xref: /openbmc/linux/arch/powerpc/kernel/misc_32.S (revision 8fa5723aa7e053d498336b48448b292fc2e0458b)
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 * kexec bits:
9 * Copyright (C) 2002-2003 Eric Biederman  <ebiederm@xmission.com>
10 * GameCube/ppc32 port Copyright (C) 2004 Albert Herranz
11 *
12 * This program is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU General Public License
14 * as published by the Free Software Foundation; either version
15 * 2 of the License, or (at your option) any later version.
16 *
17 */
18
19#include <linux/sys.h>
20#include <asm/unistd.h>
21#include <asm/errno.h>
22#include <asm/reg.h>
23#include <asm/page.h>
24#include <asm/cache.h>
25#include <asm/cputable.h>
26#include <asm/mmu.h>
27#include <asm/ppc_asm.h>
28#include <asm/thread_info.h>
29#include <asm/asm-offsets.h>
30#include <asm/processor.h>
31#include <asm/kexec.h>
32
33	.text
34
35#ifdef CONFIG_IRQSTACKS
36_GLOBAL(call_do_softirq)
37	mflr	r0
38	stw	r0,4(r1)
39	stwu	r1,THREAD_SIZE-STACK_FRAME_OVERHEAD(r3)
40	mr	r1,r3
41	bl	__do_softirq
42	lwz	r1,0(r1)
43	lwz	r0,4(r1)
44	mtlr	r0
45	blr
46
47_GLOBAL(call_handle_irq)
48	mflr	r0
49	stw	r0,4(r1)
50	mtctr	r6
51	stwu	r1,THREAD_SIZE-STACK_FRAME_OVERHEAD(r5)
52	mr	r1,r5
53	bctrl
54	lwz	r1,0(r1)
55	lwz	r0,4(r1)
56	mtlr	r0
57	blr
58#endif /* CONFIG_IRQSTACKS */
59
60/*
61 * This returns the high 64 bits of the product of two 64-bit numbers.
62 */
63_GLOBAL(mulhdu)
64	cmpwi	r6,0
65	cmpwi	cr1,r3,0
66	mr	r10,r4
67	mulhwu	r4,r4,r5
68	beq	1f
69	mulhwu	r0,r10,r6
70	mullw	r7,r10,r5
71	addc	r7,r0,r7
72	addze	r4,r4
731:	beqlr	cr1		/* all done if high part of A is 0 */
74	mr	r10,r3
75	mullw	r9,r3,r5
76	mulhwu	r3,r3,r5
77	beq	2f
78	mullw	r0,r10,r6
79	mulhwu	r8,r10,r6
80	addc	r7,r0,r7
81	adde	r4,r4,r8
82	addze	r3,r3
832:	addc	r4,r4,r9
84	addze	r3,r3
85	blr
86
87/*
88 * sub_reloc_offset(x) returns x - reloc_offset().
89 */
90_GLOBAL(sub_reloc_offset)
91	mflr	r0
92	bl	1f
931:	mflr	r5
94	lis	r4,1b@ha
95	addi	r4,r4,1b@l
96	subf	r5,r4,r5
97	subf	r3,r5,r3
98	mtlr	r0
99	blr
100
101/*
102 * reloc_got2 runs through the .got2 section adding an offset
103 * to each entry.
