blob: 03e3304d6ae58b23b238652cd9cf796018d0355b [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
Markos Chandras5bc05972014-01-07 12:57:04 +000013 * Copyright (C) 2014 Imagination Technologies Ltd.
Linus Torvalds1da177e2005-04-16 15:20:36 -070014 *
15 * Mnemonic names for arguments to memcpy/__copy_user
16 */
Ralf Baechlee5adb872005-10-20 22:55:26 +010017
18/*
19 * Hack to resolve longstanding prefetch issue
20 *
21 * Prefetching may be fatal on some systems if we're prefetching beyond the
22 * end of memory on some systems. It's also a seriously bad idea on non
23 * dma-coherent systems.
24 */
Ralf Baechle634286f2009-01-28 17:48:40 +000025#ifdef CONFIG_DMA_NONCOHERENT
Ralf Baechlee5adb872005-10-20 22:55:26 +010026#undef CONFIG_CPU_HAS_PREFETCH
27#endif
28#ifdef CONFIG_MIPS_MALTA
29#undef CONFIG_CPU_HAS_PREFETCH
30#endif
Leonid Yegoshin3daf2812017-06-19 17:50:10 +020031#ifdef CONFIG_CPU_MIPSR6
32#undef CONFIG_CPU_HAS_PREFETCH
33#endif
Ralf Baechlee5adb872005-10-20 22:55:26 +010034
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include <asm/asm.h>
Sam Ravnborg048eb582005-09-09 22:32:31 +020036#include <asm/asm-offsets.h>
Paul Burton576a2f02016-11-07 11:14:15 +000037#include <asm/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070038#include <asm/regdef.h>
39
40#define dst a0
41#define src a1
42#define len a2
43
44/*
45 * Spec
46 *
47 * memcpy copies len bytes from src to dst and sets v0 to dst.
48 * It assumes that
49 * - src and dst don't overlap
50 * - src is readable
51 * - dst is writable
52 * memcpy uses the standard calling convention
53 *
54 * __copy_user copies up to len bytes from src to dst and sets a2 (len) to
55 * the number of uncopied bytes due to an exception caused by a read or write.
56 * __copy_user assumes that src and dst don't overlap, and that the call is
57 * implementing one of the following:
58 * copy_to_user
59 * - src is readable (no exceptions when reading src)
60 * copy_from_user
61 * - dst is writable (no exceptions when writing dst)
62 * __copy_user uses a non-standard calling convention; see
63 * include/asm-mips/uaccess.h
64 *
65 * When an exception happens on a load, the handler must
66 # ensure that all of the destination buffer is overwritten to prevent
67 * leaking information to user mode programs.
68 */
69
70/*
71 * Implementation
72 */
73
74/*
75 * The exception handler for loads requires that:
76 * 1- AT contain the address of the byte just past the end of the source
77 * of the copy,
78 * 2- src_entry <= src < AT, and
79 * 3- (dst - src) == (dst_entry - src_entry),
80 * The _entry suffix denotes values when __copy_user was called.
81 *
82 * (1) is set up up by uaccess.h and maintained by not writing AT in copy_user
83 * (2) is met by incrementing src by the number of bytes copied
84 * (3) is met by not doing loads between a pair of increments of dst and src
85 *
86 * The exception handlers for stores adjust len (if necessary) and return.
87 * These handlers do not need to overwrite any data.
88 *
89 * For __rmemcpy and memmove an exception is always a kernel bug, therefore
90 * they're not protected.
91 */
92
Markos Chandras5bc05972014-01-07 12:57:04 +000093/* Instruction type */
94#define LD_INSN 1
95#define ST_INSN 2
Markos Chandrasbda4d982014-01-07 15:59:03 +000096/* Pretech type */
97#define SRC_PREFETCH 1
98#define DST_PREFETCH 2
Markos Chandrascf62a8b2014-01-07 14:34:05 +000099#define LEGACY_MODE 1
100#define EVA_MODE 2
101#define USEROP 1
102#define KERNELOP 2
Markos Chandras5bc05972014-01-07 12:57:04 +0000103
104/*
105 * Wrapper to add an entry in the exception table
106 * in case the insn causes a memory exception.
