xref: /openbmc/linux/arch/mips/lib/csum_partial.S (revision 3e30a927)
1/*
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 * Quick'n'dirty IP checksum ...
7 *
8 * Copyright (C) 1998, 1999 Ralf Baechle
9 * Copyright (C) 1999 Silicon Graphics, Inc.
10 * Copyright (C) 2007  Maciej W. Rozycki
11 * Copyright (C) 2014 Imagination Technologies Ltd.
12 */
13#include <linux/errno.h>
14#include <asm/asm.h>
15#include <asm/asm-offsets.h>
16#include <asm/export.h>
17#include <asm/regdef.h>
18
19#ifdef CONFIG_64BIT
20/*
21 * As we are sharing code base with the mips32 tree (which use the o32 ABI
22 * register definitions). We need to redefine the register definitions from
23 * the n64 ABI register naming to the o32 ABI register naming.
24 */
25#undef t0
26#undef t1
27#undef t2
28#undef t3
29#define t0	$8
30#define t1	$9
31#define t2	$10
32#define t3	$11
33#define t4	$12
34#define t5	$13
35#define t6	$14
36#define t7	$15
37
38#define USE_DOUBLE
39#endif
40
41#ifdef USE_DOUBLE
42
43#define LOAD   ld
44#define LOAD32 lwu
45#define ADD    daddu
46#define NBYTES 8
47
48#else
49
50#define LOAD   lw
51#define LOAD32 lw
52#define ADD    addu
53#define NBYTES 4
54
55#endif /* USE_DOUBLE */
56
57#define UNIT(unit)  ((unit)*NBYTES)
58
59#define ADDC(sum,reg)						\
60	.set	push;						\
61	.set	noat;						\
62	ADD	sum, reg;					\
63	sltu	v1, sum, reg;					\
64	ADD	sum, v1;					\
65	.set	pop
66
67#define ADDC32(sum,reg)						\
68	.set	push;						\
69	.set	noat;						\
70	addu	sum, reg;					\
71	sltu	v1, sum, reg;					\
72	addu	sum, v1;					\
73	.set	pop
74
75#define CSUM_BIGCHUNK1(src, offset, sum, _t0, _t1, _t2, _t3)	\
76	LOAD	_t0, (offset + UNIT(0))(src);			\
77	LOAD	_t1, (offset + UNIT(1))(src);			\
78	LOAD	_t2, (offset + UNIT(2))(src);			\
79	LOAD	_t3, (offset + UNIT(3))(src);			\
80	ADDC(_t0, _t1);						\
81	ADDC(_t2, _t3);						\
82	ADDC(sum, _t0);						\
83	ADDC(sum, _t2)
84
85#ifdef USE_DOUBLE
86#define CSUM_BIGCHUNK(src, offset, sum, _t0, _t1, _t2, _t3)	\
87	CSUM_BIGCHUNK1(src, offset, sum, _t0, _t1, _t2, _t3)
88#else
89#define CSUM_BIGCHUNK(src, offset, sum, _t0, _t1, _t2, _t3)	\
90	CSUM_BIGCHUNK1(src, offset, sum, _t0, _t1, _t2, _t3);	\
91	CSUM_BIGCHUNK1(src, offset + 0x10, sum, _t0, _t1, _t2, _t3)
92#endif
93
94/*
95 * a0: source address
96 * a1: length of the area to checksum
97 * a2: partial checksum
98 */
99
100#define src a0
101#define sum v0
102
103	.text
104	.set	noreorder
105	.align	5
106LEAF(csum_partial)
107EXPORT_SYMBOL(csum_partial)
108	move	sum, zero
109	move	t7, zero
110
111	sltiu	t8, a1, 0x8
112	bnez	t8, .Lsmall_csumcpy		/* < 8 bytes to copy */
113	 move	t2, a1
114
