xref: /openbmc/linux/arch/mips/lib/csum_partial.S (revision d2168146)
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/regdef.h>
17
18#ifdef CONFIG_64BIT
19/*
20 * As we are sharing code base with the mips32 tree (which use the o32 ABI
21 * register definitions). We need to redefine the register definitions from
22 * the n64 ABI register naming to the o32 ABI register naming.
23 */
24#undef t0
25#undef t1
26#undef t2
27#undef t3
28#define t0	$8
29#define t1	$9
30#define t2	$10
31#define t3	$11
32#define t4	$12
33#define t5	$13
34#define t6	$14
35#define t7	$15
36
37#define USE_DOUBLE
38#endif
39
40#ifdef USE_DOUBLE
41
42#define LOAD   ld
43#define LOAD32 lwu
44#define ADD    daddu
45#define NBYTES 8
46
47#else
48
49#define LOAD   lw
50#define LOAD32 lw
51#define ADD    addu
52#define NBYTES 4
53
54#endif /* USE_DOUBLE */
55
56#define UNIT(unit)  ((unit)*NBYTES)
57
58#define ADDC(sum,reg)						\
59	.set	push;						\
60	.set	noat;						\
61	ADD	sum, reg;					\
62	sltu	v1, sum, reg;					\
63	ADD	sum, v1;					\
64	.set	pop
65
66#define ADDC32(sum,reg)						\
67	.set	push;						\
68	.set	noat;						\
69	addu	sum, reg;					\
70	sltu	v1, sum, reg;					\
71	addu	sum, v1;					\
72	.set	pop
73
74#define CSUM_BIGCHUNK1(src, offset, sum, _t0, _t1, _t2, _t3)	\
75	LOAD	_t0, (offset + UNIT(0))(src);			\
76	LOAD	_t1, (offset + UNIT(1))(src);			\
77	LOAD	_t2, (offset + UNIT(2))(src);			\
78	LOAD	_t3, (offset + UNIT(3))(src);			\
79	ADDC(sum, _t0);						\
80	ADDC(sum, _t1);						\
81	ADDC(sum, _t2);						\
82	ADDC(sum, _t3)
83
84#ifdef USE_DOUBLE
85#define CSUM_BIGCHUNK(src, offset, sum, _t0, _t1, _t2, _t3)	\
86	CSUM_BIGCHUNK1(src, offset, sum, _t0, _t1, _t2, _t3)
87#else
88#define CSUM_BIGCHUNK(src, offset, sum, _t0, _t1, _t2, _t3)	\
89	CSUM_BIGCHUNK1(src, offset, sum, _t0, _t1, _t2, _t3);	\
90	CSUM_BIGCHUNK1(src, offset + 0x10, sum, _t0, _t1, _t2, _t3)
91#endif
92
93/*
94 * a0: source address
95 * a1: length of the area to checksum
96 * a2: partial checksum
97 */
98
99#define src a0
100#define sum v0
101
102	.text
103	.set	noreorder
104	.align	5
105LEAF(csum_partial)
106	move	sum, zero
107	move	t7, zero
108
109	sltiu	t8, a1, 0x8
110	bnez	t8, .Lsmall_csumcpy		/* < 8 bytes to copy */
111	 move	t2, a1
112
113	andi	t7, src, 0x1			/* odd buffer? */
114
115.Lhword_align:
116	beqz	t7, .Lword_align
117	 andi	t8, src, 0x2
118
119	lbu	t0, (src)
120	LONG_SUBU	a1, a1, 0x1
121#ifdef __MIPSEL__
122	sll	t0, t0, 8
123#endif
124	ADDC(sum, t0)
125	PTR_ADDU	src, src, 0x1
