xref: /openbmc/linux/arch/mips/lib/csum_partial.S (revision b8265621)
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_MIPSR5) || \
283    defined(CONFIG_CPU_LOONGSON64)
284	.set	push
285	.set	arch=mips32r2
286	wsbh	v1, sum
287	movn	sum, v1, t7
288	.set	pop
289#else
290	beqz	t7, 1f			/* odd buffer alignment? */
291	 lui	v1, 0x00ff
292	addu	v1, 0x00ff
293	and	t0, sum, v1
294	sll	t0, t0, 8
295	srl	sum, sum, 8
296	and	sum, sum, v1
297	or	sum, sum, t0
2981:
299#endif
300	.set	reorder
301	/* Add the passed partial csum.	 */
302	ADDC32(sum, a2)
303	jr	ra
304	.set	noreorder
305	END(csum_partial)
306
307
308/*
309 * checksum and copy routines based on memcpy.S
310 *
311 *	csum_partial_copy_nocheck(src, dst, len, sum)
312 *	__csum_partial_copy_kernel(src, dst, len, sum, errp)
313 *
314 * See "Spec" in memcpy.S for details.	Unlike __copy_user, all
315 * function in this file use the standard calling convention.
316 */
317
318#define src a0
319#define dst a1
320#define len a2
321#define psum a3
322#define sum v0
323#define odd t8
324#define errptr t9
325
326/*
327 * The exception handler for loads requires that:
328 *  1- AT contain the address of the byte just past the end of the source
329 *     of the copy,
330 *  2- src_entry <= src < AT, and
331 *  3- (dst - src) == (dst_entry - src_entry),
332 * The _entry suffix denotes values when __copy_user was called.
333 *
334 * (1) is set up up by __csum_partial_copy_from_user and maintained by
335 *	not writing AT in __csum_partial_copy
336 * (2) is met by incrementing src by the number of bytes copied
337 * (3) is met by not doing loads between a pair of increments of dst and src
338 *
339 * The exception handlers for stores stores -EFAULT to errptr and return.
340 * These handlers do not need to overwrite any data.
341 */
342
343/* Instruction type */
344#define LD_INSN 1
345#define ST_INSN 2
346#define LEGACY_MODE 1
347#define EVA_MODE    2
348#define USEROP   1
349#define KERNELOP 2
350
351/*
352 * Wrapper to add an entry in the exception table
353 * in case the insn causes a memory exception.
354 * Arguments:
355 * insn    : Load/store instruction
356 * type    : Instruction type
357 * reg     : Register
358 * addr    : Address
359 * handler : Exception handler
360 */
361#define EXC(insn, type, reg, addr, handler)	\
362	.if \mode == LEGACY_MODE;		\
3639:		insn reg, addr;			\
364		.section __ex_table,"a";	\
365		PTR	9b, handler;		\
366		.previous;			\
367	/* This is enabled in EVA mode */	\
368	.else;					\
369		/* If loading from user or storing to user */	\
370		.if ((\from == USEROP) && (type == LD_INSN)) || \
371		    ((\to == USEROP) && (type == ST_INSN));	\
3729:			__BUILD_EVA_INSN(insn##e, reg, addr);	\
373			.section __ex_table,"a";		\
374			PTR	9b, handler;			\
375			.previous;				\
376		.else;						\
377			/* EVA without exception */		\
378			insn reg, addr;				\
379		.endif;						\
380	.endif
381
382#undef LOAD
383
384#ifdef USE_DOUBLE
385
386#define LOADK	ld /* No exception */
