blob: 3114a2ed1f4e54f797233f48d8793ba4ebfca21c [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
31
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include <asm/asm.h>
Sam Ravnborg048eb582005-09-09 22:32:31 +020033#include <asm/asm-offsets.h>
Paul Burton576a2f02016-11-07 11:14:15 +000034#include <asm/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include <asm/regdef.h>
36
37#define dst a0
38#define src a1
39#define len a2
40
41/*
42 * Spec
43 *
44 * memcpy copies len bytes from src to dst and sets v0 to dst.
45 * It assumes that
46 * - src and dst don't overlap
47 * - src is readable
48 * - dst is writable
49 * memcpy uses the standard calling convention
50 *
51 * __copy_user copies up to len bytes from src to dst and sets a2 (len) to
52 * the number of uncopied bytes due to an exception caused by a read or write.
53 * __copy_user assumes that src and dst don't overlap, and that the call is
54 * implementing one of the following:
55 * copy_to_user
56 * - src is readable (no exceptions when reading src)
57 * copy_from_user
58 * - dst is writable (no exceptions when writing dst)
59 * __copy_user uses a non-standard calling convention; see
60 * include/asm-mips/uaccess.h
61 *
62 * When an exception happens on a load, the handler must
63 # ensure that all of the destination buffer is overwritten to prevent
64 * leaking information to user mode programs.
65 */
66
67/*
68 * Implementation
69 */
70
71/*
72 * The exception handler for loads requires that:
73 * 1- AT contain the address of the byte just past the end of the source
74 * of the copy,
75 * 2- src_entry <= src < AT, and
76 * 3- (dst - src) == (dst_entry - src_entry),
77 * The _entry suffix denotes values when __copy_user was called.
78 *
79 * (1) is set up up by uaccess.h and maintained by not writing AT in copy_user
80 * (2) is met by incrementing src by the number of bytes copied
81 * (3) is met by not doing loads between a pair of increments of dst and src
82 *
83 * The exception handlers for stores adjust len (if necessary) and return.
84 * These handlers do not need to overwrite any data.
85 *
86 * For __rmemcpy and memmove an exception is always a kernel bug, therefore
87 * they're not protected.
88 */
89
Markos Chandras5bc05972014-01-07 12:57:04 +000090/* Instruction type */
91#define LD_INSN 1
92#define ST_INSN 2
Markos Chandrasbda4d982014-01-07 15:59:03 +000093/* Pretech type */
94#define SRC_PREFETCH 1
95#define DST_PREFETCH 2
Markos Chandrascf62a8b2014-01-07 14:34:05 +000096#define LEGACY_MODE 1
97#define EVA_MODE 2
98#define USEROP 1
99#define KERNELOP 2
Markos Chandras5bc05972014-01-07 12:57:04 +0000100
101/*
102 * Wrapper to add an entry in the exception table
103 * in case the insn causes a memory exception.
104 * Arguments:
105 * insn : Load/store instruction
106 * type : Instruction type
107 * reg : Register
108 * addr : Address
109 * handler : Exception handler
110 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000112#define EXC(insn, type, reg, addr, handler) \
113 .if \mode == LEGACY_MODE; \
1149: insn reg, addr; \
115 .section __ex_table,"a"; \
116 PTR 9b, handler; \
117 .previous; \
Markos Chandrascd26cb42014-01-07 16:20:22 +0000118 /* This is assembled in EVA mode */ \
119 .else; \
120 /* If loading from user or storing to user */ \
121 .if ((\from == USEROP) && (type == LD_INSN)) || \
122 ((\to == USEROP) && (type == ST_INSN)); \
1239: __BUILD_EVA_INSN(insn##e, reg, addr); \
124 .section __ex_table,"a"; \
125 PTR 9b, handler; \
126 .previous; \
127 .else; \
128 /* \
129 * Still in EVA, but no need for \
130 * exception handler or EVA insn \
131 */ \
132 insn reg, addr; \
133 .endif; \
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000134 .endif
Markos Chandrascd26cb42014-01-07 16:20:22 +0000135
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136/*
137 * Only on the 64-bit kernel we can made use of 64-bit registers.
