, sum;					\
37	addxcc	t1, sum, sum;					\
38	addxcc	t2, sum, sum;					\
39	addxcc	t3, sum, sum;
40
41	/* Do end cruft out of band to get better cache patterns. */
42csum_partial_end_cruft:
43	be	1f				! caller asks %o1 & 0x8
44	 andcc	%o1, 4, %g0			! nope, check for word remaining
45	ldd	[%o0], %g2			! load two
46	addcc	%g2, %o2, %o2			! add first word to sum
47	addxcc	%g3, %o2, %o2			! add second word as well
48	add	%o0, 8, %o0			! advance buf ptr
49	addx	%g0, %o2, %o2			! add in final carry
50	andcc	%o1, 4, %g0			! check again for word remaining
511:	be	1f				! nope, skip this code
52	 andcc	%o1, 3, %o1			! check for trailing bytes
53	ld	[%o0], %g2			! load it
54	addcc	%g2, %o2, %o2			! add to sum
55	add	%o0, 4, %o0			! advance buf ptr
56	addx	%g0, %o2, %o2			! add in final carry
57	andcc	%o1, 3, %g0			! check again for trailing bytes
581:	be	1f				! no trailing bytes, return
59	 addcc	%o1, -1, %g0			! only one byte remains?
60	bne	2f				! at least two bytes more
61	 subcc	%o1, 2, %o1			! only two bytes more?
62	b	4f				! only one byte remains
63	 or	%g0, %g0, %o4			! clear fake hword value
642:	lduh	[%o0], %o4			! get hword
65	be	6f				! jmp if only hword remains
66	 add	%o0, 2, %o0			! advance buf ptr either way
67	sll	%o4, 16, %o4			! create upper hword
684:	ldub	[%o0], %o5			! get final byte
69	sll	%o5, 8, %o5			! put into place
70	or	%o5, %o4, %o4			! coalese with hword (if any)
716:	addcc	%o4, %o2, %o2			! add to sum
721:	retl					! get outta here
73	 addx	%g0, %o2, %o0			! add final carry into retval
74
75	/* Also do alignment out of band to get better cache patterns. */
76csum_partial_fix_alignment:
77	cmp	%o1, 6
78	bl	cpte - 0x4
79	 andcc	%o0, 0x2, %g0
80	be	1f
81	 andcc	%o0, 0x4, %g0
82	lduh	[%o0 + 0x00], %g2
83	sub	%o1, 2, %o1
84	add	%o0, 2, %o0
85	sll	%g2, 16, %g2
86	addcc	%g2, %o2, %o2
87	srl	%o2, 16, %g3
88	addx	%g0, %g3, %g2
89	sll	%o2, 16, %o2
90	sll	%g2, 16, %g3
91	srl	%o2, 16, %o2
92	andcc	%o0, 0x4, %g0
93	or	%g3, %o2, %o2
941:	be	cpa
95	 andcc	%o1, 0xffffff80, %o3
96	ld	[%o0 + 0x00], %g2
97	sub	%o1, 4, %o1
98	addcc	%g2, %o2, %o2
99	add	%o0, 4, %o0
100	addx	%g0, %o2, %o2
101	b	cpa
102	 andcc	%o1, 0xffffff80, %o3
103
104	/* The common case is to get called with a nicely aligned
105	 * buffer of size 0x20.  Follow the code path for that case.
106	 */
107	.globl	C_LABEL(csum_partial)
108C_LABEL(csum_partial):			/* %o0=buf, %o1=len, %o2=sum */
109	andcc	%o0, 0x7, %g0				! alignment problems?
