aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorChris Zankel <czankel@tensilica.com>2005-06-24 01:01:20 -0400
committerLinus Torvalds <torvalds@ppc970.osdl.org>2005-06-24 03:05:21 -0400
commit249ac17e96811acc3c6402317dd5d5c89d2cbf68 (patch)
tree0a174065460de196861b85f1d9a48c88b2a2675a
parent5a0015d62668e64c8b6e02e360fbbea121bfd5e6 (diff)
[PATCH] xtensa: Architecture support for Tensilica Xtensa Part 4
The attached patches provides part 4 of an architecture implementation for the Tensilica Xtensa CPU series. Signed-off-by: Chris Zankel <chris@zankel.net> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Diffstat
-rw-r--r--arch/xtensa/lib/Makefile7
-rw-r--r--arch/xtensa/lib/checksum.S410
-rw-r--r--arch/xtensa/lib/memcopy.S315
-rw-r--r--arch/xtensa/lib/memset.S160
-rw-r--r--arch/xtensa/lib/pci-auto.c352
-rw-r--r--arch/xtensa/lib/strcasecmp.c32
-rw-r--r--arch/xtensa/lib/strncpy_user.S224
-rw-r--r--arch/xtensa/lib/strnlen_user.S147
-rw-r--r--arch/xtensa/lib/usercopy.S321
9 files changed, 1968 insertions, 0 deletions
diff --git a/arch/xtensa/lib/Makefile b/arch/xtensa/lib/Makefile
new file mode 100644
index 000000000000..ed935b58e8a4
--- /dev/null
+++ b/arch/xtensa/lib/Makefile
@@ -0,0 +1,7 @@
1#
2# Makefile for Xtensa-specific library files.
3#
4
5lib-y += memcopy.o memset.o checksum.o strcasecmp.o \
6 usercopy.o strncpy_user.o strnlen_user.o
7lib-$(CONFIG_PCI) += pci-auto.o
diff --git a/arch/xtensa/lib/checksum.S b/arch/xtensa/lib/checksum.S
new file mode 100644
index 000000000000..e2d64dfd530c
--- /dev/null
+++ b/arch/xtensa/lib/checksum.S
@@ -0,0 +1,410 @@
1/*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * IP/TCP/UDP checksumming routines
7 *
8 * Xtensa version: Copyright (C) 2001 Tensilica, Inc. by Kevin Chea
9 * Optimized by Joe Taylor
10 *
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public License
13 * as published by the Free Software Foundation; either version
14 * 2 of the License, or (at your option) any later version.
15 */
16
17#include <asm/errno.h>
18#include <linux/linkage.h>
19#define _ASMLANGUAGE
20#include <xtensa/config/core.h>
21
22/*
23 * computes a partial checksum, e.g. for TCP/UDP fragments
24 */
25
26/*
27 * unsigned int csum_partial(const unsigned char *buf, int len,
28 * unsigned int sum);
29 * a2 = buf
30 * a3 = len
31 * a4 = sum
32 *
33 * This function assumes 2- or 4-byte alignment. Other alignments will fail!
34 */
35
36/* ONES_ADD converts twos-complement math to ones-complement. */
37#define ONES_ADD(sum, val) \
38 add sum, sum, val ; \
39 bgeu sum, val, 99f ; \
40 addi sum, sum, 1 ; \
4199: ;
42
43.text
44ENTRY(csum_partial)
45 /*
46 * Experiments with Ethernet and SLIP connections show that buf
47 * is aligned on either a 2-byte or 4-byte boundary.
48 */
49 entry sp, 32
50 extui a5, a2, 0, 2
51 bnez a5, 8f /* branch if 2-byte aligned */
52 /* Fall-through on common case, 4-byte alignment */
531:
54 srli a5, a3, 5 /* 32-byte chunks */
55#if XCHAL_HAVE_LOOPS
56 loopgtz a5, 2f
57#else
58 beqz a5, 2f
59 slli a5, a5, 5
60 add a5, a5, a2 /* a5 = end of last 32-byte chunk */
61.Loop1:
62#endif
63 l32i a6, a2, 0
64 l32i a7, a2, 4
65 ONES_ADD(a4, a6)
66 ONES_ADD(a4, a7)
67 l32i a6, a2, 8
68 l32i a7, a2, 12
69 ONES_ADD(a4, a6)
70 ONES_ADD(a4, a7)
71 l32i a6, a2, 16
72 l32i a7, a2, 20
73 ONES_ADD(a4, a6)
74 ONES_ADD(a4, a7)
75 l32i a6, a2, 24
76 l32i a7, a2, 28
77 ONES_ADD(a4, a6)
78 ONES_ADD(a4, a7)
79 addi a2, a2, 4*8
80#if !XCHAL_HAVE_LOOPS
81 blt a2, a5, .Loop1
82#endif
832:
84 extui a5, a3, 2, 3 /* remaining 4-byte chunks */
85#if XCHAL_HAVE_LOOPS
86 loopgtz a5, 3f
87#else
88 beqz a5, 3f
89 slli a5, a5, 2
90 add a5, a5, a2 /* a5 = end of last 4-byte chunk */
91.Loop2:
92#endif
93 l32i a6, a2, 0
94 ONES_ADD(a4, a6)
95 addi a2, a2, 4
96#if !XCHAL_HAVE_LOOPS
97 blt a2, a5, .Loop2
98#endif
993:
100 _bbci.l a3, 1, 5f /* remaining 2-byte chunk */
101 l16ui a6, a2, 0
102 ONES_ADD(a4, a6)
103 addi a2, a2, 2
1045:
105 _bbci.l a3, 0, 7f /* remaining 1-byte chunk */
1066: l8ui a6, a2, 0
107#ifdef __XTENSA_EB__
108 slli a6, a6, 8 /* load byte into bits 8..15 */
109#endif
110 ONES_ADD(a4, a6)
1117:
112 mov a2, a4
113 retw
114
115 /* uncommon case, buf is 2-byte aligned */
1168:
117 beqz a3, 7b /* branch if len == 0 */
118 beqi a3, 1, 6b /* branch if len == 1 */
119
120 extui a5, a2, 0, 1
121 bnez a5, 8f /* branch if 1-byte aligned */
122
123 l16ui a6, a2, 0 /* common case, len >= 2 */
124 ONES_ADD(a4, a6)
125 addi a2, a2, 2 /* adjust buf */
126 addi a3, a3, -2 /* adjust len */
127 j 1b /* now buf is 4-byte aligned */
128
129 /* case: odd-byte aligned, len > 1
130 * This case is dog slow, so don't give us an odd address.
131 * (I don't think this ever happens, but just in case.)
132 */
1338:
134 srli a5, a3, 2 /* 4-byte chunks */
135#if XCHAL_HAVE_LOOPS
136 loopgtz a5, 2f
137#else
138 beqz a5, 2f
139 slli a5, a5, 2
140 add a5, a5, a2 /* a5 = end of last 4-byte chunk */
141.Loop3:
142#endif
143 l8ui a6, a2, 0 /* bits 24..31 */
144 l16ui a7, a2, 1 /* bits 8..23 */
145 l8ui a8, a2, 3 /* bits 0.. 8 */
146#ifdef __XTENSA_EB__
147 slli a6, a6, 24
148#else
149 slli a8, a8, 24
150#endif
151 slli a7, a7, 8
152 or a7, a7, a6
153 or a7, a7, a8
154 ONES_ADD(a4, a7)
155 addi a2, a2, 4
156#if !XCHAL_HAVE_LOOPS
157 blt a2, a5, .Loop3
158#endif
1592:
160 _bbci.l a3, 1, 3f /* remaining 2-byte chunk, still odd addr */
161 l8ui a6, a2, 0
162 l8ui a7, a2, 1
163#ifdef __XTENSA_EB__
164 slli a6, a6, 8
165#else
166 slli a7, a7, 8
167#endif
168 or a7, a7, a6
169 ONES_ADD(a4, a7)
170 addi a2, a2, 2
1713:
172 j 5b /* branch to handle the remaining byte */
173
174
175
176/*
177 * Copy from ds while checksumming, otherwise like csum_partial
178 *
179 * The macros SRC and DST specify the type of access for the instruction.
180 * thus we can call a custom exception handler for each access type.
181 */
182
183#define SRC(y...) \
184 9999: y; \
185 .section __ex_table, "a"; \
186 .long 9999b, 6001f ; \
187 .previous
188
189#define DST(y...) \
190 9999: y; \
191 .section __ex_table, "a"; \
192 .long 9999b, 6002f ; \
193 .previous
194
195/*
196unsigned int csum_partial_copy_generic (const char *src, char *dst, int len,
197 int sum, int *src_err_ptr, int *dst_err_ptr)
198 a2 = src
199 a3 = dst
200 a4 = len
201 a5 = sum
202 a6 = src_err_ptr
203 a7 = dst_err_ptr
204 a8 = temp
205 a9 = temp
206 a10 = temp
207 a11 = original len for exception handling
208 a12 = original dst for exception handling
209
210 This function is optimized for 4-byte aligned addresses. Other
211 alignments work, but not nearly as efficiently.