115	andi	t7, src, 0x1			/* odd buffer? */
116
117.Lhword_align:
118	beqz	t7, .Lword_align
119	 andi	t8, src, 0x2
120
121	lbu	t0, (src)
122	LONG_SUBU	a1, a1, 0x1
123#ifdef __MIPSEL__
124	sll	t0, t0, 8
125#endif
126	ADDC(sum, t0)
127	PTR_ADDU	src, src, 0x1
128	andi	t8, src, 0x2
129
130.Lword_align:
131	beqz	t8, .Ldword_align
132	 sltiu	t8, a1, 56
133
134	lhu	t0, (src)
135	LONG_SUBU	a1, a1, 0x2
136	ADDC(sum, t0)
137	sltiu	t8, a1, 56
138	PTR_ADDU	src, src, 0x2
139
140.Ldword_align:
141	bnez	t8, .Ldo_end_words
142	 move	t8, a1
143
144	andi	t8, src, 0x4
145	beqz	t8, .Lqword_align
146	 andi	t8, src, 0x8
147
148	LOAD32	t0, 0x00(src)
149	LONG_SUBU	a1, a1, 0x4
150	ADDC(sum, t0)
151	PTR_ADDU	src, src, 0x4
152	andi	t8, src, 0x8
153
154.Lqword_align:
155	beqz	t8, .Loword_align
156	 andi	t8, src, 0x10
157
158#ifdef USE_DOUBLE
159	ld	t0, 0x00(src)
160	LONG_SUBU	a1, a1, 0x8
161	ADDC(sum, t0)
162#else
163	lw	t0, 0x00(src)
164	lw	t1, 0x04(src)
165	LONG_SUBU	a1, a1, 0x8
166	ADDC(sum, t0)
167	ADDC(sum, t1)
168#endif
169	PTR_ADDU	src, src, 0x8
170	andi	t8, src, 0x10
171
172.Loword_align:
173	beqz	t8, .Lbegin_movement
174	 LONG_SRL	t8, a1, 0x7
175
176#ifdef USE_DOUBLE
177	ld	t0, 0x00(src)
178	ld	t1, 0x08(src)
179	ADDC(sum, t0)
180	ADDC(sum, t1)
181#else
182	CSUM_BIGCHUNK1(src, 0x00, sum, t0, t1, t3, t4)
183#endif
184	LONG_SUBU	a1, a1, 0x10
185	PTR_ADDU	src, src, 0x10
186	LONG_SRL	t8, a1, 0x7
187
188.Lbegin_movement:
189	beqz	t8, 1f
190	 andi	t2, a1, 0x40
191
192.Lmove_128bytes:
193	CSUM_BIGCHUNK(src, 0x00, sum, t0, t1, t3, t4)
194	CSUM_BIGCHUNK(src, 0x20, sum, t0, t1, t3, t4)
195	CSUM_BIGCHUNK(src, 0x40, sum, t0, t1, t3, t4)
196	CSUM_BIGCHUNK(src, 0x60, sum, t0, t1, t3, t4)
197	LONG_SUBU	t8, t8, 0x01
198	.set	reorder				/* DADDI_WAR */
199	PTR_ADDU	src, src, 0x80
200	bnez	t8, .Lmove_128bytes
201	.set	noreorder
202
2031:
204	beqz	t2, 1f
205	 andi	t2, a1, 0x20
206
207.Lmove_64bytes:
208	CSUM_BIGCHUNK(src, 0x00, sum, t0, t1, t3, t4)
209	CSUM_BIGCHUNK(src, 0x20, sum, t0, t1, t3, t4)
210	PTR_ADDU	src, src, 0x40
211
2121:
213	beqz	t2, .Ldo_end_words
214	 andi	t8, a1, 0x1c
215
216.Lmove_32bytes:
217	CSUM_BIGCHUNK(src, 0x00, sum, t0, t1, t3, t4)
218	andi	t8, a1, 0x1c
219	PTR_ADDU	src, src, 0x20
220
221.Ldo_end_words:
222	beqz	t8, .Lsmall_csumcpy
223	 andi	t2, a1, 0x3
224	LONG_SRL	t8, t8, 0x2
225
226.Lend_words:
227	LOAD32	t0, (src)
228	LONG_SUBU	t8, t8, 0x1
229	ADDC(sum, t0)
230	.set	reorder				/* DADDI_WAR */
231	PTR_ADDU	src, src, 0x4
232	bnez	t8, .Lend_words
233	.set	noreorder
234
235/* unknown src alignment and < 8 bytes to go  */
236.Lsmall_csumcpy:
237	move	a1, t2
238
239	andi	t0, a1, 4
240	beqz	t0, 1f
241	 andi	t0, a1, 2
242