107 * Arguments:
108 * insn : Load/store instruction
109 * type : Instruction type
110 * reg : Register
111 * addr : Address
112 * handler : Exception handler
113 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000115#define EXC(insn, type, reg, addr, handler) \
116 .if \mode == LEGACY_MODE; \
1179: insn reg, addr; \
118 .section __ex_table,"a"; \
119 PTR 9b, handler; \
120 .previous; \
Markos Chandrascd26cb42014-01-07 16:20:22 +0000121 /* This is assembled in EVA mode */ \
122 .else; \
123 /* If loading from user or storing to user */ \
124 .if ((\from == USEROP) && (type == LD_INSN)) || \
125 ((\to == USEROP) && (type == ST_INSN)); \
1269: __BUILD_EVA_INSN(insn##e, reg, addr); \
127 .section __ex_table,"a"; \
128 PTR 9b, handler; \
129 .previous; \
130 .else; \
131 /* \
132 * Still in EVA, but no need for \
133 * exception handler or EVA insn \
134 */ \
135 insn reg, addr; \
136 .endif; \
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000137 .endif
Markos Chandrascd26cb42014-01-07 16:20:22 +0000138
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139/*
140 * Only on the 64-bit kernel we can made use of 64-bit registers.
141 */
Ralf Baechle875d43e2005-09-03 15:56:16 -0700142#ifdef CONFIG_64BIT
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143#define USE_DOUBLE
144#endif
145
146#ifdef USE_DOUBLE
147
Markos Chandras5bc05972014-01-07 12:57:04 +0000148#define LOADK ld /* No exception */
149#define LOAD(reg, addr, handler) EXC(ld, LD_INSN, reg, addr, handler)
150#define LOADL(reg, addr, handler) EXC(ldl, LD_INSN, reg, addr, handler)
151#define LOADR(reg, addr, handler) EXC(ldr, LD_INSN, reg, addr, handler)
152#define STOREL(reg, addr, handler) EXC(sdl, ST_INSN, reg, addr, handler)
153#define STORER(reg, addr, handler) EXC(sdr, ST_INSN, reg, addr, handler)
154#define STORE(reg, addr, handler) EXC(sd, ST_INSN, reg, addr, handler)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155#define ADD daddu
156#define SUB dsubu
157#define SRL dsrl
158#define SRA dsra
159#define SLL dsll
160#define SLLV dsllv
161#define SRLV dsrlv
162#define NBYTES 8
163#define LOG_NBYTES 3
164
Ralf Baechle42a3b4f2005-09-03 15:56:17 -0700165/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166 * As we are sharing code base with the mips32 tree (which use the o32 ABI
167 * register definitions). We need to redefine the register definitions from
168 * the n64 ABI register naming to the o32 ABI register naming.
169 */
170#undef t0
171#undef t1
172#undef t2
173#undef t3
174#define t0 $8
175#define t1 $9
176#define t2 $10
177#define t3 $11
178#define t4 $12
179#define t5 $13
180#define t6 $14
181#define t7 $15
Ralf Baechle42a3b4f2005-09-03 15:56:17 -0700182
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183#else
184
Markos Chandras5bc05972014-01-07 12:57:04 +0000185#define LOADK lw /* No exception */
186#define LOAD(reg, addr, handler) EXC(lw, LD_INSN, reg, addr, handler)
187#define LOADL(reg, addr, handler) EXC(lwl, LD_INSN, reg, addr, handler)
188#define LOADR(reg, addr, handler) EXC(lwr, LD_INSN, reg, addr, handler)
189#define STOREL(reg, addr, handler) EXC(swl, ST_INSN, reg, addr, handler)
190#define STORER(reg, addr, handler) EXC(swr, ST_INSN, reg, addr, handler)
191#define STORE(reg, addr, handler) EXC(sw, ST_INSN, reg, addr, handler)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192#define ADD addu
193#define SUB subu
194#define SRL srl
195#define SLL sll
196#define SRA sra
197#define SLLV sllv
198#define SRLV srlv
199#define NBYTES 4
200#define LOG_NBYTES 2