126	andi	t8, src, 0x2
127
128.Lword_align:
129	beqz	t8, .Ldword_align
130	 sltiu	t8, a1, 56
131
132	lhu	t0, (src)
133	LONG_SUBU	a1, a1, 0x2
134	ADDC(sum, t0)
135	sltiu	t8, a1, 56
136	PTR_ADDU	src, src, 0x2
137
138.Ldword_align:
139	bnez	t8, .Ldo_end_words
140	 move	t8, a1
141
142	andi	t8, src, 0x4
143	beqz	t8, .Lqword_align
144	 andi	t8, src, 0x8
145
146	LOAD32	t0, 0x00(src)
147	LONG_SUBU	a1, a1, 0x4
148	ADDC(sum, t0)
149	PTR_ADDU	src, src, 0x4
150	andi	t8, src, 0x8
151
152.Lqword_align:
153	beqz	t8, .Loword_align
154	 andi	t8, src, 0x10
155
156#ifdef USE_DOUBLE
157	ld	t0, 0x00(src)
158	LONG_SUBU	a1, a1, 0x8
159	ADDC(sum, t0)
160#else
161	lw	t0, 0x00(src)
162	lw	t1, 0x04(src)
163	LONG_SUBU	a1, a1, 0x8
164	ADDC(sum, t0)
165	ADDC(sum, t1)
166#endif
167	PTR_ADDU	src, src, 0x8
168	andi	t8, src, 0x10
169
170.Loword_align:
171	beqz	t8, .Lbegin_movement
172	 LONG_SRL	t8, a1, 0x7
173
174#ifdef USE_DOUBLE
175	ld	t0, 0x00(src)
176	ld	t1, 0x08(src)
177	ADDC(sum, t0)
178	ADDC(sum, t1)
179#else
180	CSUM_BIGCHUNK1(src, 0x00, sum, t0, t1, t3, t4)
181#endif
182	LONG_SUBU	a1, a1, 0x10
183	PTR_ADDU	src, src, 0x10
184	LONG_SRL	t8, a1, 0x7
185
186.Lbegin_movement:
187	beqz	t8, 1f
188	 andi	t2, a1, 0x40
189
190.Lmove_128bytes:
191	CSUM_BIGCHUNK(src, 0x00, sum, t0, t1, t3, t4)
192	CSUM_BIGCHUNK(src, 0x20, sum, t0, t1, t3, t4)
193	CSUM_BIGCHUNK(src, 0x40, sum, t0, t1, t3, t4)
194	CSUM_BIGCHUNK(src, 0x60, sum, t0, t1, t3, t4)
195	LONG_SUBU	t8, t8, 0x01
196	.set	reorder				/* DADDI_WAR */
197	PTR_ADDU	src, src, 0x80
198	bnez	t8, .Lmove_128bytes
199	.set	noreorder
200
2011:
202	beqz	t2, 1f
203	 andi	t2, a1, 0x20
204
205.Lmove_64bytes:
206	CSUM_BIGCHUNK(src, 0x00, sum, t0, t1, t3, t4)
207	CSUM_BIGCHUNK(src, 0x20, sum, t0, t1, t3, t4)
208	PTR_ADDU	src, src, 0x40
209
2101:
211	beqz	t2, .Ldo_end_words
212	 andi	t8, a1, 0x1c
213
214.Lmove_32bytes:
215	CSUM_BIGCHUNK(src, 0x00, sum, t0, t1, t3, t4)
216	andi	t8, a1, 0x1c
217	PTR_ADDU	src, src, 0x20
218
219.Ldo_end_words:
220	beqz	t8, .Lsmall_csumcpy
221	 andi	t2, a1, 0x3
222	LONG_SRL	t8, t8, 0x2
223
224.Lend_words:
225	LOAD32	t0, (src)
226	LONG_SUBU	t8, t8, 0x1
227	ADDC(sum, t0)
228	.set	reorder				/* DADDI_WAR */
229	PTR_ADDU	src, src, 0x4
230	bnez	t8, .Lend_words
231	.set	noreorder
232
233/* unknown src alignment and < 8 bytes to go  */
234.Lsmall_csumcpy:
235	move	a1, t2
236
237	andi	t0, a1, 4
238	beqz	t0, 1f
239	 andi	t0, a1, 2
240
241	/* Still a full word to go  */
242	ulw	t1, (src)
243	PTR_ADDIU	src, 4
