blob: 56a1f85a1ce807504a95d54d49b3d3cab46b33a6 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * This file is subject to the terms and conditions of the GNU General Public
3 * License. See the file "COPYING" in the main directory of this archive
4 * for more details.
5 *
6 * Unified implementation of memcpy, memmove and the __copy_user backend.
7 *
8 * Copyright (C) 1998, 99, 2000, 01, 2002 Ralf Baechle (ralf@gnu.org)
9 * Copyright (C) 1999, 2000, 01, 2002 Silicon Graphics, Inc.
10 * Copyright (C) 2002 Broadcom, Inc.
11 * memcpy/copy_user author: Mark Vandevoorde
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +010012 * Copyright (C) 2007 Maciej W. Rozycki
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 *
14 * Mnemonic names for arguments to memcpy/__copy_user
15 */
Ralf Baechlee5adb872005-10-20 22:55:26 +010016
17/*
18 * Hack to resolve longstanding prefetch issue
19 *
20 * Prefetching may be fatal on some systems if we're prefetching beyond the
21 * end of memory on some systems. It's also a seriously bad idea on non
22 * dma-coherent systems.
23 */
Ralf Baechle634286f2009-01-28 17:48:40 +000024#ifdef CONFIG_DMA_NONCOHERENT
Ralf Baechlee5adb872005-10-20 22:55:26 +010025#undef CONFIG_CPU_HAS_PREFETCH
26#endif
27#ifdef CONFIG_MIPS_MALTA
28#undef CONFIG_CPU_HAS_PREFETCH
29#endif
30
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include <asm/asm.h>
Sam Ravnborg048eb582005-09-09 22:32:31 +020032#include <asm/asm-offsets.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include <asm/regdef.h>
34
35#define dst a0
36#define src a1
37#define len a2
38
39/*
40 * Spec
41 *
42 * memcpy copies len bytes from src to dst and sets v0 to dst.
43 * It assumes that
44 * - src and dst don't overlap
45 * - src is readable
46 * - dst is writable
47 * memcpy uses the standard calling convention
48 *
49 * __copy_user copies up to len bytes from src to dst and sets a2 (len) to
50 * the number of uncopied bytes due to an exception caused by a read or write.
51 * __copy_user assumes that src and dst don't overlap, and that the call is
52 * implementing one of the following:
53 * copy_to_user
54 * - src is readable (no exceptions when reading src)
55 * copy_from_user
56 * - dst is writable (no exceptions when writing dst)
57 * __copy_user uses a non-standard calling convention; see
58 * include/asm-mips/uaccess.h
59 *
60 * When an exception happens on a load, the handler must
61 # ensure that all of the destination buffer is overwritten to prevent
62 * leaking information to user mode programs.
63 */
64
65/*
66 * Implementation
67 */
68
69/*
70 * The exception handler for loads requires that:
71 * 1- AT contain the address of the byte just past the end of the source
72 * of the copy,
73 * 2- src_entry <= src < AT, and
74 * 3- (dst - src) == (dst_entry - src_entry),
75 * The _entry suffix denotes values when __copy_user was called.
76 *
77 * (1) is set up up by uaccess.h and maintained by not writing AT in copy_user
78 * (2) is met by incrementing src by the number of bytes copied
79 * (3) is met by not doing loads between a pair of increments of dst and src
80 *
81 * The exception handlers for stores adjust len (if necessary) and return.
82 * These handlers do not need to overwrite any data.
83 *
84 * For __rmemcpy and memmove an exception is always a kernel bug, therefore
85 * they're not protected.
86 */
87
88#define EXC(inst_reg,addr,handler) \
899: inst_reg, addr; \
90 .section __ex_table,"a"; \
91 PTR 9b, handler; \
92 .previous
93
94/*
95 * Only on the 64-bit kernel we can made use of 64-bit registers.
