xref: /linux/arch/powerpc/kernel/misc_32.S (revision c7fb577e2a6cb04732541f2dc402bd46747f7558)
1/*
2 * This file contains miscellaneous low-level functions.
3 *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
4 *
5 * Largely rewritten by Cort Dougan (cort@cs.nmt.edu)
6 * and Paul Mackerras.
7 *
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License
10 * as published by the Free Software Foundation; either version
11 * 2 of the License, or (at your option) any later version.
12 *
13 */
14
15#include <linux/config.h>
16#include <linux/sys.h>
17#include <asm/unistd.h>
18#include <asm/errno.h>
19#include <asm/reg.h>
20#include <asm/page.h>
21#include <asm/cache.h>
22#include <asm/cputable.h>
23#include <asm/mmu.h>
24#include <asm/ppc_asm.h>
25#include <asm/thread_info.h>
26#include <asm/asm-offsets.h>
27
28	.text
29
30	.align	5
31_GLOBAL(__delay)
32	cmpwi	0,r3,0
33	mtctr	r3
34	beqlr
351:	bdnz	1b
36	blr
37
38/*
39 * This returns the high 64 bits of the product of two 64-bit numbers.
40 */
41_GLOBAL(mulhdu)
42	cmpwi	r6,0
43	cmpwi	cr1,r3,0
44	mr	r10,r4
45	mulhwu	r4,r4,r5
46	beq	1f
47	mulhwu	r0,r10,r6
48	mullw	r7,r10,r5
49	addc	r7,r0,r7
50	addze	r4,r4
511:	beqlr	cr1		/* all done if high part of A is 0 */
52	mr	r10,r3
53	mullw	r9,r3,r5
54	mulhwu	r3,r3,r5
55	beq	2f
56	mullw	r0,r10,r6
57	mulhwu	r8,r10,r6
58	addc	r7,r0,r7
59	adde	r4,r4,r8
60	addze	r3,r3
612:	addc	r4,r4,r9
62	addze	r3,r3
63	blr
64
65/*
66 * Returns (address we're running at) - (address we were linked at)
67 * for use before the text and data are mapped to KERNELBASE.
68 */
69_GLOBAL(reloc_offset)
70	mflr	r0
71	bl	1f
721:	mflr	r3
73	LOADADDR(r4,1b)
74	subf	r3,r4,r3
75	mtlr	r0
76	blr
77
78/*
79 * add_reloc_offset(x) returns x + reloc_offset().
80 */
81_GLOBAL(add_reloc_offset)
82	mflr	r0
83	bl	1f
841:	mflr	r5
85	LOADADDR(r4,1b)
86	subf	r5,r4,r5
87	add	r3,r3,r5
88	mtlr	r0
89	blr
90
91/*
92 * sub_reloc_offset(x) returns x - reloc_offset().
93 */
94_GLOBAL(sub_reloc_offset)
95	mflr	r0
96	bl	1f
971:	mflr	r5
98	lis	r4,1b@ha
99	addi	r4,r4,1b@l
100	subf	r5,r4,r5
101	subf	r3,r5,r3
102	mtlr	r0
103	blr
104
105/*
106 * reloc_got2 runs through the .got2 section adding an offset
107 * to each entry.
108 */
109_GLOBAL(reloc_got2)
110	mflr	r11
111	lis	r7,__got2_start@ha
112	addi	r7,r7,__got2_start@l
113	lis	r8,__got2_end@ha
114	addi	r8,r8,__got2_end@l
115	subf	r8,r7,r8
116	srwi.	r8,r8,2
117	beqlr
118	mtctr	r8
119	bl	1f
1201:	mflr	r0
121	lis	r4,1b@ha
122	addi	r4,r4,1b@l
123	subf	r0,r4,r0
124	add	r7,r0,r7
1252:	lwz	r0,0(r7)
126	add	r0,r0,r3
127	stw	r0,0(r7)
128	addi	r7,r7,4
129	bdnz	2b
130	mtlr	r11
131	blr
132
133/*
134 * identify_cpu,
135 * called with r3 = data offset and r4 = CPU number
136 * doesn't change r3
137 */
138_GLOBAL(identify_cpu)
139	addis	r8,r3,cpu_specs@ha
140	addi	r8,r8,cpu_specs@l
141	mfpvr	r7
1421:
143	lwz	r5,CPU_SPEC_PVR_MASK(r8)
144	and	r5,r5,r7
145	lwz	r6,CPU_SPEC_PVR_VALUE(r8)
146	cmplw	0,r6,r5
147	beq	1f
148	addi	r8,r8,CPU_SPEC_ENTRY_SIZE
149	b	1b
1501:
151	addis	r6,r3,cur_cpu_spec@ha
152	addi	r6,r6,cur_cpu_spec@l
153	sub	r8,r8,r3
154	stw	r8,0(r6)
155	blr
156
157/*
158 * do_cpu_ftr_fixups - goes through the list of CPU feature fixups
159 * and writes nop's over sections of code that don't apply for this cpu.