244#ifdef USE_DOUBLE
245	dsll	t1, t1, 32			/* clear lower 32bit */
246#endif
247	ADDC(sum, t1)
248
2491:	move	t1, zero
250	beqz	t0, 1f
251	 andi	t0, a1, 1
252
253	/* Still a halfword to go  */
254	ulhu	t1, (src)
255	PTR_ADDIU	src, 2
256
2571:	beqz	t0, 1f
258	 sll	t1, t1, 16
259
260	lbu	t2, (src)
261	 nop
262
263#ifdef __MIPSEB__
264	sll	t2, t2, 8
265#endif
266	or	t1, t2
267
2681:	ADDC(sum, t1)
269
270	/* fold checksum */
271#ifdef USE_DOUBLE
272	dsll32	v1, sum, 0
273	daddu	sum, v1
274	sltu	v1, sum, v1
275	dsra32	sum, sum, 0
276	addu	sum, v1
277#endif
278
279	/* odd buffer alignment? */
280#ifdef CONFIG_CPU_MIPSR2
281	wsbh	v1, sum
282	movn	sum, v1, t7
283#else
284	beqz	t7, 1f			/* odd buffer alignment? */
285	 lui	v1, 0x00ff
286	addu	v1, 0x00ff
287	and	t0, sum, v1
288	sll	t0, t0, 8
289	srl	sum, sum, 8
290	and	sum, sum, v1
291	or	sum, sum, t0
2921:
293#endif
294	.set	reorder
295	/* Add the passed partial csum.	 */
296	ADDC32(sum, a2)
297	jr	ra
298	.set	noreorder
299	END(csum_partial)
300
301
302/*
303 * checksum and copy routines based on memcpy.S
304 *
305 *	csum_partial_copy_nocheck(src, dst, len, sum)
306 *	__csum_partial_copy_kernel(src, dst, len, sum, errp)
307 *
308 * See "Spec" in memcpy.S for details.	Unlike __copy_user, all
309 * function in this file use the standard calling convention.
310 */
311
312#define src a0
313#define dst a1
314#define len a2
315#define psum a3
316#define sum v0
317#define odd t8
318#define errptr t9
319
320/*
321 * The exception handler for loads requires that:
322 *  1- AT contain the address of the byte just past the end of the source
323 *     of the copy,
324 *  2- src_entry <= src < AT, and
325 *  3- (dst - src) == (dst_entry - src_entry),
326 * The _entry suffix denotes values when __copy_user was called.
327 *
328 * (1) is set up up by __csum_partial_copy_from_user and maintained by
329 *	not writing AT in __csum_partial_copy
330 * (2) is met by incrementing src by the number of bytes copied
331 * (3) is met by not doing loads between a pair of increments of dst and src
332 *
333 * The exception handlers for stores stores -EFAULT to errptr and return.
334 * These handlers do not need to overwrite any data.
335 */
336
337/* Instruction type */
338#define LD_INSN 1
339#define ST_INSN 2
340#define LEGACY_MODE 1
341#define EVA_MODE    2
342#define USEROP   1
343#define KERNELOP 2
344
345/*
346 * Wrapper to add an entry in the exception table
347 * in case the insn causes a memory exception.