201
202#endif /* USE_DOUBLE */
203
Markos Chandras5bc05972014-01-07 12:57:04 +0000204#define LOADB(reg, addr, handler) EXC(lb, LD_INSN, reg, addr, handler)
205#define STOREB(reg, addr, handler) EXC(sb, ST_INSN, reg, addr, handler)
206
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000207#define _PREF(hint, addr, type) \
208 .if \mode == LEGACY_MODE; \
209 PREF(hint, addr); \
Markos Chandrascd26cb42014-01-07 16:20:22 +0000210 .else; \
211 .if ((\from == USEROP) && (type == SRC_PREFETCH)) || \
212 ((\to == USEROP) && (type == DST_PREFETCH)); \
213 /* \
214 * PREFE has only 9 bits for the offset \
215 * compared to PREF which has 16, so it may \
216 * need to use the $at register but this \
217 * register should remain intact because it's \
218 * used later on. Therefore use $v1. \
219 */ \
220 .set at=v1; \
221 PREFE(hint, addr); \
222 .set noat; \
223 .else; \
224 PREF(hint, addr); \
225 .endif; \
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000226 .endif
Markos Chandrasbda4d982014-01-07 15:59:03 +0000227
228#define PREFS(hint, addr) _PREF(hint, addr, SRC_PREFETCH)
229#define PREFD(hint, addr) _PREF(hint, addr, DST_PREFETCH)
230
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231#ifdef CONFIG_CPU_LITTLE_ENDIAN
232#define LDFIRST LOADR
Ralf Baechle70342282013-01-22 12:59:30 +0100233#define LDREST LOADL
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234#define STFIRST STORER
Ralf Baechle70342282013-01-22 12:59:30 +0100235#define STREST STOREL
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236#define SHIFT_DISCARD SLLV
237#else
238#define LDFIRST LOADL
Ralf Baechle70342282013-01-22 12:59:30 +0100239#define LDREST LOADR
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240#define STFIRST STOREL
Ralf Baechle70342282013-01-22 12:59:30 +0100241#define STREST STORER
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242#define SHIFT_DISCARD SRLV
243#endif
244
245#define FIRST(unit) ((unit)*NBYTES)
246#define REST(unit) (FIRST(unit)+NBYTES-1)
247#define UNIT(unit) FIRST(unit)
248
249#define ADDRMASK (NBYTES-1)
250
251 .text
252 .set noreorder
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100253#ifndef CONFIG_CPU_DADDI_WORKAROUNDS
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 .set noat
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100255#else
256 .set at=v1
257#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 .align 5
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000260
261 /*
262 * Macro to build the __copy_user common code
Andrea Gelmini209b8772016-05-21 14:01:36 +0200263 * Arguments:
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000264 * mode : LEGACY_MODE or EVA_MODE
265 * from : Source operand. USEROP or KERNELOP
266 * to : Destination operand. USEROP or KERNELOP
267 */
268 .macro __BUILD_COPY_USER mode, from, to
269
270 /* initialize __memcpy if this the first time we execute this macro */
271 .ifnotdef __memcpy
272 .set __memcpy, 1
273 .hidden __memcpy /* make sure it does not leak */
274 .endif
275
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 /*
277 * Note: dst & src may be unaligned, len may be 0
278 * Temps
279 */
280#define rem t8
281
Thomas Bogendoerfer930bff82007-11-25 11:47:56 +0100282 R10KCBARRIER(0(ra))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283 /*
284 * The "issue break"s below are very approximate.
285 * Issue delays for dcache fills will perturb the schedule, as will
286 * load queue full replay traps, etc.
287 *
288 * If len < NBYTES use byte operations.