160 * r3 = data offset (not changed)
161 */
162_GLOBAL(do_cpu_ftr_fixups)
163	/* Get CPU 0 features */
164	addis	r6,r3,cur_cpu_spec@ha
165	addi	r6,r6,cur_cpu_spec@l
166	lwz	r4,0(r6)
167	add	r4,r4,r3
168	lwz	r4,CPU_SPEC_FEATURES(r4)
169
170	/* Get the fixup table */
171	addis	r6,r3,__start___ftr_fixup@ha
172	addi	r6,r6,__start___ftr_fixup@l
173	addis	r7,r3,__stop___ftr_fixup@ha
174	addi	r7,r7,__stop___ftr_fixup@l
175
176	/* Do the fixup */
1771:	cmplw	0,r6,r7
178	bgelr
179	addi	r6,r6,16
180	lwz	r8,-16(r6)	/* mask */
181	and	r8,r8,r4
182	lwz	r9,-12(r6)	/* value */
183	cmplw	0,r8,r9
184	beq	1b
185	lwz	r8,-8(r6)	/* section begin */
186	lwz	r9,-4(r6)	/* section end */
187	subf.	r9,r8,r9
188	beq	1b
189	/* write nops over the section of code */
190	/* todo: if large section, add a branch at the start of it */
191	srwi	r9,r9,2
192	mtctr	r9
193	add	r8,r8,r3
194	lis	r0,0x60000000@h	/* nop */
1953:	stw	r0,0(r8)
196	andi.	r10,r4,CPU_FTR_SPLIT_ID_CACHE@l
197	beq	2f
198	dcbst	0,r8		/* suboptimal, but simpler */
199	sync
200	icbi	0,r8
2012:	addi	r8,r8,4
202	bdnz	3b
203	sync			/* additional sync needed on g4 */
204	isync
205	b	1b
206
207/*
208 * call_setup_cpu - call the setup_cpu function for this cpu
209 * r3 = data offset, r24 = cpu number
210 *
211 * Setup function is called with:
212 *   r3 = data offset
213 *   r4 = ptr to CPU spec (relocated)
214 */
215_GLOBAL(call_setup_cpu)
216	addis	r4,r3,cur_cpu_spec@ha
217	addi	r4,r4,cur_cpu_spec@l
218	lwz	r4,0(r4)
219	add	r4,r4,r3
220	lwz	r5,CPU_SPEC_SETUP(r4)
221	cmpi	0,r5,0
222	add	r5,r5,r3
223	beqlr
224	mtctr	r5
225	bctr
226
227#if defined(CONFIG_CPU_FREQ_PMAC) && defined(CONFIG_6xx)
228
229/* This gets called by via-pmu.c to switch the PLL selection
230 * on 750fx CPU. This function should really be moved to some
231 * other place (as most of the cpufreq code in via-pmu
232 */
233_GLOBAL(low_choose_750fx_pll)
234	/* Clear MSR:EE */
235	mfmsr	r7
236	rlwinm	r0,r7,0,17,15
237	mtmsr	r0
238
239	/* If switching to PLL1, disable HID0:BTIC */
240	cmplwi	cr0,r3,0
241	beq	1f
242	mfspr	r5,SPRN_HID0
243	rlwinm	r5,r5,0,27,25
244	sync
245	mtspr	SPRN_HID0,r5
246	isync
247	sync
248
2491:
250	/* Calc new HID1 value */
251	mfspr	r4,SPRN_HID1	/* Build a HID1:PS bit from parameter */
252	rlwinm	r5,r3,16,15,15	/* Clear out HID1:PS from value read */
253	rlwinm	r4,r4,0,16,14	/* Could have I used rlwimi here ? */
254	or	r4,r4,r5
255	mtspr	SPRN_HID1,r4
256
257	/* Store new HID1 image */
258	rlwinm	r6,r1,0,0,18
259	lwz	r6,TI_CPU(r6)
260	slwi	r6,r6,2
261	addis	r6,r6,nap_save_hid1@ha
262	stw	r4,nap_save_hid1@l(r6)
263
264	/* If switching to PLL0, enable HID0:BTIC */
265	cmplwi	cr0,r3,0
266	bne	1f
267	mfspr	r5,SPRN_HID0
268	ori	r5,r5,HID0_BTIC
269	sync
270	mtspr	SPRN_HID0,r5
271	isync
272	sync
273
2741:
275	/* Return */
276	mtmsr	r7
277	blr
278
279_GLOBAL(low_choose_7447a_dfs)
280	/* Clear MSR:EE */
281	mfmsr	r7
282	rlwinm	r0,r7,0,17,15
283	mtmsr	r0
284
285	/* Calc new HID1 value */
286	mfspr	r4,SPRN_HID1
287	insrwi	r4,r3,1,9	/* insert parameter into bit 9 */
288	sync
289	mtspr	SPRN_HID1,r4
290	sync
291	isync
292
293	/* Return */
294	mtmsr	r7
295	blr
296
297#endif /* CONFIG_CPU_FREQ_PMAC && CONFIG_6xx */
298
299/*
300 * complement mask on the msr then "or" some values on.