138 */
Ralf Baechle875d43e2005-09-03 15:56:16 -0700139#ifdef CONFIG_64BIT
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140#define USE_DOUBLE
141#endif
142
143#ifdef USE_DOUBLE
144
Markos Chandras5bc05972014-01-07 12:57:04 +0000145#define LOADK ld /* No exception */
146#define LOAD(reg, addr, handler) EXC(ld, LD_INSN, reg, addr, handler)
147#define LOADL(reg, addr, handler) EXC(ldl, LD_INSN, reg, addr, handler)
148#define LOADR(reg, addr, handler) EXC(ldr, LD_INSN, reg, addr, handler)
149#define STOREL(reg, addr, handler) EXC(sdl, ST_INSN, reg, addr, handler)
150#define STORER(reg, addr, handler) EXC(sdr, ST_INSN, reg, addr, handler)
151#define STORE(reg, addr, handler) EXC(sd, ST_INSN, reg, addr, handler)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152#define ADD daddu
153#define SUB dsubu
154#define SRL dsrl
155#define SRA dsra
156#define SLL dsll
157#define SLLV dsllv
158#define SRLV dsrlv
159#define NBYTES 8
160#define LOG_NBYTES 3
161
Ralf Baechle42a3b4f2005-09-03 15:56:17 -0700162/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163 * As we are sharing code base with the mips32 tree (which use the o32 ABI
164 * register definitions). We need to redefine the register definitions from
165 * the n64 ABI register naming to the o32 ABI register naming.
166 */
167#undef t0
168#undef t1
169#undef t2
170#undef t3
171#define t0 $8
172#define t1 $9
173#define t2 $10
174#define t3 $11
175#define t4 $12
176#define t5 $13
177#define t6 $14
178#define t7 $15
Ralf Baechle42a3b4f2005-09-03 15:56:17 -0700179
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180#else
181
Markos Chandras5bc05972014-01-07 12:57:04 +0000182#define LOADK lw /* No exception */
183#define LOAD(reg, addr, handler) EXC(lw, LD_INSN, reg, addr, handler)
184#define LOADL(reg, addr, handler) EXC(lwl, LD_INSN, reg, addr, handler)
185#define LOADR(reg, addr, handler) EXC(lwr, LD_INSN, reg, addr, handler)
186#define STOREL(reg, addr, handler) EXC(swl, ST_INSN, reg, addr, handler)
187#define STORER(reg, addr, handler) EXC(swr, ST_INSN, reg, addr, handler)
188#define STORE(reg, addr, handler) EXC(sw, ST_INSN, reg, addr, handler)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189#define ADD addu
190#define SUB subu
191#define SRL srl
192#define SLL sll
193#define SRA sra
194#define SLLV sllv
195#define SRLV srlv
196#define NBYTES 4
197#define LOG_NBYTES 2
198
199#endif /* USE_DOUBLE */
200
Markos Chandras5bc05972014-01-07 12:57:04 +0000201#define LOADB(reg, addr, handler) EXC(lb, LD_INSN, reg, addr, handler)
202#define STOREB(reg, addr, handler) EXC(sb, ST_INSN, reg, addr, handler)
203
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000204#define _PREF(hint, addr, type) \
205 .if \mode == LEGACY_MODE; \
206 PREF(hint, addr); \
Markos Chandrascd26cb42014-01-07 16:20:22 +0000207 .else; \
208 .if ((\from == USEROP) && (type == SRC_PREFETCH)) || \
209 ((\to == USEROP) && (type == DST_PREFETCH)); \
210 /* \
211 * PREFE has only 9 bits for the offset \
212 * compared to PREF which has 16, so it may \
213 * need to use the $at register but this \
214 * register should remain intact because it's \
215 * used later on. Therefore use $v1. \
216 */ \
217 .set at=v1; \
218 PREFE(hint, addr); \
219 .set noat; \
220 .else; \
221 PREF(hint, addr); \
222 .endif; \
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000223 .endif
Markos Chandrasbda4d982014-01-07 15:59:03 +0000224
225#define PREFS(hint, addr) _PREF(hint, addr, SRC_PREFETCH)
226#define PREFD(hint, addr) _PREF(hint, addr, DST_PREFETCH)
227
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228#ifdef CONFIG_CPU_LITTLE_ENDIAN
229#define LDFIRST LOADR