289 */
Markos Chandrasbda4d982014-01-07 15:59:03 +0000290 PREFS( 0, 0(src) )
291 PREFD( 1, 0(dst) )
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292 sltu t2, len, NBYTES
293 and t1, dst, ADDRMASK
Markos Chandrasbda4d982014-01-07 15:59:03 +0000294 PREFS( 0, 1*32(src) )
295 PREFD( 1, 1*32(dst) )
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000296 bnez t2, .Lcopy_bytes_checklen\@
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297 and t0, src, ADDRMASK
Markos Chandrasbda4d982014-01-07 15:59:03 +0000298 PREFS( 0, 2*32(src) )
299 PREFD( 1, 2*32(dst) )
Leonid Yegoshinb0ce4bd2014-11-14 11:55:50 +0000300#ifndef CONFIG_CPU_MIPSR6
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000301 bnez t1, .Ldst_unaligned\@
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302 nop
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000303 bnez t0, .Lsrc_unaligned_dst_aligned\@
Leonid Yegoshinb0ce4bd2014-11-14 11:55:50 +0000304#else
305 or t0, t0, t1
306 bnez t0, .Lcopy_unaligned_bytes\@
307#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308 /*
309 * use delay slot for fall-through
310 * src and dst are aligned; need to compute rem
311 */
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000312.Lboth_aligned\@:
Ralf Baechle70342282013-01-22 12:59:30 +0100313 SRL t0, len, LOG_NBYTES+3 # +3 for 8 units/iter
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000314 beqz t0, .Lcleanup_both_aligned\@ # len < 8*NBYTES
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 and rem, len, (8*NBYTES-1) # rem = len % (8*NBYTES)
Markos Chandrasbda4d982014-01-07 15:59:03 +0000316 PREFS( 0, 3*32(src) )
317 PREFD( 1, 3*32(dst) )
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318 .align 4
3191:
Thomas Bogendoerfer930bff82007-11-25 11:47:56 +0100320 R10KCBARRIER(0(ra))
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000321 LOAD(t0, UNIT(0)(src), .Ll_exc\@)
322 LOAD(t1, UNIT(1)(src), .Ll_exc_copy\@)
323 LOAD(t2, UNIT(2)(src), .Ll_exc_copy\@)
324 LOAD(t3, UNIT(3)(src), .Ll_exc_copy\@)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325 SUB len, len, 8*NBYTES
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000326 LOAD(t4, UNIT(4)(src), .Ll_exc_copy\@)
327 LOAD(t7, UNIT(5)(src), .Ll_exc_copy\@)
328 STORE(t0, UNIT(0)(dst), .Ls_exc_p8u\@)
329 STORE(t1, UNIT(1)(dst), .Ls_exc_p7u\@)
330 LOAD(t0, UNIT(6)(src), .Ll_exc_copy\@)
331 LOAD(t1, UNIT(7)(src), .Ll_exc_copy\@)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 ADD src, src, 8*NBYTES
333 ADD dst, dst, 8*NBYTES
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000334 STORE(t2, UNIT(-6)(dst), .Ls_exc_p6u\@)
335 STORE(t3, UNIT(-5)(dst), .Ls_exc_p5u\@)
336 STORE(t4, UNIT(-4)(dst), .Ls_exc_p4u\@)
337 STORE(t7, UNIT(-3)(dst), .Ls_exc_p3u\@)
338 STORE(t0, UNIT(-2)(dst), .Ls_exc_p2u\@)
339 STORE(t1, UNIT(-1)(dst), .Ls_exc_p1u\@)
Markos Chandrasbda4d982014-01-07 15:59:03 +0000340 PREFS( 0, 8*32(src) )
341 PREFD( 1, 8*32(dst) )
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 bne len, rem, 1b
343 nop
344
345 /*
346 * len == rem == the number of bytes left to copy < 8*NBYTES
347 */
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000348.Lcleanup_both_aligned\@:
349 beqz len, .Ldone\@
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 sltu t0, len, 4*NBYTES
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000351 bnez t0, .Lless_than_4units\@
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352 and rem, len, (NBYTES-1) # rem = len % NBYTES
353 /*
354 * len >= 4*NBYTES
355 */
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000356 LOAD( t0, UNIT(0)(src), .Ll_exc\@)
357 LOAD( t1, UNIT(1)(src), .Ll_exc_copy\@)
358 LOAD( t2, UNIT(2)(src), .Ll_exc_copy\@)
359 LOAD( t3, UNIT(3)(src), .Ll_exc_copy\@)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360 SUB len, len, 4*NBYTES
361 ADD src, src, 4*NBYTES
Thomas Bogendoerfer930bff82007-11-25 11:47:56 +0100362 R10KCBARRIER(0(ra))
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000363 STORE(t0, UNIT(0)(dst), .Ls_exc_p4u\@)
364 STORE(t1, UNIT(1)(dst), .Ls_exc_p3u\@)
365 STORE(t2, UNIT(2)(dst), .Ls_exc_p2u\@)
366 STORE(t3, UNIT(3)(dst), .Ls_exc_p1u\@)
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100367 .set reorder /* DADDI_WAR */
368 ADD dst, dst, 4*NBYTES
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000369 beqz len, .Ldone\@
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100370 .set noreorder