301 *     _nmask_and_or_msr(nmask, value_to_or)
302 */
303_GLOBAL(_nmask_and_or_msr)
304	mfmsr	r0		/* Get current msr */
305	andc	r0,r0,r3	/* And off the bits set in r3 (first parm) */
306	or	r0,r0,r4	/* Or on the bits in r4 (second parm) */
307	SYNC			/* Some chip revs have problems here... */
308	mtmsr	r0		/* Update machine state */
309	isync
310	blr			/* Done */
311
312
313/*
314 * Flush MMU TLB
315 */
316_GLOBAL(_tlbia)
317#if defined(CONFIG_40x)
318	sync			/* Flush to memory before changing mapping */
319	tlbia
320	isync			/* Flush shadow TLB */
321#elif defined(CONFIG_44x)
322	li	r3,0
323	sync
324
325	/* Load high watermark */
326	lis	r4,tlb_44x_hwater@ha
327	lwz	r5,tlb_44x_hwater@l(r4)
328
3291:	tlbwe	r3,r3,PPC44x_TLB_PAGEID
330	addi	r3,r3,1
331	cmpw	0,r3,r5
332	ble	1b
333
334	isync
335#elif defined(CONFIG_FSL_BOOKE)
336	/* Invalidate all entries in TLB0 */
337	li	r3, 0x04
338	tlbivax	0,3
339	/* Invalidate all entries in TLB1 */
340	li	r3, 0x0c
341	tlbivax	0,3
342	/* Invalidate all entries in TLB2 */
343	li	r3, 0x14
344	tlbivax	0,3
345	/* Invalidate all entries in TLB3 */
346	li	r3, 0x1c
347	tlbivax	0,3
348	msync
349#ifdef CONFIG_SMP
350	tlbsync
351#endif /* CONFIG_SMP */
352#else /* !(CONFIG_40x || CONFIG_44x || CONFIG_FSL_BOOKE) */
353#if defined(CONFIG_SMP)
354	rlwinm	r8,r1,0,0,18
355	lwz	r8,TI_CPU(r8)
356	oris	r8,r8,10
357	mfmsr	r10
358	SYNC
359	rlwinm	r0,r10,0,17,15		/* clear bit 16 (MSR_EE) */
360	rlwinm	r0,r0,0,28,26		/* clear DR */
361	mtmsr	r0
362	SYNC_601
363	isync
364	lis	r9,mmu_hash_lock@h
365	ori	r9,r9,mmu_hash_lock@l
366	tophys(r9,r9)
36710:	lwarx	r7,0,r9
368	cmpwi	0,r7,0
369	bne-	10b
370	stwcx.	r8,0,r9
371	bne-	10b
372	sync
373	tlbia
374	sync
375	TLBSYNC
376	li	r0,0
377	stw	r0,0(r9)		/* clear mmu_hash_lock */
378	mtmsr	r10
379	SYNC_601
380	isync
381#else /* CONFIG_SMP */
382	sync
383	tlbia
384	sync
385#endif /* CONFIG_SMP */
386#endif /* ! defined(CONFIG_40x) */
387	blr
388
389/*
390 * Flush MMU TLB for a particular address
391 */
392_GLOBAL(_tlbie)
393#if defined(CONFIG_40x)
394	tlbsx.	r3, 0, r3
395	bne	10f
396	sync
397	/* There are only 64 TLB entries, so r3 < 64, which means bit 25 is clear.