348 * Arguments:
349 * insn    : Load/store instruction
350 * type    : Instruction type
351 * reg     : Register
352 * addr    : Address
353 * handler : Exception handler
354 */
355#define EXC(insn, type, reg, addr, handler)	\
356	.if \mode == LEGACY_MODE;		\
3579:		insn reg, addr;			\
358		.section __ex_table,"a";	\
359		PTR	9b, handler;		\
360		.previous;			\
361	/* This is enabled in EVA mode */	\
362	.else;					\
363		/* If loading from user or storing to user */	\
364		.if ((\from == USEROP) && (type == LD_INSN)) || \
365		    ((\to == USEROP) && (type == ST_INSN));	\
3669:			__BUILD_EVA_INSN(insn##e, reg, addr);	\
367			.section __ex_table,"a";		\
368			PTR	9b, handler;			\
369			.previous;				\
370		.else;						\
371			/* EVA without exception */		\
372			insn reg, addr;				\
373		.endif;						\
374	.endif
375
376#undef LOAD
377
378#ifdef USE_DOUBLE
379
380#define LOADK	ld /* No exception */
381#define LOAD(reg, addr, handler)	EXC(ld, LD_INSN, reg, addr, handler)
382#define LOADBU(reg, addr, handler)	EXC(lbu, LD_INSN, reg, addr, handler)
383#define LOADL(reg, addr, handler)	EXC(ldl, LD_INSN, reg, addr, handler)
384#define LOADR(reg, addr, handler)	EXC(ldr, LD_INSN, reg, addr, handler)
385#define STOREB(reg, addr, handler)	EXC(sb, ST_INSN, reg, addr, handler)
386#define STOREL(reg, addr, handler)	EXC(sdl, ST_INSN, reg, addr, handler)
387#define STORER(reg, addr, handler)	EXC(sdr, ST_INSN, reg, addr, handler)
388#define STORE(reg, addr, handler)	EXC(sd, ST_INSN, reg, addr, handler)
389#define ADD    daddu
390#define SUB    dsubu
391#define SRL    dsrl
392#define SLL    dsll
393#define SLLV   dsllv
394#define SRLV   dsrlv
395#define NBYTES 8
396#define LOG_NBYTES 3
397
398#else
399
400#define LOADK	lw /* No exception */
401#define LOAD(reg, addr, handler)	EXC(lw, LD_INSN, reg, addr, handler)
402#define LOADBU(reg, addr, handler)	EXC(lbu, LD_INSN, reg, addr, handler)
403#define LOADL(reg, addr, handler)	EXC(lwl, LD_INSN, reg, addr, handler)
404#define LOADR(reg, addr, handler)	EXC(lwr, LD_INSN, reg, addr, handler)
405#define STOREB(reg, addr, handler)	EXC(sb, ST_INSN, reg, addr, handler)
406#define STOREL(reg, addr, handler)	EXC(swl, ST_INSN, reg, addr, handler)
407#define STORER(reg, addr, handler)	EXC(swr, ST_INSN, reg, addr, handler)
408#define STORE(reg, addr, handler)	EXC(sw, ST_INSN, reg, addr, handler)
409#define ADD    addu
410#define SUB    subu
411#define SRL    srl
412#define SLL    sll
413#define SLLV   sllv
414#define SRLV   srlv
415#define NBYTES 4
416#define LOG_NBYTES 2
417
418#endif /* USE_DOUBLE */
419
420#ifdef CONFIG_CPU_LITTLE_ENDIAN
421#define LDFIRST LOADR
422#define LDREST	LOADL
423#define STFIRST STORER
424#define STREST	STOREL
425#define SHIFT_DISCARD SLLV
426#define SHIFT_DISCARD_REVERT SRLV
427#else
428#define LDFIRST LOADL
429#define LDREST	LOADR
430#define STFIRST STOREL
431#define STREST	STORER
432#define SHIFT_DISCARD SRLV
433#define SHIFT_DISCARD_REVERT SLLV
434#endif
435
436#define FIRST(unit) ((unit)*NBYTES)
437#define REST(unit)  (FIRST(unit)+NBYTES-1)
438
439#define ADDRMASK (NBYTES-1)
440
441#ifndef CONFIG_CPU_DADDI_WORKAROUNDS
442	.set	noat
443#else
444	.set	at=v1
445#endif
446
447	.macro __BUILD_CSUM_PARTIAL_COPY_USER mode, from, to, __nocheck
448
449	PTR_ADDU	AT, src, len	/* See (1) above. */
450	/* initialize __nocheck if this the first time we execute this
451	 * macro
452	 */
453#ifdef CONFIG_64BIT
454	move	errptr, a4
455#else
456	lw	errptr, 16(sp)
457#endif
458	.if \__nocheck == 1
459	FEXPORT(csum_partial_copy_nocheck)
460	.endif
461	move	sum, zero
462	move	odd, zero
463	/*
464	 * Note: dst & src may be unaligned, len may be 0
465	 * Temps
466	 */
467	/*
468	 * The "issue break"s below are very approximate.
