1/*
2 * Copyright (C) 2013 ARM Ltd.
3 * Copyright (C) 2013 Linaro.
4 *
5 * This code is based on glibc cortex strings work originally authored by Linaro
6 * and re-licensed under GPLv2 for the Linux kernel. The original code can
7 * be found @
8 *
9 * http://bazaar.launchpad.net/~linaro-toolchain-dev/cortex-strings/trunk/
10 * files/head:/src/aarch64/
11 *
12 * This program is free software; you can redistribute it and/or modify
13 * it under the terms of the GNU General Public License version 2 as
14 * published by the Free Software Foundation.
15 *
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
19 * GNU General Public License for more details.
20 *
21 * You should have received a copy of the GNU General Public License
22 * along with this program.  If not, see <http://www.gnu.org/licenses/>.
23 */
24
25#include <linux/linkage.h>
26#include <asm/assembler.h>
27
28/*
29 * determine the length of a fixed-size string
30 *
31 * Parameters:
32 *	x0 - const string pointer
33 *	x1 - maximal string length
34 * Returns:
35 *	x0 - the return length of specific string
36 */
37
38/* Arguments and results.  */
39srcin		.req	x0
40len		.req	x0
41limit		.req	x1
42
43/* Locals and temporaries.  */
44src		.req	x2
45data1		.req	x3
46data2		.req	x4
47data2a		.req	x5
48has_nul1	.req	x6
49has_nul2	.req	x7
50tmp1		.req	x8
51tmp2		.req	x9
52tmp3		.req	x10
53tmp4		.req	x11
54zeroones	.req	x12
55pos		.req	x13
56limit_wd	.req	x14
57
58#define REP8_01 0x0101010101010101
59#define REP8_7f 0x7f7f7f7f7f7f7f7f
60#define REP8_80 0x8080808080808080
61
62ENTRY(strnlen)
63	cbz	limit, .Lhit_limit
64	mov	zeroones, #REP8_01
65	bic	src, srcin, #15
66	ands	tmp1, srcin, #15
67	b.ne	.Lmisaligned
68	/* Calculate the number of full and partial words -1.  */
69	sub	limit_wd, limit, #1 /* Limit != 0, so no underflow.  */
70	lsr	limit_wd, limit_wd, #4  /* Convert to Qwords.  */
71
72	/*
73	* NUL detection works on the principle that (X - 1) & (~X) & 0x80
74	* (=> (X - 1) & ~(X | 0x7f)) is non-zero iff a byte is zero, and
75	* can be done in parallel across the entire word.
76	*/
77	/*
78	* The inner loop deals with two Dwords at a time.  This has a
79	* slightly higher start-up cost, but we should win quite quickly,
80	* especially on cores with a high number of issue slots per
81	* cycle, as we get much better parallelism out of the operations.
82	*/
83.Lloop:
84	ldp	data1, data2, [src], #16
85.Lrealigned:
86	sub	tmp1, data1, zeroones
87	orr	tmp2, data1, #REP8_7f
88	sub	tmp3, data2, zeroones
89	orr	tmp4, data2, #REP8_7f
90	bic	has_nul1, tmp1, tmp2
91	bic	has_nul2, tmp3, tmp4
92	subs	limit_wd, limit_wd, #1
93	orr	tmp1, has_nul1, has_nul2
94	ccmp	tmp1, #0, #0, pl    /* NZCV = 0000  */
95	b.eq	.Lloop
96
97	cbz	tmp1, .Lhit_limit   /* No null in final Qword.  */
98
99	/*
100	* We know there's a null in the final Qword. The easiest thing
101	* to do now is work out the length of the string and return
102	* MIN (len, limit).
103	*/
104	sub	len, src, srcin
105	cbz	has_nul1, .Lnul_in_data2
106CPU_BE( mov	data2, data1 )	/*perpare data to re-calculate the syndrome*/
107
108	sub	len, len, #8
109	mov	has_nul2, has_nul1
110.Lnul_in_data2:
111	/*
112	* For big-endian, carry propagation (if the final byte in the
113	* string is 0x01) means we cannot use has_nul directly.  The
114	* easiest way to get the correct byte is to byte-swap the data
115	* and calculate the syndrome a second time.
116	*/
117CPU_BE( rev	data2, data2 )
118CPU_BE( sub	tmp1, data2, zeroones )
119CPU_BE( orr	tmp2, data2, #REP8_7f )
120CPU_BE( bic	has_nul2, tmp1, tmp2 )
121
122	sub	len, len, #8
123	rev	has_nul2, has_nul2
124	clz	pos, has_nul2
125	add	len, len, pos, lsr #3       /* Bits to bytes.  */
126	cmp	len, limit
127	csel	len, len, limit, ls     /* Return the lower value.  */
128	ret
129
130.Lmisaligned:
131	/*
132	* Deal with a partial first word.
133	* We're doing two things in parallel here;
134	* 1) Calculate the number of words (but avoiding overflow if
135	* limit is near ULONG_MAX) - to do this we need to work out
136	* limit + tmp1 - 1 as a 65-bit value before shifting it;
137	* 2) Load and mask the initial data words - we force the bytes
138	* before the ones we are interested in to 0xff - this ensures
139	* early bytes will not hit any zero detection.
140	*/
141	ldp	data1, data2, [src], #16
142
143	sub	limit_wd, limit, #1
144	and	tmp3, limit_wd, #15
145	lsr	limit_wd, limit_wd, #4
146
147	add	tmp3, tmp3, tmp1
148	add	limit_wd, limit_wd, tmp3, lsr #4
149
150	neg	tmp4, tmp1
151	lsl	tmp4, tmp4, #3  /* Bytes beyond alignment -> bits.  */
152
153	mov	tmp2, #~0
154	/* Big-endian.  Early bytes are at MSB.  */
155CPU_BE( lsl	tmp2, tmp2, tmp4 )	/* Shift (tmp1 & 63).  */
156	/* Little-endian.  Early bytes are at LSB.  */
157CPU_LE( lsr	tmp2, tmp2, tmp4 )	/* Shift (tmp1 & 63).  */
158
159	cmp	tmp1, #8
160
161	orr	data1, data1, tmp2
162	orr	data2a, data2, tmp2
163
164	csinv	data1, data1, xzr, le
165	csel	data2, data2, data2a, le
166	b	.Lrealigned
167
168.Lhit_limit:
169	mov	len, limit
170	ret
171ENDPROC(strnlen)
172