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000371.Lless_than_4units\@:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372 /*
373 * rem = len % NBYTES
374 */
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000375 beq rem, len, .Lcopy_bytes\@
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376 nop
3771:
Thomas Bogendoerfer930bff82007-11-25 11:47:56 +0100378 R10KCBARRIER(0(ra))
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000379 LOAD(t0, 0(src), .Ll_exc\@)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380 ADD src, src, NBYTES
381 SUB len, len, NBYTES
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000382 STORE(t0, 0(dst), .Ls_exc_p1u\@)
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100383 .set reorder /* DADDI_WAR */
384 ADD dst, dst, NBYTES
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 bne rem, len, 1b
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100386 .set noreorder
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387
Leonid Yegoshinb0ce4bd2014-11-14 11:55:50 +0000388#ifndef CONFIG_CPU_MIPSR6
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 /*
390 * src and dst are aligned, need to copy rem bytes (rem < NBYTES)
391 * A loop would do only a byte at a time with possible branch
Ralf Baechle70342282013-01-22 12:59:30 +0100392 * mispredicts. Can't do an explicit LOAD dst,mask,or,STORE
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 * because can't assume read-access to dst. Instead, use
394 * STREST dst, which doesn't require read access to dst.
395 *
396 * This code should perform better than a simple loop on modern,
397 * wide-issue mips processors because the code has fewer branches and
398 * more instruction-level parallelism.
399 */
400#define bits t2
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000401 beqz len, .Ldone\@
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 ADD t1, dst, len # t1 is just past last byte of dst
403 li bits, 8*NBYTES
404 SLL rem, len, 3 # rem = number of bits to keep
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000405 LOAD(t0, 0(src), .Ll_exc\@)
Ralf Baechle70342282013-01-22 12:59:30 +0100406 SUB bits, bits, rem # bits = number of bits to discard
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407 SHIFT_DISCARD t0, t0, bits
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000408 STREST(t0, -1(t1), .Ls_exc\@)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 jr ra
410 move len, zero
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000411.Ldst_unaligned\@:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412 /*
413 * dst is unaligned
414 * t0 = src & ADDRMASK
415 * t1 = dst & ADDRMASK; T1 > 0
416 * len >= NBYTES
417 *
418 * Copy enough bytes to align dst
419 * Set match = (src and dst have same alignment)
420 */
421#define match rem
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000422 LDFIRST(t3, FIRST(0)(src), .Ll_exc\@)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423 ADD t2, zero, NBYTES
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000424 LDREST(t3, REST(0)(src), .Ll_exc_copy\@)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425 SUB t2, t2, t1 # t2 = number of bytes copied
426 xor match, t0, t1
Thomas Bogendoerfer930bff82007-11-25 11:47:56 +0100427 R10KCBARRIER(0(ra))
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000428 STFIRST(t3, FIRST(0)(dst), .Ls_exc\@)
429 beq len, t2, .Ldone\@
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 SUB len, len, t2
431 ADD dst, dst, t2
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000432 beqz match, .Lboth_aligned\@
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 ADD src, src, t2
434
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000435.Lsrc_unaligned_dst_aligned\@:
Ralf Baechle70342282013-01-22 12:59:30 +0100436 SRL t0, len, LOG_NBYTES+2 # +2 for 4 units/iter
Markos Chandrasbda4d982014-01-07 15:59:03 +0000437 PREFS( 0, 3*32(src) )
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000438 beqz t0, .Lcleanup_src_unaligned\@
Ralf Baechle70342282013-01-22 12:59:30 +0100439 and rem, len, (4*NBYTES-1) # rem = len % 4*NBYTES
Markos Chandrasbda4d982014-01-07 15:59:03 +0000440 PREFD( 1, 3*32(dst) )
Linus Torvalds1da177e2005-04-16 15:20:36 -07004411:
442/*
443 * Avoid consecutive LD*'s to the same register since some mips
444 * implementations can't issue them in the same cycle.