104 */
105_GLOBAL(reloc_got2)
106	mflr	r11
107	lis	r7,__got2_start@ha
108	addi	r7,r7,__got2_start@l
109	lis	r8,__got2_end@ha
110	addi	r8,r8,__got2_end@l
111	subf	r8,r7,r8
112	srwi.	r8,r8,2
113	beqlr
114	mtctr	r8
115	bl	1f
1161:	mflr	r0
117	lis	r4,1b@ha
118	addi	r4,r4,1b@l
119	subf	r0,r4,r0
120	add	r7,r0,r7
1212:	lwz	r0,0(r7)
122	add	r0,r0,r3
123	stw	r0,0(r7)
124	addi	r7,r7,4
125	bdnz	2b
126	mtlr	r11
127	blr
128
129/*
130 * call_setup_cpu - call the setup_cpu function for this cpu
131 * r3 = data offset, r24 = cpu number
132 *
133 * Setup function is called with:
134 *   r3 = data offset
135 *   r4 = ptr to CPU spec (relocated)
136 */
137_GLOBAL(call_setup_cpu)
138	addis	r4,r3,cur_cpu_spec@ha
139	addi	r4,r4,cur_cpu_spec@l
140	lwz	r4,0(r4)
141	add	r4,r4,r3
142	lwz	r5,CPU_SPEC_SETUP(r4)
143	cmpwi	0,r5,0
144	add	r5,r5,r3
145	beqlr
146	mtctr	r5
147	bctr
148
149#if defined(CONFIG_CPU_FREQ_PMAC) && defined(CONFIG_6xx)
150
151/* This gets called by via-pmu.c to switch the PLL selection
152 * on 750fx CPU. This function should really be moved to some
153 * other place (as most of the cpufreq code in via-pmu
154 */
155_GLOBAL(low_choose_750fx_pll)
156	/* Clear MSR:EE */
157	mfmsr	r7
158	rlwinm	r0,r7,0,17,15
159	mtmsr	r0
160
161	/* If switching to PLL1, disable HID0:BTIC */
162	cmplwi	cr0,r3,0
163	beq	1f
164	mfspr	r5,SPRN_HID0
165	rlwinm	r5,r5,0,27,25
166	sync
167	mtspr	SPRN_HID0,r5
168	isync
169	sync
170
1711:
172	/* Calc new HID1 value */
173	mfspr	r4,SPRN_HID1	/* Build a HID1:PS bit from parameter */
174	rlwinm	r5,r3,16,15,15	/* Clear out HID1:PS from value read */
175	rlwinm	r4,r4,0,16,14	/* Could have I used rlwimi here ? */
176	or	r4,r4,r5
177	mtspr	SPRN_HID1,r4
178
179	/* Store new HID1 image */
180	rlwinm	r6,r1,0,0,(31-THREAD_SHIFT)
181	lwz	r6,TI_CPU(r6)
182	slwi	r6,r6,2
183	addis	r6,r6,nap_save_hid1@ha
184	stw	r4,nap_save_hid1@l(r6)
185
186	/* If switching to PLL0, enable HID0:BTIC */
187	cmplwi	cr0,r3,0
188	bne	1f
189	mfspr	r5,SPRN_HID0
190	ori	r5,r5,HID0_BTIC
191	sync
192	mtspr	SPRN_HID0,r5
193	isync
194	sync
195
1961:
197	/* Return */
198	mtmsr	r7
199	blr
200
201_GLOBAL(low_choose_7447a_dfs)
202	/* Clear MSR:EE */
203	mfmsr	r7
204	rlwinm	r0,r7,0,17,15
205	mtmsr	r0
206
207	/* Calc new HID1 value */
208	mfspr	r4,SPRN_HID1
209	insrwi	r4,r3,1,9	/* insert parameter into bit 9 */
210	sync
211	mtspr	SPRN_HID1,r4
212	sync
213	isync
214
215	/* Return */
216	mtmsr	r7
217	blr
218
219#endif /* CONFIG_CPU_FREQ_PMAC && CONFIG_6xx */
220
221/*
222 * complement mask on the msr then "or" some values on.
223 *     _nmask_and_or_msr(nmask, value_to_or)
224 */
225_GLOBAL(_nmask_and_or_msr)
226	mfmsr	r0		/* Get current msr */
227	andc	r0,r0,r3	/* And off the bits set in r3 (first parm) */
228	or	r0,r0,r4	/* Or on the bits in r4 (second parm) */
229	SYNC			/* Some chip revs have problems here... */
230	mtmsr	r0		/* Update machine state */
231	isync
232	blr			/* Done */
233
234#ifdef CONFIG_40x
235
236/*
237 * Do an IO access in real mode
238 */
239_GLOBAL(real_readb)
240	mfmsr	r7
241	ori	r0,r7,MSR_DR
242	xori	r0,r0,MSR_DR
243	sync
244	mtmsr	r0
245	sync
246	isync
247	lbz	r3,0(r3)
248	sync
249	mtmsr	r7
250	sync
251	isync
252	blr
253
254	/*
255 * Do an IO access in real mode
256 */