469	 * Issue delays for dcache fills will perturb the schedule, as will
470	 * load queue full replay traps, etc.
471	 *
472	 * If len < NBYTES use byte operations.
473	 */
474	sltu	t2, len, NBYTES
475	and	t1, dst, ADDRMASK
476	bnez	t2, .Lcopy_bytes_checklen\@
477	 and	t0, src, ADDRMASK
478	andi	odd, dst, 0x1			/* odd buffer? */
479	bnez	t1, .Ldst_unaligned\@
480	 nop
481	bnez	t0, .Lsrc_unaligned_dst_aligned\@
482	/*
483	 * use delay slot for fall-through
484	 * src and dst are aligned; need to compute rem
485	 */
486.Lboth_aligned\@:
487	 SRL	t0, len, LOG_NBYTES+3	 # +3 for 8 units/iter
488	beqz	t0, .Lcleanup_both_aligned\@ # len < 8*NBYTES
489	 nop
490	SUB	len, 8*NBYTES		# subtract here for bgez loop
491	.align	4
4921:
493	LOAD(t0, UNIT(0)(src), .Ll_exc\@)
494	LOAD(t1, UNIT(1)(src), .Ll_exc_copy\@)
495	LOAD(t2, UNIT(2)(src), .Ll_exc_copy\@)
496	LOAD(t3, UNIT(3)(src), .Ll_exc_copy\@)
497	LOAD(t4, UNIT(4)(src), .Ll_exc_copy\@)
498	LOAD(t5, UNIT(5)(src), .Ll_exc_copy\@)
499	LOAD(t6, UNIT(6)(src), .Ll_exc_copy\@)
500	LOAD(t7, UNIT(7)(src), .Ll_exc_copy\@)
501	SUB	len, len, 8*NBYTES
502	ADD	src, src, 8*NBYTES
503	STORE(t0, UNIT(0)(dst),	.Ls_exc\@)
504	ADDC(sum, t0)
505	STORE(t1, UNIT(1)(dst),	.Ls_exc\@)
506	ADDC(sum, t1)
507	STORE(t2, UNIT(2)(dst),	.Ls_exc\@)
508	ADDC(sum, t2)
509	STORE(t3, UNIT(3)(dst),	.Ls_exc\@)
510	ADDC(sum, t3)
511	STORE(t4, UNIT(4)(dst),	.Ls_exc\@)
512	ADDC(sum, t4)
513	STORE(t5, UNIT(5)(dst),	.Ls_exc\@)
514	ADDC(sum, t5)
515	STORE(t6, UNIT(6)(dst),	.Ls_exc\@)
516	ADDC(sum, t6)
517	STORE(t7, UNIT(7)(dst),	.Ls_exc\@)
518	ADDC(sum, t7)
519	.set	reorder				/* DADDI_WAR */
520	ADD	dst, dst, 8*NBYTES
521	bgez	len, 1b
522	.set	noreorder
523	ADD	len, 8*NBYTES		# revert len (see above)
524
525	/*
526	 * len == the number of bytes left to copy < 8*NBYTES
527	 */
528.Lcleanup_both_aligned\@:
529#define rem t7
530	beqz	len, .Ldone\@
531	 sltu	t0, len, 4*NBYTES
532	bnez	t0, .Lless_than_4units\@
533	 and	rem, len, (NBYTES-1)	# rem = len % NBYTES
534	/*
535	 * len >= 4*NBYTES
536	 */
537	LOAD(t0, UNIT(0)(src), .Ll_exc\@)
538	LOAD(t1, UNIT(1)(src), .Ll_exc_copy\@)
539	LOAD(t2, UNIT(2)(src), .Ll_exc_copy\@)
540	LOAD(t3, UNIT(3)(src), .Ll_exc_copy\@)
541	SUB	len, len, 4*NBYTES
542	ADD	src, src, 4*NBYTES
543	STORE(t0, UNIT(0)(dst),	.Ls_exc\@)
544	ADDC(sum, t0)
545	STORE(t1, UNIT(1)(dst),	.Ls_exc\@)
546	ADDC(sum, t1)
547	STORE(t2, UNIT(2)(dst),	.Ls_exc\@)
548	ADDC(sum, t2)
549	STORE(t3, UNIT(3)(dst),	.Ls_exc\@)
550	ADDC(sum, t3)
551	.set	reorder				/* DADDI_WAR */
552	ADD	dst, dst, 4*NBYTES
553	beqz	len, .Ldone\@