445 * It's OK to load FIRST(N+1) before REST(N) because the two addresses
446 * are to the same unit (unless src is aligned, but it's not).
447 */
Thomas Bogendoerfer930bff82007-11-25 11:47:56 +0100448 R10KCBARRIER(0(ra))
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000449 LDFIRST(t0, FIRST(0)(src), .Ll_exc\@)
450 LDFIRST(t1, FIRST(1)(src), .Ll_exc_copy\@)
Ralf Baechle70342282013-01-22 12:59:30 +0100451 SUB len, len, 4*NBYTES
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000452 LDREST(t0, REST(0)(src), .Ll_exc_copy\@)
453 LDREST(t1, REST(1)(src), .Ll_exc_copy\@)
454 LDFIRST(t2, FIRST(2)(src), .Ll_exc_copy\@)
455 LDFIRST(t3, FIRST(3)(src), .Ll_exc_copy\@)
456 LDREST(t2, REST(2)(src), .Ll_exc_copy\@)
457 LDREST(t3, REST(3)(src), .Ll_exc_copy\@)
Markos Chandrasbda4d982014-01-07 15:59:03 +0000458 PREFS( 0, 9*32(src) ) # 0 is PREF_LOAD (not streamed)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 ADD src, src, 4*NBYTES
460#ifdef CONFIG_CPU_SB1
461 nop # improves slotting
462#endif
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000463 STORE(t0, UNIT(0)(dst), .Ls_exc_p4u\@)
464 STORE(t1, UNIT(1)(dst), .Ls_exc_p3u\@)
465 STORE(t2, UNIT(2)(dst), .Ls_exc_p2u\@)
466 STORE(t3, UNIT(3)(dst), .Ls_exc_p1u\@)
Markos Chandrasbda4d982014-01-07 15:59:03 +0000467 PREFD( 1, 9*32(dst) ) # 1 is PREF_STORE (not streamed)
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100468 .set reorder /* DADDI_WAR */
469 ADD dst, dst, 4*NBYTES
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470 bne len, rem, 1b
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100471 .set noreorder
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000473.Lcleanup_src_unaligned\@:
474 beqz len, .Ldone\@
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 and rem, len, NBYTES-1 # rem = len % NBYTES
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000476 beq rem, len, .Lcopy_bytes\@
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 nop
4781:
Thomas Bogendoerfer930bff82007-11-25 11:47:56 +0100479 R10KCBARRIER(0(ra))
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000480 LDFIRST(t0, FIRST(0)(src), .Ll_exc\@)
481 LDREST(t0, REST(0)(src), .Ll_exc_copy\@)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 ADD src, src, NBYTES
483 SUB len, len, NBYTES
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000484 STORE(t0, 0(dst), .Ls_exc_p1u\@)
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100485 .set reorder /* DADDI_WAR */
486 ADD dst, dst, NBYTES
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487 bne len, rem, 1b
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100488 .set noreorder
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489
Leonid Yegoshinb0ce4bd2014-11-14 11:55:50 +0000490#endif /* !CONFIG_CPU_MIPSR6 */
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000491.Lcopy_bytes_checklen\@:
492 beqz len, .Ldone\@
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493 nop
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000494.Lcopy_bytes\@:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 /* 0 < len < NBYTES */
Thomas Bogendoerfer930bff82007-11-25 11:47:56 +0100496 R10KCBARRIER(0(ra))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497#define COPY_BYTE(N) \
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000498 LOADB(t0, N(src), .Ll_exc\@); \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 SUB len, len, 1; \
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000500 beqz len, .Ldone\@; \
501 STOREB(t0, N(dst), .Ls_exc_p1\@)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502
503 COPY_BYTE(0)
504 COPY_BYTE(1)
505#ifdef USE_DOUBLE
506 COPY_BYTE(2)
507 COPY_BYTE(3)
508 COPY_BYTE(4)
509 COPY_BYTE(5)
510#endif
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000511 LOADB(t0, NBYTES-2(src), .Ll_exc\@)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 SUB len, len, 1
513 jr ra