387#define LOAD(reg, addr, handler)	EXC(ld, LD_INSN, reg, addr, handler)
388#define LOADBU(reg, addr, handler)	EXC(lbu, LD_INSN, reg, addr, handler)
389#define LOADL(reg, addr, handler)	EXC(ldl, LD_INSN, reg, addr, handler)
390#define LOADR(reg, addr, handler)	EXC(ldr, LD_INSN, reg, addr, handler)
391#define STOREB(reg, addr, handler)	EXC(sb, ST_INSN, reg, addr, handler)
392#define STOREL(reg, addr, handler)	EXC(sdl, ST_INSN, reg, addr, handler)
393#define STORER(reg, addr, handler)	EXC(sdr, ST_INSN, reg, addr, handler)
394#define STORE(reg, addr, handler)	EXC(sd, ST_INSN, reg, addr, handler)
395#define ADD    daddu
396#define SUB    dsubu
397#define SRL    dsrl
398#define SLL    dsll
399#define SLLV   dsllv
400#define SRLV   dsrlv
401#define NBYTES 8
402#define LOG_NBYTES 3
403
404#else
405
406#define LOADK	lw /* No exception */
407#define LOAD(reg, addr, handler)	EXC(lw, LD_INSN, reg, addr, handler)
408#define LOADBU(reg, addr, handler)	EXC(lbu, LD_INSN, reg, addr, handler)
409#define LOADL(reg, addr, handler)	EXC(lwl, LD_INSN, reg, addr, handler)
410#define LOADR(reg, addr, handler)	EXC(lwr, LD_INSN, reg, addr, handler)
411#define STOREB(reg, addr, handler)	EXC(sb, ST_INSN, reg, addr, handler)
412#define STOREL(reg, addr, handler)	EXC(swl, ST_INSN, reg, addr, handler)
413#define STORER(reg, addr, handler)	EXC(swr, ST_INSN, reg, addr, handler)
414#define STORE(reg, addr, handler)	EXC(sw, ST_INSN, reg, addr, handler)
415#define ADD    addu
416#define SUB    subu
417#define SRL    srl
418#define SLL    sll
419#define SLLV   sllv
420#define SRLV   srlv
421#define NBYTES 4
422#define LOG_NBYTES 2
423
424#endif /* USE_DOUBLE */
425
426#ifdef CONFIG_CPU_LITTLE_ENDIAN
427#define LDFIRST LOADR
428#define LDREST	LOADL
429#define STFIRST STORER
430#define STREST	STOREL
431#define SHIFT_DISCARD SLLV
432#define SHIFT_DISCARD_REVERT SRLV
433#else
434#define LDFIRST LOADL
435#define LDREST	LOADR
436#define STFIRST STOREL
437#define STREST	STORER
438#define SHIFT_DISCARD SRLV
439#define SHIFT_DISCARD_REVERT SLLV
440#endif
441
442#define FIRST(unit) ((unit)*NBYTES)
443#define REST(unit)  (FIRST(unit)+NBYTES-1)
444
445#define ADDRMASK (NBYTES-1)
446
447#ifndef CONFIG_CPU_DADDI_WORKAROUNDS
448	.set	noat
449#else
450	.set	at=v1
451#endif
452
453	.macro __BUILD_CSUM_PARTIAL_COPY_USER mode, from, to, __nocheck
454
455	PTR_ADDU	AT, src, len	/* See (1) above. */
456	/* initialize __nocheck if this the first time we execute this
457	 * macro
458	 */
459#ifdef CONFIG_64BIT
460	move	errptr, a4
461#else
462	lw	errptr, 16(sp)
463#endif
464	.if \__nocheck == 1
465	FEXPORT(csum_partial_copy_nocheck)
466	EXPORT_SYMBOL(csum_partial_copy_nocheck)
467	.endif
468	move	sum, zero
469	move	odd, zero
470	/*
471	 * Note: dst & src may be unaligned, len may be 0
472	 * Temps
473	 */
474	/*
475	 * The "issue break"s below are very approximate.
476	 * Issue delays for dcache fills will perturb the schedule, as will
477	 * load queue full replay traps, etc.
478	 *
479	 * If len < NBYTES use byte operations.