398	 * Since 25 is the V bit in the TLB_TAG, loading this value will invalidate
399	 * the TLB entry. */
400	tlbwe	r3, r3, TLB_TAG
401	isync
40210:
403#elif defined(CONFIG_44x)
404	mfspr	r4,SPRN_MMUCR
405	mfspr	r5,SPRN_PID			/* Get PID */
406	rlwimi	r4,r5,0,24,31			/* Set TID */
407	mtspr	SPRN_MMUCR,r4
408
409	tlbsx.	r3, 0, r3
410	bne	10f
411	sync
412	/* There are only 64 TLB entries, so r3 < 64,
413	 * which means bit 22, is clear.  Since 22 is
414	 * the V bit in the TLB_PAGEID, loading this
415	 * value will invalidate the TLB entry.
416	 */
417	tlbwe	r3, r3, PPC44x_TLB_PAGEID
418	isync
41910:
420#elif defined(CONFIG_FSL_BOOKE)
421	rlwinm	r4, r3, 0, 0, 19
422	ori	r5, r4, 0x08	/* TLBSEL = 1 */
423	ori	r6, r4, 0x10	/* TLBSEL = 2 */
424	ori	r7, r4, 0x18	/* TLBSEL = 3 */
425	tlbivax	0, r4
426	tlbivax	0, r5
427	tlbivax	0, r6
428	tlbivax	0, r7
429	msync
430#if defined(CONFIG_SMP)
431	tlbsync
432#endif /* CONFIG_SMP */
433#else /* !(CONFIG_40x || CONFIG_44x || CONFIG_FSL_BOOKE) */
434#if defined(CONFIG_SMP)
435	rlwinm	r8,r1,0,0,18
436	lwz	r8,TI_CPU(r8)
437	oris	r8,r8,11
438	mfmsr	r10
439	SYNC
440	rlwinm	r0,r10,0,17,15		/* clear bit 16 (MSR_EE) */
441	rlwinm	r0,r0,0,28,26		/* clear DR */
442	mtmsr	r0
443	SYNC_601
444	isync
445	lis	r9,mmu_hash_lock@h
446	ori	r9,r9,mmu_hash_lock@l
447	tophys(r9,r9)
44810:	lwarx	r7,0,r9
449	cmpwi	0,r7,0
450	bne-	10b
451	stwcx.	r8,0,r9
452	bne-	10b
453	eieio
454	tlbie	r3
455	sync
456	TLBSYNC
457	li	r0,0
458	stw	r0,0(r9)		/* clear mmu_hash_lock */
459	mtmsr	r10
460	SYNC_601
461	isync
462#else /* CONFIG_SMP */
463	tlbie	r3
464	sync
465#endif /* CONFIG_SMP */
466#endif /* ! CONFIG_40x */
467	blr
468
469/*
470 * Flush instruction cache.
471 * This is a no-op on the 601.
472 */
473_GLOBAL(flush_instruction_cache)
474#if defined(CONFIG_8xx)
475	isync
476	lis	r5, IDC_INVALL@h
477	mtspr	SPRN_IC_CST, r5
478#elif defined(CONFIG_4xx)
479#ifdef CONFIG_403GCX
480	li      r3, 512
481	mtctr   r3
482	lis     r4, KERNELBASE@h
4831:	iccci   0, r4
484	addi    r4, r4, 16
485	bdnz    1b
486#else
487	lis	r3, KERNELBASE@h
488	iccci	0,r3
489#endif
490#elif CONFIG_FSL_BOOKE
491BEGIN_FTR_SECTION
492	mfspr   r3,SPRN_L1CSR0
493	ori     r3,r3,L1CSR0_CFI|L1CSR0_CLFC
494	/* msync; isync recommended here */
495	mtspr   SPRN_L1CSR0,r3
496	isync
497	blr
498END_FTR_SECTION_IFCLR(CPU_FTR_SPLIT_ID_CACHE)
499	mfspr	r3,SPRN_L1CSR1
500	ori	r3,r3,L1CSR1_ICFI|L1CSR1_ICLFR
501	mtspr	SPRN_L1CSR1,r3
502#else
503	mfspr	r3,SPRN_PVR
504	rlwinm	r3,r3,16,16,31
505	cmpwi	0,r3,1
506	beqlr			/* for 601, do nothing */
507	/* 603/604 processor - use invalidate-all bit in HID0 */
508	mfspr	r3,SPRN_HID0
509	ori	r3,r3,HID0_ICFI
510	mtspr	SPRN_HID0,r3
511#endif /* CONFIG_8xx/4xx */
512	isync
513	blr
514
515/*
516 * Write any modified data cache blocks out to memory
517 * and invalidate the corresponding instruction cache blocks.
518 * This is a no-op on the 601.