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000514 STOREB(t0, NBYTES-2(dst), .Ls_exc_p1\@)
515.Ldone\@:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 jr ra
Markos Chandras51b10292014-11-17 09:32:38 +0000517 nop
Leonid Yegoshinb0ce4bd2014-11-14 11:55:50 +0000518
519#ifdef CONFIG_CPU_MIPSR6
520.Lcopy_unaligned_bytes\@:
5211:
522 COPY_BYTE(0)
523 COPY_BYTE(1)
524 COPY_BYTE(2)
525 COPY_BYTE(3)
526 COPY_BYTE(4)
527 COPY_BYTE(5)
528 COPY_BYTE(6)
529 COPY_BYTE(7)
530 ADD src, src, 8
531 b 1b
532 ADD dst, dst, 8
533#endif /* CONFIG_CPU_MIPSR6 */
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000534 .if __memcpy == 1
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535 END(memcpy)
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000536 .set __memcpy, 0
537 .hidden __memcpy
538 .endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000540.Ll_exc_copy\@:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 /*
542 * Copy bytes from src until faulting load address (or until a
543 * lb faults)
544 *
545 * When reached by a faulting LDFIRST/LDREST, THREAD_BUADDR($28)
546 * may be more than a byte beyond the last address.
547 * Hence, the lb below may get an exception.
548 *
549 * Assumes src < THREAD_BUADDR($28)
550 */
Markos Chandras5bc05972014-01-07 12:57:04 +0000551 LOADK t0, TI_TASK($28)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 nop
Markos Chandras5bc05972014-01-07 12:57:04 +0000553 LOADK t0, THREAD_BUADDR(t0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005541:
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000555 LOADB(t1, 0(src), .Ll_exc\@)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 ADD src, src, 1
557 sb t1, 0(dst) # can't fault -- we're copy_from_user
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100558 .set reorder /* DADDI_WAR */
559 ADD dst, dst, 1
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 bne src, t0, 1b
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100561 .set noreorder
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000562.Ll_exc\@:
Markos Chandras5bc05972014-01-07 12:57:04 +0000563 LOADK t0, TI_TASK($28)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 nop
Markos Chandras5bc05972014-01-07 12:57:04 +0000565 LOADK t0, THREAD_BUADDR(t0) # t0 is just past last good address
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 nop
567 SUB len, AT, t0 # len number of uncopied bytes
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 jr ra
569 nop
570
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100571#define SEXC(n) \
Ralf Baechle70342282013-01-22 12:59:30 +0100572 .set reorder; /* DADDI_WAR */ \
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000573.Ls_exc_p ## n ## u\@: \
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100574 ADD len, len, n*NBYTES; \
575 jr ra; \
576 .set noreorder
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577
578SEXC(8)
579SEXC(7)
580SEXC(6)
581SEXC(5)
582SEXC(4)
583SEXC(3)
584SEXC(2)
585SEXC(1)
586
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000587.Ls_exc_p1\@:
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100588 .set reorder /* DADDI_WAR */
589 ADD len, len, 1
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 jr ra
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100591 .set noreorder
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000592.Ls_exc\@:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 jr ra
594 nop
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000595 .endm
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596
597 .align 5
598LEAF(memmove)
Paul Burton576a2f02016-11-07 11:14:15 +0000599EXPORT_SYMBOL(memmove)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 ADD t0, a0, a2
601 ADD t1, a1, a2
602 sltu t0, a1, t0 # dst + len <= src -> memcpy
603 sltu t1, a0, t1 # dst >= src + len -> memcpy
604 and t0, t1
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000605 beqz t0, .L__memcpy