243	/* Still a full word to go  */
244	ulw	t1, (src)
245	PTR_ADDIU	src, 4
246#ifdef USE_DOUBLE
247	dsll	t1, t1, 32			/* clear lower 32bit */
248#endif
249	ADDC(sum, t1)
250
2511:	move	t1, zero
252	beqz	t0, 1f
253	 andi	t0, a1, 1
254
255	/* Still a halfword to go  */
256	ulhu	t1, (src)
257	PTR_ADDIU	src, 2
258
2591:	beqz	t0, 1f
260	 sll	t1, t1, 16
261
262	lbu	t2, (src)
263	 nop
264
265#ifdef __MIPSEB__
266	sll	t2, t2, 8
267#endif
268	or	t1, t2
269
2701:	ADDC(sum, t1)
271
272	/* fold checksum */
273#ifdef USE_DOUBLE
274	dsll32	v1, sum, 0
275	daddu	sum, v1
276	sltu	v1, sum, v1
277	dsra32	sum, sum, 0
278	addu	sum, v1
279#endif
280
281	/* odd buffer alignment? */
282#if defined(CONFIG_CPU_MIPSR2) || defined(CONFIG_CPU_LOONGSON64)
283	.set	push
284	.set	arch=mips32r2
285	wsbh	v1, sum
286	movn	sum, v1, t7
287	.set	pop
288#else
289	beqz	t7, 1f			/* odd buffer alignment? */
290	 lui	v1, 0x00ff
291	addu	v1, 0x00ff
292	and	t0, sum, v1
293	sll	t0, t0, 8
294	srl	sum, sum, 8
295	and	sum, sum, v1
296	or	sum, sum, t0
2971:
298#endif
299	.set	reorder
300	/* Add the passed partial csum.	 */
301	ADDC32(sum, a2)
302	jr	ra
303	.set	noreorder
304	END(csum_partial)
305
306
307/*
308 * checksum and copy routines based on memcpy.S
309 *
310 *	csum_partial_copy_nocheck(src, dst, len, sum)
311 *	__csum_partial_copy_kernel(src, dst, len, sum, errp)
312 *
313 * See "Spec" in memcpy.S for details.	Unlike __copy_user, all
314 * function in this file use the standard calling convention.
315 */
316
317#define src a0
318#define dst a1
319#define len a2
320#define psum a3
321#define sum v0
322#define odd t8
323#define errptr t9
324
325/*
326 * The exception handler for loads requires that:
327 *  1- AT contain the address of the byte just past the end of the source
328 *     of the copy,
329 *  2- src_entry <= src < AT, and
330 *  3- (dst - src) == (dst_entry - src_entry),
331 * The _entry suffix denotes values when __copy_user was called.
332 *
333 * (1) is set up up by __csum_partial_copy_from_user and maintained by
334 *	not writing AT in __csum_partial_copy
335 * (2) is met by incrementing src by the number of bytes copied
336 * (3) is met by not doing loads between a pair of increments of dst and src
337 *
338 * The exception handlers for stores stores -EFAULT to errptr and return.
339 * These handlers do not need to overwrite any data.
340 */
341
342/* Instruction type */
343#define LD_INSN 1
344#define ST_INSN 2
345#define LEGACY_MODE 1
346#define EVA_MODE    2
347#define USEROP   1
348#define KERNELOP 2
349
350/*
351 * Wrapper to add an entry in the exception table
352 * in case the insn causes a memory exception.