212 */
213
214ENTRY(csum_partial_copy_generic)
215 entry sp, 32
216 mov a12, a3
217 mov a11, a4
218 or a10, a2, a3
219
220 /* We optimize the following alignment tests for the 4-byte
221 aligned case. Two bbsi.l instructions might seem more optimal
222 (commented out below). However, both labels 5: and 3: are out
223 of the imm8 range, so the assembler relaxes them into
224 equivalent bbci.l, j combinations, which is actually
225 slower. */
226
227 extui a9, a10, 0, 2
228 beqz a9, 1f /* branch if both are 4-byte aligned */
229 bbsi.l a10, 0, 5f /* branch if one address is odd */
230 j 3f /* one address is 2-byte aligned */
231
232/* _bbsi.l a10, 0, 5f */ /* branch if odd address */
233/* _bbsi.l a10, 1, 3f */ /* branch if 2-byte-aligned address */
234
2351:
236 /* src and dst are both 4-byte aligned */
237 srli a10, a4, 5 /* 32-byte chunks */
238#if XCHAL_HAVE_LOOPS
239 loopgtz a10, 2f
240#else
241 beqz a10, 2f
242 slli a10, a10, 5
243 add a10, a10, a2 /* a10 = end of last 32-byte src chunk */
244.Loop5:
245#endif
246SRC( l32i a9, a2, 0 )
247SRC( l32i a8, a2, 4 )
248DST( s32i a9, a3, 0 )
249DST( s32i a8, a3, 4 )
250 ONES_ADD(a5, a9)
251 ONES_ADD(a5, a8)
252SRC( l32i a9, a2, 8 )
253SRC( l32i a8, a2, 12 )
254DST( s32i a9, a3, 8 )
255DST( s32i a8, a3, 12 )
256 ONES_ADD(a5, a9)
257 ONES_ADD(a5, a8)
258SRC( l32i a9, a2, 16 )
259SRC( l32i a8, a2, 20 )
260DST( s32i a9, a3, 16 )
261DST( s32i a8, a3, 20 )
262 ONES_ADD(a5, a9)
263 ONES_ADD(a5, a8)
264SRC( l32i a9, a2, 24 )
265SRC( l32i a8, a2, 28 )
266DST( s32i a9, a3, 24 )
267DST( s32i a8, a3, 28 )
268 ONES_ADD(a5, a9)
269 ONES_ADD(a5, a8)
270 addi a2, a2, 32
271 addi a3, a3, 32
272#if !XCHAL_HAVE_LOOPS
273 blt a2, a10, .Loop5
274#endif
2752:
276 extui a10, a4, 2, 3 /* remaining 4-byte chunks */
277 extui a4, a4, 0, 2 /* reset len for general-case, 2-byte chunks */
278#if XCHAL_HAVE_LOOPS
279 loopgtz a10, 3f
280#else
281 beqz a10, 3f
282 slli a10, a10, 2
283 add a10, a10, a2 /* a10 = end of last 4-byte src chunk */
284.Loop6:
285#endif
286SRC( l32i a9, a2, 0 )
287DST( s32i a9, a3, 0 )
288 ONES_ADD(a5, a9)
289 addi a2, a2, 4
290 addi a3, a3, 4
291#if !XCHAL_HAVE_LOOPS
292 blt a2, a10, .Loop6
293#endif
2943:
295 /*
296 Control comes to here in two cases: (1) It may fall through
297 to here from the 4-byte alignment case to process, at most,
298 one 2-byte chunk. (2) It branches to here from above if
299 either src or dst is 2-byte aligned, and we process all bytes
300 here, except for perhaps a trailing odd byte. It's
301 inefficient, so align your addresses to 4-byte boundaries.
302
303 a2 = src
304 a3 = dst
305 a4 = len
306 a5 = sum
307 */
308 srli a10, a4, 1 /* 2-byte chunks */
309#if XCHAL_HAVE_LOOPS
310 loopgtz a10, 4f
311#else
312 beqz a10, 4f
313 slli a10, a10, 1
314 add a10, a10, a2 /* a10 = end of last 2-byte src chunk */
315.Loop7:
316#endif
317SRC( l16ui a9, a2, 0 )
318DST( s16i a9, a3, 0 )
319 ONES_ADD(a5, a9)
320 addi a2, a2, 2
321 addi a3, a3, 2
322#if !XCHAL_HAVE_LOOPS
323 blt a2, a10, .Loop7
324#endif
3254:
326 /* This section processes a possible trailing odd byte. */
327 _bbci.l a4, 0, 8f /* 1-byte chunk */
328SRC( l8ui a9, a2, 0 )
329DST( s8i a9, a3, 0 )
330#ifdef __XTENSA_EB__
331 slli a9, a9, 8 /* shift byte to bits 8..15 */
332#endif
333 ONES_ADD(a5, a9)
3348:
335 mov a2, a5
336 retw
337
3385:
339 /* Control branch to here when either src or dst is odd. We
340 process all bytes using 8-bit accesses. Grossly inefficient,
341 so don't feed us an odd address. */
342
343 srli a10, a4, 1 /* handle in pairs for 16-bit csum */
344#if XCHAL_HAVE_LOOPS
345 loopgtz a10, 6f
346#else
347 beqz a10, 6f
348 slli a10, a10, 1
349 add a10, a10, a2 /* a10 = end of last odd-aligned, 2-byte src chunk */
350.Loop8:
351#endif
352SRC( l8ui a9, a2, 0 )
353SRC( l8ui a8, a2, 1 )
354DST( s8i a9, a3, 0 )
355DST( s8i a8, a3, 1 )
356#ifdef __XTENSA_EB__
357 slli a9, a9, 8 /* combine into a single 16-bit value */
358#else /* for checksum computation */
359 slli a8, a8, 8
360#endif
361 or a9, a9, a8
362 ONES_ADD(a5, a9)
363 addi a2, a2, 2
364 addi a3, a3, 2
365#if !XCHAL_HAVE_LOOPS
366 blt a2, a10, .Loop8
367#endif
3686:
369 j 4b /* process the possible trailing odd byte */
370
371
372# Exception handler:
373.section .fixup, "ax"
374/*
375 a6 = src_err_ptr
376 a7 = dst_err_ptr
377 a11 = original len for exception handling
378 a12 = original dst for exception handling
379*/
380
3816001:
382 _movi a2, -EFAULT
383 s32i a2, a6, 0 /* src_err_ptr */
384
385 # clear the complete destination - computing the rest
386 # is too much work
387 movi a2, 0
388#if XCHAL_HAVE_LOOPS
389 loopgtz a11, 2f
390#else
391 beqz a11, 2f
392 add a11, a11, a12 /* a11 = ending address */
393.Leloop:
394#endif
395 s8i a2, a12, 0
396 addi a12, a12, 1
397#if !XCHAL_HAVE_LOOPS
398 blt a12, a11, .Leloop
399#endif
4002:
401 retw
402
4036002:
404 movi a2, -EFAULT
405 s32i a2, a7, 0 /* dst_err_ptr */
406 movi a2, 0
407 retw
408
409.previous
410
diff --git a/arch/xtensa/lib/memcopy.S b/arch/xtensa/lib/memcopy.S
new file mode 100644
index 000000000000..e8f6d7eb7222
--- /dev/null
+++ b/arch/xtensa/lib/memcopy.S
@@ -0,0 +1,315 @@
1/*
2 * arch/xtensa/lib/hal/memcopy.S -- Core HAL library functions
3 * xthal_memcpy and xthal_bcopy
4 *
5 * This file is subject to the terms and conditions of the GNU General Public
6 * License. See the file "COPYING" in the main directory of this archive
7 * for more details.
8 *
9 * Copyright (C) 2002 - 2005 Tensilica Inc.
10 */
11
12#include <xtensa/coreasm.h>
13
14 .macro src_b r, w0, w1
15#ifdef __XTENSA_EB__
16 src \r, \w0, \w1
17#else
18 src \r, \w1, \w0
19#endif
20 .endm
21
22 .macro ssa8 r
23#ifdef __XTENSA_EB__
24 ssa8b \r
25#else
26 ssa8l \r
27#endif
28 .endm
29
30
31/*
32 * void *memcpy(void *dst, const void *src, size_t len);
33 * void *memmove(void *dst, const void *src, size_t len);
34 * void *bcopy(const void *src, void *dst, size_t len);
35 *
36 * This function is intended to do the same thing as the standard
37 * library function memcpy() (or bcopy()) for most cases.
38 * However, where the source and/or destination references
39 * an instruction RAM or ROM or a data RAM or ROM, that
40 * source and/or destination will always be accessed with
41 * 32-bit load and store instructions (as required for these
42 * types of devices).
43 *
44 * !!!!!!! XTFIXME:
45 * !!!!!!! Handling of IRAM/IROM has not yet
46 * !!!!!!! been implemented.
47 *
48 * The bcopy version is provided here to avoid the overhead
49 * of an extra call, for callers that require this convention.
50 *
51 * The (general case) algorithm is as follows:
52 * If destination is unaligned, align it by conditionally
53 * copying 1 and 2 bytes.
54 * If source is aligned,
55 * do 16 bytes with a loop, and then finish up with
56 * 8, 4, 2, and 1 byte copies conditional on the length;
57 * else (if source is unaligned),
58 * do the same, but use SRC to align the source data.
