xref: /openbmc/linux/arch/mips/mm/c-r4k.c (revision 7fe2f639)
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  * Copyright (C) 1996 David S. Miller (davem@davemloft.net)
7  * Copyright (C) 1997, 1998, 1999, 2000, 2001, 2002 Ralf Baechle (ralf@gnu.org)
8  * Copyright (C) 1999, 2000 Silicon Graphics, Inc.
9  */
10 #include <linux/hardirq.h>
11 #include <linux/init.h>
12 #include <linux/highmem.h>
13 #include <linux/kernel.h>
14 #include <linux/linkage.h>
15 #include <linux/sched.h>
16 #include <linux/smp.h>
17 #include <linux/mm.h>
18 #include <linux/module.h>
19 #include <linux/bitops.h>
20 
21 #include <asm/bcache.h>
22 #include <asm/bootinfo.h>
23 #include <asm/cache.h>
24 #include <asm/cacheops.h>
25 #include <asm/cpu.h>
26 #include <asm/cpu-features.h>
27 #include <asm/io.h>
28 #include <asm/page.h>
29 #include <asm/pgtable.h>
30 #include <asm/r4kcache.h>
31 #include <asm/sections.h>
32 #include <asm/system.h>
33 #include <asm/mmu_context.h>
34 #include <asm/war.h>
35 #include <asm/cacheflush.h> /* for run_uncached() */
36 
37 
38 /*
39  * Special Variant of smp_call_function for use by cache functions:
40  *
41  *  o No return value
42  *  o collapses to normal function call on UP kernels
43  *  o collapses to normal function call on systems with a single shared
44  *    primary cache.
45  *  o doesn't disable interrupts on the local CPU
46  */
47 static inline void r4k_on_each_cpu(void (*func) (void *info), void *info)
48 {
49 	preempt_disable();
50 
51 #if !defined(CONFIG_MIPS_MT_SMP) && !defined(CONFIG_MIPS_MT_SMTC)
52 	smp_call_function(func, info, 1);
53 #endif
54 	func(info);
55 	preempt_enable();
56 }
57 
58 #if defined(CONFIG_MIPS_CMP)
59 #define cpu_has_safe_index_cacheops 0
60 #else
61 #define cpu_has_safe_index_cacheops 1
62 #endif
63 
64 /*
65  * Must die.
66  */
67 static unsigned long icache_size __read_mostly;
68 static unsigned long dcache_size __read_mostly;
69 static unsigned long scache_size __read_mostly;
70 
71 /*
72  * Dummy cache handling routines for machines without boardcaches
73  */
74 static void cache_noop(void) {}
75 
76 static struct bcache_ops no_sc_ops = {
77 	.bc_enable = (void *)cache_noop,
78 	.bc_disable = (void *)cache_noop,
79 	.bc_wback_inv = (void *)cache_noop,
80 	.bc_inv = (void *)cache_noop
81 };
82 
83 struct bcache_ops *bcops = &no_sc_ops;
84 
85 #define cpu_is_r4600_v1_x()	((read_c0_prid() & 0xfffffff0) == 0x00002010)
86 #define cpu_is_r4600_v2_x()	((read_c0_prid() & 0xfffffff0) == 0x00002020)
87 
88 #define R4600_HIT_CACHEOP_WAR_IMPL					\
89 do {									\
90 	if (R4600_V2_HIT_CACHEOP_WAR && cpu_is_r4600_v2_x())		\
91 		*(volatile unsigned long *)CKSEG1;			\
92 	if (R4600_V1_HIT_CACHEOP_WAR)					\
93 		__asm__ __volatile__("nop;nop;nop;nop");		\
94 } while (0)
95 
96 static void (*r4k_blast_dcache_page)(unsigned long addr);
97 
98 static inline void r4k_blast_dcache_page_dc32(unsigned long addr)
99 {
100 	R4600_HIT_CACHEOP_WAR_IMPL;
101 	blast_dcache32_page(addr);
102 }
103 
104 static inline void r4k_blast_dcache_page_dc64(unsigned long addr)
105 {
106 	R4600_HIT_CACHEOP_WAR_IMPL;
107 	blast_dcache64_page(addr);
108 }
109 
110 static void __cpuinit r4k_blast_dcache_page_setup(void)
111 {
112 	unsigned long  dc_lsize = cpu_dcache_line_size();
113 
114 	if (dc_lsize == 0)
