1/*
2 *  arch/xtensa/lib/strncpy_user.S
3 *
4 *  This file is subject to the terms and conditions of the GNU General
5 *  Public License.  See the file "COPYING" in the main directory of
6 *  this archive for more details.
7 *
8 *  Returns: -EFAULT if exception before terminator, N if the entire
9 *  buffer filled, else strlen.
10 *
11 *  Copyright (C) 2002 Tensilica Inc.
12 */
13
14#include <variant/core.h>
15#include <linux/errno.h>
16
17/* Load or store instructions that may cause exceptions use the EX macro. */
18
19#define EX(insn,reg1,reg2,offset,handler)	\
209:	insn	reg1, reg2, offset;		\
21	.section __ex_table, "a";		\
22	.word	9b, handler;			\
23	.previous
24
25/*
26 * char *__strncpy_user(char *dst, const char *src, size_t len)
27 */
28
29#ifdef __XTENSA_EB__
30# define MASK0 0xff000000
31# define MASK1 0x00ff0000
32# define MASK2 0x0000ff00
33# define MASK3 0x000000ff
34#else
35# define MASK0 0x000000ff
36# define MASK1 0x0000ff00
37# define MASK2 0x00ff0000
38# define MASK3 0xff000000
39#endif
40
41# Register use
42#   a0/ return address
43#   a1/ stack pointer
44#   a2/ return value
45#   a3/ src
46#   a4/ len
47#   a5/ mask0
48#   a6/ mask1
49#   a7/ mask2
50#   a8/ mask3
51#   a9/ tmp
52#   a10/ tmp
53#   a11/ dst
54#   a12/ tmp
55
56.text
57.align	4
58.global	__strncpy_user
59.type	__strncpy_user,@function
60__strncpy_user:
61	entry	sp, 16		# minimal stack frame
62	# a2/ dst, a3/ src, a4/ len
63	mov	a11, a2		# leave dst in return value register
64	beqz	a4, .Lret	# if len is zero
65	movi	a5, MASK0	# mask for byte 0
66	movi	a6, MASK1	# mask for byte 1
67	movi	a7, MASK2	# mask for byte 2
68	movi	a8, MASK3	# mask for byte 3
69	bbsi.l	a3, 0, .Lsrc1mod2 # if only  8-bit aligned
70	bbsi.l	a3, 1, .Lsrc2mod4 # if only 16-bit aligned
71.Lsrcaligned:	# return here when src is word-aligned
72	srli	a12, a4, 2	# number of loop iterations with 4B per loop
73	movi	a9, 3
74	bnone	a11, a9, .Laligned
75	j	.Ldstunaligned
76
77.Lsrc1mod2:	# src address is odd
78	EX(l8ui, a9, a3, 0, fixup_l)	# get byte 0
79	addi	a3, a3, 1		# advance src pointer
80	EX(s8i, a9, a11, 0, fixup_s)	# store byte 0
81	beqz	a9, .Lret		# if byte 0 is zero
82	addi	a11, a11, 1		# advance dst pointer
83	addi	a4, a4, -1		# decrement len
84	beqz	a4, .Lret		# if len is zero
85	bbci.l	a3, 1, .Lsrcaligned	# if src is now word-aligned
86
87.Lsrc2mod4:	# src address is 2 mod 4
88	EX(l8ui, a9, a3, 0, fixup_l)	# get byte 0
89	/* 1-cycle interlock */
90	EX(s8i, a9, a11, 0, fixup_s)	# store byte 0
91	beqz	a9, .Lret		# if byte 0 is zero
92	addi	a11, a11, 1		# advance dst pointer
93	addi	a4, a4, -1		# decrement len
94	beqz	a4, .Lret		# if len is zero
95	EX(l8ui, a9, a3, 1, fixup_l)	# get byte 0
96	addi	a3, a3, 2		# advance src pointer
97	EX(s8i, a9, a11, 0, fixup_s)	# store byte 0
98	beqz	a9, .Lret		# if byte 0 is zero
99	addi	a11, a11, 1		# advance dst pointer
100	addi	a4, a4, -1		# decrement len
101	bnez	a4, .Lsrcaligned	# if len is nonzero
102.Lret:
103	sub	a2, a11, a2		# compute strlen
104	retw
105
106/*
107 * dst is word-aligned, src is word-aligned
108 */
109	.align	4		# 1 mod 4 alignment for LOOPNEZ
110	.byte	0		# (0 mod 4 alignment for LBEG)