480	 */
481	sltu	t2, len, NBYTES
482	and	t1, dst, ADDRMASK
483	bnez	t2, .Lcopy_bytes_checklen\@
484	 and	t0, src, ADDRMASK
485	andi	odd, dst, 0x1			/* odd buffer? */
486	bnez	t1, .Ldst_unaligned\@
487	 nop
488	bnez	t0, .Lsrc_unaligned_dst_aligned\@
489	/*
490	 * use delay slot for fall-through
491	 * src and dst are aligned; need to compute rem
492	 */
493.Lboth_aligned\@:
494	 SRL	t0, len, LOG_NBYTES+3	 # +3 for 8 units/iter
495	beqz	t0, .Lcleanup_both_aligned\@ # len < 8*NBYTES
496	 nop
497	SUB	len, 8*NBYTES		# subtract here for bgez loop
498	.align	4
4991:
500	LOAD(t0, UNIT(0)(src), .Ll_exc\@)
501	LOAD(t1, UNIT(1)(src), .Ll_exc_copy\@)
502	LOAD(t2, UNIT(2)(src), .Ll_exc_copy\@)
503	LOAD(t3, UNIT(3)(src), .Ll_exc_copy\@)
504	LOAD(t4, UNIT(4)(src), .Ll_exc_copy\@)
505	LOAD(t5, UNIT(5)(src), .Ll_exc_copy\@)
506	LOAD(t6, UNIT(6)(src), .Ll_exc_copy\@)
507	LOAD(t7, UNIT(7)(src), .Ll_exc_copy\@)
508	SUB	len, len, 8*NBYTES
509	ADD	src, src, 8*NBYTES
510	STORE(t0, UNIT(0)(dst),	.Ls_exc\@)
511	ADDC(t0, t1)
512	STORE(t1, UNIT(1)(dst),	.Ls_exc\@)
513	ADDC(sum, t0)
514	STORE(t2, UNIT(2)(dst),	.Ls_exc\@)
515	ADDC(t2, t3)
516	STORE(t3, UNIT(3)(dst),	.Ls_exc\@)
517	ADDC(sum, t2)
518	STORE(t4, UNIT(4)(dst),	.Ls_exc\@)
519	ADDC(t4, t5)
520	STORE(t5, UNIT(5)(dst),	.Ls_exc\@)
521	ADDC(sum, t4)
522	STORE(t6, UNIT(6)(dst),	.Ls_exc\@)
523	ADDC(t6, t7)
524	STORE(t7, UNIT(7)(dst),	.Ls_exc\@)
525	ADDC(sum, t6)
526	.set	reorder				/* DADDI_WAR */
527	ADD	dst, dst, 8*NBYTES
528	bgez	len, 1b
529	.set	noreorder
530	ADD	len, 8*NBYTES		# revert len (see above)
531
532	/*
533	 * len == the number of bytes left to copy < 8*NBYTES
534	 */
535.Lcleanup_both_aligned\@:
536#define rem t7
537	beqz	len, .Ldone\@
538	 sltu	t0, len, 4*NBYTES
539	bnez	t0, .Lless_than_4units\@
540	 and	rem, len, (NBYTES-1)	# rem = len % NBYTES
541	/*
542	 * len >= 4*NBYTES
543	 */
544	LOAD(t0, UNIT(0)(src), .Ll_exc\@)
545	LOAD(t1, UNIT(1)(src), .Ll_exc_copy\@)
546	LOAD(t2, UNIT(2)(src), .Ll_exc_copy\@)
547	LOAD(t3, UNIT(3)(src), .Ll_exc_copy\@)
548	SUB	len, len, 4*NBYTES
549	ADD	src, src, 4*NBYTES
550	STORE(t0, UNIT(0)(dst),	.Ls_exc\@)
551	ADDC(t0, t1)
552	STORE(t1, UNIT(1)(dst),	.Ls_exc\@)
553	ADDC(sum, t0)
554	STORE(t2, UNIT(2)(dst),	.Ls_exc\@)
555	ADDC(t2, t3)
556	STORE(t3, UNIT(3)(dst),	.Ls_exc\@)
557	ADDC(sum, t2)
558	.set	reorder				/* DADDI_WAR */
559	ADD	dst, dst, 4*NBYTES
560	beqz	len, .Ldone\@
561	.set	noreorder
562.Lless_than_4units\@:
563	/*
564	 * rem = len % NBYTES
565	 */
566	beq	rem, len, .Lcopy_bytes\@
567	 nop
5681:
569	LOAD(t0, 0(src), .Ll_exc\@)
570	ADD	src, src, NBYTES
571	SUB	len, len, NBYTES
572	STORE(t0, 0(dst), .Ls_exc\@)
573	ADDC(sum, t0)
574	.set	reorder				/* DADDI_WAR */
575	ADD	dst, dst, NBYTES
576	bne	rem, len, 1b
577	.set	noreorder
578
579	/*
580	 * src and dst are aligned, need to copy rem bytes (rem < NBYTES)
581	 * A loop would do only a byte at a time with possible branch
582	 * mispredicts.	 Can't do an explicit LOAD dst,mask,or,STORE
583	 * because can't assume read-access to dst.  Instead, use
584	 * STREST dst, which doesn't require read access to dst.