519 *
520 * flush_icache_range(unsigned long start, unsigned long stop)
521 */
522_GLOBAL(flush_icache_range)
523BEGIN_FTR_SECTION
524	blr				/* for 601, do nothing */
525END_FTR_SECTION_IFCLR(CPU_FTR_SPLIT_ID_CACHE)
526	li	r5,L1_CACHE_BYTES-1
527	andc	r3,r3,r5
528	subf	r4,r3,r4
529	add	r4,r4,r5
530	srwi.	r4,r4,L1_CACHE_SHIFT
531	beqlr
532	mtctr	r4
533	mr	r6,r3
5341:	dcbst	0,r3
535	addi	r3,r3,L1_CACHE_BYTES
536	bdnz	1b
537	sync				/* wait for dcbst's to get to ram */
538	mtctr	r4
5392:	icbi	0,r6
540	addi	r6,r6,L1_CACHE_BYTES
541	bdnz	2b
542	sync				/* additional sync needed on g4 */
543	isync
544	blr
545/*
546 * Write any modified data cache blocks out to memory.
547 * Does not invalidate the corresponding cache lines (especially for
548 * any corresponding instruction cache).
549 *
550 * clean_dcache_range(unsigned long start, unsigned long stop)
551 */
552_GLOBAL(clean_dcache_range)
553	li	r5,L1_CACHE_BYTES-1
554	andc	r3,r3,r5
555	subf	r4,r3,r4
556	add	r4,r4,r5
557	srwi.	r4,r4,L1_CACHE_SHIFT
558	beqlr
559	mtctr	r4
560
5611:	dcbst	0,r3
562	addi	r3,r3,L1_CACHE_BYTES
563	bdnz	1b
564	sync				/* wait for dcbst's to get to ram */
565	blr
566
567/*
568 * Write any modified data cache blocks out to memory and invalidate them.
569 * Does not invalidate the corresponding instruction cache blocks.
570 *
571 * flush_dcache_range(unsigned long start, unsigned long stop)
572 */
573_GLOBAL(flush_dcache_range)
574	li	r5,L1_CACHE_BYTES-1
575	andc	r3,r3,r5
576	subf	r4,r3,r4
577	add	r4,r4,r5
578	srwi.	r4,r4,L1_CACHE_SHIFT
579	beqlr
580	mtctr	r4
581
5821:	dcbf	0,r3
583	addi	r3,r3,L1_CACHE_BYTES
584	bdnz	1b
585	sync				/* wait for dcbst's to get to ram */
586	blr
587
588/*
589 * Like above, but invalidate the D-cache.  This is used by the 8xx
590 * to invalidate the cache so the PPC core doesn't get stale data
591 * from the CPM (no cache snooping here :-).
592 *
593 * invalidate_dcache_range(unsigned long start, unsigned long stop)
594 */
595_GLOBAL(invalidate_dcache_range)
596	li	r5,L1_CACHE_BYTES-1
597	andc	r3,r3,r5
598	subf	r4,r3,r4
599	add	r4,r4,r5
600	srwi.	r4,r4,L1_CACHE_SHIFT
601	beqlr
602	mtctr	r4
603
6041:	dcbi	0,r3
605	addi	r3,r3,L1_CACHE_BYTES
606	bdnz	1b
607	sync				/* wait for dcbi's to get to ram */
608	blr
609
610#ifdef CONFIG_NOT_COHERENT_CACHE
611/*
612 * 40x cores have 8K or 16K dcache and 32 byte line size.
613 * 44x has a 32K dcache and 32 byte line size.
614 * 8xx has 1, 2, 4, 8K variants.
615 * For now, cover the worst case of the 44x.
616 * Must be called with external interrupts disabled.
617 */
618#define CACHE_NWAYS	64
619#define CACHE_NLINES	16
620
621_GLOBAL(flush_dcache_all)
622	li	r4, (2 * CACHE_NWAYS * CACHE_NLINES)
623	mtctr	r4
624	lis     r5, KERNELBASE@h
6251:	lwz	r3, 0(r5)		/* Load one word from every line */
626	addi	r5, r5, L1_CACHE_BYTES
627	bdnz    1b
628	blr
629#endif /* CONFIG_NOT_COHERENT_CACHE */
630
631/*
632 * Flush a particular page from the data cache to RAM.
633 * Note: this is necessary because the instruction cache does *not*
634 * snoop from the data cache.
635 * This is a no-op on the 601 which has a unified cache.