257_GLOBAL(real_writeb)
258	mfmsr	r7
259	ori	r0,r7,MSR_DR
260	xori	r0,r0,MSR_DR
261	sync
262	mtmsr	r0
263	sync
264	isync
265	stb	r3,0(r4)
266	sync
267	mtmsr	r7
268	sync
269	isync
270	blr
271
272#endif /* CONFIG_40x */
273
274/*
275 * Flush MMU TLB
276 */
277#ifndef CONFIG_FSL_BOOKE
278_GLOBAL(_tlbil_all)
279_GLOBAL(_tlbil_pid)
280#endif
281_GLOBAL(_tlbia)
282#if defined(CONFIG_40x)
283	sync			/* Flush to memory before changing mapping */
284	tlbia
285	isync			/* Flush shadow TLB */
286#elif defined(CONFIG_44x)
287	li	r3,0
288	sync
289
290	/* Load high watermark */
291	lis	r4,tlb_44x_hwater@ha
292	lwz	r5,tlb_44x_hwater@l(r4)
293
2941:	tlbwe	r3,r3,PPC44x_TLB_PAGEID
295	addi	r3,r3,1
296	cmpw	0,r3,r5
297	ble	1b
298
299	isync
300#elif defined(CONFIG_FSL_BOOKE)
301	/* Invalidate all entries in TLB0 */
302	li	r3, 0x04
303	tlbivax	0,3
304	/* Invalidate all entries in TLB1 */
305	li	r3, 0x0c
306	tlbivax	0,3
307	msync
308#ifdef CONFIG_SMP
309	tlbsync
310#endif /* CONFIG_SMP */
311#else /* !(CONFIG_40x || CONFIG_44x || CONFIG_FSL_BOOKE) */
312#if defined(CONFIG_SMP)
313	rlwinm	r8,r1,0,0,(31-THREAD_SHIFT)
314	lwz	r8,TI_CPU(r8)
315	oris	r8,r8,10
316	mfmsr	r10
317	SYNC
318	rlwinm	r0,r10,0,17,15		/* clear bit 16 (MSR_EE) */
319	rlwinm	r0,r0,0,28,26		/* clear DR */
320	mtmsr	r0
321	SYNC_601
322	isync
323	lis	r9,mmu_hash_lock@h
324	ori	r9,r9,mmu_hash_lock@l
325	tophys(r9,r9)
32610:	lwarx	r7,0,r9
327	cmpwi	0,r7,0
328	bne-	10b
329	stwcx.	r8,0,r9
330	bne-	10b
331	sync
332	tlbia
333	sync
334	TLBSYNC
335	li	r0,0
336	stw	r0,0(r9)		/* clear mmu_hash_lock */
337	mtmsr	r10
338	SYNC_601
339	isync
340#else /* CONFIG_SMP */
341	sync
342	tlbia
343	sync
344#endif /* CONFIG_SMP */
345#endif /* ! defined(CONFIG_40x) */
346	blr
347
348/*
349 * Flush MMU TLB for a particular address
350 */
351#ifndef CONFIG_FSL_BOOKE
352_GLOBAL(_tlbil_va)
353#endif
354_GLOBAL(_tlbie)
355#if defined(CONFIG_40x)
356	/* We run the search with interrupts disabled because we have to change
357	 * the PID and I don't want to preempt when that happens.
358	 */
359	mfmsr	r5
360	mfspr	r6,SPRN_PID
361	wrteei	0
362	mtspr	SPRN_PID,r4
363	tlbsx.	r3, 0, r3
364	mtspr	SPRN_PID,r6
365	wrtee	r5
366	bne	10f
367	sync
368	/* There are only 64 TLB entries, so r3 < 64, which means bit 25 is clear.
369	 * Since 25 is the V bit in the TLB_TAG, loading this value will invalidate
370	 * the TLB entry. */
371	tlbwe	r3, r3, TLB_TAG
372	isync
37310:
374
375#elif defined(CONFIG_44x)
376	mfspr	r5,SPRN_MMUCR
377	rlwimi	r5,r4,0,24,31			/* Set TID */
378
379	/* We have to run the search with interrupts disabled, even critical
380	 * and debug interrupts (in fact the only critical exceptions we have
381	 * are debug and machine check).  Otherwise  an interrupt which causes
382	 * a TLB miss can clobber the MMUCR between the mtspr and the tlbsx. */
383	mfmsr	r4
384	lis	r6,(MSR_EE|MSR_CE|MSR_ME|MSR_DE)@ha
385	addi	r6,r6,(MSR_EE|MSR_CE|MSR_ME|MSR_DE)@l
386	andc	r6,r4,r6
387	mtmsr	r6
388	mtspr	SPRN_MMUCR,r5
389	tlbsx.	r3, 0, r3
390	mtmsr	r4
391	bne	10f
392	sync
393	/* There are only 64 TLB entries, so r3 < 64,
394	 * which means bit 22, is clear.  Since 22 is
395	 * the V bit in the TLB_PAGEID, loading this
396	 * value will invalidate the TLB entry.