353 * Arguments:
354 * insn    : Load/store instruction
355 * type    : Instruction type
356 * reg     : Register
357 * addr    : Address
358 * handler : Exception handler
359 */
360#define EXC(insn, type, reg, addr, handler)	\
361	.if \mode == LEGACY_MODE;		\
3629:		insn reg, addr;			\
363		.section __ex_table,"a";	\
364		PTR	9b, handler;		\
365		.previous;			\
366	/* This is enabled in EVA mode */	\
367	.else;					\
368		/* If loading from user or storing to user */	\
369		.if ((\from == USEROP) && (type == LD_INSN)) || \
370		    ((\to == USEROP) && (type == ST_INSN));	\
3719:			__BUILD_EVA_INSN(insn##e, reg, addr);	\
372			.section __ex_table,"a";		\
373			PTR	9b, handler;			\
374			.previous;				\
375		.else;						\
376			/* EVA without exception */		\
377			insn reg, addr;				\
378		.endif;						\
379	.endif
380
381#undef LOAD
382
383#ifdef USE_DOUBLE
384
385#define LOADK	ld /* No exception */
386#define LOAD(reg, addr, handler)	EXC(ld, LD_INSN, reg, addr, handler)
387#define LOADBU(reg, addr, handler)	EXC(lbu, LD_INSN, reg, addr, handler)
388#define LOADL(reg, addr, handler)	EXC(ldl, LD_INSN, reg, addr, handler)
389#define LOADR(reg, addr, handler)	EXC(ldr, LD_INSN, reg, addr, handler)
390#define STOREB(reg, addr, handler)	EXC(sb, ST_INSN, reg, addr, handler)
391#define STOREL(reg, addr, handler)	EXC(sdl, ST_INSN, reg, addr, handler)
392#define STORER(reg, addr, handler)	EXC(sdr, ST_INSN, reg, addr, handler)
393#define STORE(reg, addr, handler)	EXC(sd, ST_INSN, reg, addr, handler)
394#define ADD    daddu
395#define SUB    dsubu
396#define SRL    dsrl
397#define SLL    dsll
398#define SLLV   dsllv
399#define SRLV   dsrlv
400#define NBYTES 8
401#define LOG_NBYTES 3
402
403#else
404
405#define LOADK	lw /* No exception */
406#define LOAD(reg, addr, handler)	EXC(lw, LD_INSN, reg, addr, handler)
407#define LOADBU(reg, addr, handler)	EXC(lbu, LD_INSN, reg, addr, handler)
408#define LOADL(reg, addr, handler)	EXC(lwl, LD_INSN, reg, addr, handler)
409#define LOADR(reg, addr, handler)	EXC(lwr, LD_INSN, reg, addr, handler)
410#define STOREB(reg, addr, handler)	EXC(sb, ST_INSN, reg, addr, handler)
411#define STOREL(reg, addr, handler)	EXC(swl, ST_INSN, reg, addr, handler)
412#define STORER(reg, addr, handler)	EXC(swr, ST_INSN, reg, addr, handler)
413#define STORE(reg, addr, handler)	EXC(sw, ST_INSN, reg, addr, handler)
414#define ADD    addu
415#define SUB    subu
416#define SRL    srl
417#define SLL    sll
418#define SLLV   sllv
419#define SRLV   srlv
420#define NBYTES 4
421#define LOG_NBYTES 2
422
423#endif /* USE_DOUBLE */
424
425#ifdef CONFIG_CPU_LITTLE_ENDIAN
426#define LDFIRST LOADR
427#define LDREST	LOADL
428#define STFIRST STORER
429#define STREST	STOREL
430#define SHIFT_DISCARD SLLV
431#define SHIFT_DISCARD_REVERT SRLV
432#else
433#define LDFIRST LOADL
434#define LDREST	LOADR
435#define STFIRST STOREL
436#define STREST	STORER
437#define SHIFT_DISCARD SRLV
438#define SHIFT_DISCARD_REVERT SLLV
439#endif
440
441#define FIRST(unit) ((unit)*NBYTES)
442#define REST(unit)  (FIRST(unit)+NBYTES-1)
443
444#define ADDRMASK (NBYTES-1)
445
446#ifndef CONFIG_CPU_DADDI_WORKAROUNDS
447	.set	noat
448#else
449	.set	at=v1
450#endif
451
452	.macro __BUILD_CSUM_PARTIAL_COPY_USER mode, from, to, __nocheck
453
454	PTR_ADDU	AT, src, len	/* See (1) above. */
455	/* initialize __nocheck if this the first time we execute this
456	 * macro
457	 */
458#ifdef CONFIG_64BIT
459	move	errptr, a4
460#else
461	lw	errptr, 16(sp)
462#endif
463	.if \__nocheck == 1
464	FEXPORT(csum_partial_copy_nocheck)
465	EXPORT_SYMBOL(csum_partial_copy_nocheck)
466	.endif
467	move	sum, zero
468	move	odd, zero
469	/*
470	 * Note: dst & src may be unaligned, len may be 0
471	 * Temps
472	 */
473	/*
474	 * The "issue break"s below are very approximate.