59 * This code tries to use fall-through branches for the common
60 * case of aligned source and destination and multiple
61 * of 4 (or 8) length.
62 *
63 * Register use:
64 * a0/ return address
65 * a1/ stack pointer
66 * a2/ return value
67 * a3/ src
68 * a4/ length
69 * a5/ dst
70 * a6/ tmp
71 * a7/ tmp
72 * a8/ tmp
73 * a9/ tmp
74 * a10/ tmp
75 * a11/ tmp
76 */
77
78 .text
79 .align 4
80 .global bcopy
81 .type bcopy,@function
82bcopy:
83 entry sp, 16 # minimal stack frame
84 # a2=src, a3=dst, a4=len
85 mov a5, a3 # copy dst so that a2 is return value
86 mov a3, a2
87 mov a2, a5
88 j .Lcommon # go to common code for memcpy+bcopy
89
90
91/*
92 * Byte by byte copy
93 */
94 .align 4
95 .byte 0 # 1 mod 4 alignment for LOOPNEZ
96 # (0 mod 4 alignment for LBEG)
97.Lbytecopy:
98#if XCHAL_HAVE_LOOPS
99 loopnez a4, .Lbytecopydone
100#else /* !XCHAL_HAVE_LOOPS */
101 beqz a4, .Lbytecopydone
102 add a7, a3, a4 # a7 = end address for source
103#endif /* !XCHAL_HAVE_LOOPS */
104.Lnextbyte:
105 l8ui a6, a3, 0
106 addi a3, a3, 1
107 s8i a6, a5, 0
108 addi a5, a5, 1
109#if !XCHAL_HAVE_LOOPS
110 blt a3, a7, .Lnextbyte
111#endif /* !XCHAL_HAVE_LOOPS */
112.Lbytecopydone:
113 retw
114
115/*
116 * Destination is unaligned
117 */
118
119 .align 4
120.Ldst1mod2: # dst is only byte aligned
121 _bltui a4, 7, .Lbytecopy # do short copies byte by byte
122
123 # copy 1 byte
124 l8ui a6, a3, 0
125 addi a3, a3, 1
126 addi a4, a4, -1
127 s8i a6, a5, 0
128 addi a5, a5, 1
129 _bbci.l a5, 1, .Ldstaligned # if dst is now aligned, then
130 # return to main algorithm
131.Ldst2mod4: # dst 16-bit aligned
132 # copy 2 bytes
133 _bltui a4, 6, .Lbytecopy # do short copies byte by byte
134 l8ui a6, a3, 0
135 l8ui a7, a3, 1
136 addi a3, a3, 2
137 addi a4, a4, -2
138 s8i a6, a5, 0
139 s8i a7, a5, 1
140 addi a5, a5, 2
141 j .Ldstaligned # dst is now aligned, return to main algorithm
142
143 .align 4
144 .global memcpy
145 .type memcpy,@function
146memcpy:
147 .global memmove
148 .type memmove,@function
149memmove:
150
151 entry sp, 16 # minimal stack frame
152 # a2/ dst, a3/ src, a4/ len
153 mov a5, a2 # copy dst so that a2 is return value
154.Lcommon:
155 _bbsi.l a2, 0, .Ldst1mod2 # if dst is 1 mod 2
156 _bbsi.l a2, 1, .Ldst2mod4 # if dst is 2 mod 4
157.Ldstaligned: # return here from .Ldst?mod? once dst is aligned
158 srli a7, a4, 4 # number of loop iterations with 16B
159 # per iteration
160 movi a8, 3 # if source is not aligned,
161 _bany a3, a8, .Lsrcunaligned # then use shifting copy
162 /*
163 * Destination and source are word-aligned, use word copy.
164 */
165 # copy 16 bytes per iteration for word-aligned dst and word-aligned src
166#if XCHAL_HAVE_LOOPS
167 loopnez a7, .Loop1done
168#else /* !XCHAL_HAVE_LOOPS */
169 beqz a7, .Loop1done
170 slli a8, a7, 4
171 add a8, a8, a3 # a8 = end of last 16B source chunk
172#endif /* !XCHAL_HAVE_LOOPS */
173.Loop1:
174 l32i a6, a3, 0
175 l32i a7, a3, 4
176 s32i a6, a5, 0
177 l32i a6, a3, 8
178 s32i a7, a5, 4
179 l32i a7, a3, 12
180 s32i a6, a5, 8
181 addi a3, a3, 16
182 s32i a7, a5, 12
183 addi a5, a5, 16
184#if !XCHAL_HAVE_LOOPS
185 blt a3, a8, .Loop1
186#endif /* !XCHAL_HAVE_LOOPS */
187.Loop1done:
188 bbci.l a4, 3, .L2
189 # copy 8 bytes
190 l32i a6, a3, 0
191 l32i a7, a3, 4
192 addi a3, a3, 8
193 s32i a6, a5, 0
194 s32i a7, a5, 4
195 addi a5, a5, 8
196.L2:
197 bbsi.l a4, 2, .L3
198 bbsi.l a4, 1, .L4
199 bbsi.l a4, 0, .L5
200 retw
201.L3:
202 # copy 4 bytes
203 l32i a6, a3, 0
204 addi a3, a3, 4
205 s32i a6, a5, 0
206 addi a5, a5, 4
207 bbsi.l a4, 1, .L4
208 bbsi.l a4, 0, .L5
209 retw
210.L4:
211 # copy 2 bytes
212 l16ui a6, a3, 0
213 addi a3, a3, 2
214 s16i a6, a5, 0
215 addi a5, a5, 2
216 bbsi.l a4, 0, .L5
217 retw
218.L5:
219 # copy 1 byte
220 l8ui a6, a3, 0
221 s8i a6, a5, 0
222 retw
223
224/*
225 * Destination is aligned, Source is unaligned
226 */
227
228 .align 4
229.Lsrcunaligned:
230 _beqz a4, .Ldone # avoid loading anything for zero-length copies
231 # copy 16 bytes per iteration for word-aligned dst and unaligned src
232 ssa8 a3 # set shift amount from byte offset
233#define SIM_CHECKS_ALIGNMENT 1 /* set to 1 when running on ISS (simulator) with the
234 lint or ferret client, or 0 to save a few cycles */
235#if XCHAL_UNALIGNED_LOAD_EXCEPTION || SIM_CHECKS_ALIGNMENT
236 and a11, a3, a8 # save unalignment offset for below
237 sub a3, a3, a11 # align a3
238#endif
239 l32i a6, a3, 0 # load first word
240#if XCHAL_HAVE_LOOPS
241 loopnez a7, .Loop2done
242#else /* !XCHAL_HAVE_LOOPS */
243 beqz a7, .Loop2done
244 slli a10, a7, 4
245 add a10, a10, a3 # a10 = end of last 16B source chunk
246#endif /* !XCHAL_HAVE_LOOPS */
247.Loop2:
248 l32i a7, a3, 4
249 l32i a8, a3, 8
250 src_b a6, a6, a7
251 s32i a6, a5, 0
252 l32i a9, a3, 12
253 src_b a7, a7, a8
254 s32i a7, a5, 4
255 l32i a6, a3, 16
256 src_b a8, a8, a9
257 s32i a8, a5, 8
258 addi a3, a3, 16
259 src_b a9, a9, a6
260 s32i a9, a5, 12
261 addi a5, a5, 16
262#if !XCHAL_HAVE_LOOPS
263 blt a3, a10, .Loop2
264#endif /* !XCHAL_HAVE_LOOPS */
265.Loop2done:
266 bbci.l a4, 3, .L12
267 # copy 8 bytes
268 l32i a7, a3, 4
269 l32i a8, a3, 8
270 src_b a6, a6, a7
271 s32i a6, a5, 0
272 addi a3, a3, 8
273 src_b a7, a7, a8
274 s32i a7, a5, 4
275 addi a5, a5, 8
276 mov a6, a8
277.L12:
278 bbci.l a4, 2, .L13
279 # copy 4 bytes
280 l32i a7, a3, 4
281 addi a3, a3, 4
282 src_b a6, a6, a7
283 s32i a6, a5, 0
284 addi a5, a5, 4
285 mov a6, a7
286.L13:
287#if XCHAL_UNALIGNED_LOAD_EXCEPTION || SIM_CHECKS_ALIGNMENT
288 add a3, a3, a11 # readjust a3 with correct misalignment
289#endif
290 bbsi.l a4, 1, .L14
291 bbsi.l a4, 0, .L15
292.Ldone: retw
293.L14:
294 # copy 2 bytes
295 l8ui a6, a3, 0
296 l8ui a7, a3, 1
297 addi a3, a3, 2
298 s8i a6, a5, 0
299 s8i a7, a5, 1
300 addi a5, a5, 2
301 bbsi.l a4, 0, .L15
302 retw
303.L15:
304 # copy 1 byte
305 l8ui a6, a3, 0
306 s8i a6, a5, 0
307 retw
308
309/*
310 * Local Variables:
311 * mode:fundamental
312 * comment-start: "# "
313 * comment-start-skip: "# *"
314 * End:
315 */
diff --git a/arch/xtensa/lib/memset.S b/arch/xtensa/lib/memset.S
new file mode 100644
index 000000000000..4de25134bc62
--- /dev/null
+++ b/arch/xtensa/lib/memset.S
@@ -0,0 +1,160 @@
1/*
2 * arch/xtensa/lib/memset.S
3 *
4 * ANSI C standard library function memset
5 * (Well, almost. .fixup code might return zero.)
