1/*
2 *  arch/xtensa/lib/usercopy.S
3 *
4 *  Copy to/from user space (derived from arch/xtensa/lib/hal/memcopy.S)
5 *
6 *  DO NOT COMBINE this function with <arch/xtensa/lib/hal/memcopy.S>.
7 *  It needs to remain separate and distinct.  The hal files are part
8 *  of the Xtensa link-time HAL, and those files may differ per
9 *  processor configuration.  Patching the kernel for another
10 *  processor configuration includes replacing the hal files, and we
11 *  could lose the special functionality for accessing user-space
12 *  memory during such a patch.  We sacrifice a little code space here
13 *  in favor to simplify code maintenance.
14 *
15 *  This file is subject to the terms and conditions of the GNU General
16 *  Public License.  See the file "COPYING" in the main directory of
17 *  this archive for more details.
18 *
19 *  Copyright (C) 2002 Tensilica Inc.
20 */
21
22
23/*
24 * size_t __xtensa_copy_user (void *dst, const void *src, size_t len);
25 *
26 * The returned value is the number of bytes not copied.  Implies zero
27 * is success.
28 *
29 * The general case algorithm is as follows:
30 *   If the destination and source are both aligned,
31 *     do 16B chunks with a loop, and then finish up with
32 *     8B, 4B, 2B, and 1B copies conditional on the length.
33 *   If destination is aligned and source unaligned,
34 *     do the same, but use SRC to align the source data.
35 *   If destination is unaligned, align it by conditionally
36 *     copying 1B and 2B and then retest.
37 *   This code tries to use fall-through braches for the common
38 *     case of aligned destinations (except for the branches to
39 *     the alignment label).
40 *
41 * Register use:
42 *	a0/ return address
43 *	a1/ stack pointer
44 *	a2/ return value
45 *	a3/ src
46 *	a4/ length
47 *	a5/ dst
48 *	a6/ tmp
49 *	a7/ tmp
50 *	a8/ tmp
51 *	a9/ tmp
52 *	a10/ tmp
53 *	a11/ original length
54 */
55
56#include <linux/linkage.h>
57#include <asm/asmmacro.h>
58#include <asm/core.h>
59
60	.text
61ENTRY(__xtensa_copy_user)
62
63#if !XCHAL_HAVE_LOOPS && defined(__XTENSA_CALL0_ABI__)
64#define STACK_SIZE 4
65#else
66#define STACK_SIZE 0
67#endif
68	abi_entry(STACK_SIZE)
69	# a2/ dst, a3/ src, a4/ len
70	mov	a5, a2		# copy dst so that a2 is return value
71	mov	a11, a4		# preserve original len for error case
72.Lcommon:
73	bbsi.l	a2, 0, .Ldst1mod2 # if dst is 1 mod 2
74	bbsi.l	a2, 1, .Ldst2mod4 # if dst is 2 mod 4
75.Ldstaligned:	# return here from .Ldstunaligned when dst is aligned
76	srli	a7, a4, 4	# number of loop iterations with 16B
77				# per iteration
78	movi	a8, 3		  # if source is also aligned,
79	bnone	a3, a8, .Laligned # then use word copy
80	__ssa8	a3		# set shift amount from byte offset
81	bnez	a4, .Lsrcunaligned
82	movi	a2, 0		# return success for len==0