Ralf Baechle70342282013-01-22 12:59:30 +0100230#define LDREST LOADL
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231#define STFIRST STORER
Ralf Baechle70342282013-01-22 12:59:30 +0100232#define STREST STOREL
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233#define SHIFT_DISCARD SLLV
234#else
235#define LDFIRST LOADL
Ralf Baechle70342282013-01-22 12:59:30 +0100236#define LDREST LOADR
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237#define STFIRST STOREL
Ralf Baechle70342282013-01-22 12:59:30 +0100238#define STREST STORER
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239#define SHIFT_DISCARD SRLV
240#endif
241
242#define FIRST(unit) ((unit)*NBYTES)
243#define REST(unit) (FIRST(unit)+NBYTES-1)
244#define UNIT(unit) FIRST(unit)
245
246#define ADDRMASK (NBYTES-1)
247
248 .text
249 .set noreorder
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100250#ifndef CONFIG_CPU_DADDI_WORKAROUNDS
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251 .set noat
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100252#else
253 .set at=v1
254#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 .align 5
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000257
258 /*
259 * Macro to build the __copy_user common code
Andrea Gelmini209b8772016-05-21 14:01:36 +0200260 * Arguments:
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000261 * mode : LEGACY_MODE or EVA_MODE
262 * from : Source operand. USEROP or KERNELOP
263 * to : Destination operand. USEROP or KERNELOP
264 */
265 .macro __BUILD_COPY_USER mode, from, to
266
267 /* initialize __memcpy if this the first time we execute this macro */
268 .ifnotdef __memcpy
269 .set __memcpy, 1
270 .hidden __memcpy /* make sure it does not leak */
271 .endif
272
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 /*
274 * Note: dst & src may be unaligned, len may be 0
275 * Temps
276 */
277#define rem t8
278
Thomas Bogendoerfer930bff82007-11-25 11:47:56 +0100279 R10KCBARRIER(0(ra))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280 /*
281 * The "issue break"s below are very approximate.
282 * Issue delays for dcache fills will perturb the schedule, as will
283 * load queue full replay traps, etc.
284 *
285 * If len < NBYTES use byte operations.
286 */
Markos Chandrasbda4d982014-01-07 15:59:03 +0000287 PREFS( 0, 0(src) )
288 PREFD( 1, 0(dst) )
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289 sltu t2, len, NBYTES
290 and t1, dst, ADDRMASK
Markos Chandrasbda4d982014-01-07 15:59:03 +0000291 PREFS( 0, 1*32(src) )
292 PREFD( 1, 1*32(dst) )
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000293 bnez t2, .Lcopy_bytes_checklen\@
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294 and t0, src, ADDRMASK
Markos Chandrasbda4d982014-01-07 15:59:03 +0000295 PREFS( 0, 2*32(src) )
296 PREFD( 1, 2*32(dst) )
Leonid Yegoshinb0ce4bd2014-11-14 11:55:50 +0000297#ifndef CONFIG_CPU_MIPSR6
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000298 bnez t1, .Ldst_unaligned\@
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299 nop
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000300 bnez t0, .Lsrc_unaligned_dst_aligned\@
Leonid Yegoshinb0ce4bd2014-11-14 11:55:50 +0000301#else
302 or t0, t0, t1
303 bnez t0, .Lcopy_unaligned_bytes\@
304#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305 /*
306 * use delay slot for fall-through
307 * src and dst are aligned; need to compute rem
308 */
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000309.Lboth_aligned\@:
Ralf Baechle70342282013-01-22 12:59:30 +0100310 SRL t0, len, LOG_NBYTES+3 # +3 for 8 units/iter
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000311 beqz t0, .Lcleanup_both_aligned\@ # len < 8*NBYTES