6 *
7 * This file is subject to the terms and conditions of the GNU General
8 * Public License. See the file "COPYING" in the main directory of
9 * this archive for more details.
10 *
11 * Copyright (C) 2002 Tensilica Inc.
12 */
13
14#include <xtensa/coreasm.h>
15
16/*
17 * void *memset(void *dst, int c, size_t length)
18 *
19 * The algorithm is as follows:
20 * Create a word with c in all byte positions
21 * If the destination is aligned,
22 * do 16B chucks with a loop, and then finish up with
23 * 8B, 4B, 2B, and 1B stores conditional on the length.
24 * If destination is unaligned, align it by conditionally
25 * setting 1B and 2B and then go to aligned case.
26 * This code tries to use fall-through branches for the common
27 * case of an aligned destination (except for the branches to
28 * the alignment labels).
29 */
30
31/* Load or store instructions that may cause exceptions use the EX macro. */
32
33#define EX(insn,reg1,reg2,offset,handler) \
349: insn reg1, reg2, offset; \
35 .section __ex_table, "a"; \
36 .word 9b, handler; \
37 .previous
38
39
40.text
41.align 4
42.global memset
43.type memset,@function
44memset:
45 entry sp, 16 # minimal stack frame
46 # a2/ dst, a3/ c, a4/ length
47 extui a3, a3, 0, 8 # mask to just 8 bits
48 slli a7, a3, 8 # duplicate character in all bytes of word
49 or a3, a3, a7 # ...
50 slli a7, a3, 16 # ...
51 or a3, a3, a7 # ...
52 mov a5, a2 # copy dst so that a2 is return value
53 movi a6, 3 # for alignment tests
54 bany a2, a6, .Ldstunaligned # if dst is unaligned
55.L0: # return here from .Ldstunaligned when dst is aligned
56 srli a7, a4, 4 # number of loop iterations with 16B
57 # per iteration
58 bnez a4, .Laligned
59 retw
60
61/*
62 * Destination is word-aligned.
63 */
64 # set 16 bytes per iteration for word-aligned dst
65 .align 4 # 1 mod 4 alignment for LOOPNEZ
66 .byte 0 # (0 mod 4 alignment for LBEG)
67.Laligned:
68#if XCHAL_HAVE_LOOPS
69 loopnez a7, .Loop1done
70#else /* !XCHAL_HAVE_LOOPS */
71 beqz a7, .Loop1done
72 slli a6, a7, 4
73 add a6, a6, a5 # a6 = end of last 16B chunk
74#endif /* !XCHAL_HAVE_LOOPS */
75.Loop1:
76 EX(s32i, a3, a5, 0, memset_fixup)
77 EX(s32i, a3, a5, 4, memset_fixup)
78 EX(s32i, a3, a5, 8, memset_fixup)
79 EX(s32i, a3, a5, 12, memset_fixup)
80 addi a5, a5, 16
81#if !XCHAL_HAVE_LOOPS
82 blt a5, a6, .Loop1
83#endif /* !XCHAL_HAVE_LOOPS */
84.Loop1done:
85 bbci.l a4, 3, .L2
86 # set 8 bytes
87 EX(s32i, a3, a5, 0, memset_fixup)
88 EX(s32i, a3, a5, 4, memset_fixup)
89 addi a5, a5, 8
90.L2:
91 bbci.l a4, 2, .L3
92 # set 4 bytes
93 EX(s32i, a3, a5, 0, memset_fixup)
94 addi a5, a5, 4
95.L3:
96 bbci.l a4, 1, .L4
97 # set 2 bytes
98 EX(s16i, a3, a5, 0, memset_fixup)
99 addi a5, a5, 2
100.L4:
101 bbci.l a4, 0, .L5
102 # set 1 byte
103 EX(s8i, a3, a5, 0, memset_fixup)
104.L5:
105.Lret1:
106 retw
107
108/*
109 * Destination is unaligned
110 */
111
112.Ldstunaligned:
113 bltui a4, 8, .Lbyteset # do short copies byte by byte
114 bbci.l a5, 0, .L20 # branch if dst alignment half-aligned
115 # dst is only byte aligned
116 # set 1 byte
117 EX(s8i, a3, a5, 0, memset_fixup)
118 addi a5, a5, 1
119 addi a4, a4, -1
120 # now retest if dst aligned
121 bbci.l a5, 1, .L0 # if now aligned, return to main algorithm
122.L20:
123 # dst half-aligned
124 # set 2 bytes
125 EX(s16i, a3, a5, 0, memset_fixup)
126 addi a5, a5, 2
127 addi a4, a4, -2
128 j .L0 # dst is now aligned, return to main algorithm
129
130/*
131 * Byte by byte set
132 */
133 .align 4
134 .byte 0 # 1 mod 4 alignment for LOOPNEZ
135 # (0 mod 4 alignment for LBEG)
136.Lbyteset:
137#if XCHAL_HAVE_LOOPS
138 loopnez a4, .Lbytesetdone
139#else /* !XCHAL_HAVE_LOOPS */
140 beqz a4, .Lbytesetdone
141 add a6, a5, a4 # a6 = ending address
142#endif /* !XCHAL_HAVE_LOOPS */
143.Lbyteloop:
144 EX(s8i, a3, a5, 0, memset_fixup)
145 addi a5, a5, 1
146#if !XCHAL_HAVE_LOOPS
147 blt a5, a6, .Lbyteloop
148#endif /* !XCHAL_HAVE_LOOPS */
149.Lbytesetdone:
150 retw
151
152
153 .section .fixup, "ax"
154 .align 4
155
156/* We return zero if a failure occurred. */
157
158memset_fixup:
159 movi a2, 0
160 retw
diff --git a/arch/xtensa/lib/pci-auto.c b/arch/xtensa/lib/pci-auto.c
new file mode 100644
index 000000000000..90c790f6123b
--- /dev/null
+++ b/arch/xtensa/lib/pci-auto.c
@@ -0,0 +1,352 @@
1/*
2 * arch/xtensa/kernel/pci-auto.c
3 *
4 * PCI autoconfiguration library
5 *
6 * Copyright (C) 2001 - 2005 Tensilica Inc.
7 *
8 * Chris Zankel <zankel@tensilica.com, cez@zankel.net>
9 *
10 * Based on work from Matt Porter <mporter@mvista.com>
11 *
12 * This program is free software; you can redistribute it and/or modify it
13 * under the terms of the GNU General Public License as published by the
14 * Free Software Foundation; either version 2 of the License, or (at your
15 * option) any later version.
16 */
17
18#include <linux/kernel.h>
19#include <linux/init.h>
20#include <linux/pci.h>
21
22#include <asm/pci-bridge.h>
23
24
25/*
26 *
27 * Setting up a PCI
28 *
29 * pci_ctrl->first_busno = <first bus number (0)>
30 * pci_ctrl->last_busno = <last bus number (0xff)>
31 * pci_ctrl->ops = <PCI config operations>
32 * pci_ctrl->map_irq = <function to return the interrupt number for a device>
33 *
34 * pci_ctrl->io_space.start = <IO space start address (PCI view)>
35 * pci_ctrl->io_space.end = <IO space end address (PCI view)>
36 * pci_ctrl->io_space.base = <IO space offset: address 0 from CPU space>
37 * pci_ctrl->mem_space.start = <MEM space start address (PCI view)>
38 * pci_ctrl->mem_space.end = <MEM space end address (PCI view)>
39 * pci_ctrl->mem_space.base = <MEM space offset: address 0 from CPU space>
40 *
41 * pcibios_init_resource(&pci_ctrl->io_resource, <IO space start>,
42 * <IO space end>, IORESOURCE_IO, "PCI host bridge");
43 * pcibios_init_resource(&pci_ctrl->mem_resources[0], <MEM space start>,
44 * <MEM space end>, IORESOURCE_MEM, "PCI host bridge");
45 *
46 * pci_ctrl->last_busno = pciauto_bus_scan(pci_ctrl,pci_ctrl->first_busno);
47 *
48 * int __init pciauto_bus_scan(struct pci_controller *pci_ctrl, int current_bus)
49 *
50 */
51
52
53/* define DEBUG to print some debugging messages. */
54
55#undef DEBUG
56
57#ifdef DEBUG
58# define DBG(x...) printk(x)
59#else
60# define DBG(x...)