111.Laligned:
112#if XCHAL_HAVE_LOOPS
113	loopnez	a12, .Loop1done
114#else
115	beqz	a12, .Loop1done
116	slli	a12, a12, 2
117	add	a12, a12, a11	# a12 = end of last 4B chunck
118#endif
119.Loop1:
120	EX(l32i, a9, a3, 0, fixup_l)	# get word from src
121	addi	a3, a3, 4		# advance src pointer
122	bnone	a9, a5, .Lz0		# if byte 0 is zero
123	bnone	a9, a6, .Lz1		# if byte 1 is zero
124	bnone	a9, a7, .Lz2		# if byte 2 is zero
125	EX(s32i, a9, a11, 0, fixup_s)	# store word to dst
126	bnone	a9, a8, .Lz3		# if byte 3 is zero
127	addi	a11, a11, 4		# advance dst pointer
128#if !XCHAL_HAVE_LOOPS
129	blt	a11, a12, .Loop1
130#endif
131
132.Loop1done:
133	bbci.l	a4, 1, .L100
134	# copy 2 bytes
135	EX(l16ui, a9, a3, 0, fixup_l)
136	addi	a3, a3, 2		# advance src pointer
137#ifdef __XTENSA_EB__
138	bnone	a9, a7, .Lz0		# if byte 2 is zero
139	bnone	a9, a8, .Lz1		# if byte 3 is zero
140#else
141	bnone	a9, a5, .Lz0		# if byte 0 is zero
142	bnone	a9, a6, .Lz1		# if byte 1 is zero
143#endif
144	EX(s16i, a9, a11, 0, fixup_s)
145	addi	a11, a11, 2		# advance dst pointer
146.L100:
147	bbci.l	a4, 0, .Lret
148	EX(l8ui, a9, a3, 0, fixup_l)
149	/* slot */
150	EX(s8i, a9, a11, 0, fixup_s)
151	beqz	a9, .Lret		# if byte is zero
152	addi	a11, a11, 1-3		# advance dst ptr 1, but also cancel
153					# the effect of adding 3 in .Lz3 code
154	/* fall thru to .Lz3 and "retw" */
155
156.Lz3:	# byte 3 is zero
157	addi	a11, a11, 3		# advance dst pointer
158	sub	a2, a11, a2		# compute strlen
159	retw
160.Lz0:	# byte 0 is zero
161#ifdef __XTENSA_EB__
162	movi	a9, 0
163#endif /* __XTENSA_EB__ */
164	EX(s8i, a9, a11, 0, fixup_s)
165	sub	a2, a11, a2		# compute strlen
166	retw
167.Lz1:	# byte 1 is zero
168#ifdef __XTENSA_EB__
169	extui   a9, a9, 16, 16
170#endif /* __XTENSA_EB__ */
171	EX(s16i, a9, a11, 0, fixup_s)
172	addi	a11, a11, 1		# advance dst pointer
173	sub	a2, a11, a2		# compute strlen
174	retw
175.Lz2:	# byte 2 is zero
176#ifdef __XTENSA_EB__
177	extui   a9, a9, 16, 16
178#endif /* __XTENSA_EB__ */
179	EX(s16i, a9, a11, 0, fixup_s)
180	movi	a9, 0
181	EX(s8i, a9, a11, 2, fixup_s)
182	addi	a11, a11, 2		# advance dst pointer
183	sub	a2, a11, a2		# compute strlen
184	retw
185
186	.align	4		# 1 mod 4 alignment for LOOPNEZ
187	.byte	0		# (0 mod 4 alignment for LBEG)
188.Ldstunaligned:
189/*
190 * for now just use byte copy loop
191 */
192#if XCHAL_HAVE_LOOPS
193	loopnez	a4, .Lunalignedend
194#else
195	beqz	a4, .Lunalignedend
196	add	a12, a11, a4		# a12 = ending address
197#endif /* XCHAL_HAVE_LOOPS */
198.Lnextbyte:
199	EX(l8ui, a9, a3, 0, fixup_l)
200	addi	a3, a3, 1
201	EX(s8i, a9, a11, 0, fixup_s)
202	beqz	a9, .Lunalignedend
203	addi	a11, a11, 1
204#if !XCHAL_HAVE_LOOPS
205	blt	a11, a12, .Lnextbyte
206#endif
207
208.Lunalignedend:
209	sub	a2, a11, a2		# compute strlen
210	retw
211
212
213	.section .fixup, "ax"
214	.align	4
215
216	/* For now, just return -EFAULT.  Future implementations might
217	 * like to clear remaining kernel space, like the fixup
218	 * implementation in memset().  Thus, we differentiate between
219	 * load/store fixups. */
220
221fixup_s:
222fixup_l:
223	movi	a2, -EFAULT
224	retw
225
226