2 * arch/xtensa/lib/usercopy.S
4 * Copy to/from user space (derived from arch/xtensa/lib/hal/memcopy.S)
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.
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.
19 * Copyright (C) 2002 Tensilica Inc.
24 * size_t __xtensa_copy_user (void *dst, const void *src, size_t len);
26 * The returned value is the number of bytes not copied. Implies zero
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).
53 * a11/ original length
56 #include <linux/linkage.h>
57 #include <asm/asmmacro.h>
61 ENTRY(__xtensa_copy_user)
64 # a2/ dst, a3/ src, a4/ len
65 mov a5, a2 # copy dst so that a2 is return value
66 mov a11, a4 # preserve original len for error case
68 bbsi.l a2, 0, .Ldst1mod2 # if dst is 1 mod 2
69 bbsi.l a2, 1, .Ldst2mod4 # if dst is 2 mod 4
70 .Ldstaligned: # return here from .Ldstunaligned when dst is aligned
71 srli a7, a4, 4 # number of loop iterations with 16B
73 movi a8, 3 # if source is also aligned,
74 bnone a3, a8, .Laligned # then use word copy
75 __ssa8 a3 # set shift amount from byte offset
76 bnez a4, .Lsrcunaligned
77 movi a2, 0 # return success for len==0
81 * Destination is unaligned
84 .Ldst1mod2: # dst is only byte aligned
85 bltui a4, 7, .Lbytecopy # do short copies byte by byte
88 EX(10f) l8ui a6, a3, 0
93 bbci.l a5, 1, .Ldstaligned # if dst is now aligned, then
94 # return to main algorithm
95 .Ldst2mod4: # dst 16-bit aligned
97 bltui a4, 6, .Lbytecopy # do short copies byte by byte
98 EX(10f) l8ui a6, a3, 0
99 EX(10f) l8ui a7, a3, 1
101 EX(10f) s8i a6, a5, 0
102 EX(10f) s8i a7, a5, 1
105 j .Ldstaligned # dst is now aligned, return to main algorithm
111 .byte 0 # 1 mod 4 alignment for LOOPNEZ
112 # (0 mod 4 alignment for LBEG)
115 loopnez a4, .Lbytecopydone
116 #else /* !XCHAL_HAVE_LOOPS */
117 beqz a4, .Lbytecopydone
118 add a7, a3, a4 # a7 = end address for source
119 #endif /* !XCHAL_HAVE_LOOPS */
121 EX(10f) l8ui a6, a3, 0
123 EX(10f) s8i a6, a5, 0
125 #if !XCHAL_HAVE_LOOPS
126 blt a3, a7, .Lnextbyte
127 #endif /* !XCHAL_HAVE_LOOPS */
129 movi a2, 0 # return success for len bytes copied
133 * Destination and source are word-aligned.
135 # copy 16 bytes per iteration for word-aligned dst and word-aligned src
136 .align 4 # 1 mod 4 alignment for LOOPNEZ
137 .byte 0 # (0 mod 4 alignment for LBEG)
140 loopnez a7, .Loop1done
141 #else /* !XCHAL_HAVE_LOOPS */
144 add a8, a8, a3 # a8 = end of last 16B source chunk
145 #endif /* !XCHAL_HAVE_LOOPS */
147 EX(10f) l32i a6, a3, 0
148 EX(10f) l32i a7, a3, 4
149 EX(10f) s32i a6, a5, 0
150 EX(10f) l32i a6, a3, 8
151 EX(10f) s32i a7, a5, 4
152 EX(10f) l32i a7, a3, 12
153 EX(10f) s32i a6, a5, 8
155 EX(10f) s32i a7, a5, 12
157 #if !XCHAL_HAVE_LOOPS
159 #endif /* !XCHAL_HAVE_LOOPS */
163 EX(10f) l32i a6, a3, 0
164 EX(10f) l32i a7, a3, 4
166 EX(10f) s32i a6, a5, 0
167 EX(10f) s32i a7, a5, 4
172 EX(10f) l32i a6, a3, 0
174 EX(10f) s32i a6, a5, 0
179 EX(10f) l16ui a6, a3, 0
181 EX(10f) s16i a6, a5, 0
186 EX(10f) l8ui a6, a3, 0
187 EX(10f) s8i a6, a5, 0
189 movi a2, 0 # return success for len bytes copied
193 * Destination is aligned, Source is unaligned
197 .byte 0 # 1 mod 4 alignement for LOOPNEZ
198 # (0 mod 4 alignment for LBEG)
200 # copy 16 bytes per iteration for word-aligned dst and unaligned src
201 and a10, a3, a8 # save unalignment offset for below
202 sub a3, a3, a10 # align a3 (to avoid sim warnings only; not needed for hardware)
203 EX(10f) l32i a6, a3, 0 # load first word
205 loopnez a7, .Loop2done
206 #else /* !XCHAL_HAVE_LOOPS */
209 add a12, a12, a3 # a12 = end of last 16B source chunk
210 #endif /* !XCHAL_HAVE_LOOPS */
212 EX(10f) l32i a7, a3, 4
213 EX(10f) l32i a8, a3, 8
215 EX(10f) s32i a6, a5, 0
216 EX(10f) l32i a9, a3, 12
218 EX(10f) s32i a7, a5, 4
219 EX(10f) l32i a6, a3, 16
221 EX(10f) s32i a8, a5, 8
224 EX(10f) s32i a9, a5, 12
226 #if !XCHAL_HAVE_LOOPS
228 #endif /* !XCHAL_HAVE_LOOPS */
232 EX(10f) l32i a7, a3, 4
233 EX(10f) l32i a8, a3, 8
235 EX(10f) s32i a6, a5, 0
238 EX(10f) s32i a7, a5, 4
244 EX(10f) l32i a7, a3, 4
247 EX(10f) s32i a6, a5, 0
251 add a3, a3, a10 # readjust a3 with correct misalignment
254 EX(10f) l8ui a6, a3, 0
255 EX(10f) l8ui a7, a3, 1
257 EX(10f) s8i a6, a5, 0
258 EX(10f) s8i a7, a5, 1
263 EX(10f) l8ui a6, a3, 0
264 EX(10f) s8i a6, a5, 0
266 movi a2, 0 # return success for len bytes copied
269 ENDPROC(__xtensa_copy_user)
271 .section .fixup, "ax"
274 /* a2 = original dst; a5 = current dst; a11= original len
275 * bytes_copied = a5 - a2
276 * retval = bytes_not_copied = original len - bytes_copied
277 * retval = a11 - (a5 - a2)
282 sub a2, a5, a2 /* a2 <-- bytes copied */
283 sub a2, a11, a2 /* a2 <-- bytes not copied */