115 		r4k_blast_dcache_page = (void *)cache_noop;
116 	else if (dc_lsize == 16)
117 		r4k_blast_dcache_page = blast_dcache16_page;
118 	else if (dc_lsize == 32)
119 		r4k_blast_dcache_page = r4k_blast_dcache_page_dc32;
120 	else if (dc_lsize == 64)
121 		r4k_blast_dcache_page = r4k_blast_dcache_page_dc64;
122 }
123 
124 static void (* r4k_blast_dcache_page_indexed)(unsigned long addr);
125 
126 static void __cpuinit r4k_blast_dcache_page_indexed_setup(void)
127 {
128 	unsigned long dc_lsize = cpu_dcache_line_size();
129 
130 	if (dc_lsize == 0)
131 		r4k_blast_dcache_page_indexed = (void *)cache_noop;
132 	else if (dc_lsize == 16)
133 		r4k_blast_dcache_page_indexed = blast_dcache16_page_indexed;
134 	else if (dc_lsize == 32)
135 		r4k_blast_dcache_page_indexed = blast_dcache32_page_indexed;
136 	else if (dc_lsize == 64)
137 		r4k_blast_dcache_page_indexed = blast_dcache64_page_indexed;
138 }
139 
140 static void (* r4k_blast_dcache)(void);
141 
142 static void __cpuinit r4k_blast_dcache_setup(void)
143 {
144 	unsigned long dc_lsize = cpu_dcache_line_size();
145 
146 	if (dc_lsize == 0)
147 		r4k_blast_dcache = (void *)cache_noop;
148 	else if (dc_lsize == 16)
149 		r4k_blast_dcache = blast_dcache16;
150 	else if (dc_lsize == 32)
151 		r4k_blast_dcache = blast_dcache32;
152 	else if (dc_lsize == 64)
153 		r4k_blast_dcache = blast_dcache64;
154 }
155 
156 /* force code alignment (used for TX49XX_ICACHE_INDEX_INV_WAR) */
157 #define JUMP_TO_ALIGN(order) \
158 	__asm__ __volatile__( \
159 		"b\t1f\n\t" \
160 		".align\t" #order "\n\t" \
161 		"1:\n\t" \
162 		)
163 #define CACHE32_UNROLL32_ALIGN	JUMP_TO_ALIGN(10) /* 32 * 32 = 1024 */
164 #define CACHE32_UNROLL32_ALIGN2	JUMP_TO_ALIGN(11)
165 
166 static inline void blast_r4600_v1_icache32(void)
167 {
168 	unsigned long flags;
169 
170 	local_irq_save(flags);
171 	blast_icache32();
172 	local_irq_restore(flags);
173 }
174 
175 static inline void tx49_blast_icache32(void)
176 {
177 	unsigned long start = INDEX_BASE;
178 	unsigned long end = start + current_cpu_data.icache.waysize;
179 	unsigned long ws_inc = 1UL << current_cpu_data.icache.waybit;
180 	unsigned long ws_end = current_cpu_data.icache.ways <<
181 	                       current_cpu_data.icache.waybit;
182 	unsigned long ws, addr;
183 
184 	CACHE32_UNROLL32_ALIGN2;
185 	/* I'm in even chunk.  blast odd chunks */
186 	for (ws = 0; ws < ws_end; ws += ws_inc)
187 		for (addr = start + 0x400; addr < end; addr += 0x400 * 2)
188 			cache32_unroll32(addr|ws, Index_Invalidate_I);
189 	CACHE32_UNROLL32_ALIGN;
190 	/* I'm in odd chunk.  blast even chunks */
191 	for (ws = 0; ws < ws_end; ws += ws_inc)
192 		for (addr = start; addr < end; addr += 0x400 * 2)
193 			cache32_unroll32(addr|ws, Index_Invalidate_I);
194 }
195 
196 static inline void blast_icache32_r4600_v1_page_indexed(unsigned long page)
197 {
198 	unsigned long flags;
199 
200 	local_irq_save(flags);
201 	blast_icache32_page_indexed(page);
202 	local_irq_restore(flags);
203 }
204 
205 static inline void tx49_blast_icache32_page_indexed(unsigned long page)
206 {
207 	unsigned long indexmask = current_cpu_data.icache.waysize - 1;
208 	unsigned long start = INDEX_BASE + (page & indexmask);
209 	unsigned long end = start + PAGE_SIZE;
210 	unsigned long ws_inc = 1UL << current_cpu_data.icache.waybit;