554	.set	noreorder
555.Lless_than_4units\@:
556	/*
557	 * rem = len % NBYTES
558	 */
559	beq	rem, len, .Lcopy_bytes\@
560	 nop
5611:
562	LOAD(t0, 0(src), .Ll_exc\@)
563	ADD	src, src, NBYTES
564	SUB	len, len, NBYTES
565	STORE(t0, 0(dst), .Ls_exc\@)
566	ADDC(sum, t0)
567	.set	reorder				/* DADDI_WAR */
568	ADD	dst, dst, NBYTES
569	bne	rem, len, 1b
570	.set	noreorder
571
572	/*
573	 * src and dst are aligned, need to copy rem bytes (rem < NBYTES)
574	 * A loop would do only a byte at a time with possible branch
575	 * mispredicts.	 Can't do an explicit LOAD dst,mask,or,STORE
576	 * because can't assume read-access to dst.  Instead, use
577	 * STREST dst, which doesn't require read access to dst.
578	 *
579	 * This code should perform better than a simple loop on modern,
580	 * wide-issue mips processors because the code has fewer branches and
581	 * more instruction-level parallelism.
582	 */
583#define bits t2
584	beqz	len, .Ldone\@
585	 ADD	t1, dst, len	# t1 is just past last byte of dst
586	li	bits, 8*NBYTES
587	SLL	rem, len, 3	# rem = number of bits to keep
588	LOAD(t0, 0(src), .Ll_exc\@)
589	SUB	bits, bits, rem # bits = number of bits to discard
590	SHIFT_DISCARD t0, t0, bits
591	STREST(t0, -1(t1), .Ls_exc\@)
592	SHIFT_DISCARD_REVERT t0, t0, bits
593	.set reorder
594	ADDC(sum, t0)
595	b	.Ldone\@
596	.set noreorder
597.Ldst_unaligned\@:
598	/*
599	 * dst is unaligned
600	 * t0 = src & ADDRMASK
601	 * t1 = dst & ADDRMASK; T1 > 0
602	 * len >= NBYTES
603	 *
604	 * Copy enough bytes to align dst
605	 * Set match = (src and dst have same alignment)
606	 */
607#define match rem
608	LDFIRST(t3, FIRST(0)(src), .Ll_exc\@)
609	ADD	t2, zero, NBYTES
610	LDREST(t3, REST(0)(src), .Ll_exc_copy\@)
611	SUB	t2, t2, t1	# t2 = number of bytes copied
612	xor	match, t0, t1
613	STFIRST(t3, FIRST(0)(dst), .Ls_exc\@)
614	SLL	t4, t1, 3		# t4 = number of bits to discard
615	SHIFT_DISCARD t3, t3, t4
616	/* no SHIFT_DISCARD_REVERT to handle odd buffer properly */
617	ADDC(sum, t3)
618	beq	len, t2, .Ldone\@
619	 SUB	len, len, t2
620	ADD	dst, dst, t2
621	beqz	match, .Lboth_aligned\@
622	 ADD	src, src, t2
623
624.Lsrc_unaligned_dst_aligned\@:
625	SRL	t0, len, LOG_NBYTES+2	 # +2 for 4 units/iter
626	beqz	t0, .Lcleanup_src_unaligned\@
627	 and	rem, len, (4*NBYTES-1)	 # rem = len % 4*NBYTES
6281:
629/*
630 * Avoid consecutive LD*'s to the same register since some mips
631 * implementations can't issue them in the same cycle.
632 * It's OK to load FIRST(N+1) before REST(N) because the two addresses
633 * are to the same unit (unless src is aligned, but it's not).