96 */
Ralf Baechle875d43e2005-09-03 15:56:16 -070097#ifdef CONFIG_64BIT
Linus Torvalds1da177e2005-04-16 15:20:36 -070098#define USE_DOUBLE
99#endif
100
101#ifdef USE_DOUBLE
102
103#define LOAD ld
104#define LOADL ldl
105#define LOADR ldr
106#define STOREL sdl
107#define STORER sdr
108#define STORE sd
109#define ADD daddu
110#define SUB dsubu
111#define SRL dsrl
112#define SRA dsra
113#define SLL dsll
114#define SLLV dsllv
115#define SRLV dsrlv
116#define NBYTES 8
117#define LOG_NBYTES 3
118
Ralf Baechle42a3b4f2005-09-03 15:56:17 -0700119/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120 * As we are sharing code base with the mips32 tree (which use the o32 ABI
121 * register definitions). We need to redefine the register definitions from
122 * the n64 ABI register naming to the o32 ABI register naming.
123 */
124#undef t0
125#undef t1
126#undef t2
127#undef t3
128#define t0 $8
129#define t1 $9
130#define t2 $10
131#define t3 $11
132#define t4 $12
133#define t5 $13
134#define t6 $14
135#define t7 $15
Ralf Baechle42a3b4f2005-09-03 15:56:17 -0700136
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137#else
138
139#define LOAD lw
140#define LOADL lwl
141#define LOADR lwr
142#define STOREL swl
143#define STORER swr
144#define STORE sw
145#define ADD addu
146#define SUB subu
147#define SRL srl
148#define SLL sll
149#define SRA sra
150#define SLLV sllv
151#define SRLV srlv
152#define NBYTES 4
153#define LOG_NBYTES 2
154
155#endif /* USE_DOUBLE */
156
157#ifdef CONFIG_CPU_LITTLE_ENDIAN
158#define LDFIRST LOADR
159#define LDREST LOADL
160#define STFIRST STORER
161#define STREST STOREL
162#define SHIFT_DISCARD SLLV
163#else
164#define LDFIRST LOADL
165#define LDREST LOADR
166#define STFIRST STOREL
167#define STREST STORER
168#define SHIFT_DISCARD SRLV
169#endif
170
171#define FIRST(unit) ((unit)*NBYTES)
172#define REST(unit) (FIRST(unit)+NBYTES-1)
173#define UNIT(unit) FIRST(unit)
174
175#define ADDRMASK (NBYTES-1)
176
177 .text
178 .set noreorder
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100179#ifndef CONFIG_CPU_DADDI_WORKAROUNDS
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180 .set noat
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100181#else
182 .set at=v1
183#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184
185/*
186 * A combined memcpy/__copy_user
187 * __copy_user sets len to 0 for success; else to an upper bound of
188 * the number of uncopied bytes.
189 * memcpy sets v0 to dst.
190 */
191 .align 5
192LEAF(memcpy) /* a0=dst a1=src a2=len */
193 move v0, dst /* return value */
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000194.L__memcpy:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195FEXPORT(__copy_user)
196 /*
197 * Note: dst & src may be unaligned, len may be 0
198 * Temps
199 */
200#define rem t8
201
Thomas Bogendoerfer930bff82007-11-25 11:47:56 +0100202 R10KCBARRIER(0(ra))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 /*
204 * The "issue break"s below are very approximate.
205 * Issue delays for dcache fills will perturb the schedule, as will
206 * load queue full replay traps, etc.
207 *
208 * If len < NBYTES use byte operations.