585	 *
586	 * This code should perform better than a simple loop on modern,
587	 * wide-issue mips processors because the code has fewer branches and
588	 * more instruction-level parallelism.
589	 */
590#define bits t2
591	beqz	len, .Ldone\@
592	 ADD	t1, dst, len	# t1 is just past last byte of dst
593	li	bits, 8*NBYTES
594	SLL	rem, len, 3	# rem = number of bits to keep
595	LOAD(t0, 0(src), .Ll_exc\@)
596	SUB	bits, bits, rem # bits = number of bits to discard
597	SHIFT_DISCARD t0, t0, bits
598	STREST(t0, -1(t1), .Ls_exc\@)
599	SHIFT_DISCARD_REVERT t0, t0, bits
600	.set reorder
601	ADDC(sum, t0)
602	b	.Ldone\@
603	.set noreorder
604.Ldst_unaligned\@:
605	/*
606	 * dst is unaligned
607	 * t0 = src & ADDRMASK
608	 * t1 = dst & ADDRMASK; T1 > 0
609	 * len >= NBYTES
610	 *
611	 * Copy enough bytes to align dst
612	 * Set match = (src and dst have same alignment)
613	 */
614#define match rem
615	LDFIRST(t3, FIRST(0)(src), .Ll_exc\@)
616	ADD	t2, zero, NBYTES
617	LDREST(t3, REST(0)(src), .Ll_exc_copy\@)
618	SUB	t2, t2, t1	# t2 = number of bytes copied
619	xor	match, t0, t1
620	STFIRST(t3, FIRST(0)(dst), .Ls_exc\@)
621	SLL	t4, t1, 3		# t4 = number of bits to discard
622	SHIFT_DISCARD t3, t3, t4
623	/* no SHIFT_DISCARD_REVERT to handle odd buffer properly */
624	ADDC(sum, t3)
625	beq	len, t2, .Ldone\@
626	 SUB	len, len, t2
627	ADD	dst, dst, t2
628	beqz	match, .Lboth_aligned\@
629	 ADD	src, src, t2
630
631.Lsrc_unaligned_dst_aligned\@:
632	SRL	t0, len, LOG_NBYTES+2	 # +2 for 4 units/iter
633	beqz	t0, .Lcleanup_src_unaligned\@
634	 and	rem, len, (4*NBYTES-1)	 # rem = len % 4*NBYTES
6351:
636/*
637 * Avoid consecutive LD*'s to the same register since some mips
638 * implementations can't issue them in the same cycle.
639 * It's OK to load FIRST(N+1) before REST(N) because the two addresses
640 * are to the same unit (unless src is aligned, but it's not).