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312 and rem, len, (8*NBYTES-1) # rem = len % (8*NBYTES)
Markos Chandrasbda4d982014-01-07 15:59:03 +0000313 PREFS( 0, 3*32(src) )
314 PREFD( 1, 3*32(dst) )
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 .align 4
3161:
Thomas Bogendoerfer930bff82007-11-25 11:47:56 +0100317 R10KCBARRIER(0(ra))
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000318 LOAD(t0, UNIT(0)(src), .Ll_exc\@)
319 LOAD(t1, UNIT(1)(src), .Ll_exc_copy\@)
320 LOAD(t2, UNIT(2)(src), .Ll_exc_copy\@)
321 LOAD(t3, UNIT(3)(src), .Ll_exc_copy\@)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322 SUB len, len, 8*NBYTES
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000323 LOAD(t4, UNIT(4)(src), .Ll_exc_copy\@)
324 LOAD(t7, UNIT(5)(src), .Ll_exc_copy\@)
325 STORE(t0, UNIT(0)(dst), .Ls_exc_p8u\@)
326 STORE(t1, UNIT(1)(dst), .Ls_exc_p7u\@)
327 LOAD(t0, UNIT(6)(src), .Ll_exc_copy\@)
328 LOAD(t1, UNIT(7)(src), .Ll_exc_copy\@)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 ADD src, src, 8*NBYTES
330 ADD dst, dst, 8*NBYTES
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000331 STORE(t2, UNIT(-6)(dst), .Ls_exc_p6u\@)
332 STORE(t3, UNIT(-5)(dst), .Ls_exc_p5u\@)
333 STORE(t4, UNIT(-4)(dst), .Ls_exc_p4u\@)
334 STORE(t7, UNIT(-3)(dst), .Ls_exc_p3u\@)
335 STORE(t0, UNIT(-2)(dst), .Ls_exc_p2u\@)
336 STORE(t1, UNIT(-1)(dst), .Ls_exc_p1u\@)
Markos Chandrasbda4d982014-01-07 15:59:03 +0000337 PREFS( 0, 8*32(src) )
338 PREFD( 1, 8*32(dst) )
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 bne len, rem, 1b
340 nop
341
342 /*
343 * len == rem == the number of bytes left to copy < 8*NBYTES
344 */
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000345.Lcleanup_both_aligned\@:
346 beqz len, .Ldone\@
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 sltu t0, len, 4*NBYTES
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000348 bnez t0, .Lless_than_4units\@
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 and rem, len, (NBYTES-1) # rem = len % NBYTES
350 /*
351 * len >= 4*NBYTES
352 */
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000353 LOAD( t0, UNIT(0)(src), .Ll_exc\@)
354 LOAD( t1, UNIT(1)(src), .Ll_exc_copy\@)
355 LOAD( t2, UNIT(2)(src), .Ll_exc_copy\@)
356 LOAD( t3, UNIT(3)(src), .Ll_exc_copy\@)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 SUB len, len, 4*NBYTES
358 ADD src, src, 4*NBYTES
Thomas Bogendoerfer930bff82007-11-25 11:47:56 +0100359 R10KCBARRIER(0(ra))
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000360 STORE(t0, UNIT(0)(dst), .Ls_exc_p4u\@)
361 STORE(t1, UNIT(1)(dst), .Ls_exc_p3u\@)
362 STORE(t2, UNIT(2)(dst), .Ls_exc_p2u\@)
363 STORE(t3, UNIT(3)(dst), .Ls_exc_p1u\@)
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100364 .set reorder /* DADDI_WAR */
365 ADD dst, dst, 4*NBYTES
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000366 beqz len, .Ldone\@
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100367 .set noreorder
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000368.Lless_than_4units\@:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 /*
370 * rem = len % NBYTES
371 */
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000372 beq rem, len, .Lcopy_bytes\@
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373 nop
3741:
Thomas Bogendoerfer930bff82007-11-25 11:47:56 +0100375 R10KCBARRIER(0(ra))
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000376 LOAD(t0, 0(src), .Ll_exc\@)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377 ADD src, src, NBYTES
378 SUB len, len, NBYTES
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000379 STORE(t0, 0(dst), .Ls_exc_p1u\@)