83	abi_ret(STACK_SIZE)
84
85/*
86 * Destination is unaligned
87 */
88
89.Ldst1mod2:	# dst is only byte aligned
90	bltui	a4, 7, .Lbytecopy	# do short copies byte by byte
91
92	# copy 1 byte
93EX(10f)	l8ui	a6, a3, 0
94	addi	a3, a3,  1
95EX(10f)	s8i	a6, a5,  0
96	addi	a5, a5,  1
97	addi	a4, a4, -1
98	bbci.l	a5, 1, .Ldstaligned	# if dst is now aligned, then
99					# return to main algorithm
100.Ldst2mod4:	# dst 16-bit aligned
101	# copy 2 bytes
102	bltui	a4, 6, .Lbytecopy	# do short copies byte by byte
103EX(10f)	l8ui	a6, a3, 0
104EX(10f)	l8ui	a7, a3, 1
105	addi	a3, a3,  2
106EX(10f)	s8i	a6, a5,  0
107EX(10f)	s8i	a7, a5,  1
108	addi	a5, a5,  2
109	addi	a4, a4, -2
110	j	.Ldstaligned	# dst is now aligned, return to main algorithm
111
112/*
113 * Byte by byte copy
114 */
115	.align	4
116	.byte	0		# 1 mod 4 alignment for LOOPNEZ
117				# (0 mod 4 alignment for LBEG)
118.Lbytecopy:
119#if XCHAL_HAVE_LOOPS
120	loopnez	a4, .Lbytecopydone
121#else /* !XCHAL_HAVE_LOOPS */
122	beqz	a4, .Lbytecopydone
123	add	a7, a3, a4	# a7 = end address for source
124#endif /* !XCHAL_HAVE_LOOPS */
125.Lnextbyte:
126EX(10f)	l8ui	a6, a3, 0
127	addi	a3, a3, 1
128EX(10f)	s8i	a6, a5, 0
129	addi	a5, a5, 1
130#if !XCHAL_HAVE_LOOPS
131	blt	a3, a7, .Lnextbyte
132#endif /* !XCHAL_HAVE_LOOPS */
133.Lbytecopydone:
134	movi	a2, 0		# return success for len bytes copied
135	abi_ret(STACK_SIZE)
136
137/*
138 * Destination and source are word-aligned.
139 */
140	# copy 16 bytes per iteration for word-aligned dst and word-aligned src
141	.align	4		# 1 mod 4 alignment for LOOPNEZ
142	.byte	0		# (0 mod 4 alignment for LBEG)
143.Laligned:
144#if XCHAL_HAVE_LOOPS
145	loopnez	a7, .Loop1done
146#else /* !XCHAL_HAVE_LOOPS */
147	beqz	a7, .Loop1done
148	slli	a8, a7, 4
149	add	a8, a8, a3	# a8 = end of last 16B source chunk
150#endif /* !XCHAL_HAVE_LOOPS */
151.Loop1:
152EX(10f)	l32i	a6, a3,  0
153EX(10f)	l32i	a7, a3,  4
154EX(10f)	s32i	a6, a5,  0
155EX(10f)	l32i	a6, a3,  8
156EX(10f)	s32i	a7, a5,  4
157EX(10f)	l32i	a7, a3, 12
158EX(10f)	s32i	a6, a5,  8
159	addi	a3, a3, 16
160EX(10f)	s32i	a7, a5, 12
161	addi	a5, a5, 16
162#if !XCHAL_HAVE_LOOPS
163	blt	a3, a8, .Loop1
164#endif /* !XCHAL_HAVE_LOOPS */
165.Loop1done:
166	bbci.l	a4, 3, .L2
167	# copy 8 bytes
168EX(10f)	l32i	a6, a3,  0
169EX(10f)	l32i	a7, a3,  4
170	addi	a3, a3,  8
171EX(10f)	s32i	a6, a5,  0
172EX(10f)	s32i	a7, a5,  4
173	addi	a5, a5,  8
174.L2:
175	bbci.l	a4, 2, .L3
176	# copy 4 bytes
177EX(10f)	l32i	a6, a3,  0
178	addi	a3, a3,  4
179EX(10f)	s32i	a6, a5,  0
180	addi	a5, a5,  4
181.L3:
182	bbci.l	a4, 1, .L4
183	# copy 2 bytes
184EX(10f)	l16ui	a6, a3,  0
185	addi	a3, a3,  2
186EX(10f)	s16i	a6, a5,  0
187	addi	a5, a5,  2
188.L4:
189	bbci.l	a4, 0, .L5
190	# copy 1 byte
191EX(10f)	l8ui	a6, a3,  0
192EX(10f)	s8i	a6, a5,  0
193.L5:
194	movi	a2, 0		# return success for len bytes copied
195	abi_ret(STACK_SIZE)
196
197/*
198 * Destination is aligned, Source is unaligned
199 */
200
201	.align	4
202	.byte	0		# 1 mod 4 alignement for LOOPNEZ
203				# (0 mod 4 alignment for LBEG)
204.Lsrcunaligned:
205	# copy 16 bytes per iteration for word-aligned dst and unaligned src
206	and	a10, a3, a8	# save unalignment offset for below
207	sub	a3, a3, a10	# align a3 (to avoid sim warnings only; not needed for hardware)
208EX(10f)	l32i	a6, a3, 0	# load first word
209#if XCHAL_HAVE_LOOPS
210	loopnez	a7, .Loop2done
211#else /* !XCHAL_HAVE_LOOPS */
212	beqz	a7, .Loop2done
213#if defined(__XTENSA_CALL0_ABI__)
214	s32i	a10, a1, 0
215	slli	a10, a7, 4
216	add	a10, a10, a3	# a10 = end of last 16B source chunk
217#else
218	slli	a12, a7, 4
219	add	a12, a12, a3	# a12 = end of last 16B source chunk
220#endif
221#endif /* !XCHAL_HAVE_LOOPS */
222.Loop2:
223EX(10f)	l32i	a7, a3,  4
224EX(10f)	l32i	a8, a3,  8
225	__src_b	a6, a6, a7
226EX(10f)	s32i	a6, a5,  0
227EX(10f)	l32i	a9, a3, 12
228	__src_b	a7, a7, a8
229EX(10f)	s32i	a7, a5,  4
230EX(10f)	l32i	a6, a3, 16
231	__src_b	a8, a8, a9
232EX(10f)	s32i	a8, a5,  8
233	addi	a3, a3, 16
234	__src_b	a9, a9, a6
235EX(10f)	s32i	a9, a5, 12
236	addi	a5, a5, 16
237#if !XCHAL_HAVE_LOOPS
238#if defined(__XTENSA_CALL0_ABI__)
239	blt	a3, a10, .Loop2
240	l32i	a10, a1, 0
241#else
242	blt	a3, a12, .Loop2
243#endif
244#endif /* !XCHAL_HAVE_LOOPS */
245.Loop2done:
246	bbci.l	a4, 3, .L12
247	# copy 8 bytes
248EX(10f)	l32i	a7, a3,  4
249EX(10f)	l32i	a8, a3,  8
250	__src_b	a6, a6, a7
251EX(10f)	s32i	a6, a5,  0
252	addi	a3, a3,  8
253	__src_b	a7, a7, a8
254EX(10f)	s32i	a7, a5,  4
255	addi	a5, a5,  8
256	mov	a6, a8
257.L12:
258	bbci.l	a4, 2, .L13
259	# copy 4 bytes
260EX(10f)	l32i	a7, a3,  4
261	addi	a3, a3,  4
262	__src_b	a6, a6, a7
263EX(10f)	s32i	a6, a5,  0
264	addi	a5, a5,  4
265	mov	a6, a7
266.L13:
267	add	a3, a3, a10	# readjust a3 with correct misalignment
268	bbci.l	a4, 1, .L14
269	# copy 2 bytes
270EX(10f)	l8ui	a6, a3,  0
271EX(10f)	l8ui	a7, a3,  1
272	addi	a3, a3,  2
273EX(10f)	s8i	a6, a5,  0
274EX(10f)	s8i	a7, a5,  1
275	addi	a5, a5,  2
276.L14:
277	bbci.l	a4, 0, .L15
278	# copy 1 byte
279EX(10f)	l8ui	a6, a3,  0
280EX(10f)	s8i	a6, a5,  0
281.L15:
282	movi	a2, 0		# return success for len bytes copied
283	abi_ret(STACK_SIZE)
284
285ENDPROC(__xtensa_copy_user)
286EXPORT_SYMBOL(__xtensa_copy_user)
287
288	.section .fixup, "ax"
289	.align	4
290
291/* a2 = original dst; a5 = current dst; a11= original len
292 * bytes_copied = a5 - a2
293 * retval = bytes_not_copied = original len - bytes_copied
294 * retval = a11 - (a5 - a2)
295 */
296
297
29810:
299	sub	a2, a5, a2	/* a2 <-- bytes copied */
300	sub	a2, a11, a2	/* a2 <-- bytes not copied */
301	abi_ret(STACK_SIZE)
302