110	bne	csum_partial_fix_alignment		! yep, handle it
111	 sethi	%hi(cpte - 8), %g7			! prepare table jmp ptr
112	andcc	%o1, 0xffffff80, %o3			! num loop iterations
113cpa:	be	3f					! none to do
114	 andcc	%o1, 0x70, %g1				! clears carry flag too
1155:	CSUM_BIGCHUNK(%o0, 0x00, %o2, %o4, %o5, %g2, %g3, %g4, %g5)
116	CSUM_BIGCHUNK(%o0, 0x20, %o2, %o4, %o5, %g2, %g3, %g4, %g5)
117	CSUM_BIGCHUNK(%o0, 0x40, %o2, %o4, %o5, %g2, %g3, %g4, %g5)
118	CSUM_BIGCHUNK(%o0, 0x60, %o2, %o4, %o5, %g2, %g3, %g4, %g5)
119	addx	%g0, %o2, %o2				! sink in final carry
120	subcc	%o3, 128, %o3				! detract from loop iters
121	bne	5b					! more to do
122	 add	%o0, 128, %o0				! advance buf ptr
123	andcc	%o1, 0x70, %g1				! clears carry flag too
1243:	be	cpte					! nope
125	 andcc	%o1, 0xf, %g0				! anything left at all?
126	srl	%g1, 1, %o4				! compute offset
127	sub	%g7, %g1, %g7				! adjust jmp ptr
128	sub	%g7, %o4, %g7				! final jmp ptr adjust
129	jmp	%g7 + %lo(cpte - 8)			! enter the table
130	 add	%o0, %g1, %o0				! advance buf ptr
131cptbl:	CSUM_LASTCHUNK(%o0, 0x68, %o2, %g2, %g3, %g4, %g5)
132	CSUM_LASTCHUNK(%o0, 0x58, %o2, %g2, %g3, %g4, %g5)
133	CSUM_LASTCHUNK(%o0, 0x48, %o2, %g2, %g3, %g4, %g5)
134	CSUM_LASTCHUNK(%o0, 0x38, %o2, %g2, %g3, %g4, %g5)
135	CSUM_LASTCHUNK(%o0, 0x28, %o2, %g2, %g3, %g4, %g5)
136	CSUM_LASTCHUNK(%o0, 0x18, %o2, %g2, %g3, %g4, %g5)
137	CSUM_LASTCHUNK(%o0, 0x08, %o2, %g2, %g3, %g4, %g5)
138	addx	%g0, %o2, %o2				! fetch final carry
139	andcc	%o1, 0xf, %g0				! anything left at all?
140cpte:	bne	csum_partial_end_cruft			! yep, handle it
141	 andcc	%o1, 8, %g0				! check how much
142cpout:	retl						! get outta here
143	 mov	%o2, %o0				! return computed csum
144
145	.globl C_LABEL(__csum_partial_copy_start), C_LABEL(__csum_partial_copy_end)
146C_LABEL(__csum_partial_copy_start):
147
148/* Work around cpp -rob */
149#define ALLOC #alloc
150#define EXECINSTR #execinstr
151#define EX(x,y,a,b)				\
15298:     x,y;                                    \
153        .section .fixup,ALLOC,EXECINSTR;	\
154        .align  4;                              \
15599:     ba 30f;                                 \
156         a, b, %o3;                             \
157        .section __ex_table,ALLOC;		\
158        .align  4;                              \
159        .word   98b, 99b;                       \
160        .text;                                  \
161        .align  4
162
163#define EX2(x,y)				\
16498:     x,y;                                    \
165        .section __ex_table,ALLOC;		\
166        .align  4;                              \
167        .word   98b, 30f;                       \
168        .text;                                  \
169        .align  4
170
171#define EX3(x,y)				\
17298:     x,y;                                    \
173        .section __ex_table,ALLOC;		\
174        .align  4;                              \
175        .word   98b, 96f;                       \
176        .text;                                  \
177        .align  4
178
179#define EXT(start,end,handler)			\
180        .section __ex_table,ALLOC;		\
181        .align  4;                              \
182        .word   start, 0, end, handler;         \
183        .text;                                  \
184        .align  4
185
186	/* This aligned version executes typically in 8.5 superscalar cycles, this
187	 * is the best I can do.  I say 8.5 because the final add will pair with
188	 * the next ldd in the main unrolled loop.  Thus the pipe is always full.