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100380 .set reorder /* DADDI_WAR */
381 ADD dst, dst, NBYTES
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 bne rem, len, 1b
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100383 .set noreorder
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384
Leonid Yegoshinb0ce4bd2014-11-14 11:55:50 +0000385#ifndef CONFIG_CPU_MIPSR6
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386 /*
387 * src and dst are aligned, need to copy rem bytes (rem < NBYTES)
388 * A loop would do only a byte at a time with possible branch
Ralf Baechle70342282013-01-22 12:59:30 +0100389 * mispredicts. Can't do an explicit LOAD dst,mask,or,STORE
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390 * because can't assume read-access to dst. Instead, use
391 * STREST dst, which doesn't require read access to dst.
392 *
393 * This code should perform better than a simple loop on modern,
394 * wide-issue mips processors because the code has fewer branches and
395 * more instruction-level parallelism.
396 */
397#define bits t2
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000398 beqz len, .Ldone\@
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 ADD t1, dst, len # t1 is just past last byte of dst
400 li bits, 8*NBYTES
401 SLL rem, len, 3 # rem = number of bits to keep
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000402 LOAD(t0, 0(src), .Ll_exc\@)
Ralf Baechle70342282013-01-22 12:59:30 +0100403 SUB bits, bits, rem # bits = number of bits to discard
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 SHIFT_DISCARD t0, t0, bits
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000405 STREST(t0, -1(t1), .Ls_exc\@)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406 jr ra
407 move len, zero
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000408.Ldst_unaligned\@:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 /*
410 * dst is unaligned
411 * t0 = src & ADDRMASK
412 * t1 = dst & ADDRMASK; T1 > 0
413 * len >= NBYTES
414 *
415 * Copy enough bytes to align dst
416 * Set match = (src and dst have same alignment)
417 */
418#define match rem
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000419 LDFIRST(t3, FIRST(0)(src), .Ll_exc\@)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 ADD t2, zero, NBYTES
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000421 LDREST(t3, REST(0)(src), .Ll_exc_copy\@)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422 SUB t2, t2, t1 # t2 = number of bytes copied
423 xor match, t0, t1
Thomas Bogendoerfer930bff82007-11-25 11:47:56 +0100424 R10KCBARRIER(0(ra))
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000425 STFIRST(t3, FIRST(0)(dst), .Ls_exc\@)
426 beq len, t2, .Ldone\@
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 SUB len, len, t2
428 ADD dst, dst, t2
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000429 beqz match, .Lboth_aligned\@
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 ADD src, src, t2
431
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000432.Lsrc_unaligned_dst_aligned\@:
Ralf Baechle70342282013-01-22 12:59:30 +0100433 SRL t0, len, LOG_NBYTES+2 # +2 for 4 units/iter
Markos Chandrasbda4d982014-01-07 15:59:03 +0000434 PREFS( 0, 3*32(src) )
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000435 beqz t0, .Lcleanup_src_unaligned\@
Ralf Baechle70342282013-01-22 12:59:30 +0100436 and rem, len, (4*NBYTES-1) # rem = len % 4*NBYTES
Markos Chandrasbda4d982014-01-07 15:59:03 +0000437 PREFD( 1, 3*32(dst) )
Linus Torvalds1da177e2005-04-16 15:20:36 -07004381:
439/*
440 * Avoid consecutive LD*'s to the same register since some mips
441 * implementations can't issue them in the same cycle.