61#endif
62
63static int pciauto_upper_iospc;
64static int pciauto_upper_memspc;
65
66static struct pci_dev pciauto_dev;
67static struct pci_bus pciauto_bus;
68
69/*
70 * Helper functions
71 */
72
73/* Initialize the bars of a PCI device. */
74
75static void __init
76pciauto_setup_bars(struct pci_dev *dev, int bar_limit)
77{
78 int bar_size;
79 int bar, bar_nr;
80 int *upper_limit;
81 int found_mem64 = 0;
82
83 for (bar = PCI_BASE_ADDRESS_0, bar_nr = 0;
84 bar <= bar_limit;
85 bar+=4, bar_nr++)
86 {
87 /* Tickle the BAR and get the size */
88 pci_write_config_dword(dev, bar, 0xffffffff);
89 pci_read_config_dword(dev, bar, &bar_size);
90
91 /* If BAR is not implemented go to the next BAR */
92 if (!bar_size)
93 continue;
94
95 /* Check the BAR type and set our address mask */
96 if (bar_size & PCI_BASE_ADDRESS_SPACE_IO)
97 {
98 bar_size &= PCI_BASE_ADDRESS_IO_MASK;
99 upper_limit = &pciauto_upper_iospc;
100 DBG("PCI Autoconfig: BAR %d, I/O, ", bar_nr);
101 }
102 else
103 {
104 if ((bar_size & PCI_BASE_ADDRESS_MEM_TYPE_MASK) ==
105 PCI_BASE_ADDRESS_MEM_TYPE_64)
106 found_mem64 = 1;
107
108 bar_size &= PCI_BASE_ADDRESS_MEM_MASK;
109 upper_limit = &pciauto_upper_memspc;
110 DBG("PCI Autoconfig: BAR %d, Mem, ", bar_nr);
111 }
112
113 /* Allocate a base address (bar_size is negative!) */
114 *upper_limit = (*upper_limit + bar_size) & bar_size;
115
116 /* Write it out and update our limit */
117 pci_write_config_dword(dev, bar, *upper_limit);
118
119 /*
120 * If we are a 64-bit decoder then increment to the
121 * upper 32 bits of the bar and force it to locate
122 * in the lower 4GB of memory.
123 */
124
125 if (found_mem64)
126 pci_write_config_dword(dev, (bar+=4), 0x00000000);
127
128 DBG("size=0x%x, address=0x%x\n", ~bar_size + 1, *upper_limit);
129 }
130}
131
132/* Initialize the interrupt number. */
133
134static void __init
135pciauto_setup_irq(struct pci_controller* pci_ctrl,struct pci_dev *dev,int devfn)
136{
137 u8 pin;
138 int irq = 0;
139
140 pci_read_config_byte(dev, PCI_INTERRUPT_PIN, &pin);
141
142 /* Fix illegal pin numbers. */
143
144 if (pin == 0 || pin > 4)
145 pin = 1;
146
147 if (pci_ctrl->map_irq)
148 irq = pci_ctrl->map_irq(dev, PCI_SLOT(devfn), pin);
149
150 if (irq == -1)
151 irq = 0;
152
153 DBG("PCI Autoconfig: Interrupt %d, pin %d\n", irq, pin);
154
155 pci_write_config_byte(dev, PCI_INTERRUPT_LINE, irq);
156}
157
158
159static void __init
160pciauto_prescan_setup_bridge(struct pci_dev *dev, int current_bus,
161 int sub_bus, int *iosave, int *memsave)
162{
163 /* Configure bus number registers */
164 pci_write_config_byte(dev, PCI_PRIMARY_BUS, current_bus);
165 pci_write_config_byte(dev, PCI_SECONDARY_BUS, sub_bus + 1);
166 pci_write_config_byte(dev, PCI_SUBORDINATE_BUS, 0xff);
167
168 /* Round memory allocator to 1MB boundary */
169 pciauto_upper_memspc &= ~(0x100000 - 1);
170 *memsave = pciauto_upper_memspc;
171
172 /* Round I/O allocator to 4KB boundary */
173 pciauto_upper_iospc &= ~(0x1000 - 1);
174 *iosave = pciauto_upper_iospc;
175
176 /* Set up memory and I/O filter limits, assume 32-bit I/O space */
177 pci_write_config_word(dev, PCI_MEMORY_LIMIT,
178 ((pciauto_upper_memspc - 1) & 0xfff00000) >> 16);
179 pci_write_config_byte(dev, PCI_IO_LIMIT,
180 ((pciauto_upper_iospc - 1) & 0x0000f000) >> 8);
181 pci_write_config_word(dev, PCI_IO_LIMIT_UPPER16,
182 ((pciauto_upper_iospc - 1) & 0xffff0000) >> 16);
183}
184
185static void __init
186pciauto_postscan_setup_bridge(struct pci_dev *dev, int current_bus, int sub_bus,
187 int *iosave, int *memsave)
188{
189 int cmdstat;
190
191 /* Configure bus number registers */
192 pci_write_config_byte(dev, PCI_SUBORDINATE_BUS, sub_bus);
193
194 /*
195 * Round memory allocator to 1MB boundary.
196 * If no space used, allocate minimum.
197 */
198 pciauto_upper_memspc &= ~(0x100000 - 1);
199 if (*memsave == pciauto_upper_memspc)
200 pciauto_upper_memspc -= 0x00100000;
201
202 pci_write_config_word(dev, PCI_MEMORY_BASE, pciauto_upper_memspc >> 16);
203
204 /* Allocate 1MB for pre-fretch */
205 pci_write_config_word(dev, PCI_PREF_MEMORY_LIMIT,
206 ((pciauto_upper_memspc - 1) & 0xfff00000) >> 16);
207
208 pciauto_upper_memspc -= 0x100000;
209
210 pci_write_config_word(dev, PCI_PREF_MEMORY_BASE,
211 pciauto_upper_memspc >> 16);
212
213 /* Round I/O allocator to 4KB boundary */
214 pciauto_upper_iospc &= ~(0x1000 - 1);
215 if (*iosave == pciauto_upper_iospc)
216 pciauto_upper_iospc -= 0x1000;
217
218 pci_write_config_byte(dev, PCI_IO_BASE,
219 (pciauto_upper_iospc & 0x0000f000) >> 8);
220 pci_write_config_word(dev, PCI_IO_BASE_UPPER16,
221 pciauto_upper_iospc >> 16);
222
223 /* Enable memory and I/O accesses, enable bus master */
224 pci_read_config_dword(dev, PCI_COMMAND, &cmdstat);
225 pci_write_config_dword(dev, PCI_COMMAND,
226 cmdstat |
227 PCI_COMMAND_IO |
228 PCI_COMMAND_MEMORY |
229 PCI_COMMAND_MASTER);
230}
231
232/*
233 * Scan the current PCI bus.
234 */
235
236
237int __init pciauto_bus_scan(struct pci_controller *pci_ctrl, int current_bus)
238{
239 int sub_bus, pci_devfn, pci_class, cmdstat, found_multi=0;
240 unsigned short vid;
241 unsigned char header_type;
242 struct pci_dev *dev = &pciauto_dev;
243
244 pciauto_dev.bus = &pciauto_bus;
245 pciauto_dev.sysdata = pci_ctrl;
246 pciauto_bus.ops = pci_ctrl->ops;
247
248 /*
249 * Fetch our I/O and memory space upper boundaries used
250 * to allocated base addresses on this pci_controller.
251 */
252
253 if (current_bus == pci_ctrl->first_busno)
254 {
255 pciauto_upper_iospc = pci_ctrl->io_resource.end + 1;
256 pciauto_upper_memspc = pci_ctrl->mem_resources[0].end + 1;
257 }
258
259 sub_bus = current_bus;
260
261 for (pci_devfn = 0; pci_devfn < 0xff; pci_devfn++)
262 {
263 /* Skip our host bridge */
264 if ((current_bus == pci_ctrl->first_busno) && (pci_devfn == 0))
265 continue;
266
267 if (PCI_FUNC(pci_devfn) && !found_multi)
268 continue;
269
270 pciauto_bus.number = current_bus;
271 pciauto_dev.devfn = pci_devfn;
272
273 /* If config space read fails from this device, move on */
274 if (pci_read_config_byte(dev, PCI_HEADER_TYPE, &header_type))
275 continue;
276
277 if (!PCI_FUNC(pci_devfn))
278 found_multi = header_type & 0x80;
279 pci_read_config_word(dev, PCI_VENDOR_ID, &vid);
280
281 if (vid == 0xffff || vid == 0x0000) {
282 found_multi = 0;
283 continue;
284 }
285
286 pci_read_config_dword(dev, PCI_CLASS_REVISION, &pci_class);
287
288 if ((pci_class >> 16) == PCI_CLASS_BRIDGE_PCI) {
289
290 int iosave, memsave;
291
292 DBG("PCI Autoconfig: Found P2P bridge, device %d\n",
293 PCI_SLOT(pci_devfn));
294
295 /* Allocate PCI I/O and/or memory space */
296 pciauto_setup_bars(dev, PCI_BASE_ADDRESS_1);
297
298 pciauto_prescan_setup_bridge(dev, current_bus, sub_bus,
299 &iosave, &memsave);
300 sub_bus = pciauto_bus_scan(pci_ctrl, sub_bus+1);
301 pciauto_postscan_setup_bridge(dev, current_bus, sub_bus,
302 &iosave, &memsave);
303 pciauto_bus.number = current_bus;
304
305 continue;
306
307 }
308
309
310#if 0
311 /* Skip legacy mode IDE controller */
312
313 if ((pci_class >> 16) == PCI_CLASS_STORAGE_IDE) {
314
315 unsigned char prg_iface;
316 pci_read_config_byte(dev, PCI_CLASS_PROG, &prg_iface);
317
318 if (!(prg_iface & PCIAUTO_IDE_MODE_MASK)) {
319 DBG("PCI Autoconfig: Skipping legacy mode "
320 "IDE controller\n");
321 continue;
322 }
323 }
324#endif
325
326 /*
327 * Found a peripheral, enable some standard
328 * settings
329 */
330
331 pci_read_config_dword(dev, PCI_COMMAND, &cmdstat);
332 pci_write_config_dword(dev, PCI_COMMAND,
333 cmdstat |
334 PCI_COMMAND_IO |
335 PCI_COMMAND_MEMORY |
336 PCI_COMMAND_MASTER);
337 pci_write_config_byte(dev, PCI_LATENCY_TIMER, 0x80);
338
339 /* Allocate PCI I/O and/or memory space */
340 DBG("PCI Autoconfig: Found Bus %d, Device %d, Function %d\n",
341 current_bus, PCI_SLOT(pci_devfn), PCI_FUNC(pci_devfn) );
342
343 pciauto_setup_bars(dev, PCI_BASE_ADDRESS_5);
344 pciauto_setup_irq(pci_ctrl, dev, pci_devfn);
345 }
346 return sub_bus;
347}
348
349
350
351
352
diff --git a/arch/xtensa/lib/strcasecmp.c b/arch/xtensa/lib/strcasecmp.c
new file mode 100644
index 000000000000..165b2d6effa5
--- /dev/null
+++ b/arch/xtensa/lib/strcasecmp.c
@@ -0,0 +1,32 @@
1/*
2 * linux/arch/xtensa/lib/strcasecmp.c
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 * Copyright (C) 2002 Tensilica Inc.