397	 */
398	tlbwe	r3, r3, PPC44x_TLB_PAGEID
399	isync
40010:
401#elif defined(CONFIG_FSL_BOOKE)
402	rlwinm	r4, r3, 0, 0, 19
403	ori	r5, r4, 0x08	/* TLBSEL = 1 */
404	tlbivax	0, r4
405	tlbivax	0, r5
406	msync
407#if defined(CONFIG_SMP)
408	tlbsync
409#endif /* CONFIG_SMP */
410#else /* !(CONFIG_40x || CONFIG_44x || CONFIG_FSL_BOOKE) */
411#if defined(CONFIG_SMP)
412	rlwinm	r8,r1,0,0,(31-THREAD_SHIFT)
413	lwz	r8,TI_CPU(r8)
414	oris	r8,r8,11
415	mfmsr	r10
416	SYNC
417	rlwinm	r0,r10,0,17,15		/* clear bit 16 (MSR_EE) */
418	rlwinm	r0,r0,0,28,26		/* clear DR */
419	mtmsr	r0
420	SYNC_601
421	isync
422	lis	r9,mmu_hash_lock@h
423	ori	r9,r9,mmu_hash_lock@l
424	tophys(r9,r9)
42510:	lwarx	r7,0,r9
426	cmpwi	0,r7,0
427	bne-	10b
428	stwcx.	r8,0,r9
429	bne-	10b
430	eieio
431	tlbie	r3
432	sync
433	TLBSYNC
434	li	r0,0
435	stw	r0,0(r9)		/* clear mmu_hash_lock */
436	mtmsr	r10
437	SYNC_601
438	isync
439#else /* CONFIG_SMP */
440	tlbie	r3
441	sync
442#endif /* CONFIG_SMP */
443#endif /* ! CONFIG_40x */
444	blr
445
446#if defined(CONFIG_FSL_BOOKE)
447/*
448 * Flush MMU TLB, but only on the local processor (no broadcast)
449 */
450_GLOBAL(_tlbil_all)
451#define MMUCSR0_TLBFI	(MMUCSR0_TLB0FI | MMUCSR0_TLB1FI | \
452			 MMUCSR0_TLB2FI | MMUCSR0_TLB3FI)
453	li	r3,(MMUCSR0_TLBFI)@l
454	mtspr	SPRN_MMUCSR0, r3
4551:
456	mfspr	r3,SPRN_MMUCSR0
457	andi.	r3,r3,MMUCSR0_TLBFI@l
458	bne	1b
459	blr
460
461/*
462 * Flush MMU TLB for a particular process id, but only on the local processor
463 * (no broadcast)
464 */
465_GLOBAL(_tlbil_pid)
466/* we currently do an invalidate all since we don't have per pid invalidate */
467	li	r3,(MMUCSR0_TLBFI)@l
468	mtspr	SPRN_MMUCSR0, r3
4691:
470	mfspr	r3,SPRN_MMUCSR0
471	andi.	r3,r3,MMUCSR0_TLBFI@l
472	bne	1b
473	blr
474
475/*
476 * Flush MMU TLB for a particular address, but only on the local processor
477 * (no broadcast)
478 */
479_GLOBAL(_tlbil_va)
480	slwi	r4,r4,16
481	mtspr	SPRN_MAS6,r4		/* assume AS=0 for now */
482	tlbsx	0,r3
483	mfspr	r4,SPRN_MAS1		/* check valid */
484	andis.	r3,r4,MAS1_VALID@h
485	beqlr
486	rlwinm	r4,r4,0,1,31
487	mtspr	SPRN_MAS1,r4
488	tlbwe
489	blr
490#endif /* CONFIG_FSL_BOOKE */
491
492
493/*
494 * Flush instruction cache.
495 * This is a no-op on the 601.
496 */
497_GLOBAL(flush_instruction_cache)
498#if defined(CONFIG_8xx)
499	isync
500	lis	r5, IDC_INVALL@h
501	mtspr	SPRN_IC_CST, r5
502#elif defined(CONFIG_4xx)
503#ifdef CONFIG_403GCX
504	li      r3, 512
505	mtctr   r3
506	lis     r4, KERNELBASE@h
5071:	iccci   0, r4
508	addi    r4, r4, 16
509	bdnz    1b
510#else
511	lis	r3, KERNELBASE@h
512	iccci	0,r3
513#endif
514#elif CONFIG_FSL_BOOKE
515BEGIN_FTR_SECTION
516	mfspr   r3,SPRN_L1CSR0
517	ori     r3,r3,L1CSR0_CFI|L1CSR0_CLFC
518	/* msync; isync recommended here */
519	mtspr   SPRN_L1CSR0,r3
520	isync
521	blr
522END_FTR_SECTION_IFSET(CPU_FTR_UNIFIED_ID_CACHE)
523	mfspr	r3,SPRN_L1CSR1
524	ori	r3,r3,L1CSR1_ICFI|L1CSR1_ICLFR
525	mtspr	SPRN_L1CSR1,r3
526#else
527	mfspr	r3,SPRN_PVR
528	rlwinm	r3,r3,16,16,31
529	cmpwi	0,r3,1
530	beqlr			/* for 601, do nothing */
531	/* 603/604 processor - use invalidate-all bit in HID0 */
532	mfspr	r3,SPRN_HID0
533	ori	r3,r3,HID0_ICFI
534	mtspr	SPRN_HID0,r3
535#endif /* CONFIG_8xx/4xx */
536	isync
537	blr
538
539/*
540 * Write any modified data cache blocks out to memory
541 * and invalidate the corresponding instruction cache blocks.