636 *
637 *	void __flush_dcache_icache(void *page)
638 */
639_GLOBAL(__flush_dcache_icache)
640BEGIN_FTR_SECTION
641	blr					/* for 601, do nothing */
642END_FTR_SECTION_IFCLR(CPU_FTR_SPLIT_ID_CACHE)
643	rlwinm	r3,r3,0,0,19			/* Get page base address */
644	li	r4,4096/L1_CACHE_BYTES	/* Number of lines in a page */
645	mtctr	r4
646	mr	r6,r3
6470:	dcbst	0,r3				/* Write line to ram */
648	addi	r3,r3,L1_CACHE_BYTES
649	bdnz	0b
650	sync
651	mtctr	r4
6521:	icbi	0,r6
653	addi	r6,r6,L1_CACHE_BYTES
654	bdnz	1b
655	sync
656	isync
657	blr
658
659/*
660 * Flush a particular page from the data cache to RAM, identified
661 * by its physical address.  We turn off the MMU so we can just use
662 * the physical address (this may be a highmem page without a kernel
663 * mapping).
664 *
665 *	void __flush_dcache_icache_phys(unsigned long physaddr)
666 */
667_GLOBAL(__flush_dcache_icache_phys)
668BEGIN_FTR_SECTION
669	blr					/* for 601, do nothing */
670END_FTR_SECTION_IFCLR(CPU_FTR_SPLIT_ID_CACHE)
671	mfmsr	r10
672	rlwinm	r0,r10,0,28,26			/* clear DR */
673	mtmsr	r0
674	isync
675	rlwinm	r3,r3,0,0,19			/* Get page base address */
676	li	r4,4096/L1_CACHE_BYTES	/* Number of lines in a page */
677	mtctr	r4
678	mr	r6,r3
6790:	dcbst	0,r3				/* Write line to ram */
680	addi	r3,r3,L1_CACHE_BYTES
681	bdnz	0b
682	sync
683	mtctr	r4
6841:	icbi	0,r6
685	addi	r6,r6,L1_CACHE_BYTES
686	bdnz	1b
687	sync
688	mtmsr	r10				/* restore DR */
689	isync
690	blr
691
692/*
693 * Clear pages using the dcbz instruction, which doesn't cause any
694 * memory traffic (except to write out any cache lines which get
695 * displaced).  This only works on cacheable memory.
696 *
697 * void clear_pages(void *page, int order) ;
698 */
699_GLOBAL(clear_pages)
700	li	r0,4096/L1_CACHE_BYTES
701	slw	r0,r0,r4
702	mtctr	r0
703#ifdef CONFIG_8xx
704	li	r4, 0
7051:	stw	r4, 0(r3)
706	stw	r4, 4(r3)
707	stw	r4, 8(r3)
708	stw	r4, 12(r3)
709#else
7101:	dcbz	0,r3
711#endif
712	addi	r3,r3,L1_CACHE_BYTES
713	bdnz	1b
714	blr
715
716/*
717 * Copy a whole page.  We use the dcbz instruction on the destination
718 * to reduce memory traffic (it eliminates the unnecessary reads of
719 * the destination into cache).  This requires that the destination
720 * is cacheable.
721 */
722#define COPY_16_BYTES		\
723	lwz	r6,4(r4);	\
724	lwz	r7,8(r4);	\
725	lwz	r8,12(r4);	\
726	lwzu	r9,16(r4);	\
727	stw	r6,4(r3);	\
728	stw	r7,8(r3);	\
729	stw	r8,12(r3);	\
730	stwu	r9,16(r3)
731
732_GLOBAL(copy_page)
733	addi	r3,r3,-4
734	addi	r4,r4,-4
735
736#ifdef CONFIG_8xx
737	/* don't use prefetch on 8xx */
738    	li	r0,4096/L1_CACHE_BYTES
739	mtctr	r0
7401:	COPY_16_BYTES
741	bdnz	1b
742	blr
743
744#else	/* not 8xx, we can prefetch */
745	li	r5,4
746
747#if MAX_COPY_PREFETCH > 1
748	li	r0,MAX_COPY_PREFETCH
749	li	r11,4
750	mtctr	r0
75111:	dcbt	r11,r4
752	addi	r11,r11,L1_CACHE_BYTES
753	bdnz	11b
754#else /* MAX_COPY_PREFETCH == 1 */
755	dcbt	r5,r4
756	li	r11,L1_CACHE_BYTES+4
757#endif /* MAX_COPY_PREFETCH */