442 * It's OK to load FIRST(N+1) before REST(N) because the two addresses
443 * are to the same unit (unless src is aligned, but it's not).
444 */
Thomas Bogendoerfer930bff82007-11-25 11:47:56 +0100445 R10KCBARRIER(0(ra))
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000446 LDFIRST(t0, FIRST(0)(src), .Ll_exc\@)
447 LDFIRST(t1, FIRST(1)(src), .Ll_exc_copy\@)
Ralf Baechle70342282013-01-22 12:59:30 +0100448 SUB len, len, 4*NBYTES
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000449 LDREST(t0, REST(0)(src), .Ll_exc_copy\@)
450 LDREST(t1, REST(1)(src), .Ll_exc_copy\@)
451 LDFIRST(t2, FIRST(2)(src), .Ll_exc_copy\@)
452 LDFIRST(t3, FIRST(3)(src), .Ll_exc_copy\@)
453 LDREST(t2, REST(2)(src), .Ll_exc_copy\@)
454 LDREST(t3, REST(3)(src), .Ll_exc_copy\@)
Markos Chandrasbda4d982014-01-07 15:59:03 +0000455 PREFS( 0, 9*32(src) ) # 0 is PREF_LOAD (not streamed)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 ADD src, src, 4*NBYTES
457#ifdef CONFIG_CPU_SB1
458 nop # improves slotting
459#endif
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000460 STORE(t0, UNIT(0)(dst), .Ls_exc_p4u\@)
461 STORE(t1, UNIT(1)(dst), .Ls_exc_p3u\@)
462 STORE(t2, UNIT(2)(dst), .Ls_exc_p2u\@)
463 STORE(t3, UNIT(3)(dst), .Ls_exc_p1u\@)
Markos Chandrasbda4d982014-01-07 15:59:03 +0000464 PREFD( 1, 9*32(dst) ) # 1 is PREF_STORE (not streamed)
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100465 .set reorder /* DADDI_WAR */
466 ADD dst, dst, 4*NBYTES
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467 bne len, rem, 1b
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100468 .set noreorder
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000470.Lcleanup_src_unaligned\@:
471 beqz len, .Ldone\@
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 and rem, len, NBYTES-1 # rem = len % NBYTES
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000473 beq rem, len, .Lcopy_bytes\@
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 nop
4751:
Thomas Bogendoerfer930bff82007-11-25 11:47:56 +0100476 R10KCBARRIER(0(ra))
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000477 LDFIRST(t0, FIRST(0)(src), .Ll_exc\@)
478 LDREST(t0, REST(0)(src), .Ll_exc_copy\@)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 ADD src, src, NBYTES
480 SUB len, len, NBYTES
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000481 STORE(t0, 0(dst), .Ls_exc_p1u\@)
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100482 .set reorder /* DADDI_WAR */
483 ADD dst, dst, NBYTES
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 bne len, rem, 1b
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100485 .set noreorder
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486
Leonid Yegoshinb0ce4bd2014-11-14 11:55:50 +0000487#endif /* !CONFIG_CPU_MIPSR6 */
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000488.Lcopy_bytes_checklen\@:
489 beqz len, .Ldone\@
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 nop
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000491.Lcopy_bytes\@:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 /* 0 < len < NBYTES */
Thomas Bogendoerfer930bff82007-11-25 11:47:56 +0100493 R10KCBARRIER(0(ra))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494#define COPY_BYTE(N) \
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000495 LOADB(t0, N(src), .Ll_exc\@); \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 SUB len, len, 1; \
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000497 beqz len, .Ldone\@; \
498 STOREB(t0, N(dst), .Ls_exc_p1\@)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499
500 COPY_BYTE(0)
501 COPY_BYTE(1)
502#ifdef USE_DOUBLE
503 COPY_BYTE(2)
504 COPY_BYTE(3)
505 COPY_BYTE(4)
506 COPY_BYTE(5)
507#endif
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000508 LOADB(t0, NBYTES-2(src), .Ll_exc\@)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 SUB len, len, 1
510 jr ra