542 * This is a no-op on the 601.
543 *
544 * flush_icache_range(unsigned long start, unsigned long stop)
545 */
546_KPROBE(__flush_icache_range)
547BEGIN_FTR_SECTION
548	blr				/* for 601, do nothing */
549END_FTR_SECTION_IFSET(CPU_FTR_COHERENT_ICACHE)
550	li	r5,L1_CACHE_BYTES-1
551	andc	r3,r3,r5
552	subf	r4,r3,r4
553	add	r4,r4,r5
554	srwi.	r4,r4,L1_CACHE_SHIFT
555	beqlr
556	mtctr	r4
557	mr	r6,r3
5581:	dcbst	0,r3
559	addi	r3,r3,L1_CACHE_BYTES
560	bdnz	1b
561	sync				/* wait for dcbst's to get to ram */
562	mtctr	r4
5632:	icbi	0,r6
564	addi	r6,r6,L1_CACHE_BYTES
565	bdnz	2b
566	sync				/* additional sync needed on g4 */
567	isync
568	blr
569/*
570 * Write any modified data cache blocks out to memory.
571 * Does not invalidate the corresponding cache lines (especially for
572 * any corresponding instruction cache).
573 *
574 * clean_dcache_range(unsigned long start, unsigned long stop)
575 */
576_GLOBAL(clean_dcache_range)
577	li	r5,L1_CACHE_BYTES-1
578	andc	r3,r3,r5
579	subf	r4,r3,r4
580	add	r4,r4,r5
581	srwi.	r4,r4,L1_CACHE_SHIFT
582	beqlr
583	mtctr	r4
584
5851:	dcbst	0,r3
586	addi	r3,r3,L1_CACHE_BYTES
587	bdnz	1b
588	sync				/* wait for dcbst's to get to ram */
589	blr
590
591/*
592 * Write any modified data cache blocks out to memory and invalidate them.
593 * Does not invalidate the corresponding instruction cache blocks.
594 *
595 * flush_dcache_range(unsigned long start, unsigned long stop)
596 */
597_GLOBAL(flush_dcache_range)
598	li	r5,L1_CACHE_BYTES-1
599	andc	r3,r3,r5
600	subf	r4,r3,r4
601	add	r4,r4,r5
602	srwi.	r4,r4,L1_CACHE_SHIFT
603	beqlr
604	mtctr	r4
605
6061:	dcbf	0,r3
607	addi	r3,r3,L1_CACHE_BYTES
608	bdnz	1b
609	sync				/* wait for dcbst's to get to ram */
610	blr
611
612/*
613 * Like above, but invalidate the D-cache.  This is used by the 8xx
614 * to invalidate the cache so the PPC core doesn't get stale data
615 * from the CPM (no cache snooping here :-).
616 *
617 * invalidate_dcache_range(unsigned long start, unsigned long stop)
618 */
619_GLOBAL(invalidate_dcache_range)
620	li	r5,L1_CACHE_BYTES-1
621	andc	r3,r3,r5
622	subf	r4,r3,r4
623	add	r4,r4,r5
624	srwi.	r4,r4,L1_CACHE_SHIFT
625	beqlr
626	mtctr	r4
627
6281:	dcbi	0,r3
629	addi	r3,r3,L1_CACHE_BYTES
630	bdnz	1b
631	sync				/* wait for dcbi's to get to ram */
632	blr
633
634/*
635 * Flush a particular page from the data cache to RAM.
636 * Note: this is necessary because the instruction cache does *not*
637 * snoop from the data cache.
638 * This is a no-op on the 601 which has a unified cache.