758	li	r0,4096/L1_CACHE_BYTES - MAX_COPY_PREFETCH
759	crclr	4*cr0+eq
7602:
761	mtctr	r0
7621:
763	dcbt	r11,r4
764	dcbz	r5,r3
765	COPY_16_BYTES
766#if L1_CACHE_BYTES >= 32
767	COPY_16_BYTES
768#if L1_CACHE_BYTES >= 64
769	COPY_16_BYTES
770	COPY_16_BYTES
771#if L1_CACHE_BYTES >= 128
772	COPY_16_BYTES
773	COPY_16_BYTES
774	COPY_16_BYTES
775	COPY_16_BYTES
776#endif
777#endif
778#endif
779	bdnz	1b
780	beqlr
781	crnot	4*cr0+eq,4*cr0+eq
782	li	r0,MAX_COPY_PREFETCH
783	li	r11,4
784	b	2b
785#endif	/* CONFIG_8xx */
786
787/*
788 * void atomic_clear_mask(atomic_t mask, atomic_t *addr)
789 * void atomic_set_mask(atomic_t mask, atomic_t *addr);
790 */
791_GLOBAL(atomic_clear_mask)
79210:	lwarx	r5,0,r4
793	andc	r5,r5,r3
794	PPC405_ERR77(0,r4)
795	stwcx.	r5,0,r4
796	bne-	10b
797	blr
798_GLOBAL(atomic_set_mask)
79910:	lwarx	r5,0,r4
800	or	r5,r5,r3
801	PPC405_ERR77(0,r4)
802	stwcx.	r5,0,r4
803	bne-	10b
804	blr
805
806/*
807 * I/O string operations
808 *
809 * insb(port, buf, len)
810 * outsb(port, buf, len)
811 * insw(port, buf, len)
812 * outsw(port, buf, len)
813 * insl(port, buf, len)
814 * outsl(port, buf, len)
815 * insw_ns(port, buf, len)
816 * outsw_ns(port, buf, len)
817 * insl_ns(port, buf, len)
818 * outsl_ns(port, buf, len)
819 *
820 * The *_ns versions don't do byte-swapping.
821 */
822_GLOBAL(_insb)
823	cmpwi	0,r5,0
824	mtctr	r5
825	subi	r4,r4,1
826	blelr-
82700:	lbz	r5,0(r3)
828	eieio
829	stbu	r5,1(r4)
830	bdnz	00b
831	blr
832
833_GLOBAL(_outsb)
834	cmpwi	0,r5,0
835	mtctr	r5
836	subi	r4,r4,1
837	blelr-
83800:	lbzu	r5,1(r4)
839	stb	r5,0(r3)
840	eieio
841	bdnz	00b
842	blr
843
844_GLOBAL(_insw)
845	cmpwi	0,r5,0
846	mtctr	r5
847	subi	r4,r4,2
848	blelr-
84900:	lhbrx	r5,0,r3
850	eieio
851	sthu	r5,2(r4)
852	bdnz	00b
853	blr
854
855_GLOBAL(_outsw)
856	cmpwi	0,r5,0
857	mtctr	r5
858	subi	r4,r4,2
859	blelr-
86000:	lhzu	r5,2(r4)
861	eieio
862	sthbrx	r5,0,r3
863	bdnz	00b
864	blr
865
866_GLOBAL(_insl)
867	cmpwi	0,r5,0
868	mtctr	r5
869	subi	r4,r4,4
870	blelr-
87100:	lwbrx	r5,0,r3
872	eieio
873	stwu	r5,4(r4)
874	bdnz	00b
875	blr
876
877_GLOBAL(_outsl)
878	cmpwi	0,r5,0
879	mtctr	r5
880	subi	r4,r4,4
881	blelr-
88200:	lwzu	r5,4(r4)
883	stwbrx	r5,0,r3
884	eieio
885	bdnz	00b
886	blr
887
888_GLOBAL(__ide_mm_insw)
889_GLOBAL(_insw_ns)
890	cmpwi	0,r5,0
891	mtctr	r5
892	subi	r4,r4,2
893	blelr-
89400:	lhz	r5,0(r3)
895	eieio
896	sthu	r5,2(r4)
897	bdnz	00b
898	blr
899
900_GLOBAL(__ide_mm_outsw)
901_GLOBAL(_outsw_ns)
902	cmpwi	0,r5,0
903	mtctr	r5
904	subi	r4,r4,2
905	blelr-
90600:	lhzu	r5,2(r4)
907	sth	r5,0(r3)
908	eieio
909	bdnz	00b
910	blr
911
912_GLOBAL(__ide_mm_insl)
913_GLOBAL(_insl_ns)
914	cmpwi	0,r5,0
915	mtctr	r5
916	subi	r4,r4,4
917	blelr-
91800:	lwz	r5,0(r3)
919	eieio
920	stwu	r5,4(r4)
921	bdnz	00b
922	blr
923
924_GLOBAL(__ide_mm_outsl)
925_GLOBAL(_outsl_ns)
926	cmpwi	0,r5,0
927	mtctr	r5
928	subi	r4,r4,4
929	blelr-
93000:	lwzu	r5,4(r4)
931	stw	r5,0(r3)
932	eieio
933	bdnz	00b
934	blr
935
936/*
937 * Extended precision shifts.