189	 * If you change these macros (including order of instructions),
190	 * please check the fixup code below as well.
191	 */
192#define CSUMCOPY_BIGCHUNK_ALIGNED(src, dst, sum, off, t0, t1, t2, t3, t4, t5, t6, t7)	\
193	ldd	[src + off + 0x00], t0;							\
194	ldd	[src + off + 0x08], t2;							\
195	addxcc	t0, sum, sum;								\
196	ldd	[src + off + 0x10], t4;							\
197	addxcc	t1, sum, sum;								\
198	ldd	[src + off + 0x18], t6;							\
199	addxcc	t2, sum, sum;								\
200	std	t0, [dst + off + 0x00];							\
201	addxcc	t3, sum, sum;								\
202	std	t2, [dst + off + 0x08];							\
203	addxcc	t4, sum, sum;								\
204	std	t4, [dst + off + 0x10];							\
205	addxcc	t5, sum, sum;								\
206	std	t6, [dst + off + 0x18];							\
207	addxcc	t6, sum, sum;								\
208	addxcc	t7, sum, sum;
209
210	/* 12 superscalar cycles seems to be the limit for this case,
211	 * because of this we thus do all the ldd's together to get
212	 * Viking MXCC into streaming mode.  Ho hum...
213	 */
214#define CSUMCOPY_BIGCHUNK(src, dst, sum, off, t0, t1, t2, t3, t4, t5, t6, t7)	\
215	ldd	[src + off + 0x00], t0;						\
216	ldd	[src + off + 0x08], t2;						\
217	ldd	[src + off + 0x10], t4;						\
218	ldd	[src + off + 0x18], t6;						\
219	st	t0, [dst + off + 0x00];						\
220	addxcc	t0, sum, sum;							\
221	st	t1, [dst + off + 0x04];						\
222	addxcc	t1, sum, sum;							\
223	st	t2, [dst + off + 0x08];						\
224	addxcc	t2, sum, sum;							\
225	st	t3, [dst + off + 0x0c];						\
226	addxcc	t3, sum, sum;							\
227	st	t4, [dst + off + 0x10];						\
228	addxcc	t4, sum, sum;							\
229	st	t5, [dst + off + 0x14];						\
230	addxcc	t5, sum, sum;							\
231	st	t6, [dst + off + 0x18];						\
232	addxcc	t6, sum, sum;							\
233	st	t7, [dst + off + 0x1c];						\
234	addxcc	t7, sum, sum;
235
236	/* Yuck, 6 superscalar cycles... */
237#define CSUMCOPY_LASTCHUNK(src, dst, sum, off, t0, t1, t2, t3)	\
238	ldd	[src - off - 0x08], t0;				\
239	ldd	[src - off - 0x00], t2;				\
240	addxcc	t0, sum, sum;					\
241	st	t0, [dst - off - 0x08];				\
242	addxcc	t1, sum, sum;					\
243	st	t1, [dst - off - 0x04];				\
244	addxcc	t2, sum, sum;					\
245	st	t2, [dst - off - 0x00];				\
246	addxcc	t3, sum, sum;					\
247	st	t3, [dst - off + 0x04];
248
249	/* Handle the end cruft code out of band for better cache patterns. */
250cc_end_cruft:
251	be	1f
252	 andcc	%o3, 4, %g0
253	EX(ldd	[%o0 + 0x00], %g2, and %o3, 0xf)
254	add	%o1, 8, %o1