209 */
210 PREF( 0, 0(src) )
211 PREF( 1, 0(dst) )
212 sltu t2, len, NBYTES
213 and t1, dst, ADDRMASK
214 PREF( 0, 1*32(src) )
215 PREF( 1, 1*32(dst) )
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000216 bnez t2, .Lcopy_bytes_checklen
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217 and t0, src, ADDRMASK
218 PREF( 0, 2*32(src) )
219 PREF( 1, 2*32(dst) )
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000220 bnez t1, .Ldst_unaligned
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221 nop
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000222 bnez t0, .Lsrc_unaligned_dst_aligned
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223 /*
224 * use delay slot for fall-through
225 * src and dst are aligned; need to compute rem
226 */
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000227.Lboth_aligned:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 SRL t0, len, LOG_NBYTES+3 # +3 for 8 units/iter
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000229 beqz t0, .Lcleanup_both_aligned # len < 8*NBYTES
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 and rem, len, (8*NBYTES-1) # rem = len % (8*NBYTES)
231 PREF( 0, 3*32(src) )
232 PREF( 1, 3*32(dst) )
233 .align 4
2341:
Thomas Bogendoerfer930bff82007-11-25 11:47:56 +0100235 R10KCBARRIER(0(ra))
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000236EXC( LOAD t0, UNIT(0)(src), .Ll_exc)
237EXC( LOAD t1, UNIT(1)(src), .Ll_exc_copy)
238EXC( LOAD t2, UNIT(2)(src), .Ll_exc_copy)
239EXC( LOAD t3, UNIT(3)(src), .Ll_exc_copy)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 SUB len, len, 8*NBYTES
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000241EXC( LOAD t4, UNIT(4)(src), .Ll_exc_copy)
242EXC( LOAD t7, UNIT(5)(src), .Ll_exc_copy)
243EXC( STORE t0, UNIT(0)(dst), .Ls_exc_p8u)
244EXC( STORE t1, UNIT(1)(dst), .Ls_exc_p7u)
245EXC( LOAD t0, UNIT(6)(src), .Ll_exc_copy)
246EXC( LOAD t1, UNIT(7)(src), .Ll_exc_copy)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 ADD src, src, 8*NBYTES
248 ADD dst, dst, 8*NBYTES
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000249EXC( STORE t2, UNIT(-6)(dst), .Ls_exc_p6u)
250EXC( STORE t3, UNIT(-5)(dst), .Ls_exc_p5u)
251EXC( STORE t4, UNIT(-4)(dst), .Ls_exc_p4u)
252EXC( STORE t7, UNIT(-3)(dst), .Ls_exc_p3u)
253EXC( STORE t0, UNIT(-2)(dst), .Ls_exc_p2u)
254EXC( STORE t1, UNIT(-1)(dst), .Ls_exc_p1u)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 PREF( 0, 8*32(src) )
256 PREF( 1, 8*32(dst) )
257 bne len, rem, 1b
258 nop
259
260 /*
261 * len == rem == the number of bytes left to copy < 8*NBYTES
262 */
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000263.Lcleanup_both_aligned:
264 beqz len, .Ldone
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 sltu t0, len, 4*NBYTES
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000266 bnez t0, .Lless_than_4units
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267 and rem, len, (NBYTES-1) # rem = len % NBYTES
268 /*
269 * len >= 4*NBYTES
270 */
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000271EXC( LOAD t0, UNIT(0)(src), .Ll_exc)
272EXC( LOAD t1, UNIT(1)(src), .Ll_exc_copy)
273EXC( LOAD t2, UNIT(2)(src), .Ll_exc_copy)
274EXC( LOAD t3, UNIT(3)(src), .Ll_exc_copy)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 SUB len, len, 4*NBYTES
276 ADD src, src, 4*NBYTES
Thomas Bogendoerfer930bff82007-11-25 11:47:56 +0100277 R10KCBARRIER(0(ra))
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000278EXC( STORE t0, UNIT(0)(dst), .Ls_exc_p4u)
279EXC( STORE t1, UNIT(1)(dst), .Ls_exc_p3u)
280EXC( STORE t2, UNIT(2)(dst), .Ls_exc_p2u)
281EXC( STORE t3, UNIT(3)(dst), .Ls_exc_p1u)
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100282 .set reorder /* DADDI_WAR */
283 ADD dst, dst, 4*NBYTES
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000284 beqz len, .Ldone
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100285 .set noreorder
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000286.Lless_than_4units:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 /*
288 * rem = len % NBYTES
289 */
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000290 beq rem, len, .Lcopy_bytes
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 nop
2921:
Thomas Bogendoerfer930bff82007-11-25 11:47:56 +0100293 R10KCBARRIER(0(ra))
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000294EXC( LOAD t0, 0(src), .Ll_exc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 ADD src, src, NBYTES
296 SUB len, len, NBYTES
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000297EXC( STORE t0, 0(dst), .Ls_exc_p1u)
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100298 .set reorder /* DADDI_WAR */
299 ADD dst, dst, NBYTES
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300 bne rem, len, 1b
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100301 .set noreorder
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302
303 /*
304 * src and dst are aligned, need to copy rem bytes (rem < NBYTES)
305 * A loop would do only a byte at a time with possible branch
306 * mispredicts. Can't do an explicit LOAD dst,mask,or,STORE
307 * because can't assume read-access to dst. Instead, use
308 * STREST dst, which doesn't require read access to dst.