639 *
640 *	void __flush_dcache_icache(void *page)
641 */
642_GLOBAL(__flush_dcache_icache)
643BEGIN_FTR_SECTION
644	blr
645END_FTR_SECTION_IFSET(CPU_FTR_COHERENT_ICACHE)
646	rlwinm	r3,r3,0,0,19			/* Get page base address */
647	li	r4,4096/L1_CACHE_BYTES	/* Number of lines in a page */
648	mtctr	r4
649	mr	r6,r3
6500:	dcbst	0,r3				/* Write line to ram */
651	addi	r3,r3,L1_CACHE_BYTES
652	bdnz	0b
653	sync
654#ifndef CONFIG_44x
655	/* We don't flush the icache on 44x. Those have a virtual icache
656	 * and we don't have access to the virtual address here (it's
657	 * not the page vaddr but where it's mapped in user space). The
658	 * flushing of the icache on these is handled elsewhere, when
659	 * a change in the address space occurs, before returning to
660	 * user space
661	 */
662	mtctr	r4
6631:	icbi	0,r6
664	addi	r6,r6,L1_CACHE_BYTES
665	bdnz	1b
666	sync
667	isync
668#endif /* CONFIG_44x */
669	blr
670
671/*
672 * Flush a particular page from the data cache to RAM, identified
673 * by its physical address.  We turn off the MMU so we can just use
674 * the physical address (this may be a highmem page without a kernel
675 * mapping).
676 *
677 *	void __flush_dcache_icache_phys(unsigned long physaddr)
678 */
679_GLOBAL(__flush_dcache_icache_phys)
680BEGIN_FTR_SECTION
681	blr					/* for 601, do nothing */
682END_FTR_SECTION_IFSET(CPU_FTR_COHERENT_ICACHE)
683	mfmsr	r10
684	rlwinm	r0,r10,0,28,26			/* clear DR */
685	mtmsr	r0
686	isync
687	rlwinm	r3,r3,0,0,19			/* Get page base address */
688	li	r4,4096/L1_CACHE_BYTES	/* Number of lines in a page */
689	mtctr	r4
690	mr	r6,r3
6910:	dcbst	0,r3				/* Write line to ram */
692	addi	r3,r3,L1_CACHE_BYTES
693	bdnz	0b
694	sync
695	mtctr	r4
6961:	icbi	0,r6
697	addi	r6,r6,L1_CACHE_BYTES
698	bdnz	1b
699	sync
700	mtmsr	r10				/* restore DR */
701	isync
702	blr
703
704/*
705 * Clear pages using the dcbz instruction, which doesn't cause any
706 * memory traffic (except to write out any cache lines which get
707 * displaced).  This only works on cacheable memory.
708 *
709 * void clear_pages(void *page, int order) ;
710 */
711_GLOBAL(clear_pages)
712	li	r0,4096/L1_CACHE_BYTES
713	slw	r0,r0,r4
714	mtctr	r0
715#ifdef CONFIG_8xx
716	li	r4, 0
7171:	stw	r4, 0(r3)
718	stw	r4, 4(r3)
719	stw	r4, 8(r3)
720	stw	r4, 12(r3)
721#else
7221:	dcbz	0,r3
723#endif
724	addi	r3,r3,L1_CACHE_BYTES
725	bdnz	1b
726	blr
727
728/*
729 * Copy a whole page.  We use the dcbz instruction on the destination
730 * to reduce memory traffic (it eliminates the unnecessary reads of
731 * the destination into cache).  This requires that the destination
732 * is cacheable.
733 */
734#define COPY_16_BYTES		\
735	lwz	r6,4(r4);	\
736	lwz	r7,8(r4);	\
737	lwz	r8,12(r4);	\
738	lwzu	r9,16(r4);	\
739	stw	r6,4(r3);	\
740	stw	r7,8(r3);	\
741	stw	r8,12(r3);	\
742	stwu	r9,16(r3)
743
744_GLOBAL(copy_page)
745	addi	r3,r3,-4
746	addi	r4,r4,-4
747
748#ifdef CONFIG_8xx
749	/* don't use prefetch on 8xx */
750    	li	r0,4096/L1_CACHE_BYTES
751	mtctr	r0
7521:	COPY_16_BYTES
753	bdnz	1b
754	blr
755
756#else	/* not 8xx, we can prefetch */
757	li	r5,4
758
759#if MAX_COPY_PREFETCH > 1
760	li	r0,MAX_COPY_PREFETCH
761	li	r11,4
762	mtctr	r0
76311:	dcbt	r11,r4
764	addi	r11,r11,L1_CACHE_BYTES
765	bdnz	11b
766#else /* MAX_COPY_PREFETCH == 1 */
767	dcbt	r5,r4
768	li	r11,L1_CACHE_BYTES+4
769#endif /* MAX_COPY_PREFETCH */
770	li	r0,4096/L1_CACHE_BYTES - MAX_COPY_PREFETCH
771	crclr	4*cr0+eq
7722:
773	mtctr	r0
7741:
775	dcbt	r11,r4
776	dcbz	r5,r3
777	COPY_16_BYTES
778#if L1_CACHE_BYTES >= 32
779	COPY_16_BYTES
780#if L1_CACHE_BYTES >= 64
781	COPY_16_BYTES
782	COPY_16_BYTES
783#if L1_CACHE_BYTES >= 128
784	COPY_16_BYTES
785	COPY_16_BYTES
786	COPY_16_BYTES
787	COPY_16_BYTES
788#endif
789#endif
790#endif
791	bdnz	1b
792	beqlr
793	crnot	4*cr0+eq,4*cr0+eq
794	li	r0,MAX_COPY_PREFETCH
795	li	r11,4
796	b	2b
797#endif	/* CONFIG_8xx */
798
799/*
800 * void atomic_clear_mask(atomic_t mask, atomic_t *addr)
801 * void atomic_set_mask(atomic_t mask, atomic_t *addr);
802 */
803_GLOBAL(atomic_clear_mask)
80410:	lwarx	r5,0,r4
805	andc	r5,r5,r3
806	PPC405_ERR77(0,r4)
807	stwcx.	r5,0,r4
808	bne-	10b
809	blr
810_GLOBAL(atomic_set_mask)
81110:	lwarx	r5,0,r4
812	or	r5,r5,r3
813	PPC405_ERR77(0,r4)
814	stwcx.	r5,0,r4
815	bne-	10b
816	blr
817
818/*
819 * Extended precision shifts.