9 */
10
11#include <linux/string.h>
12
13
14/* We handle nothing here except the C locale. Since this is used in
15 only one place, on strings known to contain only 7 bit ASCII, this
16 is ok. */
17
18int strcasecmp(const char *a, const char *b)
19{
20 int ca, cb;
21
22 do {
23 ca = *a++ & 0xff;
24 cb = *b++ & 0xff;
25 if (ca >= 'A' && ca <= 'Z')
26 ca += 'a' - 'A';
27 if (cb >= 'A' && cb <= 'Z')
28 cb += 'a' - 'A';
29 } while (ca == cb && ca != '\0');
30
31 return ca - cb;
32}
diff --git a/arch/xtensa/lib/strncpy_user.S b/arch/xtensa/lib/strncpy_user.S
new file mode 100644
index 000000000000..71d55df43893
--- /dev/null
+++ b/arch/xtensa/lib/strncpy_user.S
@@ -0,0 +1,224 @@
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 <xtensa/coreasm.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.text
29.begin literal
30.align 4
31.Lmask0:
32 .byte 0xff, 0x00, 0x00, 0x00
33.Lmask1:
34 .byte 0x00, 0xff, 0x00, 0x00
35.Lmask2:
36 .byte 0x00, 0x00, 0xff, 0x00
37.Lmask3:
38 .byte 0x00, 0x00, 0x00, 0xff
39.end literal
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.align 4
57.global __strncpy_user
58.type __strncpy_user,@function
59__strncpy_user:
60 entry sp, 16 # minimal stack frame
61 # a2/ dst, a3/ src, a4/ len
62 mov a11, a2 # leave dst in return value register
63 beqz a4, .Lret # if len is zero
64 l32r a5, .Lmask0 # mask for byte 0
65 l32r a6, .Lmask1 # mask for byte 1
66 l32r a7, .Lmask2 # mask for byte 2
67 l32r a8, .Lmask3 # mask for byte 3
68 bbsi.l a3, 0, .Lsrc1mod2 # if only 8-bit aligned
69 bbsi.l a3, 1, .Lsrc2mod4 # if only 16-bit aligned
70.Lsrcaligned: # return here when src is word-aligned
71 srli a12, a4, 2 # number of loop iterations with 4B per loop
72 movi a9, 3
73 bnone a11, a9, .Laligned
74 j .Ldstunaligned
75
76.Lsrc1mod2: # src address is odd
77 EX(l8ui, a9, a3, 0, fixup_l) # get byte 0
78 addi a3, a3, 1 # advance src pointer
79 EX(s8i, a9, a11, 0, fixup_s) # store byte 0
80 beqz a9, .Lret # if byte 0 is zero
81 addi a11, a11, 1 # advance dst pointer
82 addi a4, a4, -1 # decrement len
83 beqz a4, .Lret # if len is zero
84 bbci.l a3, 1, .Lsrcaligned # if src is now word-aligned
85
86.Lsrc2mod4: # src address is 2 mod 4
87 EX(l8ui, a9, a3, 0, fixup_l) # get byte 0
88 /* 1-cycle interlock */
89 EX(s8i, a9, a11, 0, fixup_s) # store byte 0
90 beqz a9, .Lret # if byte 0 is zero
91 addi a11, a11, 1 # advance dst pointer
92 addi a4, a4, -1 # decrement len
93 beqz a4, .Lret # if len is zero
94 EX(l8ui, a9, a3, 1, fixup_l) # get byte 0
95 addi a3, a3, 2 # advance src pointer
96 EX(s8i, a9, a11, 0, fixup_s) # store byte 0
97 beqz a9, .Lret # if byte 0 is zero
98 addi a11, a11, 1 # advance dst pointer
99 addi a4, a4, -1 # decrement len
100 bnez a4, .Lsrcaligned # if len is nonzero
101.Lret:
102 sub a2, a11, a2 # compute strlen
103 retw
104
105/*
106 * dst is word-aligned, src is word-aligned
107 */
108 .align 4 # 1 mod 4 alignment for LOOPNEZ
109 .byte 0 # (0 mod 4 alignment for LBEG)
110.Laligned:
111#if XCHAL_HAVE_LOOPS
112 loopnez a12, .Loop1done
113#else
114 beqz a12, .Loop1done
115 slli a12, a12, 2
116 add a12, a12, a11 # a12 = end of last 4B chunck
117#endif
118.Loop1:
119 EX(l32i, a9, a3, 0, fixup_l) # get word from src
120 addi a3, a3, 4 # advance src pointer
121 bnone a9, a5, .Lz0 # if byte 0 is zero
122 bnone a9, a6, .Lz1 # if byte 1 is zero
123 bnone a9, a7, .Lz2 # if byte 2 is zero
124 EX(s32i, a9, a11, 0, fixup_s) # store word to dst
125 bnone a9, a8, .Lz3 # if byte 3 is zero
126 addi a11, a11, 4 # advance dst pointer
127#if !XCHAL_HAVE_LOOPS
128 blt a11, a12, .Loop1
129#endif
130
131.Loop1done:
132 bbci.l a4, 1, .L100
133 # copy 2 bytes
134 EX(l16ui, a9, a3, 0, fixup_l)
135 addi a3, a3, 2 # advance src pointer
136#ifdef __XTENSA_EB__
137 bnone a9, a7, .Lz0 # if byte 2 is zero
138 bnone a9, a8, .Lz1 # if byte 3 is zero
139#else
140 bnone a9, a5, .Lz0 # if byte 0 is zero
141 bnone a9, a6, .Lz1 # if byte 1 is zero
142#endif
143 EX(s16i, a9, a11, 0, fixup_s)
144 addi a11, a11, 2 # advance dst pointer
145.L100:
146 bbci.l a4, 0, .Lret
147 EX(l8ui, a9, a3, 0, fixup_l)
148 /* slot */
149 EX(s8i, a9, a11, 0, fixup_s)
150 beqz a9, .Lret # if byte is zero
151 addi a11, a11, 1-3 # advance dst ptr 1, but also cancel
152 # the effect of adding 3 in .Lz3 code
153 /* fall thru to .Lz3 and "retw" */
154
155.Lz3: # byte 3 is zero
156 addi a11, a11, 3 # advance dst pointer
157 sub a2, a11, a2 # compute strlen
158 retw
159.Lz0: # byte 0 is zero
160#ifdef __XTENSA_EB__
161 movi a9, 0
162#endif /* __XTENSA_EB__ */
163 EX(s8i, a9, a11, 0, fixup_s)
164 sub a2, a11, a2 # compute strlen
165 retw
166.Lz1: # byte 1 is zero
167#ifdef __XTENSA_EB__
168 extui a9, a9, 16, 16
169#endif /* __XTENSA_EB__ */
170 EX(s16i, a9, a11, 0, fixup_s)
171 addi a11, a11, 1 # advance dst pointer
172 sub a2, a11, a2 # compute strlen
173 retw
174.Lz2: # byte 2 is zero
175#ifdef __XTENSA_EB__
176 extui a9, a9, 16, 16
177#endif /* __XTENSA_EB__ */
178 EX(s16i, a9, a11, 0, fixup_s)
179 movi a9, 0
180 EX(s8i, a9, a11, 2, fixup_s)
181 addi a11, a11, 2 # advance dst pointer
182 sub a2, a11, a2 # compute strlen
183 retw
184
185 .align 4 # 1 mod 4 alignment for LOOPNEZ
186 .byte 0 # (0 mod 4 alignment for LBEG)
187.Ldstunaligned:
188/*
189 * for now just use byte copy loop
190 */
191#if XCHAL_HAVE_LOOPS
192 loopnez a4, .Lunalignedend
193#else
194 beqz a4, .Lunalignedend
195 add a12, a11, a4 # a12 = ending address
196#endif /* XCHAL_HAVE_LOOPS */
197.Lnextbyte:
198 EX(l8ui, a9, a3, 0, fixup_l)
199 addi a3, a3, 1
200 EX(s8i, a9, a11, 0, fixup_s)
201 beqz a9, .Lunalignedend
202 addi a11, a11, 1
203#if !XCHAL_HAVE_LOOPS
204 blt a11, a12, .Lnextbyte
205#endif
206
207.Lunalignedend:
208 sub a2, a11, a2 # compute strlen
209 retw
210
211
212 .section .fixup, "ax"
213 .align 4
214
215 /* For now, just return -EFAULT. Future implementations might
216 * like to clear remaining kernel space, like the fixup
217 * implementation in memset(). Thus, we differentiate between
218 * load/store fixups. */
219
220fixup_s:
221fixup_l:
222 movi a2, -EFAULT
223 retw
224
diff --git a/arch/xtensa/lib/strnlen_user.S b/arch/xtensa/lib/strnlen_user.S
new file mode 100644
index 000000000000..cdff4d670f3b
--- /dev/null
+++ b/arch/xtensa/lib/strnlen_user.S
@@ -0,0 +1,147 @@
1/*
2 * arch/xtensa/lib/strnlen_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 strnlen, including trailing zero terminator.