309 *
310 * This code should perform better than a simple loop on modern,
311 * wide-issue mips processors because the code has fewer branches and
312 * more instruction-level parallelism.
313 */
314#define bits t2
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000315 beqz len, .Ldone
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316 ADD t1, dst, len # t1 is just past last byte of dst
317 li bits, 8*NBYTES
318 SLL rem, len, 3 # rem = number of bits to keep
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000319EXC( LOAD t0, 0(src), .Ll_exc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 SUB bits, bits, rem # bits = number of bits to discard
321 SHIFT_DISCARD t0, t0, bits
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000322EXC( STREST t0, -1(t1), .Ls_exc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 jr ra
324 move len, zero
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000325.Ldst_unaligned:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326 /*
327 * dst is unaligned
328 * t0 = src & ADDRMASK
329 * t1 = dst & ADDRMASK; T1 > 0
330 * len >= NBYTES
331 *
332 * Copy enough bytes to align dst
333 * Set match = (src and dst have same alignment)
334 */
335#define match rem
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000336EXC( LDFIRST t3, FIRST(0)(src), .Ll_exc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 ADD t2, zero, NBYTES
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000338EXC( LDREST t3, REST(0)(src), .Ll_exc_copy)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 SUB t2, t2, t1 # t2 = number of bytes copied
340 xor match, t0, t1
Thomas Bogendoerfer930bff82007-11-25 11:47:56 +0100341 R10KCBARRIER(0(ra))
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000342EXC( STFIRST t3, FIRST(0)(dst), .Ls_exc)
343 beq len, t2, .Ldone
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344 SUB len, len, t2
345 ADD dst, dst, t2
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000346 beqz match, .Lboth_aligned
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 ADD src, src, t2
348
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000349.Lsrc_unaligned_dst_aligned:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 SRL t0, len, LOG_NBYTES+2 # +2 for 4 units/iter
351 PREF( 0, 3*32(src) )
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000352 beqz t0, .Lcleanup_src_unaligned
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353 and rem, len, (4*NBYTES-1) # rem = len % 4*NBYTES
354 PREF( 1, 3*32(dst) )
3551:
356/*
357 * Avoid consecutive LD*'s to the same register since some mips
358 * implementations can't issue them in the same cycle.
359 * It's OK to load FIRST(N+1) before REST(N) because the two addresses
360 * are to the same unit (unless src is aligned, but it's not).
