xref: /linux/arch/arm64/lib/strncmp.S (revision 088e88be5a380cc4e81963a9a02815da465d144f)
1/* SPDX-License-Identifier: GPL-2.0-only */
2/*
3 * Copyright (C) 2013 ARM Ltd.
4 * Copyright (C) 2013 Linaro.
5 *
6 * This code is based on glibc cortex strings work originally authored by Linaro
7 * be found @
8 *
9 * http://bazaar.launchpad.net/~linaro-toolchain-dev/cortex-strings/trunk/
10 * files/head:/src/aarch64/
11 */
12
13#include <linux/linkage.h>
14#include <asm/assembler.h>
15
16/*
17 * compare two strings
18 *
19 * Parameters:
20 *  x0 - const string 1 pointer
21 *  x1 - const string 2 pointer
22 *  x2 - the maximal length to be compared
23 * Returns:
24 *  x0 - an integer less than, equal to, or greater than zero if s1 is found,
25 *     respectively, to be less than, to match, or be greater than s2.
26 */
27
28#define REP8_01 0x0101010101010101
29#define REP8_7f 0x7f7f7f7f7f7f7f7f
30#define REP8_80 0x8080808080808080
31
32/* Parameters and result.  */
33src1		.req	x0
34src2		.req	x1
35limit		.req	x2
36result		.req	x0
37
38/* Internal variables.  */
39data1		.req	x3
40data1w		.req	w3
41data2		.req	x4
42data2w		.req	w4
43has_nul		.req	x5
44diff		.req	x6
45syndrome	.req	x7
46tmp1		.req	x8
47tmp2		.req	x9
48tmp3		.req	x10
49zeroones	.req	x11
50pos		.req	x12
51limit_wd	.req	x13
52mask		.req	x14
53endloop		.req	x15
54
55WEAK(strncmp)
56	cbz	limit, .Lret0
57	eor	tmp1, src1, src2
58	mov	zeroones, #REP8_01
59	tst	tmp1, #7
60	b.ne	.Lmisaligned8
61	ands	tmp1, src1, #7
62	b.ne	.Lmutual_align
63	/* Calculate the number of full and partial words -1.  */
64	/*
65	* when limit is mulitply of 8, if not sub 1,
66	* the judgement of last dword will wrong.
67	*/
68	sub	limit_wd, limit, #1 /* limit != 0, so no underflow.  */
69	lsr	limit_wd, limit_wd, #3  /* Convert to Dwords.  */
70
71	/*
72	* NUL detection works on the principle that (X - 1) & (~X) & 0x80
73	* (=> (X - 1) & ~(X | 0x7f)) is non-zero iff a byte is zero, and
74	* can be done in parallel across the entire word.
75	*/
76.Lloop_aligned:
77	ldr	data1, [src1], #8
78	ldr	data2, [src2], #8
79.Lstart_realigned:
80	subs	limit_wd, limit_wd, #1
81	sub	tmp1, data1, zeroones
82	orr	tmp2, data1, #REP8_7f
83	eor	diff, data1, data2  /* Non-zero if differences found.  */
84	csinv	endloop, diff, xzr, pl  /* Last Dword or differences.*/
85	bics	has_nul, tmp1, tmp2 /* Non-zero if NUL terminator.  */
86	ccmp	endloop, #0, #0, eq
87	b.eq	.Lloop_aligned
88
89	/*Not reached the limit, must have found the end or a diff.  */
90	tbz	limit_wd, #63, .Lnot_limit
91
92	/* Limit % 8 == 0 => all bytes significant.  */
93	ands	limit, limit, #7
94	b.eq	.Lnot_limit
95
96	lsl	limit, limit, #3    /* Bits -> bytes.  */
97	mov	mask, #~0
98CPU_BE( lsr	mask, mask, limit )
99CPU_LE( lsl	mask, mask, limit )
100	bic	data1, data1, mask
101	bic	data2, data2, mask
102
103	/* Make sure that the NUL byte is marked in the syndrome.  */
104	orr	has_nul, has_nul, mask
105
106.Lnot_limit:
107	orr	syndrome, diff, has_nul
108	b	.Lcal_cmpresult
109
110.Lmutual_align:
111	/*
112	* Sources are mutually aligned, but are not currently at an
113	* alignment boundary.  Round down the addresses and then mask off
114	* the bytes that precede the start point.