9 * Zero indicates error.
10 *
11 * Copyright (C) 2002 Tensilica Inc.
12 */
13
14#include <xtensa/coreasm.h>
15
16/* Load or store instructions that may cause exceptions use the EX macro. */
17
18#define EX(insn,reg1,reg2,offset,handler) \
199: insn reg1, reg2, offset; \
20 .section __ex_table, "a"; \
21 .word 9b, handler; \
22 .previous
23
24/*
25 * size_t __strnlen_user(const char *s, size_t len)
26 */
27.text
28.begin literal
29.align 4
30.Lmask0:
31 .byte 0xff, 0x00, 0x00, 0x00
32.Lmask1:
33 .byte 0x00, 0xff, 0x00, 0x00
34.Lmask2:
35 .byte 0x00, 0x00, 0xff, 0x00
36.Lmask3:
37 .byte 0x00, 0x00, 0x00, 0xff
38.end literal
39
40# Register use:
41# a2/ src
42# a3/ len
43# a4/ tmp
44# a5/ mask0
45# a6/ mask1
46# a7/ mask2
47# a8/ mask3
48# a9/ tmp
49# a10/ tmp
50
51.align 4
52.global __strnlen_user
53.type __strnlen_user,@function
54__strnlen_user:
55 entry sp, 16 # minimal stack frame
56 # a2/ s, a3/ len
57 addi a4, a2, -4 # because we overincrement at the end;
58 # we compensate with load offsets of 4
59 l32r a5, .Lmask0 # mask for byte 0
60 l32r a6, .Lmask1 # mask for byte 1
61 l32r a7, .Lmask2 # mask for byte 2
62 l32r a8, .Lmask3 # mask for byte 3
63 bbsi.l a2, 0, .L1mod2 # if only 8-bit aligned
64 bbsi.l a2, 1, .L2mod4 # if only 16-bit aligned
65
66/*
67 * String is word-aligned.
68 */
69.Laligned:
70 srli a10, a3, 2 # number of loop iterations with 4B per loop
71#if XCHAL_HAVE_LOOPS
72 loopnez a10, .Ldone
73#else
74 beqz a10, .Ldone
75 slli a10, a10, 2
76 add a10, a10, a4 # a10 = end of last 4B chunk
77#endif /* XCHAL_HAVE_LOOPS */
78.Loop:
79 EX(l32i, a9, a4, 4, lenfixup) # get next word of string
80 addi a4, a4, 4 # advance string pointer
81 bnone a9, a5, .Lz0 # if byte 0 is zero
82 bnone a9, a6, .Lz1 # if byte 1 is zero
83 bnone a9, a7, .Lz2 # if byte 2 is zero
84 bnone a9, a8, .Lz3 # if byte 3 is zero
85#if !XCHAL_HAVE_LOOPS
86 blt a4, a10, .Loop
87#endif
88
89.Ldone:
90 EX(l32i, a9, a4, 4, lenfixup) # load 4 bytes for remaining checks
91
92 bbci.l a3, 1, .L100
93 # check two more bytes (bytes 0, 1 of word)
94 addi a4, a4, 2 # advance string pointer
95 bnone a9, a5, .Lz0 # if byte 0 is zero
96 bnone a9, a6, .Lz1 # if byte 1 is zero
97.L100:
98 bbci.l a3, 0, .L101
99 # check one more byte (byte 2 of word)
100 # Actually, we don't need to check. Zero or nonzero, we'll add one.
101 # Do not add an extra one for the NULL terminator since we have
102 # exhausted the original len parameter.
103 addi a4, a4, 1 # advance string pointer
104.L101:
105 sub a2, a4, a2 # compute length
106 retw
107
108# NOTE that in several places below, we point to the byte just after
109# the zero byte in order to include the NULL terminator in the count.
110
111.Lz3: # byte 3 is zero
112 addi a4, a4, 3 # point to zero byte
113.Lz0: # byte 0 is zero
114 addi a4, a4, 1 # point just beyond zero byte
115 sub a2, a4, a2 # subtract to get length
116 retw
117.Lz1: # byte 1 is zero
118 addi a4, a4, 1+1 # point just beyond zero byte
119 sub a2, a4, a2 # subtract to get length
120 retw
121.Lz2: # byte 2 is zero
122 addi a4, a4, 2+1 # point just beyond zero byte
123 sub a2, a4, a2 # subtract to get length
124 retw
125
126.L1mod2: # address is odd
127 EX(l8ui, a9, a4, 4, lenfixup) # get byte 0
128 addi a4, a4, 1 # advance string pointer
129 beqz a9, .Lz3 # if byte 0 is zero
130 bbci.l a4, 1, .Laligned # if string pointer is now word-aligned
131
132.L2mod4: # address is 2 mod 4
133 addi a4, a4, 2 # advance ptr for aligned access
134 EX(l32i, a9, a4, 0, lenfixup) # get word with first two bytes of string
135 bnone a9, a7, .Lz2 # if byte 2 (of word, not string) is zero
136 bany a9, a8, .Laligned # if byte 3 (of word, not string) is nonzero
137 # byte 3 is zero
138 addi a4, a4, 3+1 # point just beyond zero byte
139 sub a2, a4, a2 # subtract to get length
140 retw
141
142 .section .fixup, "ax"
143 .align 4
144lenfixup:
145 movi a2, 0
146 retw
147
diff --git a/arch/xtensa/lib/usercopy.S b/arch/xtensa/lib/usercopy.S
new file mode 100644
index 000000000000..265db2693cbd
--- /dev/null
+++ b/arch/xtensa/lib/usercopy.S
@@ -0,0 +1,321 @@
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 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 loose 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 <xtensa/coreasm.h>
57
58#ifdef __XTENSA_EB__
59#define ALIGN(R, W0, W1) src R, W0, W1
60#define SSA8(R) ssa8b R
61#else
62#define ALIGN(R, W0, W1) src R, W1, W0
63#define SSA8(R) ssa8l R
64#endif
65
66/* Load or store instructions that may cause exceptions use the EX macro. */
67
68#define EX(insn,reg1,reg2,offset,handler) \
699: insn reg1, reg2, offset; \
70 .section __ex_table, "a"; \
71 .word 9b, handler; \
72 .previous
73
74
75 .text
76 .align 4
77 .global __xtensa_copy_user
78 .type __xtensa_copy_user,@function
79__xtensa_copy_user:
80 entry sp, 16 # minimal stack frame
81 # a2/ dst, a3/ src, a4/ len
82 mov a5, a2 # copy dst so that a2 is return value
83 mov a11, a4 # preserve original len for error case
84.Lcommon:
85 bbsi.l a2, 0, .Ldst1mod2 # if dst is 1 mod 2
86 bbsi.l a2, 1, .Ldst2mod4 # if dst is 2 mod 4
87.Ldstaligned: # return here from .Ldstunaligned when dst is aligned
88 srli a7, a4, 4 # number of loop iterations with 16B
89 # per iteration
90 movi a8, 3 # if source is also aligned,
91 bnone a3, a8, .Laligned # then use word copy
92 SSA8( a3) # set shift amount from byte offset
93 bnez a4, .Lsrcunaligned
94 movi a2, 0 # return success for len==0
95 retw
96
97/*
98 * Destination is unaligned
99 */
100
101.Ldst1mod2: # dst is only byte aligned
102 bltui a4, 7, .Lbytecopy # do short copies byte by byte
103
104 # copy 1 byte
105 EX(l8ui, a6, a3, 0, l_fixup)
106 addi a3, a3, 1
107 EX(s8i, a6, a5, 0, s_fixup)
108 addi a5, a5, 1
109 addi a4, a4, -1
110 bbci.l a5, 1, .Ldstaligned # if dst is now aligned, then
111 # return to main algorithm
112.Ldst2mod4: # dst 16-bit aligned
113 # copy 2 bytes
114 bltui a4, 6, .Lbytecopy # do short copies byte by byte
115 EX(l8ui, a6, a3, 0, l_fixup)
116 EX(l8ui, a7, a3, 1, l_fixup)
117 addi a3, a3, 2
118 EX(s8i, a6, a5, 0, s_fixup)
119 EX(s8i, a7, a5, 1, s_fixup)
120 addi a5, a5, 2
121 addi a4, a4, -2
122 j .Ldstaligned # dst is now aligned, return to main algorithm
123
124/*
125 * Byte by byte copy
126 */
127 .align 4
128 .byte 0 # 1 mod 4 alignment for LOOPNEZ
129 # (0 mod 4 alignment for LBEG)
130.Lbytecopy:
131#if XCHAL_HAVE_LOOPS
132 loopnez a4, .Lbytecopydone
133#else /* !XCHAL_HAVE_LOOPS */
134 beqz a4, .Lbytecopydone
135 add a7, a3, a4 # a7 = end address for source
136#endif /* !XCHAL_HAVE_LOOPS */
137.Lnextbyte:
138 EX(l8ui, a6, a3, 0, l_fixup)
139 addi a3, a3, 1
140 EX(s8i, a6, a5, 0, s_fixup)
141 addi a5, a5, 1
142#if !XCHAL_HAVE_LOOPS
143 blt a3, a7, .Lnextbyte
144#endif /* !XCHAL_HAVE_LOOPS */
145.Lbytecopydone:
146 movi a2, 0 # return success for len bytes copied
147 retw
148
149/*
150 * Destination and source are word-aligned.