255	addcc	%g2, %g7, %g7
256	add	%o0, 8, %o0
257	addxcc	%g3, %g7, %g7
258	EX2(st	%g2, [%o1 - 0x08])
259	addx	%g0, %g7, %g7
260	andcc	%o3, 4, %g0
261	EX2(st	%g3, [%o1 - 0x04])
2621:	be	1f
263	 andcc	%o3, 3, %o3
264	EX(ld	[%o0 + 0x00], %g2, add %o3, 4)
265	add	%o1, 4, %o1
266	addcc	%g2, %g7, %g7
267	EX2(st	%g2, [%o1 - 0x04])
268	addx	%g0, %g7, %g7
269	andcc	%o3, 3, %g0
270	add	%o0, 4, %o0
2711:	be	1f
272	 addcc	%o3, -1, %g0
273	bne	2f
274	 subcc	%o3, 2, %o3
275	b	4f
276	 or	%g0, %g0, %o4
2772:	EX(lduh	[%o0 + 0x00], %o4, add %o3, 2)
278	add	%o0, 2, %o0
279	EX2(sth	%o4, [%o1 + 0x00])
280	be	6f
281	 add	%o1, 2, %o1
282	sll	%o4, 16, %o4
2834:	EX(ldub	[%o0 + 0x00], %o5, add %g0, 1)
284	EX2(stb	%o5, [%o1 + 0x00])
285	sll	%o5, 8, %o5
286	or	%o5, %o4, %o4
2876:	addcc	%o4, %g7, %g7
2881:	retl
289	 addx	%g0, %g7, %o0
290
291	/* Also, handle the alignment code out of band. */
292cc_dword_align:
293	cmp	%g1, 6
294	bl,a	ccte
295	 andcc	%g1, 0xf, %o3
296	andcc	%o0, 0x1, %g0
297	bne	ccslow
298	 andcc	%o0, 0x2, %g0
299	be	1f
300	 andcc	%o0, 0x4, %g0
301	EX(lduh	[%o0 + 0x00], %g4, add %g1, 0)
302	sub	%g1, 2, %g1
303	EX2(sth	%g4, [%o1 + 0x00])
304	add	%o0, 2, %o0
305	sll	%g4, 16, %g4
306	addcc	%g4, %g7, %g7
307	add	%o1, 2, %o1
308	srl	%g7, 16, %g3
309	addx	%g0, %g3, %g4
310	sll	%g7, 16, %g7
311	sll	%g4, 16, %g3
312	srl	%g7, 16, %g7
313	andcc	%o0, 0x4, %g0
314	or	%g3, %g7, %g7
3151:	be	3f
316	 andcc	%g1, 0xffffff80, %g0
317	EX(ld	[%o0 + 0x00], %g4, add %g1, 0)
318	sub	%g1, 4, %g1
319	EX2(st	%g4, [%o1 + 0x00])
320	add	%o0, 4, %o0
321	addcc	%g4, %g7, %g7
322	add	%o1, 4, %o1
323	addx	%g0, %g7, %g7
324	b	3f
325	 andcc	%g1, 0xffffff80, %g0
326
327	/* Sun, you just can't beat me, you just can't.  Stop trying,
328	 * give up.  I'm serious, I am going to kick the living shit
329	 * out of you, game over, lights out.
330	 */
331	.align	8
332	.globl	C_LABEL(__csum_partial_copy_sparc_generic)
333C_LABEL(__csum_partial_copy_sparc_generic):
334					/* %o0=src, %o1=dest, %g1=len, %g7=sum */
335	xor	%o0, %o1, %o4		! get changing bits
336	andcc	%o4, 3, %g0		! check for mismatched alignment
337	bne	ccslow			! better this than unaligned/fixups
338	 andcc	%o0, 7, %g0		! need to align things?
339	bne	cc_dword_align		! yes, we check for short lengths there
340	 andcc	%g1, 0xffffff80, %g0	! can we use unrolled loop?
3413:	be	3f			! nope, less than one loop remains
342	 andcc	%o1, 4, %g0		! dest aligned on 4 or 8 byte boundry?
