1/* SPDX-License-Identifier: GPL-2.0 */
2/*
3 * arch/alpha/lib/ev67-strncat.S
4 * 21264 version contributed by Rick Gorton <rick.gorton@api-networks.com>
5 *
6 * Append no more than COUNT characters from the null-terminated string SRC
7 * to the null-terminated string DST.  Always null-terminate the new DST.
8 *
9 * This differs slightly from the semantics in libc in that we never write
10 * past count, whereas libc may write to count+1.  This follows the generic
11 * implementation in lib/string.c and is, IMHO, more sensible.
12 *
13 * Much of the information about 21264 scheduling/coding comes from:
14 *	Compiler Writer's Guide for the Alpha 21264
15 *	abbreviated as 'CWG' in other comments here
16 *	ftp.digital.com/pub/Digital/info/semiconductor/literature/dsc-library.html
17 * Scheduling notation:
18 *	E	- either cluster
19 *	U	- upper subcluster; U0 - subcluster U0; U1 - subcluster U1
20 *	L	- lower subcluster; L0 - subcluster L0; L1 - subcluster L1
21 * Try not to change the actual algorithm if possible for consistency.
22 */
23
24#include <linux/export.h>
25	.text
26
27	.align 4
28	.globl strncat
29	.ent strncat
30strncat:
31	.frame $30, 0, $26
32	.prologue 0
33
34	mov	$16, $0		# set up return value
35	beq	$18, $zerocount	# U :
36	/* Find the end of the string.  */
37	ldq_u   $1, 0($16)	# L : load first quadword ($16 may be misaligned)
38	lda     $2, -1($31)	# E :
39
40	insqh   $2, $0, $2	# U :
41	andnot  $16, 7, $16	# E :
42	nop			# E :
43	or      $2, $1, $1	# E :
44
45	nop			# E :
46	nop			# E :
47	cmpbge  $31, $1, $2	# E : bits set iff byte == 0
48	bne     $2, $found	# U :
49
50$loop:	ldq     $1, 8($16)	# L :
51	addq    $16, 8, $16	# E :
52	cmpbge  $31, $1, $2	# E :
53	beq     $2, $loop	# U :
54
55$found:	cttz	$2, $3		# U0 :
56	addq	$16, $3, $16	# E :
57	nop			# E :
58	bsr	$23, __stxncpy	# L0 :/* Now do the append.  */
59
60	/* Worry about the null termination.  */
61
62	zapnot	$1, $27, $2	# U : was last byte a null?
63	cmplt	$27, $24, $5	# E : did we fill the buffer completely?
64	bne	$2, 0f		# U :
65	ret			# L0 :
66
670:	or	$5, $18, $2	# E :
68	nop
69	bne	$2, 2f		# U :
70	and	$24, 0x80, $3	# E : no zero next byte
71
72	nop			# E :
73	bne	$3, 1f		# U :
74	/* Here there are bytes left in the current word.  Clear one.  */
75	addq	$24, $24, $24	# E : end-of-count bit <<= 1
76	nop			# E :
77
782:	zap	$1, $24, $1	# U :
79	nop			# E :
80	stq_u	$1, 0($16)	# L :
81	ret			# L0 :
82
831:	/* Here we must clear the first byte of the next DST word */
84	stb	$31, 8($16)	# L :
85	nop			# E :
86	nop			# E :
87	ret			# L0 :
88
89$zerocount:
90	nop			# E :
91	nop			# E :
92	nop			# E :
93	ret			# L0 :
94
95	.end strncat
96	EXPORT_SYMBOL(strncat)
97