151 */
152 # copy 16 bytes per iteration for word-aligned dst and word-aligned src
153 .align 4 # 1 mod 4 alignment for LOOPNEZ
154 .byte 0 # (0 mod 4 alignment for LBEG)
155.Laligned:
156#if XCHAL_HAVE_LOOPS
157 loopnez a7, .Loop1done
158#else /* !XCHAL_HAVE_LOOPS */
159 beqz a7, .Loop1done
160 slli a8, a7, 4
161 add a8, a8, a3 # a8 = end of last 16B source chunk
162#endif /* !XCHAL_HAVE_LOOPS */
163.Loop1:
164 EX(l32i, a6, a3, 0, l_fixup)
165 EX(l32i, a7, a3, 4, l_fixup)
166 EX(s32i, a6, a5, 0, s_fixup)
167 EX(l32i, a6, a3, 8, l_fixup)
168 EX(s32i, a7, a5, 4, s_fixup)
169 EX(l32i, a7, a3, 12, l_fixup)
170 EX(s32i, a6, a5, 8, s_fixup)
171 addi a3, a3, 16
172 EX(s32i, a7, a5, 12, s_fixup)
173 addi a5, a5, 16
174#if !XCHAL_HAVE_LOOPS
175 blt a3, a8, .Loop1
176#endif /* !XCHAL_HAVE_LOOPS */
177.Loop1done:
178 bbci.l a4, 3, .L2
179 # copy 8 bytes
180 EX(l32i, a6, a3, 0, l_fixup)
181 EX(l32i, a7, a3, 4, l_fixup)
182 addi a3, a3, 8
183 EX(s32i, a6, a5, 0, s_fixup)
184 EX(s32i, a7, a5, 4, s_fixup)
185 addi a5, a5, 8
186.L2:
187 bbci.l a4, 2, .L3
188 # copy 4 bytes
189 EX(l32i, a6, a3, 0, l_fixup)
190 addi a3, a3, 4
191 EX(s32i, a6, a5, 0, s_fixup)
192 addi a5, a5, 4
193.L3:
194 bbci.l a4, 1, .L4
195 # copy 2 bytes
196 EX(l16ui, a6, a3, 0, l_fixup)
197 addi a3, a3, 2
198 EX(s16i, a6, a5, 0, s_fixup)
199 addi a5, a5, 2
200.L4:
201 bbci.l a4, 0, .L5
202 # copy 1 byte
203 EX(l8ui, a6, a3, 0, l_fixup)
204 EX(s8i, a6, a5, 0, s_fixup)
205.L5:
206 movi a2, 0 # return success for len bytes copied
207 retw
208
209/*
210 * Destination is aligned, Source is unaligned
211 */
212
213 .align 4
214 .byte 0 # 1 mod 4 alignement for LOOPNEZ
215 # (0 mod 4 alignment for LBEG)
216.Lsrcunaligned:
217 # copy 16 bytes per iteration for word-aligned dst and unaligned src
218 and a10, a3, a8 # save unalignment offset for below
219 sub a3, a3, a10 # align a3 (to avoid sim warnings only; not needed for hardware)
220 EX(l32i, a6, a3, 0, l_fixup) # load first word
221#if XCHAL_HAVE_LOOPS
222 loopnez a7, .Loop2done
223#else /* !XCHAL_HAVE_LOOPS */
224 beqz a7, .Loop2done
225 slli a10, a7, 4
226 add a10, a10, a3 # a10 = end of last 16B source chunk
227#endif /* !XCHAL_HAVE_LOOPS */
228.Loop2:
229 EX(l32i, a7, a3, 4, l_fixup)
230 EX(l32i, a8, a3, 8, l_fixup)
231 ALIGN( a6, a6, a7)
232 EX(s32i, a6, a5, 0, s_fixup)
233 EX(l32i, a9, a3, 12, l_fixup)
234 ALIGN( a7, a7, a8)
235 EX(s32i, a7, a5, 4, s_fixup)
236 EX(l32i, a6, a3, 16, l_fixup)
237 ALIGN( a8, a8, a9)
238 EX(s32i, a8, a5, 8, s_fixup)
239 addi a3, a3, 16
240 ALIGN( a9, a9, a6)
241 EX(s32i, a9, a5, 12, s_fixup)
242 addi a5, a5, 16
243#if !XCHAL_HAVE_LOOPS
244 blt a3, a10, .Loop2
245#endif /* !XCHAL_HAVE_LOOPS */
246.Loop2done:
247 bbci.l a4, 3, .L12
248 # copy 8 bytes
249 EX(l32i, a7, a3, 4, l_fixup)
250 EX(l32i, a8, a3, 8, l_fixup)
251 ALIGN( a6, a6, a7)
252 EX(s32i, a6, a5, 0, s_fixup)
253 addi a3, a3, 8
254 ALIGN( a7, a7, a8)
255 EX(s32i, a7, a5, 4, s_fixup)
256 addi a5, a5, 8
257 mov a6, a8
258.L12:
259 bbci.l a4, 2, .L13
260 # copy 4 bytes
261 EX(l32i, a7, a3, 4, l_fixup)
262 addi a3, a3, 4
263 ALIGN( a6, a6, a7)
264 EX(s32i, a6, a5, 0, s_fixup)
265 addi a5, a5, 4
266 mov a6, a7
267.L13:
268 add a3, a3, a10 # readjust a3 with correct misalignment
269 bbci.l a4, 1, .L14
270 # copy 2 bytes
271 EX(l8ui, a6, a3, 0, l_fixup)
272 EX(l8ui, a7, a3, 1, l_fixup)
273 addi a3, a3, 2
274 EX(s8i, a6, a5, 0, s_fixup)
275 EX(s8i, a7, a5, 1, s_fixup)
276 addi a5, a5, 2
277.L14:
278 bbci.l a4, 0, .L15
279 # copy 1 byte
280 EX(l8ui, a6, a3, 0, l_fixup)
281 EX(s8i, a6, a5, 0, s_fixup)
282.L15:
283 movi a2, 0 # return success for len bytes copied
284 retw
285
286
287 .section .fixup, "ax"
288 .align 4
289
290/* a2 = original dst; a5 = current dst; a11= original len
291 * bytes_copied = a5 - a2
292 * retval = bytes_not_copied = original len - bytes_copied
293 * retval = a11 - (a5 - a2)
294 *
295 * Clearing the remaining pieces of kernel memory plugs security
296 * holes. This functionality is the equivalent of the *_zeroing
297 * functions that some architectures provide.
298 */
299
300.Lmemset:
301 .word memset
302
303s_fixup:
304 sub a2, a5, a2 /* a2 <-- bytes copied */
305 sub a2, a11, a2 /* a2 <-- bytes not copied */
306 retw
307
308l_fixup:
309 sub a2, a5, a2 /* a2 <-- bytes copied */
310 sub a2, a11, a2 /* a2 <-- bytes not copied == return value */
311
312 /* void *memset(void *s, int c, size_t n); */
313 mov a6, a5 /* s */
314 movi a7, 0 /* c */
315 mov a8, a2 /* n */
316 l32r a4, .Lmemset
317 callx4 a4
318 /* Ignore memset return value in a6. */
319 /* a2 still contains bytes not copied. */
320 retw
321
generated by cgit v1.2.2 (git 2.25.0) at 2024-12-02 09:35:41 -0500