641 */
642	LDFIRST(t0, FIRST(0)(src), .Ll_exc\@)
643	LDFIRST(t1, FIRST(1)(src), .Ll_exc_copy\@)
644	SUB	len, len, 4*NBYTES
645	LDREST(t0, REST(0)(src), .Ll_exc_copy\@)
646	LDREST(t1, REST(1)(src), .Ll_exc_copy\@)
647	LDFIRST(t2, FIRST(2)(src), .Ll_exc_copy\@)
648	LDFIRST(t3, FIRST(3)(src), .Ll_exc_copy\@)
649	LDREST(t2, REST(2)(src), .Ll_exc_copy\@)
650	LDREST(t3, REST(3)(src), .Ll_exc_copy\@)
651	ADD	src, src, 4*NBYTES
652#ifdef CONFIG_CPU_SB1
653	nop				# improves slotting
654#endif
655	STORE(t0, UNIT(0)(dst),	.Ls_exc\@)
656	ADDC(t0, t1)
657	STORE(t1, UNIT(1)(dst),	.Ls_exc\@)
658	ADDC(sum, t0)
659	STORE(t2, UNIT(2)(dst),	.Ls_exc\@)
660	ADDC(t2, t3)
661	STORE(t3, UNIT(3)(dst),	.Ls_exc\@)
662	ADDC(sum, t2)
663	.set	reorder				/* DADDI_WAR */
664	ADD	dst, dst, 4*NBYTES
665	bne	len, rem, 1b
666	.set	noreorder
667
668.Lcleanup_src_unaligned\@:
669	beqz	len, .Ldone\@
670	 and	rem, len, NBYTES-1  # rem = len % NBYTES
671	beq	rem, len, .Lcopy_bytes\@
672	 nop
6731:
674	LDFIRST(t0, FIRST(0)(src), .Ll_exc\@)
675	LDREST(t0, REST(0)(src), .Ll_exc_copy\@)
676	ADD	src, src, NBYTES
677	SUB	len, len, NBYTES
678	STORE(t0, 0(dst), .Ls_exc\@)
679	ADDC(sum, t0)
680	.set	reorder				/* DADDI_WAR */
681	ADD	dst, dst, NBYTES
682	bne	len, rem, 1b
683	.set	noreorder
684
685.Lcopy_bytes_checklen\@:
686	beqz	len, .Ldone\@
687	 nop
688.Lcopy_bytes\@:
689	/* 0 < len < NBYTES  */
690#ifdef CONFIG_CPU_LITTLE_ENDIAN
691#define SHIFT_START 0
692#define SHIFT_INC 8
693#else
694#define SHIFT_START 8*(NBYTES-1)
695#define SHIFT_INC -8
696#endif
697	move	t2, zero	# partial word
698	li	t3, SHIFT_START # shift
699/* use .Ll_exc_copy here to return correct sum on fault */
700#define COPY_BYTE(N)			\
701	LOADBU(t0, N(src), .Ll_exc_copy\@);	\
702	SUB	len, len, 1;		\
703	STOREB(t0, N(dst), .Ls_exc\@);	\
704	SLLV	t0, t0, t3;		\
705	addu	t3, SHIFT_INC;		\
706	beqz	len, .Lcopy_bytes_done\@; \
707	 or	t2, t0
708
709	COPY_BYTE(0)
710	COPY_BYTE(1)
711#ifdef USE_DOUBLE
712	COPY_BYTE(2)
713	COPY_BYTE(3)
714	COPY_BYTE(4)
715	COPY_BYTE(5)
716#endif
717	LOADBU(t0, NBYTES-2(src), .Ll_exc_copy\@)
718	SUB	len, len, 1
719	STOREB(t0, NBYTES-2(dst), .Ls_exc\@)
720	SLLV	t0, t0, t3
721	or	t2, t0
722.Lcopy_bytes_done\@:
723	ADDC(sum, t2)
724.Ldone\@:
725	/* fold checksum */
726	.set	push
727	.set	noat
728#ifdef USE_DOUBLE
729	dsll32	v1, sum, 0
730	daddu	sum, v1
731	sltu	v1, sum, v1
732	dsra32	sum, sum, 0
733	addu	sum, v1
734#endif
735
736#if defined(CONFIG_CPU_MIPSR2) || defined(CONFIG_CPU_MIPSR5) || \
737    defined(CONFIG_CPU_LOONGSON64)
738	.set	push
739	.set	arch=mips32r2
740	wsbh	v1, sum
741	movn	sum, v1, odd
742	.set	pop
743#else
744	beqz	odd, 1f			/* odd buffer alignment? */
745	 lui	v1, 0x00ff
746	addu	v1, 0x00ff
747	and	t0, sum, v1
748	sll	t0, t0, 8
749	srl	sum, sum, 8
750	and	sum, sum, v1
751	or	sum, sum, t0
7521:
753#endif
754	.set	pop
755	.set reorder
756	ADDC32(sum, psum)
757	jr	ra
758	.set noreorder
759
760.Ll_exc_copy\@:
761	/*
762	 * Copy bytes from src until faulting load address (or until a
763	 * lb faults)
764	 *
765	 * When reached by a faulting LDFIRST/LDREST, THREAD_BUADDR($28)
766	 * may be more than a byte beyond the last address.