115	* We also need to adjust the limit calculations, but without
116	* overflowing if the limit is near ULONG_MAX.
117	*/
118	bic	src1, src1, #7
119	bic	src2, src2, #7
120	ldr	data1, [src1], #8
121	neg	tmp3, tmp1, lsl #3  /* 64 - bits(bytes beyond align). */
122	ldr	data2, [src2], #8
123	mov	tmp2, #~0
124	sub	limit_wd, limit, #1 /* limit != 0, so no underflow.  */
125	/* Big-endian.  Early bytes are at MSB.  */
126CPU_BE( lsl	tmp2, tmp2, tmp3 )	/* Shift (tmp1 & 63).  */
127	/* Little-endian.  Early bytes are at LSB.  */
128CPU_LE( lsr	tmp2, tmp2, tmp3 )	/* Shift (tmp1 & 63).  */
129
130	and	tmp3, limit_wd, #7
131	lsr	limit_wd, limit_wd, #3
132	/* Adjust the limit. Only low 3 bits used, so overflow irrelevant.*/
133	add	limit, limit, tmp1
134	add	tmp3, tmp3, tmp1
135	orr	data1, data1, tmp2
136	orr	data2, data2, tmp2
137	add	limit_wd, limit_wd, tmp3, lsr #3
138	b	.Lstart_realigned
139
140/*when src1 offset is not equal to src2 offset...*/
141.Lmisaligned8:
142	cmp	limit, #8
143	b.lo	.Ltiny8proc /*limit < 8... */
144	/*
145	* Get the align offset length to compare per byte first.
146	* After this process, one string's address will be aligned.*/
147	and	tmp1, src1, #7
148	neg	tmp1, tmp1
149	add	tmp1, tmp1, #8
150	and	tmp2, src2, #7
151	neg	tmp2, tmp2
152	add	tmp2, tmp2, #8
153	subs	tmp3, tmp1, tmp2
154	csel	pos, tmp1, tmp2, hi /*Choose the maximum. */
155	/*
156	* Here, limit is not less than 8, so directly run .Ltinycmp
157	* without checking the limit.*/
158	sub	limit, limit, pos
159.Ltinycmp:
160	ldrb	data1w, [src1], #1
161	ldrb	data2w, [src2], #1
162	subs	pos, pos, #1
163	ccmp	data1w, #1, #0, ne  /* NZCV = 0b0000.  */
164	ccmp	data1w, data2w, #0, cs  /* NZCV = 0b0000.  */
165	b.eq	.Ltinycmp
166	cbnz	pos, 1f /*find the null or unequal...*/
167	cmp	data1w, #1
168	ccmp	data1w, data2w, #0, cs
169	b.eq	.Lstart_align /*the last bytes are equal....*/
1701:
171	sub	result, data1, data2
172	ret
173
174.Lstart_align:
175	lsr	limit_wd, limit, #3
176	cbz	limit_wd, .Lremain8
177	/*process more leading bytes to make str1 aligned...*/
178	ands	xzr, src1, #7
179	b.eq	.Lrecal_offset
180	add	src1, src1, tmp3	/*tmp3 is positive in this branch.*/
181	add	src2, src2, tmp3
182	ldr	data1, [src1], #8
183	ldr	data2, [src2], #8
184
185	sub	limit, limit, tmp3
186	lsr	limit_wd, limit, #3
187	subs	limit_wd, limit_wd, #1
188
189	sub	tmp1, data1, zeroones
190	orr	tmp2, data1, #REP8_7f
191	eor	diff, data1, data2  /* Non-zero if differences found.  */
192	csinv	endloop, diff, xzr, ne/*if limit_wd is 0,will finish the cmp*/
193	bics	has_nul, tmp1, tmp2
194	ccmp	endloop, #0, #0, eq /*has_null is ZERO: no null byte*/
195	b.ne	.Lunequal_proc
196	/*How far is the current str2 from the alignment boundary...*/
197	and	tmp3, tmp3, #7
198.Lrecal_offset:
199	neg	pos, tmp3
200.Lloopcmp_proc:
201	/*
202	* Divide the eight bytes into two parts. First,backwards the src2
203	* to an alignment boundary,load eight bytes from the SRC2 alignment
204	* boundary,then compare with the relative bytes from SRC1.