938 *
939 * Updated to be valid for shift counts from 0 to 63 inclusive.
940 * -- Gabriel
941 *
942 * R3/R4 has 64 bit value
943 * R5    has shift count
944 * result in R3/R4
945 *
946 *  ashrdi3: arithmetic right shift (sign propagation)
947 *  lshrdi3: logical right shift
948 *  ashldi3: left shift
949 */
950_GLOBAL(__ashrdi3)
951	subfic	r6,r5,32
952	srw	r4,r4,r5	# LSW = count > 31 ? 0 : LSW >> count
953	addi	r7,r5,32	# could be xori, or addi with -32
954	slw	r6,r3,r6	# t1 = count > 31 ? 0 : MSW << (32-count)
955	rlwinm	r8,r7,0,32	# t3 = (count < 32) ? 32 : 0
956	sraw	r7,r3,r7	# t2 = MSW >> (count-32)
957	or	r4,r4,r6	# LSW |= t1
958	slw	r7,r7,r8	# t2 = (count < 32) ? 0 : t2
959	sraw	r3,r3,r5	# MSW = MSW >> count
960	or	r4,r4,r7	# LSW |= t2
961	blr
962
963_GLOBAL(__ashldi3)
964	subfic	r6,r5,32
965	slw	r3,r3,r5	# MSW = count > 31 ? 0 : MSW << count
966	addi	r7,r5,32	# could be xori, or addi with -32
967	srw	r6,r4,r6	# t1 = count > 31 ? 0 : LSW >> (32-count)
968	slw	r7,r4,r7	# t2 = count < 32 ? 0 : LSW << (count-32)
969	or	r3,r3,r6	# MSW |= t1
970	slw	r4,r4,r5	# LSW = LSW << count
971	or	r3,r3,r7	# MSW |= t2
972	blr
973
974_GLOBAL(__lshrdi3)
975	subfic	r6,r5,32
976	srw	r4,r4,r5	# LSW = count > 31 ? 0 : LSW >> count
977	addi	r7,r5,32	# could be xori, or addi with -32
978	slw	r6,r3,r6	# t1 = count > 31 ? 0 : MSW << (32-count)
979	srw	r7,r3,r7	# t2 = count < 32 ? 0 : MSW >> (count-32)
980	or	r4,r4,r6	# LSW |= t1
981	srw	r3,r3,r5	# MSW = MSW >> count
982	or	r4,r4,r7	# LSW |= t2
983	blr
984
985_GLOBAL(abs)
986	srawi	r4,r3,31
987	xor	r3,r3,r4
988	sub	r3,r3,r4
989	blr
990
991_GLOBAL(_get_SP)
992	mr	r3,r1		/* Close enough */
993	blr
994
995/*
996 * Create a kernel thread
997 *   kernel_thread(fn, arg, flags)
998 */
999_GLOBAL(kernel_thread)
1000	stwu	r1,-16(r1)
1001	stw	r30,8(r1)
1002	stw	r31,12(r1)
1003	mr	r30,r3		/* function */
1004	mr	r31,r4		/* argument */
1005	ori	r3,r5,CLONE_VM	/* flags */
1006	oris	r3,r3,CLONE_UNTRACED>>16
1007	li	r4,0		/* new sp (unused) */
1008	li	r0,__NR_clone
1009	sc
1010	cmpwi	0,r3,0		/* parent or child? */
1011	bne	1f		/* return if parent */
1012	li	r0,0		/* make top-level stack frame */
1013	stwu	r0,-16(r1)
1014	mtlr	r30		/* fn addr in lr */
1015	mr	r3,r31		/* load arg and call fn */
1016	PPC440EP_ERR42
1017	blrl
1018	li	r0,__NR_exit	/* exit if function returns */
1019	li	r3,0
1020	sc
10211:	lwz	r30,8(r1)
1022	lwz	r31,12(r1)
1023	addi	r1,r1,16
1024	blr
1025
1026_GLOBAL(execve)
1027	li	r0,__NR_execve
1028	sc
1029	bnslr
1030	neg	r3,r3
1031	blr
1032
1033/*
1034 * This routine is just here to keep GCC happy - sigh...
1035 */
1036_GLOBAL(__main)
1037	blr
1038