343	be	ccdbl + 4		! 8 byte aligned, kick ass
3445:	CSUMCOPY_BIGCHUNK(%o0,%o1,%g7,0x00,%o4,%o5,%g2,%g3,%g4,%g5,%o2,%o3)
345	CSUMCOPY_BIGCHUNK(%o0,%o1,%g7,0x20,%o4,%o5,%g2,%g3,%g4,%g5,%o2,%o3)
346	CSUMCOPY_BIGCHUNK(%o0,%o1,%g7,0x40,%o4,%o5,%g2,%g3,%g4,%g5,%o2,%o3)
347	CSUMCOPY_BIGCHUNK(%o0,%o1,%g7,0x60,%o4,%o5,%g2,%g3,%g4,%g5,%o2,%o3)
34810:	EXT(5b, 10b, 20f)		! note for exception handling
349	sub	%g1, 128, %g1		! detract from length
350	addx	%g0, %g7, %g7		! add in last carry bit
351	andcc	%g1, 0xffffff80, %g0	! more to csum?
352	add	%o0, 128, %o0		! advance src ptr
353	bne	5b			! we did not go negative, continue looping
354	 add	%o1, 128, %o1		! advance dest ptr
3553:	andcc	%g1, 0x70, %o2		! can use table?
356ccmerge:be	ccte			! nope, go and check for end cruft
357	 andcc	%g1, 0xf, %o3		! get low bits of length (clears carry btw)
358	srl	%o2, 1, %o4		! begin negative offset computation
359	sethi	%hi(12f), %o5		! set up table ptr end
360	add	%o0, %o2, %o0		! advance src ptr
361	sub	%o5, %o4, %o5		! continue table calculation
362	sll	%o2, 1, %g2		! constant multiplies are fun...
363	sub	%o5, %g2, %o5		! some more adjustments
364	jmp	%o5 + %lo(12f)		! jump into it, duff style, wheee...
365	 add	%o1, %o2, %o1		! advance dest ptr (carry is clear btw)
366cctbl:	CSUMCOPY_LASTCHUNK(%o0,%o1,%g7,0x68,%g2,%g3,%g4,%g5)
367	CSUMCOPY_LASTCHUNK(%o0,%o1,%g7,0x58,%g2,%g3,%g4,%g5)
368	CSUMCOPY_LASTCHUNK(%o0,%o1,%g7,0x48,%g2,%g3,%g4,%g5)
369	CSUMCOPY_LASTCHUNK(%o0,%o1,%g7,0x38,%g2,%g3,%g4,%g5)
370	CSUMCOPY_LASTCHUNK(%o0,%o1,%g7,0x28,%g2,%g3,%g4,%g5)
371	CSUMCOPY_LASTCHUNK(%o0,%o1,%g7,0x18,%g2,%g3,%g4,%g5)
372	CSUMCOPY_LASTCHUNK(%o0,%o1,%g7,0x08,%g2,%g3,%g4,%g5)
37312:	EXT(cctbl, 12b, 22f)		! note for exception table handling
374	addx	%g0, %g7, %g7
375	andcc	%o3, 0xf, %g0		! check for low bits set
376ccte:	bne	cc_end_cruft		! something left, handle it out of band
377	 andcc	%o3, 8, %g0		! begin checks for that code
378	retl				! return
379	 mov	%g7, %o0		! give em the computed checksum
380ccdbl:	CSUMCOPY_BIGCHUNK_ALIGNED(%o0,%o1,%g7,0x00,%o4,%o5,%g2,%g3,%g4,%g5,%o2,%o3)
381	CSUMCOPY_BIGCHUNK_ALIGNED(%o0,%o1,%g7,0x20,%o4,%o5,%g2,%g3,%g4,%g5,%o2,%o3)
382	CSUMCOPY_BIGCHUNK_ALIGNED(%o0,%o1,%g7,0x40,%o4,%o5,%g2,%g3,%g4,%g5,%o2,%o3)
383	CSUMCOPY_BIGCHUNK_ALIGNED(%o0,%o1,%g7,0x60,%o4,%o5,%g2,%g3,%g4,%g5,%o2,%o3)
38411:	EXT(ccdbl, 11b, 21f)		! note for exception table handling
385	sub	%g1, 128, %g1		! detract from length
386	addx	%g0, %g7, %g7		! add in last carry bit
387	andcc	%g1, 0xffffff80, %g0	! more to csum?