634 */
635	LDFIRST(t0, FIRST(0)(src), .Ll_exc\@)
636	LDFIRST(t1, FIRST(1)(src), .Ll_exc_copy\@)
637	SUB	len, len, 4*NBYTES
638	LDREST(t0, REST(0)(src), .Ll_exc_copy\@)
639	LDREST(t1, REST(1)(src), .Ll_exc_copy\@)
640	LDFIRST(t2, FIRST(2)(src), .Ll_exc_copy\@)
641	LDFIRST(t3, FIRST(3)(src), .Ll_exc_copy\@)
642	LDREST(t2, REST(2)(src), .Ll_exc_copy\@)
643	LDREST(t3, REST(3)(src), .Ll_exc_copy\@)
644	ADD	src, src, 4*NBYTES
645#ifdef CONFIG_CPU_SB1
646	nop				# improves slotting
647#endif
648	STORE(t0, UNIT(0)(dst),	.Ls_exc\@)
649	ADDC(sum, t0)
650	STORE(t1, UNIT(1)(dst),	.Ls_exc\@)
651	ADDC(sum, t1)
652	STORE(t2, UNIT(2)(dst),	.Ls_exc\@)
653	ADDC(sum, t2)
654	STORE(t3, UNIT(3)(dst),	.Ls_exc\@)
655	ADDC(sum, t3)
656	.set	reorder				/* DADDI_WAR */
657	ADD	dst, dst, 4*NBYTES
658	bne	len, rem, 1b
659	.set	noreorder
660
661.Lcleanup_src_unaligned\@:
662	beqz	len, .Ldone\@
663	 and	rem, len, NBYTES-1  # rem = len % NBYTES
664	beq	rem, len, .Lcopy_bytes\@
665	 nop
6661:
667	LDFIRST(t0, FIRST(0)(src), .Ll_exc\@)
668	LDREST(t0, REST(0)(src), .Ll_exc_copy\@)
669	ADD	src, src, NBYTES
670	SUB	len, len, NBYTES
671	STORE(t0, 0(dst), .Ls_exc\@)
672	ADDC(sum, t0)
673	.set	reorder				/* DADDI_WAR */
674	ADD	dst, dst, NBYTES
675	bne	len, rem, 1b
676	.set	noreorder
677
678.Lcopy_bytes_checklen\@:
679	beqz	len, .Ldone\@
680	 nop
681.Lcopy_bytes\@:
682	/* 0 < len < NBYTES  */
683#ifdef CONFIG_CPU_LITTLE_ENDIAN
684#define SHIFT_START 0
685#define SHIFT_INC 8
686#else
687#define SHIFT_START 8*(NBYTES-1)
688#define SHIFT_INC -8
689#endif
690	move	t2, zero	# partial word
691	li	t3, SHIFT_START # shift
692/* use .Ll_exc_copy here to return correct sum on fault */
693#define COPY_BYTE(N)			\
694	LOADBU(t0, N(src), .Ll_exc_copy\@);	\
695	SUB	len, len, 1;		\
696	STOREB(t0, N(dst), .Ls_exc\@);	\
697	SLLV	t0, t0, t3;		\
698	addu	t3, SHIFT_INC;		\
699	beqz	len, .Lcopy_bytes_done\@; \
700	 or	t2, t0
701
702	COPY_BYTE(0)
703	COPY_BYTE(1)
704#ifdef USE_DOUBLE
705	COPY_BYTE(2)
706	COPY_BYTE(3)
707	COPY_BYTE(4)
708	COPY_BYTE(5)
709#endif
710	LOADBU(t0, NBYTES-2(src), .Ll_exc_copy\@)
711	SUB	len, len, 1
712	STOREB(t0, NBYTES-2(dst), .Ls_exc\@)
713	SLLV	t0, t0, t3
714	or	t2, t0
715.Lcopy_bytes_done\@:
716	ADDC(sum, t2)
717.Ldone\@:
718	/* fold checksum */
719	.set	push
720	.set	noat
721#ifdef USE_DOUBLE
722	dsll32	v1, sum, 0
723	daddu	sum, v1
724	sltu	v1, sum, v1
725	dsra32	sum, sum, 0
726	addu	sum, v1
727#endif
728
729#ifdef CONFIG_CPU_MIPSR2
730	wsbh	v1, sum
731	movn	sum, v1, odd
732#else
733	beqz	odd, 1f			/* odd buffer alignment? */
734	 lui	v1, 0x00ff
735	addu	v1, 0x00ff
736	and	t0, sum, v1
737	sll	t0, t0, 8
738	srl	sum, sum, 8
739	and	sum, sum, v1
740	or	sum, sum, t0
7411:
742#endif
743	.set	pop
744	.set reorder
745	ADDC32(sum, psum)
746	jr	ra
747	.set noreorder
748
749.Ll_exc_copy\@:
750	/*
751	 * Copy bytes from src until faulting load address (or until a
752	 * lb faults)
753	 *
754	 * When reached by a faulting LDFIRST/LDREST, THREAD_BUADDR($28)
755	 * may be more than a byte beyond the last address.