211 	unsigned long ws_end = current_cpu_data.icache.ways <<
212 	                       current_cpu_data.icache.waybit;
213 	unsigned long ws, addr;
214 
215 	CACHE32_UNROLL32_ALIGN2;
216 	/* I'm in even chunk.  blast odd chunks */
217 	for (ws = 0; ws < ws_end; ws += ws_inc)
218 		for (addr = start + 0x400; addr < end; addr += 0x400 * 2)
219 			cache32_unroll32(addr|ws, Index_Invalidate_I);
220 	CACHE32_UNROLL32_ALIGN;
221 	/* I'm in odd chunk.  blast even chunks */
222 	for (ws = 0; ws < ws_end; ws += ws_inc)
223 		for (addr = start; addr < end; addr += 0x400 * 2)
224 			cache32_unroll32(addr|ws, Index_Invalidate_I);
225 }
226 
227 static void (* r4k_blast_icache_page)(unsigned long addr);
228 
229 static void __cpuinit r4k_blast_icache_page_setup(void)
230 {
231 	unsigned long ic_lsize = cpu_icache_line_size();
232 
233 	if (ic_lsize == 0)
234 		r4k_blast_icache_page = (void *)cache_noop;
235 	else if (ic_lsize == 16)
236 		r4k_blast_icache_page = blast_icache16_page;
237 	else if (ic_lsize == 32)
238 		r4k_blast_icache_page = blast_icache32_page;
239 	else if (ic_lsize == 64)
240 		r4k_blast_icache_page = blast_icache64_page;
241 }
242 
243 
244 static void (* r4k_blast_icache_page_indexed)(unsigned long addr);
245 
246 static void __cpuinit r4k_blast_icache_page_indexed_setup(void)
247 {
248 	unsigned long ic_lsize = cpu_icache_line_size();
249 
250 	if (ic_lsize == 0)
251 		r4k_blast_icache_page_indexed = (void *)cache_noop;
252 	else if (ic_lsize == 16)
253 		r4k_blast_icache_page_indexed = blast_icache16_page_indexed;
254 	else if (ic_lsize == 32) {
255 		if (R4600_V1_INDEX_ICACHEOP_WAR && cpu_is_r4600_v1_x())
256 			r4k_blast_icache_page_indexed =
257 				blast_icache32_r4600_v1_page_indexed;
258 		else if (TX49XX_ICACHE_INDEX_INV_WAR)
259 			r4k_blast_icache_page_indexed =
260 				tx49_blast_icache32_page_indexed;
261 		else
262 			r4k_blast_icache_page_indexed =
263 				blast_icache32_page_indexed;
264 	} else if (ic_lsize == 64)
265 		r4k_blast_icache_page_indexed = blast_icache64_page_indexed;
266 }
267 
268 static void (* r4k_blast_icache)(void);
269 
270 static void __cpuinit r4k_blast_icache_setup(void)
271 {
272 	unsigned long ic_lsize = cpu_icache_line_size();
273 
274 	if (ic_lsize == 0)
275 		r4k_blast_icache = (void *)cache_noop;
276 	else if (ic_lsize == 16)
277 		r4k_blast_icache = blast_icache16;
278 	else if (ic_lsize == 32) {
279 		if (R4600_V1_INDEX_ICACHEOP_WAR && cpu_is_r4600_v1_x())
280 			r4k_blast_icache = blast_r4600_v1_icache32;
281 		else if (TX49XX_ICACHE_INDEX_INV_WAR)
282 			r4k_blast_icache = tx49_blast_icache32;
283 		else
284 			r4k_blast_icache = blast_icache32;
285 	} else if (ic_lsize == 64)
286 		r4k_blast_icache = blast_icache64;
287 }
288 
289 static void (* r4k_blast_scache_page)(unsigned long addr);
290 
291 static void __cpuinit r4k_blast_scache_page_setup(void)
292 {
293 	unsigned long sc_lsize = cpu_scache_line_size();
294 
295 	if (scache_size == 0)
296 		r4k_blast_scache_page = (void *)cache_noop;
297 	else if (sc_lsize == 16)
298 		r4k_blast_scache_page = blast_scache16_page;
299 	else if (sc_lsize == 32)
300 		r4k_blast_scache_page = blast_scache32_page;
301 	else if (sc_lsize == 64)
302 		r4k_blast_scache_page = blast_scache64_page;