388	add	%o0, 128, %o0		! advance src ptr
389	bne	ccdbl			! we did not go negative, continue looping
390	 add	%o1, 128, %o1		! advance dest ptr
391	b	ccmerge			! finish it off, above
392	 andcc	%g1, 0x70, %o2		! can use table? (clears carry btw)
393
394ccslow:	cmp	%g1, 0
395	mov	0, %g5
396	bleu	4f
397	 andcc	%o0, 1, %o5
398	be,a	1f
399	 srl	%g1, 1, %g4
400	sub	%g1, 1, %g1
401	EX(ldub	[%o0], %g5, add %g1, 1)
402	add	%o0, 1, %o0
403	EX2(stb	%g5, [%o1])
404	srl	%g1, 1, %g4
405	add	%o1, 1, %o1
4061:	cmp	%g4, 0
407	be,a	3f
408	 andcc	%g1, 1, %g0
409	andcc	%o0, 2, %g0
410	be,a	1f
411	 srl	%g4, 1, %g4
412	EX(lduh	[%o0], %o4, add %g1, 0)
413	sub	%g1, 2, %g1
414	srl	%o4, 8, %g2
415	sub	%g4, 1, %g4
416	EX2(stb	%g2, [%o1])
417	add	%o4, %g5, %g5
418	EX2(stb	%o4, [%o1 + 1])
419	add	%o0, 2, %o0
420	srl	%g4, 1, %g4
421	add	%o1, 2, %o1
4221:	cmp	%g4, 0
423	be,a	2f
424	 andcc	%g1, 2, %g0
425	EX3(ld	[%o0], %o4)
4265:	srl	%o4, 24, %g2
427	srl	%o4, 16, %g3
428	EX2(stb	%g2, [%o1])
429	srl	%o4, 8, %g2
430	EX2(stb	%g3, [%o1 + 1])
431	add	%o0, 4, %o0
432	EX2(stb	%g2, [%o1 + 2])
433	addcc	%o4, %g5, %g5
434	EX2(stb	%o4, [%o1 + 3])
435	addx	%g5, %g0, %g5	! I am now to lazy to optimize this (question it
436	add	%o1, 4, %o1	! is worthy). Maybe some day - with the sll/srl
437	subcc	%g4, 1, %g4	! tricks
438	bne,a	5b
439	 EX3(ld	[%o0], %o4)
440	sll	%g5, 16, %g2
441	srl	%g5, 16, %g5
442	srl	%g2, 16, %g2
443	andcc	%g1, 2, %g0
444	add	%g2, %g5, %g5
4452:	be,a	3f
446	 andcc	%g1, 1, %g0
447	EX(lduh	[%o0], %o4, and %g1, 3)
448	andcc	%g1, 1, %g0
449	srl	%o4, 8, %g2
450	add	%o0, 2, %o0
451	EX2(stb	%g2, [%o1])
452	add	%g5, %o4, %g5
453	EX2(stb	%o4, [%o1 + 1])
454	add	%o1, 2, %o1
4553:	be,a	1f
456	 sll	%g5, 16, %o4
457	EX(ldub	[%o0], %g2, add %g0, 1)
458	sll	%g2, 8, %o4
459	EX2(stb	%g2, [%o1])
460	add	%g5, %o4, %g5
461	sll	%g5, 16, %o4
4621:	addcc	%o4, %g5, %g5
463	srl	%g5, 16, %o4
464	addx	%g0, %o4, %g5
465	orcc	%o5, %g0, %g0
466	be	4f
467	 srl	%g5, 8, %o4
468	and	%g5, 0xff, %g2
469	and	%o4, 0xff, %o4
470	sll	%g2, 8, %g2
471	or	%g2, %o4, %g5
4724:	addcc	%g7, %g5, %g7
473	retl
474	 addx	%g0, %g7, %o0
475C_LABEL(__csum_partial_copy_end):
476
477/* We do these strange calculations for the csum_*_from_user case only, ie.