820 *
821 * Updated to be valid for shift counts from 0 to 63 inclusive.
822 * -- Gabriel
823 *
824 * R3/R4 has 64 bit value
825 * R5    has shift count
826 * result in R3/R4
827 *
828 *  ashrdi3: arithmetic right shift (sign propagation)
829 *  lshrdi3: logical right shift
830 *  ashldi3: left shift
831 */
832_GLOBAL(__ashrdi3)
833	subfic	r6,r5,32
834	srw	r4,r4,r5	# LSW = count > 31 ? 0 : LSW >> count
835	addi	r7,r5,32	# could be xori, or addi with -32
836	slw	r6,r3,r6	# t1 = count > 31 ? 0 : MSW << (32-count)
837	rlwinm	r8,r7,0,32	# t3 = (count < 32) ? 32 : 0
838	sraw	r7,r3,r7	# t2 = MSW >> (count-32)
839	or	r4,r4,r6	# LSW |= t1
840	slw	r7,r7,r8	# t2 = (count < 32) ? 0 : t2
841	sraw	r3,r3,r5	# MSW = MSW >> count
842	or	r4,r4,r7	# LSW |= t2
843	blr
844
845_GLOBAL(__ashldi3)
846	subfic	r6,r5,32
847	slw	r3,r3,r5	# MSW = count > 31 ? 0 : MSW << count
848	addi	r7,r5,32	# could be xori, or addi with -32
849	srw	r6,r4,r6	# t1 = count > 31 ? 0 : LSW >> (32-count)
850	slw	r7,r4,r7	# t2 = count < 32 ? 0 : LSW << (count-32)
851	or	r3,r3,r6	# MSW |= t1
852	slw	r4,r4,r5	# LSW = LSW << count
853	or	r3,r3,r7	# MSW |= t2
854	blr
855
856_GLOBAL(__lshrdi3)
857	subfic	r6,r5,32
858	srw	r4,r4,r5	# LSW = count > 31 ? 0 : LSW >> count
859	addi	r7,r5,32	# could be xori, or addi with -32
860	slw	r6,r3,r6	# t1 = count > 31 ? 0 : MSW << (32-count)
861	srw	r7,r3,r7	# t2 = count < 32 ? 0 : MSW >> (count-32)
862	or	r4,r4,r6	# LSW |= t1
863	srw	r3,r3,r5	# MSW = MSW >> count
864	or	r4,r4,r7	# LSW |= t2
865	blr
866
867/*
868 * 64-bit comparison: __ucmpdi2(u64 a, u64 b)
869 * Returns 0 if a < b, 1 if a == b, 2 if a > b.