303 	else if (sc_lsize == 128)
304 		r4k_blast_scache_page = blast_scache128_page;
305 }
306 
307 static void (* r4k_blast_scache_page_indexed)(unsigned long addr);
308 
309 static void __cpuinit r4k_blast_scache_page_indexed_setup(void)
310 {
311 	unsigned long sc_lsize = cpu_scache_line_size();
312 
313 	if (scache_size == 0)
314 		r4k_blast_scache_page_indexed = (void *)cache_noop;
315 	else if (sc_lsize == 16)
316 		r4k_blast_scache_page_indexed = blast_scache16_page_indexed;
317 	else if (sc_lsize == 32)
318 		r4k_blast_scache_page_indexed = blast_scache32_page_indexed;
319 	else if (sc_lsize == 64)
320 		r4k_blast_scache_page_indexed = blast_scache64_page_indexed;
321 	else if (sc_lsize == 128)
322 		r4k_blast_scache_page_indexed = blast_scache128_page_indexed;
323 }
324 
325 static void (* r4k_blast_scache)(void);
326 
327 static void __cpuinit r4k_blast_scache_setup(void)
328 {
329 	unsigned long sc_lsize = cpu_scache_line_size();
330 
331 	if (scache_size == 0)
332 		r4k_blast_scache = (void *)cache_noop;
333 	else if (sc_lsize == 16)
334 		r4k_blast_scache = blast_scache16;
335 	else if (sc_lsize == 32)
336 		r4k_blast_scache = blast_scache32;
337 	else if (sc_lsize == 64)
338 		r4k_blast_scache = blast_scache64;
339 	else if (sc_lsize == 128)
340 		r4k_blast_scache = blast_scache128;
341 }
342 
343 static inline void local_r4k___flush_cache_all(void * args)
344 {
345 #if defined(CONFIG_CPU_LOONGSON2)
346 	r4k_blast_scache();
347 	return;
348 #endif
349 	r4k_blast_dcache();
350 	r4k_blast_icache();
351 
352 	switch (current_cpu_type()) {
353 	case CPU_R4000SC:
354 	case CPU_R4000MC:
355 	case CPU_R4400SC:
356 	case CPU_R4400MC:
357 	case CPU_R10000:
358 	case CPU_R12000:
359 	case CPU_R14000:
360 		r4k_blast_scache();
361 	}
362 }
363 
364 static void r4k___flush_cache_all(void)
365 {
366 	r4k_on_each_cpu(local_r4k___flush_cache_all, NULL);
367 }
368 
369 static inline int has_valid_asid(const struct mm_struct *mm)
370 {
371 #if defined(CONFIG_MIPS_MT_SMP) || defined(CONFIG_MIPS_MT_SMTC)
372 	int i;
373 
374 	for_each_online_cpu(i)
375 		if (cpu_context(i, mm))
376 			return 1;
377 
378 	return 0;
379 #else
380 	return cpu_context(smp_processor_id(), mm);
381 #endif
382 }
383 
384 static void r4k__flush_cache_vmap(void)
385 {
386 	r4k_blast_dcache();
387 }
388 
389 static void r4k__flush_cache_vunmap(void)
390 {
391 	r4k_blast_dcache();
392 }
393 
394 static inline void local_r4k_flush_cache_range(void * args)
395 {
396 	struct vm_area_struct *vma = args;
397 	int exec = vma->vm_flags & VM_EXEC;
398 
399 	if (!(has_valid_asid(vma->vm_mm)))
400 		return;
401 
402 	r4k_blast_dcache();
403 	if (exec)
404 		r4k_blast_icache();
405 }
406 
407 static void r4k_flush_cache_range(struct vm_area_struct *vma,
408 	unsigned long start, unsigned long end)
409 {
410 	int exec = vma->vm_flags & VM_EXEC;
411 
412 	if (cpu_has_dc_aliases || (exec && !cpu_has_ic_fills_f_dc))
413 		r4k_on_each_cpu(local_r4k_flush_cache_range, vma);
414 }
415 
416 static inline void local_r4k_flush_cache_mm(void * args)
417 {
418 	struct mm_struct *mm = args;
419 
420 	if (!has_valid_asid(mm))
421 		return;
422 
423 	/*
424 	 * Kludge alert.  For obscure reasons R4000SC and R4400SC go nuts if we
425 	 * only flush the primary caches but R10000 and R12000 behave sane ...