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000511 STOREB(t0, NBYTES-2(dst), .Ls_exc_p1\@)
512.Ldone\@:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 jr ra
Markos Chandras51b10292014-11-17 09:32:38 +0000514 nop
Leonid Yegoshinb0ce4bd2014-11-14 11:55:50 +0000515
516#ifdef CONFIG_CPU_MIPSR6
517.Lcopy_unaligned_bytes\@:
5181:
519 COPY_BYTE(0)
520 COPY_BYTE(1)
521 COPY_BYTE(2)
522 COPY_BYTE(3)
523 COPY_BYTE(4)
524 COPY_BYTE(5)
525 COPY_BYTE(6)
526 COPY_BYTE(7)
527 ADD src, src, 8
528 b 1b
529 ADD dst, dst, 8
530#endif /* CONFIG_CPU_MIPSR6 */
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000531 .if __memcpy == 1
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 END(memcpy)
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000533 .set __memcpy, 0
534 .hidden __memcpy
535 .endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000537.Ll_exc_copy\@:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 /*
539 * Copy bytes from src until faulting load address (or until a
540 * lb faults)
541 *
542 * When reached by a faulting LDFIRST/LDREST, THREAD_BUADDR($28)
543 * may be more than a byte beyond the last address.
544 * Hence, the lb below may get an exception.
545 *
546 * Assumes src < THREAD_BUADDR($28)
547 */
Markos Chandras5bc05972014-01-07 12:57:04 +0000548 LOADK t0, TI_TASK($28)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 nop
Markos Chandras5bc05972014-01-07 12:57:04 +0000550 LOADK t0, THREAD_BUADDR(t0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005511:
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000552 LOADB(t1, 0(src), .Ll_exc\@)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553 ADD src, src, 1
554 sb t1, 0(dst) # can't fault -- we're copy_from_user
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100555 .set reorder /* DADDI_WAR */
556 ADD dst, dst, 1
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 bne src, t0, 1b
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100558 .set noreorder
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000559.Ll_exc\@:
Markos Chandras5bc05972014-01-07 12:57:04 +0000560 LOADK t0, TI_TASK($28)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561 nop
Markos Chandras5bc05972014-01-07 12:57:04 +0000562 LOADK t0, THREAD_BUADDR(t0) # t0 is just past last good address
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 nop
564 SUB len, AT, t0 # len number of uncopied bytes
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 jr ra
566 nop
567
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100568#define SEXC(n) \
Ralf Baechle70342282013-01-22 12:59:30 +0100569 .set reorder; /* DADDI_WAR */ \
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000570.Ls_exc_p ## n ## u\@: \
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100571 ADD len, len, n*NBYTES; \
572 jr ra; \
573 .set noreorder
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574
575SEXC(8)
576SEXC(7)
577SEXC(6)
578SEXC(5)
579SEXC(4)
580SEXC(3)
581SEXC(2)
582SEXC(1)
583
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000584.Ls_exc_p1\@:
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100585 .set reorder /* DADDI_WAR */
586 ADD len, len, 1
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 jr ra
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100588 .set noreorder
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000589.Ls_exc\@:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 jr ra
591 nop
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000592 .endm
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593
594 .align 5
595LEAF(memmove)
Paul Burton576a2f02016-11-07 11:14:15 +0000596EXPORT_SYMBOL(memmove)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 ADD t0, a0, a2
598 ADD t1, a1, a2
599 sltu t0, a1, t0 # dst + len <= src -> memcpy
600 sltu t1, a0, t1 # dst >= src + len -> memcpy
601 and t0, t1
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000602 beqz t0, .L__memcpy