475	 * Issue delays for dcache fills will perturb the schedule, as will
476	 * load queue full replay traps, etc.
477	 *
478	 * If len < NBYTES use byte operations.
479	 */
480	sltu	t2, len, NBYTES
481	and	t1, dst, ADDRMASK
482	bnez	t2, .Lcopy_bytes_checklen\@
483	 and	t0, src, ADDRMASK
484	andi	odd, dst, 0x1			/* odd buffer? */
485	bnez	t1, .Ldst_unaligned\@
486	 nop
487	bnez	t0, .Lsrc_unaligned_dst_aligned\@
488	/*
489	 * use delay slot for fall-through
490	 * src and dst are aligned; need to compute rem
491	 */
492.Lboth_aligned\@:
493	 SRL	t0, len, LOG_NBYTES+3	 # +3 for 8 units/iter
494	beqz	t0, .Lcleanup_both_aligned\@ # len < 8*NBYTES
495	 nop
496	SUB	len, 8*NBYTES		# subtract here for bgez loop
497	.align	4
4981:
499	LOAD(t0, UNIT(0)(src), .Ll_exc\@)
500	LOAD(t1, UNIT(1)(src), .Ll_exc_copy\@)
501	LOAD(t2, UNIT(2)(src), .Ll_exc_copy\@)
502	LOAD(t3, UNIT(3)(src), .Ll_exc_copy\@)
503	LOAD(t4, UNIT(4)(src), .Ll_exc_copy\@)
504	LOAD(t5, UNIT(5)(src), .Ll_exc_copy\@)
505	LOAD(t6, UNIT(6)(src), .Ll_exc_copy\@)
506	LOAD(t7, UNIT(7)(src), .Ll_exc_copy\@)
507	SUB	len, len, 8*NBYTES
508	ADD	src, src, 8*NBYTES
509	STORE(t0, UNIT(0)(dst),	.Ls_exc\@)
510	ADDC(t0, t1)
511	STORE(t1, UNIT(1)(dst),	.Ls_exc\@)
512	ADDC(sum, t0)
513	STORE(t2, UNIT(2)(dst),	.Ls_exc\@)
514	ADDC(t2, t3)
515	STORE(t3, UNIT(3)(dst),	.Ls_exc\@)
516	ADDC(sum, t2)
517	STORE(t4, UNIT(4)(dst),	.Ls_exc\@)
518	ADDC(t4, t5)
519	STORE(t5, UNIT(5)(dst),	.Ls_exc\@)
520	ADDC(sum, t4)
521	STORE(t6, UNIT(6)(dst),	.Ls_exc\@)
522	ADDC(t6, t7)
523	STORE(t7, UNIT(7)(dst),	.Ls_exc\@)
524	ADDC(sum, t6)
525	.set	reorder				/* DADDI_WAR */
526	ADD	dst, dst, 8*NBYTES
527	bgez	len, 1b
528	.set	noreorder
529	ADD	len, 8*NBYTES		# revert len (see above)
530
531	/*
532	 * len == the number of bytes left to copy < 8*NBYTES
533	 */
534.Lcleanup_both_aligned\@:
535#define rem t7
536	beqz	len, .Ldone\@
537	 sltu	t0, len, 4*NBYTES
538	bnez	t0, .Lless_than_4units\@
539	 and	rem, len, (NBYTES-1)	# rem = len % NBYTES
540	/*
541	 * len >= 4*NBYTES
542	 */
543	LOAD(t0, UNIT(0)(src), .Ll_exc\@)
544	LOAD(t1, UNIT(1)(src), .Ll_exc_copy\@)
545	LOAD(t2, UNIT(2)(src), .Ll_exc_copy\@)
546	LOAD(t3, UNIT(3)(src), .Ll_exc_copy\@)
547	SUB	len, len, 4*NBYTES
548	ADD	src, src, 4*NBYTES
549	STORE(t0, UNIT(0)(dst),	.Ls_exc\@)
550	ADDC(t0, t1)
551	STORE(t1, UNIT(1)(dst),	.Ls_exc\@)
552	ADDC(sum, t0)
553	STORE(t2, UNIT(2)(dst),	.Ls_exc\@)
554	ADDC(t2, t3)
555	STORE(t3, UNIT(3)(dst),	.Ls_exc\@)
556	ADDC(sum, t2)
557	.set	reorder				/* DADDI_WAR */
558	ADD	dst, dst, 4*NBYTES
559	beqz	len, .Ldone\@
560	.set	noreorder