767	 * Hence, the lb below may get an exception.
768	 *
769	 * Assumes src < THREAD_BUADDR($28)
770	 */
771	LOADK	t0, TI_TASK($28)
772	 li	t2, SHIFT_START
773	LOADK	t0, THREAD_BUADDR(t0)
7741:
775	LOADBU(t1, 0(src), .Ll_exc\@)
776	ADD	src, src, 1
777	sb	t1, 0(dst)	# can't fault -- we're copy_from_user
778	SLLV	t1, t1, t2
779	addu	t2, SHIFT_INC
780	ADDC(sum, t1)
781	.set	reorder				/* DADDI_WAR */
782	ADD	dst, dst, 1
783	bne	src, t0, 1b
784	.set	noreorder
785.Ll_exc\@:
786	LOADK	t0, TI_TASK($28)
787	 nop
788	LOADK	t0, THREAD_BUADDR(t0)	# t0 is just past last good address
789	 nop
790	SUB	len, AT, t0		# len number of uncopied bytes
791	/*
792	 * Here's where we rely on src and dst being incremented in tandem,
793	 *   See (3) above.
794	 * dst += (fault addr - src) to put dst at first byte to clear
795	 */
796	ADD	dst, t0			# compute start address in a1
797	SUB	dst, src
798	/*
799	 * Clear len bytes starting at dst.  Can't call __bzero because it
800	 * might modify len.  An inefficient loop for these rare times...
801	 */
802	.set	reorder				/* DADDI_WAR */
803	SUB	src, len, 1
804	beqz	len, .Ldone\@
805	.set	noreorder
8061:	sb	zero, 0(dst)
807	ADD	dst, dst, 1
808	.set	push
809	.set	noat
810#ifndef CONFIG_CPU_DADDI_WORKAROUNDS
811	bnez	src, 1b
812	 SUB	src, src, 1
813#else
814	li	v1, 1
815	bnez	src, 1b
816	 SUB	src, src, v1
817#endif
818	li	v1, -EFAULT
819	b	.Ldone\@
820	 sw	v1, (errptr)
821
822.Ls_exc\@:
823	li	v0, -1 /* invalid checksum */
824	li	v1, -EFAULT
825	jr	ra
826	 sw	v1, (errptr)
827	.set	pop
828	.endm
829
830LEAF(__csum_partial_copy_kernel)
831EXPORT_SYMBOL(__csum_partial_copy_kernel)
832#ifndef CONFIG_EVA
833FEXPORT(__csum_partial_copy_to_user)
834EXPORT_SYMBOL(__csum_partial_copy_to_user)
835FEXPORT(__csum_partial_copy_from_user)
836EXPORT_SYMBOL(__csum_partial_copy_from_user)
837#endif
838__BUILD_CSUM_PARTIAL_COPY_USER LEGACY_MODE USEROP USEROP 1
839END(__csum_partial_copy_kernel)
840
841#ifdef CONFIG_EVA
842LEAF(__csum_partial_copy_to_user)
843__BUILD_CSUM_PARTIAL_COPY_USER EVA_MODE KERNELOP USEROP 0
844END(__csum_partial_copy_to_user)
845
846LEAF(__csum_partial_copy_from_user)
847__BUILD_CSUM_PARTIAL_COPY_USER EVA_MODE USEROP KERNELOP 0
848END(__csum_partial_copy_from_user)
849#endif
850