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 move v0, a0 /* return value */
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000604 beqz a2, .Lr_out
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 END(memmove)
606
607 /* fall through to __rmemcpy */
608LEAF(__rmemcpy) /* a0=dst a1=src a2=len */
609 sltu t0, a1, a0
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000610 beqz t0, .Lr_end_bytes_up # src >= dst
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 nop
612 ADD a0, a2 # dst = dst + len
613 ADD a1, a2 # src = src + len
614
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000615.Lr_end_bytes:
Thomas Bogendoerfer930bff82007-11-25 11:47:56 +0100616 R10KCBARRIER(0(ra))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 lb t0, -1(a1)
618 SUB a2, a2, 0x1
619 sb t0, -1(a0)
620 SUB a1, a1, 0x1
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100621 .set reorder /* DADDI_WAR */
622 SUB a0, a0, 0x1
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000623 bnez a2, .Lr_end_bytes
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100624 .set noreorder
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000626.Lr_out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 jr ra
628 move a2, zero
629
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000630.Lr_end_bytes_up:
Thomas Bogendoerfer930bff82007-11-25 11:47:56 +0100631 R10KCBARRIER(0(ra))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632 lb t0, (a1)
633 SUB a2, a2, 0x1
634 sb t0, (a0)
635 ADD a1, a1, 0x1
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100636 .set reorder /* DADDI_WAR */
637 ADD a0, a0, 0x1
Ralf Baechlec5ec1982008-01-29 10:14:59 +0000638 bnez a2, .Lr_end_bytes_up
Maciej W. Rozycki619b6e12007-10-23 12:43:25 +0100639 .set noreorder
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640
641 jr ra
642 move a2, zero
643 END(__rmemcpy)
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000644
645/*
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000646 * A combined memcpy/__copy_user
647 * __copy_user sets len to 0 for success; else to an upper bound of
648 * the number of uncopied bytes.
649 * memcpy sets v0 to dst.
650 */
651 .align 5
652LEAF(memcpy) /* a0=dst a1=src a2=len */
Paul Burton576a2f02016-11-07 11:14:15 +0000653EXPORT_SYMBOL(memcpy)
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000654 move v0, dst /* return value */
655.L__memcpy:
656FEXPORT(__copy_user)
Paul Burton576a2f02016-11-07 11:14:15 +0000657EXPORT_SYMBOL(__copy_user)
Markos Chandrascf62a8b2014-01-07 14:34:05 +0000658 /* Legacy Mode, user <-> user */
659 __BUILD_COPY_USER LEGACY_MODE USEROP USEROP
Markos Chandrascd26cb42014-01-07 16:20:22 +0000660
661#ifdef CONFIG_EVA
662
663/*
664 * For EVA we need distinct symbols for reading and writing to user space.
665 * This is because we need to use specific EVA instructions to perform the
666 * virtual <-> physical translation when a virtual address is actually in user
667 * space
668 */
669
Markos Chandrascd26cb42014-01-07 16:20:22 +0000670/*
671 * __copy_from_user (EVA)
672 */
673
674LEAF(__copy_from_user_eva)
Paul Burton576a2f02016-11-07 11:14:15 +0000675EXPORT_SYMBOL(__copy_from_user_eva)
Markos Chandrascd26cb42014-01-07 16:20:22 +0000676 __BUILD_COPY_USER EVA_MODE USEROP KERNELOP
677END(__copy_from_user_eva)
678
679
680
681/*
682 * __copy_to_user (EVA)
683 */
684
685LEAF(__copy_to_user_eva)
Paul Burton576a2f02016-11-07 11:14:15 +0000686EXPORT_SYMBOL(__copy_to_user_eva)
Markos Chandrascd26cb42014-01-07 16:20:22 +0000687__BUILD_COPY_USER EVA_MODE KERNELOP USEROP
688END(__copy_to_user_eva)
689
690/*
691 * __copy_in_user (EVA)
692 */
693
694LEAF(__copy_in_user_eva)
Paul Burton576a2f02016-11-07 11:14:15 +0000695EXPORT_SYMBOL(__copy_in_user_eva)
Markos Chandrascd26cb42014-01-07 16:20:22 +0000696__BUILD_COPY_USER EVA_MODE USEROP USEROP
697END(__copy_in_user_eva)
698
699#endif