561.Lless_than_4units\@:
562	/*
563	 * rem = len % NBYTES
564	 */
565	beq	rem, len, .Lcopy_bytes\@
566	 nop
5671:
568	LOAD(t0, 0(src), .Ll_exc\@)
569	ADD	src, src, NBYTES
570	SUB	len, len, NBYTES
571	STORE(t0, 0(dst), .Ls_exc\@)
572	ADDC(sum, t0)
573	.set	reorder				/* DADDI_WAR */
574	ADD	dst, dst, NBYTES
575	bne	rem, len, 1b
576	.set	noreorder
577
578	/*
579	 * src and dst are aligned, need to copy rem bytes (rem < NBYTES)
580	 * A loop would do only a byte at a time with possible branch
581	 * mispredicts.	 Can't do an explicit LOAD dst,mask,or,STORE
582	 * because can't assume read-access to dst.  Instead, use
583	 * STREST dst, which doesn't require read access to dst.
584	 *
585	 * This code should perform better than a simple loop on modern,
586	 * wide-issue mips processors because the code has fewer branches and
587	 * more instruction-level parallelism.
588	 */
589#define bits t2
590	beqz	len, .Ldone\@
591	 ADD	t1, dst, len	# t1 is just past last byte of dst
592	li	bits, 8*NBYTES
593	SLL	rem, len, 3	# rem = number of bits to keep
594	LOAD(t0, 0(src), .Ll_exc\@)
595	SUB	bits, bits, rem # bits = number of bits to discard
596	SHIFT_DISCARD t0, t0, bits
597	STREST(t0, -1(t1), .Ls_exc\@)
598	SHIFT_DISCARD_REVERT t0, t0, bits
599	.set reorder
600	ADDC(sum, t0)
601	b	.Ldone\@
602	.set noreorder
603.Ldst_unaligned\@:
604	/*
605	 * dst is unaligned
606	 * t0 = src & ADDRMASK
607	 * t1 = dst & ADDRMASK; T1 > 0
608	 * len >= NBYTES
609	 *
610	 * Copy enough bytes to align dst
611	 * Set match = (src and dst have same alignment)
612	 */
613#define match rem
614	LDFIRST(t3, FIRST(0)(src), .Ll_exc\@)
615	ADD	t2, zero, NBYTES
616	LDREST(t3, REST(0)(src), .Ll_exc_copy\@)
617	SUB	t2, t2, t1	# t2 = number of bytes copied
618	xor	match, t0, t1
619	STFIRST(t3, FIRST(0)(dst), .Ls_exc\@)
620	SLL	t4, t1, 3		# t4 = number of bits to discard
621	SHIFT_DISCARD t3, t3, t4
622	/* no SHIFT_DISCARD_REVERT to handle odd buffer properly */
623	ADDC(sum, t3)
624	beq	len, t2, .Ldone\@
625	 SUB	len, len, t2
626	ADD	dst, dst, t2
627	beqz	match, .Lboth_aligned\@
628	 ADD	src, src, t2
629
630.Lsrc_unaligned_dst_aligned\@:
631	SRL	t0, len, LOG_NBYTES+2	 # +2 for 4 units/iter
632	beqz	t0, .Lcleanup_src_unaligned\@
633	 and	rem, len, (4*NBYTES-1)	 # rem = len % 4*NBYTES
6341:
635/*
636 * Avoid consecutive LD*'s to the same register since some mips
637 * implementations can't issue them in the same cycle.
638 * It's OK to load FIRST(N+1) before REST(N) because the two addresses
639 * are to the same unit (unless src is aligned, but it's not).