426 	 * R4000SC and R4400SC indexed S-cache ops also invalidate primary
427 	 * caches, so we can bail out early.
428 	 */
429 	if (current_cpu_type() == CPU_R4000SC ||
430 	    current_cpu_type() == CPU_R4000MC ||
431 	    current_cpu_type() == CPU_R4400SC ||
432 	    current_cpu_type() == CPU_R4400MC) {
433 		r4k_blast_scache();
434 		return;
435 	}
436 
437 	r4k_blast_dcache();
438 }
439 
440 static void r4k_flush_cache_mm(struct mm_struct *mm)
441 {
442 	if (!cpu_has_dc_aliases)
443 		return;
444 
445 	r4k_on_each_cpu(local_r4k_flush_cache_mm, mm);
446 }
447 
448 struct flush_cache_page_args {
449 	struct vm_area_struct *vma;
450 	unsigned long addr;
451 	unsigned long pfn;
452 };
453 
454 static inline void local_r4k_flush_cache_page(void *args)
455 {
456 	struct flush_cache_page_args *fcp_args = args;
457 	struct vm_area_struct *vma = fcp_args->vma;
458 	unsigned long addr = fcp_args->addr;
459 	struct page *page = pfn_to_page(fcp_args->pfn);
460 	int exec = vma->vm_flags & VM_EXEC;
461 	struct mm_struct *mm = vma->vm_mm;
462 	int map_coherent = 0;
463 	pgd_t *pgdp;
464 	pud_t *pudp;
465 	pmd_t *pmdp;
466 	pte_t *ptep;
467 	void *vaddr;
468 
469 	/*
470 	 * If ownes no valid ASID yet, cannot possibly have gotten
471 	 * this page into the cache.
472 	 */
473 	if (!has_valid_asid(mm))
474 		return;
475 
476 	addr &= PAGE_MASK;
477 	pgdp = pgd_offset(mm, addr);
478 	pudp = pud_offset(pgdp, addr);
479 	pmdp = pmd_offset(pudp, addr);
480 	ptep = pte_offset(pmdp, addr);
481 
482 	/*
483 	 * If the page isn't marked valid, the page cannot possibly be
484 	 * in the cache.
485 	 */
486 	if (!(pte_present(*ptep)))
487 		return;
488 
489 	if ((mm == current->active_mm) && (pte_val(*ptep) & _PAGE_VALID))
490 		vaddr = NULL;
491 	else {
492 		/*
493 		 * Use kmap_coherent or kmap_atomic to do flushes for
494 		 * another ASID than the current one.