361 */
Thomas Bogendoerfer930bff82007-11-25 11:47:56 +0100362 R10KCBARRIER(0(ra))
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000363EXC( LDFIRST t0, FIRST(0)(src), .Ll_exc)
364EXC( LDFIRST t1, FIRST(1)(src), .Ll_exc_copy)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 SUB len, len, 4*NBYTES
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000366EXC( LDREST t0, REST(0)(src), .Ll_exc_copy)
367EXC( LDREST t1, REST(1)(src), .Ll_exc_copy)
368EXC( LDFIRST t2, FIRST(2)(src), .Ll_exc_copy)
369EXC( LDFIRST t3, FIRST(3)(src), .Ll_exc_copy)
370EXC( LDREST t2, REST(2)(src), .Ll_exc_copy)
371EXC( LDREST t3, REST(3)(src), .Ll_exc_copy)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372 PREF( 0, 9*32(src) ) # 0 is PREF_LOAD (not streamed)
373 ADD src, src, 4*NBYTES
374#ifdef CONFIG_CPU_SB1
375 nop # improves slotting
376#endif
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000377EXC( STORE t0, UNIT(0)(dst), .Ls_exc_p4u)
378EXC( STORE t1, UNIT(1)(dst), .Ls_exc_p3u)
379EXC( STORE t2, UNIT(2)(dst), .Ls_exc_p2u)
380EXC( STORE t3, UNIT(3)(dst), .Ls_exc_p1u)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381 PREF( 1, 9*32(dst) ) # 1 is PREF_STORE (not streamed)
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100382 .set reorder /* DADDI_WAR */
383 ADD dst, dst, 4*NBYTES
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 bne len, rem, 1b
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100385 .set noreorder
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000387.Lcleanup_src_unaligned:
388 beqz len, .Ldone
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 and rem, len, NBYTES-1 # rem = len % NBYTES
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000390 beq rem, len, .Lcopy_bytes
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 nop
3921:
Thomas Bogendoerfer930bff82007-11-25 11:47:56 +0100393 R10KCBARRIER(0(ra))
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000394EXC( LDFIRST t0, FIRST(0)(src), .Ll_exc)
395EXC( LDREST t0, REST(0)(src), .Ll_exc_copy)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396 ADD src, src, NBYTES
397 SUB len, len, NBYTES
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000398EXC( STORE t0, 0(dst), .Ls_exc_p1u)
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100399 .set reorder /* DADDI_WAR */
400 ADD dst, dst, NBYTES
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401 bne len, rem, 1b
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100402 .set noreorder
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000404.Lcopy_bytes_checklen:
405 beqz len, .Ldone
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406 nop
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000407.Lcopy_bytes:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 /* 0 < len < NBYTES */
Thomas Bogendoerfer930bff82007-11-25 11:47:56 +0100409 R10KCBARRIER(0(ra))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410#define COPY_BYTE(N) \
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000411EXC( lb t0, N(src), .Ll_exc); \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412 SUB len, len, 1; \
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000413 beqz len, .Ldone; \
414EXC( sb t0, N(dst), .Ls_exc_p1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415
416 COPY_BYTE(0)
417 COPY_BYTE(1)
418#ifdef USE_DOUBLE
419 COPY_BYTE(2)
420 COPY_BYTE(3)
421 COPY_BYTE(4)
422 COPY_BYTE(5)
423#endif
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000424EXC( lb t0, NBYTES-2(src), .Ll_exc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425 SUB len, len, 1
426 jr ra
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000427EXC( sb t0, NBYTES-2(dst), .Ls_exc_p1)
428.Ldone:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429 jr ra
430 nop
431 END(memcpy)
432
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000433.Ll_exc_copy:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434 /*
435 * Copy bytes from src until faulting load address (or until a
436 * lb faults)
437 *
438 * When reached by a faulting LDFIRST/LDREST, THREAD_BUADDR($28)
439 * may be more than a byte beyond the last address.
440 * Hence, the lb below may get an exception.