640 */
641	LDFIRST(t0, FIRST(0)(src), .Ll_exc\@)
642	LDFIRST(t1, FIRST(1)(src), .Ll_exc_copy\@)
643	SUB	len, len, 4*NBYTES
644	LDREST(t0, REST(0)(src), .Ll_exc_copy\@)
645	LDREST(t1, REST(1)(src), .Ll_exc_copy\@)
646	LDFIRST(t2, FIRST(2)(src), .Ll_exc_copy\@)
647	LDFIRST(t3, FIRST(3)(src), .Ll_exc_copy\@)
648	LDREST(t2, REST(2)(src), .Ll_exc_copy\@)
649	LDREST(t3, REST(3)(src), .Ll_exc_copy\@)
650	ADD	src, src, 4*NBYTES
651#ifdef CONFIG_CPU_SB1
652	nop				# improves slotting
653#endif
654	STORE(t0, UNIT(0)(dst),	.Ls_exc\@)
655	ADDC(t0, t1)
656	STORE(t1, UNIT(1)(dst),	.Ls_exc\@)
657	ADDC(sum, t0)
658	STORE(t2, UNIT(2)(dst),	.Ls_exc\@)
659	ADDC(t2, t3)
660	STORE(t3, UNIT(3)(dst),	.Ls_exc\@)
661	ADDC(sum, t2)
662	.set	reorder				/* DADDI_WAR */
663	ADD	dst, dst, 4*NBYTES
664	bne	len, rem, 1b
665	.set	noreorder
666
667.Lcleanup_src_unaligned\@:
668	beqz	len, .Ldone\@
669	 and	rem, len, NBYTES-1  # rem = len % NBYTES
670	beq	rem, len, .Lcopy_bytes\@
671	 nop
6721:
673	LDFIRST(t0, FIRST(0)(src), .Ll_exc\@)
674	LDREST(t0, REST(0)(src), .Ll_exc_copy\@)
675	ADD	src, src, NBYTES
676	SUB	len, len, NBYTES
677	STORE(t0, 0(dst), .Ls_exc\@)
678	ADDC(sum, t0)
679	.set	reorder				/* DADDI_WAR */
680	ADD	dst, dst, NBYTES
681	bne	len, rem, 1b
682	.set	noreorder
683
684.Lcopy_bytes_checklen\@:
685	beqz	len, .Ldone\@
686	 nop
687.Lcopy_bytes\@:
688	/* 0 < len < NBYTES  */
689#ifdef CONFIG_CPU_LITTLE_ENDIAN
690#define SHIFT_START 0
691#define SHIFT_INC 8
692#else
693#define SHIFT_START 8*(NBYTES-1)
694#define SHIFT_INC -8
695#endif
696	move	t2, zero	# partial word
697	li	t3, SHIFT_START # shift
698/* use .Ll_exc_copy here to return correct sum on fault */
699#define COPY_BYTE(N)			\
700	LOADBU(t0, N(src), .Ll_exc_copy\@);	\
701	SUB	len, len, 1;		\
702	STOREB(t0, N(dst), .Ls_exc\@);	\
703	SLLV	t0, t0, t3;		\
704	addu	t3, SHIFT_INC;		\
705	beqz	len, .Lcopy_bytes_done\@; \
706	 or	t2, t0
707
708	COPY_BYTE(0)
709	COPY_BYTE(1)
710#ifdef USE_DOUBLE
711	COPY_BYTE(2)
712	COPY_BYTE(3)
713	COPY_BYTE(4)
714	COPY_BYTE(5)
715#endif
716	LOADBU(t0, NBYTES-2(src), .Ll_exc_copy\@)
717	SUB	len, len, 1
718	STOREB(t0, NBYTES-2(dst), .Ls_exc\@)
719	SLLV	t0, t0, t3
720	or	t2, t0
721.Lcopy_bytes_done\@:
722	ADDC(sum, t2)
723.Ldone\@:
724	/* fold checksum */
725	.set	push
726	.set	noat
727#ifdef USE_DOUBLE
728	dsll32	v1, sum, 0
729	daddu	sum, v1
730	sltu	v1, sum, v1
731	dsra32	sum, sum, 0
732	addu	sum, v1
733#endif
734
735#if defined(CONFIG_CPU_MIPSR2) || defined(CONFIG_CPU_LOONGSON64)
736	.set	push
737	.set	arch=mips32r2
738	wsbh	v1, sum
739	movn	sum, v1, odd
740	.set	pop
741#else
742	beqz	odd, 1f			/* odd buffer alignment? */
743	 lui	v1, 0x00ff
744	addu	v1, 0x00ff
745	and	t0, sum, v1
746	sll	t0, t0, 8
747	srl	sum, sum, 8
748	and	sum, sum, v1
749	or	sum, sum, t0
7501:
751#endif
752	.set	pop
753	.set reorder
754	ADDC32(sum, psum)
755	jr	ra
756	.set noreorder
757
758.Ll_exc_copy\@:
759	/*
760	 * Copy bytes from src until faulting load address (or until a
761	 * lb faults)
762	 *
763	 * When reached by a faulting LDFIRST/LDREST, THREAD_BUADDR($28)
764	 * may be more than a byte beyond the last address.