478 * we only bother with faults on loads... */
479
480/* o2 = ((g2%20)&3)*8
481 * o3 = g1 - (g2/20)*32 - o2 */
48220:
483	cmp	%g2, 20
484	blu,a	1f
485	 and	%g2, 3, %o2
486	sub	%g1, 32, %g1
487	b	20b
488	 sub	%g2, 20, %g2
4891:
490	sll	%o2, 3, %o2
491	b	31f
492	 sub	%g1, %o2, %o3
493
494/* o2 = (!(g2 & 15) ? 0 : (((g2 & 15) + 1) & ~1)*8)
495 * o3 = g1 - (g2/16)*32 - o2 */
49621:
497	andcc	%g2, 15, %o3
498	srl	%g2, 4, %g2
499	be,a	1f
500	 clr	%o2
501	add	%o3, 1, %o3
502	and	%o3, 14, %o3
503	sll	%o3, 3, %o2
5041:
505	sll	%g2, 5, %g2
506	sub	%g1, %g2, %o3
507	b	31f
508	 sub	%o3, %o2, %o3
509
510/* o0 += (g2/10)*16 - 0x70
511 * 01 += (g2/10)*16 - 0x70
512 * o2 = (g2 % 10) ? 8 : 0
513 * o3 += 0x70 - (g2/10)*16 - o2 */
51422:
515	cmp	%g2, 10
516	blu,a	1f
517	 sub	%o0, 0x70, %o0
518	add	%o0, 16, %o0
519	add	%o1, 16, %o1
520	sub	%o3, 16, %o3
521	b	22b
522	 sub	%g2, 10, %g2
5231:
524	sub	%o1, 0x70, %o1
525	add	%o3, 0x70, %o3
526	clr	%o2
527	tst	%g2
528	bne,a	1f
529	 mov	8, %o2
5301:
531	b	31f
532	 sub	%o3, %o2, %o3
53396:
534	and	%g1, 3, %g1
535	sll	%g4, 2, %g4
536	add	%g1, %g4, %o3
53730:
538/* %o1 is dst
539 * %o3 is # bytes to zero out
540 * %o4 is faulting address
541 * %o5 is %pc where fault occurred */
542	clr	%o2
54331:
544/* %o0 is src
545 * %o1 is dst
546 * %o2 is # of bytes to copy from src to dst
547 * %o3 is # bytes to zero out
548 * %o4 is faulting address
549 * %o5 is %pc where fault occurred */
550	save	%sp, -104, %sp
551        mov     %i5, %o0
552        mov     %i7, %o1
553        mov	%i4, %o2
554        call    C_LABEL(lookup_fault)
555	 mov	%g7, %i4
556	cmp	%o0, 2
557	bne	1f
558	 add	%g0, -EFAULT, %i5
559	tst	%i2
560	be	2f
561	 mov	%i0, %o1
562	mov	%i1, %o0
5635:
564	call	C_LABEL(__memcpy)
565	 mov	%i2, %o2
566	tst	%o0
567	bne,a	2f
568	 add	%i3, %i2, %i3
569	add	%i1, %i2, %i1
5702:
571	mov	%i1, %o0
5726:
573	call	C_LABEL(__bzero)
574	 mov	%i3, %o1
5751:
576	ld	[%sp + 168], %o2		! struct_ptr of parent
577	st	%i5, [%o2]
578	ret
579	 restore
580
581        .section __ex_table,#alloc
582        .align 4
583        .word 5b,2
584	.word 6b,2
585