441 *
442 * Assumes src < THREAD_BUADDR($28)
443 */
444 LOAD t0, TI_TASK($28)
445 nop
446 LOAD t0, THREAD_BUADDR(t0)
4471:
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000448EXC( lb t1, 0(src), .Ll_exc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 ADD src, src, 1
450 sb t1, 0(dst) # can't fault -- we're copy_from_user
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100451 .set reorder /* DADDI_WAR */
452 ADD dst, dst, 1
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 bne src, t0, 1b
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100454 .set noreorder
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000455.Ll_exc:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 LOAD t0, TI_TASK($28)
457 nop
458 LOAD t0, THREAD_BUADDR(t0) # t0 is just past last good address
459 nop
460 SUB len, AT, t0 # len number of uncopied bytes
461 /*
462 * Here's where we rely on src and dst being incremented in tandem,
463 * See (3) above.
464 * dst += (fault addr - src) to put dst at first byte to clear
465 */
466 ADD dst, t0 # compute start address in a1
467 SUB dst, src
468 /*
469 * Clear len bytes starting at dst. Can't call __bzero because it
470 * might modify len. An inefficient loop for these rare times...
471 */
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100472 .set reorder /* DADDI_WAR */
473 SUB src, len, 1
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000474 beqz len, .Ldone
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100475 .set noreorder
Linus Torvalds1da177e2005-04-16 15:20:36 -07004761: sb zero, 0(dst)
477 ADD dst, dst, 1
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100478#ifndef CONFIG_CPU_DADDI_WORKAROUNDS
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 bnez src, 1b
480 SUB src, src, 1
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100481#else
482 .set push
483 .set noat
484 li v1, 1
485 bnez src, 1b
486 SUB src, src, v1
487 .set pop
488#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 jr ra
490 nop
491
492
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100493#define SEXC(n) \
494 .set reorder; /* DADDI_WAR */ \
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000495.Ls_exc_p ## n ## u: \
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100496 ADD len, len, n*NBYTES; \
497 jr ra; \
498 .set noreorder
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499
500SEXC(8)
501SEXC(7)
502SEXC(6)
503SEXC(5)
504SEXC(4)
505SEXC(3)
506SEXC(2)
507SEXC(1)
508
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000509.Ls_exc_p1:
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100510 .set reorder /* DADDI_WAR */
511 ADD len, len, 1
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 jr ra
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100513 .set noreorder
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000514.Ls_exc:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 jr ra
516 nop
517
518 .align 5
519LEAF(memmove)
520 ADD t0, a0, a2
521 ADD t1, a1, a2
522 sltu t0, a1, t0 # dst + len <= src -> memcpy
523 sltu t1, a0, t1 # dst >= src + len -> memcpy
524 and t0, t1
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000525 beqz t0, .L__memcpy
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 move v0, a0 /* return value */
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000527 beqz a2, .Lr_out
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 END(memmove)
529
530 /* fall through to __rmemcpy */
531LEAF(__rmemcpy) /* a0=dst a1=src a2=len */
532 sltu t0, a1, a0
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000533 beqz t0, .Lr_end_bytes_up # src >= dst
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 nop
535 ADD a0, a2 # dst = dst + len
536 ADD a1, a2 # src = src + len
537
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000538.Lr_end_bytes:
Thomas Bogendoerfer930bff82007-11-25 11:47:56 +0100539 R10KCBARRIER(0(ra))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 lb t0, -1(a1)
541 SUB a2, a2, 0x1
542 sb t0, -1(a0)
543 SUB a1, a1, 0x1
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100544 .set reorder /* DADDI_WAR */
545 SUB a0, a0, 0x1
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000546 bnez a2, .Lr_end_bytes
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100547 .set noreorder
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000549.Lr_out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 jr ra
551 move a2, zero
552
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000553.Lr_end_bytes_up:
Thomas Bogendoerfer930bff82007-11-25 11:47:56 +0100554 R10KCBARRIER(0(ra))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 lb t0, (a1)
556 SUB a2, a2, 0x1
557 sb t0, (a0)
558 ADD a1, a1, 0x1
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100559 .set reorder /* DADDI_WAR */
560 ADD a0, a0, 0x1
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000561 bnez a2, .Lr_end_bytes_up
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100562 .set noreorder
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563
564 jr ra
565 move a2, zero
566 END(__rmemcpy)