765	 * Hence, the lb below may get an exception.
766	 *
767	 * Assumes src < THREAD_BUADDR($28)
768	 */
769	LOADK	t0, TI_TASK($28)
770	 li	t2, SHIFT_START
771	LOADK	t0, THREAD_BUADDR(t0)
7721:
773	LOADBU(t1, 0(src), .Ll_exc\@)
774	ADD	src, src, 1
775	sb	t1, 0(dst)	# can't fault -- we're copy_from_user
776	SLLV	t1, t1, t2
777	addu	t2, SHIFT_INC
778	ADDC(sum, t1)
779	.set	reorder				/* DADDI_WAR */
780	ADD	dst, dst, 1
781	bne	src, t0, 1b
782	.set	noreorder
783.Ll_exc\@:
784	LOADK	t0, TI_TASK($28)
785	 nop
786	LOADK	t0, THREAD_BUADDR(t0)	# t0 is just past last good address
787	 nop
788	SUB	len, AT, t0		# len number of uncopied bytes
789	/*
790	 * Here's where we rely on src and dst being incremented in tandem,
791	 *   See (3) above.
792	 * dst += (fault addr - src) to put dst at first byte to clear
793	 */
794	ADD	dst, t0			# compute start address in a1
795	SUB	dst, src
796	/*
797	 * Clear len bytes starting at dst.  Can't call __bzero because it
798	 * might modify len.  An inefficient loop for these rare times...
799	 */
800	.set	reorder				/* DADDI_WAR */
801	SUB	src, len, 1
802	beqz	len, .Ldone\@
803	.set	noreorder
8041:	sb	zero, 0(dst)
805	ADD	dst, dst, 1
806	.set	push
807	.set	noat
808#ifndef CONFIG_CPU_DADDI_WORKAROUNDS
809	bnez	src, 1b
810	 SUB	src, src, 1
811#else
812	li	v1, 1
813	bnez	src, 1b
814	 SUB	src, src, v1
815#endif
816	li	v1, -EFAULT
817	b	.Ldone\@
818	 sw	v1, (errptr)
819
820.Ls_exc\@:
821	li	v0, -1 /* invalid checksum */
822	li	v1, -EFAULT
823	jr	ra
824	 sw	v1, (errptr)
825	.set	pop
826	.endm
827
828LEAF(__csum_partial_copy_kernel)
829EXPORT_SYMBOL(__csum_partial_copy_kernel)
830#ifndef CONFIG_EVA
831FEXPORT(__csum_partial_copy_to_user)
832EXPORT_SYMBOL(__csum_partial_copy_to_user)
833FEXPORT(__csum_partial_copy_from_user)
834EXPORT_SYMBOL(__csum_partial_copy_from_user)
835#endif
836__BUILD_CSUM_PARTIAL_COPY_USER LEGACY_MODE USEROP USEROP 1
837END(__csum_partial_copy_kernel)
838
839#ifdef CONFIG_EVA
840LEAF(__csum_partial_copy_to_user)
841__BUILD_CSUM_PARTIAL_COPY_USER EVA_MODE KERNELOP USEROP 0
842END(__csum_partial_copy_to_user)
843
844LEAF(__csum_partial_copy_from_user)
845__BUILD_CSUM_PARTIAL_COPY_USER EVA_MODE USEROP KERNELOP 0
846END(__csum_partial_copy_from_user)
847#endif
848