495 		 */
496 		map_coherent = (cpu_has_dc_aliases &&
497 				page_mapped(page) && !Page_dcache_dirty(page));
498 		if (map_coherent)
499 			vaddr = kmap_coherent(page, addr);
500 		else
501 			vaddr = kmap_atomic(page, KM_USER0);
502 		addr = (unsigned long)vaddr;
503 	}
504 
505 	if (cpu_has_dc_aliases || (exec && !cpu_has_ic_fills_f_dc)) {
506 		r4k_blast_dcache_page(addr);
507 		if (exec && !cpu_icache_snoops_remote_store)
508 			r4k_blast_scache_page(addr);
509 	}
510 	if (exec) {
511 		if (vaddr && cpu_has_vtag_icache && mm == current->active_mm) {
512 			int cpu = smp_processor_id();
513 
514 			if (cpu_context(cpu, mm) != 0)
515 				drop_mmu_context(mm, cpu);
516 		} else
517 			r4k_blast_icache_page(addr);
518 	}
519 
520 	if (vaddr) {
521 		if (map_coherent)
522 			kunmap_coherent();
523 		else
524 			kunmap_atomic(vaddr, KM_USER0);
525 	}
526 }
527 
528 static void r4k_flush_cache_page(struct vm_area_struct *vma,
529 	unsigned long addr, unsigned long pfn)
530 {
531 	struct flush_cache_page_args args;
532 
533 	args.vma = vma;
534 	args.addr = addr;
535 	args.pfn = pfn;
536 
537 	r4k_on_each_cpu(local_r4k_flush_cache_page, &args);
538 }
539 
540 static inline void local_r4k_flush_data_cache_page(void * addr)
541 {
542 	r4k_blast_dcache_page((unsigned long) addr);
543 }
544 
545 static void r4k_flush_data_cache_page(unsigned long addr)
546 {
547 	if (in_atomic())
548 		local_r4k_flush_data_cache_page((void *)addr);
549 	else
550 		r4k_on_each_cpu(local_r4k_flush_data_cache_page, (void *) addr);
551 }
552 
553 struct flush_icache_range_args {
554 	unsigned long start;
555 	unsigned long end;
556 };
557 
558 static inline void local_r4k_flush_icache_range(unsigned long start, unsigned long end)
559 {
560 	if (!cpu_has_ic_fills_f_dc) {
561 		if (end - start >= dcache_size) {
562 			r4k_blast_dcache();
563 		} else {
564 			R4600_HIT_CACHEOP_WAR_IMPL;
565 			protected_blast_dcache_range(start, end);
566 		}
567 	}
568 
569 	if (end - start > icache_size)
570 		r4k_blast_icache();
571 	else
572 		protected_blast_icache_range(start, end);
573 }
574 
575 static inline void local_r4k_flush_icache_range_ipi(void *args)
576 {
577 	struct flush_icache_range_args *fir_args = args;
578 	unsigned long start = fir_args->start;
579 	unsigned long end = fir_args->end;
580 
581 	local_r4k_flush_icache_range(start, end);
582 }
583 
584 static void r4k_flush_icache_range(unsigned long start, unsigned long end)
585 {
586 	struct flush_icache_range_args args;
587 
588 	args.start = start;
589 	args.end = end;
590 
591 	r4k_on_each_cpu(local_r4k_flush_icache_range_ipi, &args);
592 	instruction_hazard();
593 }
594 
595 #ifdef CONFIG_DMA_NONCOHERENT
596 
597 static void r4k_dma_cache_wback_inv(unsigned long addr, unsigned long size)
598 {
599 	/* Catch bad driver code */
600 	BUG_ON(size == 0);
601 
602 	if (cpu_has_inclusive_pcaches) {
603 		if (size >= scache_size)
604 			r4k_blast_scache();
605 		else
606 			blast_scache_range(addr, addr + size);
607 		return;
608 	}
609 
610 	/*
611 	 * Either no secondary cache or the available caches don't have the
612 	 * subset property so we have to flush the primary caches
613 	 * explicitly
614 	 */
615 	if (cpu_has_safe_index_cacheops && size >= dcache_size) {
616 		r4k_blast_dcache();
617 	} else {
618 		R4600_HIT_CACHEOP_WAR_IMPL;
619 		blast_dcache_range(addr, addr + size);
620 	}
621 
622 	bc_wback_inv(addr, size);
623 }
624 
625 static void r4k_dma_cache_inv(unsigned long addr, unsigned long size)
626 {
627 	/* Catch bad driver code */
628 	BUG_ON(size == 0);
629 
630 	if (cpu_has_inclusive_pcaches) {
631 		if (size >= scache_size)
632 			r4k_blast_scache();
633 		else {
634 			unsigned long lsize = cpu_scache_line_size();
635 			unsigned long almask = ~(lsize - 1);
636 
637 			/*
638 			 * There is no clearly documented alignment requirement
639 			 * for the cache instruction on MIPS processors and
640 			 * some processors, among them the RM5200 and RM7000
641 			 * QED processors will throw an address error for cache
642 			 * hit ops with insufficient alignment.  Solved by
643 			 * aligning the address to cache line size.