756	 * Hence, the lb below may get an exception.
757	 *
758	 * Assumes src < THREAD_BUADDR($28)
759	 */
760	LOADK	t0, TI_TASK($28)
761	 li	t2, SHIFT_START
762	LOADK	t0, THREAD_BUADDR(t0)
7631:
764	LOADBU(t1, 0(src), .Ll_exc\@)
765	ADD	src, src, 1
766	sb	t1, 0(dst)	# can't fault -- we're copy_from_user
767	SLLV	t1, t1, t2
768	addu	t2, SHIFT_INC
769	ADDC(sum, t1)
770	.set	reorder				/* DADDI_WAR */
771	ADD	dst, dst, 1
772	bne	src, t0, 1b
773	.set	noreorder
774.Ll_exc\@:
775	LOADK	t0, TI_TASK($28)
776	 nop
777	LOADK	t0, THREAD_BUADDR(t0)	# t0 is just past last good address
778	 nop
779	SUB	len, AT, t0		# len number of uncopied bytes
780	/*
781	 * Here's where we rely on src and dst being incremented in tandem,
782	 *   See (3) above.
783	 * dst += (fault addr - src) to put dst at first byte to clear
784	 */
785	ADD	dst, t0			# compute start address in a1
786	SUB	dst, src
787	/*
788	 * Clear len bytes starting at dst.  Can't call __bzero because it
789	 * might modify len.  An inefficient loop for these rare times...
790	 */
791	.set	reorder				/* DADDI_WAR */
792	SUB	src, len, 1
793	beqz	len, .Ldone\@
794	.set	noreorder
7951:	sb	zero, 0(dst)
796	ADD	dst, dst, 1
797	.set	push
798	.set	noat
799#ifndef CONFIG_CPU_DADDI_WORKAROUNDS
800	bnez	src, 1b
801	 SUB	src, src, 1
802#else
803	li	v1, 1
804	bnez	src, 1b
805	 SUB	src, src, v1
806#endif
807	li	v1, -EFAULT
808	b	.Ldone\@
809	 sw	v1, (errptr)
810
811.Ls_exc\@:
812	li	v0, -1 /* invalid checksum */
813	li	v1, -EFAULT
814	jr	ra
815	 sw	v1, (errptr)
816	.set	pop
817	.endm
818
819LEAF(__csum_partial_copy_kernel)
820#ifndef CONFIG_EVA
821FEXPORT(__csum_partial_copy_to_user)
822FEXPORT(__csum_partial_copy_from_user)
823#endif
824__BUILD_CSUM_PARTIAL_COPY_USER LEGACY_MODE USEROP USEROP 1
825END(__csum_partial_copy_kernel)
826
827#ifdef CONFIG_EVA
828LEAF(__csum_partial_copy_to_user)
829__BUILD_CSUM_PARTIAL_COPY_USER EVA_MODE KERNELOP USEROP 0
830END(__csum_partial_copy_to_user)
831
832LEAF(__csum_partial_copy_from_user)
833__BUILD_CSUM_PARTIAL_COPY_USER EVA_MODE USEROP KERNELOP 0
834END(__csum_partial_copy_from_user)
835#endif
836