205	* If all 8 bytes are equal,then start the second part's comparison.
206	* Otherwise finish the comparison.
207	* This special handle can garantee all the accesses are in the
208	* thread/task space in avoid to overrange access.
209	*/
210	ldr	data1, [src1,pos]
211	ldr	data2, [src2,pos]
212	sub	tmp1, data1, zeroones
213	orr	tmp2, data1, #REP8_7f
214	bics	has_nul, tmp1, tmp2 /* Non-zero if NUL terminator.  */
215	eor	diff, data1, data2  /* Non-zero if differences found.  */
216	csinv	endloop, diff, xzr, eq
217	cbnz	endloop, .Lunequal_proc
218
219	/*The second part process*/
220	ldr	data1, [src1], #8
221	ldr	data2, [src2], #8
222	subs	limit_wd, limit_wd, #1
223	sub	tmp1, data1, zeroones
224	orr	tmp2, data1, #REP8_7f
225	eor	diff, data1, data2  /* Non-zero if differences found.  */
226	csinv	endloop, diff, xzr, ne/*if limit_wd is 0,will finish the cmp*/
227	bics	has_nul, tmp1, tmp2
228	ccmp	endloop, #0, #0, eq /*has_null is ZERO: no null byte*/
229	b.eq	.Lloopcmp_proc
230
231.Lunequal_proc:
232	orr	syndrome, diff, has_nul
233	cbz	syndrome, .Lremain8
234.Lcal_cmpresult:
235	/*
236	* reversed the byte-order as big-endian,then CLZ can find the most
237	* significant zero bits.
238	*/
239CPU_LE( rev	syndrome, syndrome )
240CPU_LE( rev	data1, data1 )
241CPU_LE( rev	data2, data2 )
242	/*
243	* For big-endian we cannot use the trick with the syndrome value
244	* as carry-propagation can corrupt the upper bits if the trailing
245	* bytes in the string contain 0x01.
246	* However, if there is no NUL byte in the dword, we can generate
247	* the result directly.  We can't just subtract the bytes as the
248	* MSB might be significant.
249	*/
250CPU_BE( cbnz	has_nul, 1f )
251CPU_BE( cmp	data1, data2 )
252CPU_BE( cset	result, ne )
253CPU_BE( cneg	result, result, lo )
254CPU_BE( ret )
255CPU_BE( 1: )
256	/* Re-compute the NUL-byte detection, using a byte-reversed value.*/
257CPU_BE( rev	tmp3, data1 )
258CPU_BE( sub	tmp1, tmp3, zeroones )
259CPU_BE( orr	tmp2, tmp3, #REP8_7f )
260CPU_BE( bic	has_nul, tmp1, tmp2 )
261CPU_BE( rev	has_nul, has_nul )
262CPU_BE( orr	syndrome, diff, has_nul )
263	/*
264	* The MS-non-zero bit of the syndrome marks either the first bit
265	* that is different, or the top bit of the first zero byte.
266	* Shifting left now will bring the critical information into the
267	* top bits.
268	*/
269	clz	pos, syndrome
270	lsl	data1, data1, pos
271	lsl	data2, data2, pos
272	/*
273	* But we need to zero-extend (char is unsigned) the value and then
274	* perform a signed 32-bit subtraction.
275	*/
276	lsr	data1, data1, #56
277	sub	result, data1, data2, lsr #56
278	ret
279
280.Lremain8:
281	/* Limit % 8 == 0 => all bytes significant.  */
282	ands	limit, limit, #7
283	b.eq	.Lret0
284.Ltiny8proc:
285	ldrb	data1w, [src1], #1
286	ldrb	data2w, [src2], #1
287	subs	limit, limit, #1
288
289	ccmp	data1w, #1, #0, ne  /* NZCV = 0b0000.  */
290	ccmp	data1w, data2w, #0, cs  /* NZCV = 0b0000.  */
291	b.eq	.Ltiny8proc
292	sub	result, data1, data2
293	ret
294
295.Lret0:
296	mov	result, #0
297	ret
298ENDPIPROC(strncmp)
299EXPORT_SYMBOL_NOKASAN(strncmp)
300