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 move v0, a0 /* return value */
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000607 beqz a2, .Lr_out
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 END(memmove)
609
610 /* fall through to __rmemcpy */
611LEAF(__rmemcpy) /* a0=dst a1=src a2=len */
612 sltu t0, a1, a0
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000613 beqz t0, .Lr_end_bytes_up # src >= dst
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 nop
615 ADD a0, a2 # dst = dst + len
616 ADD a1, a2 # src = src + len
617
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000618.Lr_end_bytes:
Thomas Bogendoerfer930bff82007-11-25 11:47:56 +0100619 R10KCBARRIER(0(ra))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 lb t0, -1(a1)
621 SUB a2, a2, 0x1
622 sb t0, -1(a0)
623 SUB a1, a1, 0x1
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100624 .set reorder /* DADDI_WAR */
625 SUB a0, a0, 0x1
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000626 bnez a2, .Lr_end_bytes
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100627 .set noreorder
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000629.Lr_out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630 jr ra
631 move a2, zero
632
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000633.Lr_end_bytes_up:
Thomas Bogendoerfer930bff82007-11-25 11:47:56 +0100634 R10KCBARRIER(0(ra))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 lb t0, (a1)
636 SUB a2, a2, 0x1
637 sb t0, (a0)
638 ADD a1, a1, 0x1
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100639 .set reorder /* DADDI_WAR */
640 ADD a0, a0, 0x1
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000641 bnez a2, .Lr_end_bytes_up
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100642 .set noreorder
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643
644 jr ra
645 move a2, zero
646 END(__rmemcpy)
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000647
648/*
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000649 * A combined memcpy/__copy_user
650 * __copy_user sets len to 0 for success; else to an upper bound of
651 * the number of uncopied bytes.
652 * memcpy sets v0 to dst.
653 */
654 .align 5
655LEAF(memcpy) /* a0=dst a1=src a2=len */
Paul Burton576a2f02016-11-07 11:14:15 +0000656EXPORT_SYMBOL(memcpy)
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000657 move v0, dst /* return value */
658.L__memcpy:
659FEXPORT(__copy_user)
Paul Burton576a2f02016-11-07 11:14:15 +0000660EXPORT_SYMBOL(__copy_user)
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000661 /* Legacy Mode, user <-> user */
662 __BUILD_COPY_USER LEGACY_MODE USEROP USEROP
Markos Chandrascd26cb42014-01-07 16:20:22 +0000663
664#ifdef CONFIG_EVA
665
666/*
667 * For EVA we need distinct symbols for reading and writing to user space.
668 * This is because we need to use specific EVA instructions to perform the
669 * virtual <-> physical translation when a virtual address is actually in user
670 * space
671 */
672
Markos Chandrascd26cb42014-01-07 16:20:22 +0000673/*
674 * __copy_from_user (EVA)
675 */
676
677LEAF(__copy_from_user_eva)
Paul Burton576a2f02016-11-07 11:14:15 +0000678EXPORT_SYMBOL(__copy_from_user_eva)
Markos Chandrascd26cb42014-01-07 16:20:22 +0000679 __BUILD_COPY_USER EVA_MODE USEROP KERNELOP
680END(__copy_from_user_eva)
681
682
683
684/*
685 * __copy_to_user (EVA)
686 */
687
688LEAF(__copy_to_user_eva)
Paul Burton576a2f02016-11-07 11:14:15 +0000689EXPORT_SYMBOL(__copy_to_user_eva)
Markos Chandrascd26cb42014-01-07 16:20:22 +0000690__BUILD_COPY_USER EVA_MODE KERNELOP USEROP
691END(__copy_to_user_eva)
692
693/*
694 * __copy_in_user (EVA)
695 */
696
697LEAF(__copy_in_user_eva)
Paul Burton576a2f02016-11-07 11:14:15 +0000698EXPORT_SYMBOL(__copy_in_user_eva)
Markos Chandrascd26cb42014-01-07 16:20:22 +0000699__BUILD_COPY_USER EVA_MODE USEROP USEROP
700END(__copy_in_user_eva)
701
702#endif