870 */
871_GLOBAL(__ucmpdi2)
872	cmplw	r3,r5
873	li	r3,1
874	bne	1f
875	cmplw	r4,r6
876	beqlr
8771:	li	r3,0
878	bltlr
879	li	r3,2
880	blr
881
882_GLOBAL(abs)
883	srawi	r4,r3,31
884	xor	r3,r3,r4
885	sub	r3,r3,r4
886	blr
887
888/*
889 * Create a kernel thread
890 *   kernel_thread(fn, arg, flags)
891 */
892_GLOBAL(kernel_thread)
893	stwu	r1,-16(r1)
894	stw	r30,8(r1)
895	stw	r31,12(r1)
896	mr	r30,r3		/* function */
897	mr	r31,r4		/* argument */
898	ori	r3,r5,CLONE_VM	/* flags */
899	oris	r3,r3,CLONE_UNTRACED>>16
900	li	r4,0		/* new sp (unused) */
901	li	r0,__NR_clone
902	sc
903	bns+	1f		/* did system call indicate error? */
904	neg	r3,r3		/* if so, make return code negative */
9051:	cmpwi	0,r3,0		/* parent or child? */
906	bne	2f		/* return if parent */
907	li	r0,0		/* make top-level stack frame */
908	stwu	r0,-16(r1)
909	mtlr	r30		/* fn addr in lr */
910	mr	r3,r31		/* load arg and call fn */
911	PPC440EP_ERR42
912	blrl
913	li	r0,__NR_exit	/* exit if function returns */
914	li	r3,0
915	sc
9162:	lwz	r30,8(r1)
917	lwz	r31,12(r1)
918	addi	r1,r1,16
919	blr
920
921/*
922 * This routine is just here to keep GCC happy - sigh...
923 */
924_GLOBAL(__main)
925	blr
926
927#ifdef CONFIG_KEXEC
928	/*
929	 * Must be relocatable PIC code callable as a C function.
930	 */
931	.globl relocate_new_kernel
932relocate_new_kernel:
933	/* r3 = page_list   */
934	/* r4 = reboot_code_buffer */
935	/* r5 = start_address      */
936
937	li	r0, 0
938
939	/*
940	 * Set Machine Status Register to a known status,
941	 * switch the MMU off and jump to 1: in a single step.
942	 */
943
944	mr	r8, r0
945	ori     r8, r8, MSR_RI|MSR_ME
946	mtspr	SPRN_SRR1, r8
947	addi	r8, r4, 1f - relocate_new_kernel
948	mtspr	SPRN_SRR0, r8
949	sync
950	rfi
951
9521:
953	/* from this point address translation is turned off */
954	/* and interrupts are disabled */
955
956	/* set a new stack at the bottom of our page... */
957	/* (not really needed now) */
958	addi	r1, r4, KEXEC_CONTROL_PAGE_SIZE - 8 /* for LR Save+Back Chain */
959	stw	r0, 0(r1)
960
961	/* Do the copies */
962	li	r6, 0 /* checksum */
963	mr	r0, r3
964	b	1f
965
9660:	/* top, read another word for the indirection page */
967	lwzu	r0, 4(r3)
968
9691:
970	/* is it a destination page? (r8) */
971	rlwinm.	r7, r0, 0, 31, 31 /* IND_DESTINATION (1<<0) */
972	beq	2f
973
974	rlwinm	r8, r0, 0, 0, 19 /* clear kexec flags, page align */
975	b	0b
976
9772:	/* is it an indirection page? (r3) */
978	rlwinm.	r7, r0, 0, 30, 30 /* IND_INDIRECTION (1<<1) */
979	beq	2f
980
981	rlwinm	r3, r0, 0, 0, 19 /* clear kexec flags, page align */
982	subi	r3, r3, 4
983	b	0b
984
9852:	/* are we done? */
986	rlwinm.	r7, r0, 0, 29, 29 /* IND_DONE (1<<2) */
987	beq	2f
988	b	3f
989
9902:	/* is it a source page? (r9) */
991	rlwinm.	r7, r0, 0, 28, 28 /* IND_SOURCE (1<<3) */
992	beq	0b
993
994	rlwinm	r9, r0, 0, 0, 19 /* clear kexec flags, page align */
995
996	li	r7, PAGE_SIZE / 4
997	mtctr   r7
998	subi    r9, r9, 4
999	subi    r8, r8, 4
10009:
1001	lwzu    r0, 4(r9)  /* do the copy */
1002	xor	r6, r6, r0
1003	stwu    r0, 4(r8)
1004	dcbst	0, r8
1005	sync
1006	icbi	0, r8
1007	bdnz    9b
1008
1009	addi    r9, r9, 4
1010	addi    r8, r8, 4
1011	b	0b
1012
10133:
1014
1015	/* To be certain of avoiding problems with self-modifying code
1016	 * execute a serializing instruction here.
1017	 */
1018	isync
1019	sync
1020
1021	/* jump to the entry point, usually the setup routine */
1022	mtlr	r5
1023	blrl
1024
10251:	b	1b
1026
1027relocate_new_kernel_end:
1028
1029	.globl relocate_new_kernel_size
1030relocate_new_kernel_size:
1031	.long relocate_new_kernel_end - relocate_new_kernel
1032#endif
1033