644 			 */
645 			cache_op(Hit_Writeback_Inv_SD, addr & almask);
646 			cache_op(Hit_Writeback_Inv_SD,
647 				 (addr + size - 1) & almask);
648 			blast_inv_scache_range(addr, addr + size);
649 		}
650 		return;
651 	}
652 
653 	if (cpu_has_safe_index_cacheops && size >= dcache_size) {
654 		r4k_blast_dcache();
655 	} else {
656 		unsigned long lsize = cpu_dcache_line_size();
657 		unsigned long almask = ~(lsize - 1);
658 
659 		R4600_HIT_CACHEOP_WAR_IMPL;
660 		cache_op(Hit_Writeback_Inv_D, addr & almask);
661 		cache_op(Hit_Writeback_Inv_D, (addr + size - 1)  & almask);
662 		blast_inv_dcache_range(addr, addr + size);
663 	}
664 
665 	bc_inv(addr, size);
666 }
667 #endif /* CONFIG_DMA_NONCOHERENT */
668 
669 /*
670  * While we're protected against bad userland addresses we don't care
671  * very much about what happens in that case.  Usually a segmentation
672  * fault will dump the process later on anyway ...
673  */
674 static void local_r4k_flush_cache_sigtramp(void * arg)
675 {
676 	unsigned long ic_lsize = cpu_icache_line_size();
677 	unsigned long dc_lsize = cpu_dcache_line_size();
678 	unsigned long sc_lsize = cpu_scache_line_size();
679 	unsigned long addr = (unsigned long) arg;
680 
681 	R4600_HIT_CACHEOP_WAR_IMPL;
682 	if (dc_lsize)
683 		protected_writeback_dcache_line(addr & ~(dc_lsize - 1));
684 	if (!cpu_icache_snoops_remote_store && scache_size)
685 		protected_writeback_scache_line(addr & ~(sc_lsize - 1));
686 	if (ic_lsize)
687 		protected_flush_icache_line(addr & ~(ic_lsize - 1));
688 	if (MIPS4K_ICACHE_REFILL_WAR) {
689 		__asm__ __volatile__ (
690 			".set push\n\t"
691 			".set noat\n\t"
692 			".set mips3\n\t"
693 #ifdef CONFIG_32BIT
694 			"la	$at,1f\n\t"
695 #endif
696 #ifdef CONFIG_64BIT
697 			"dla	$at,1f\n\t"
698 #endif
699 			"cache	%0,($at)\n\t"
700 			"nop; nop; nop\n"
701 			"1:\n\t"
702 			".set pop"
703 			:
704 			: "i" (Hit_Invalidate_I));
705 	}
706 	if (MIPS_CACHE_SYNC_WAR)
707 		__asm__ __volatile__ ("sync");
708 }
709 
710 static void r4k_flush_cache_sigtramp(unsigned long addr)
711 {
712 	r4k_on_each_cpu(local_r4k_flush_cache_sigtramp, (void *) addr);
713 }
714 
715 static void r4k_flush_icache_all(void)
716 {
717 	if (cpu_has_vtag_icache)
718 		r4k_blast_icache();
719 }
720 
721 static inline void rm7k_erratum31(void)
722 {
723 	const unsigned long ic_lsize = 32;
724 	unsigned long addr;
725 
726 	/* RM7000 erratum #31. The icache is screwed at startup. */
727 	write_c0_taglo(0);
728 	write_c0_taghi(0);
729 
730 	for (addr = INDEX_BASE; addr <= INDEX_BASE + 4096; addr += ic_lsize) {
731 		__asm__ __volatile__ (
732 			".set push\n\t"
733 			".set noreorder\n\t"
734 			".set mips3\n\t"
735 			"cache\t%1, 0(%0)\n\t"
736 			"cache\t%1, 0x1000(%0)\n\t"
737 			"cache\t%1, 0x2000(%0)\n\t"
738 			"cache\t%1, 0x3000(%0)\n\t"
739 			"cache\t%2, 0(%0)\n\t"
740 			"cache\t%2, 0x1000(%0)\n\t"
741 			"cache\t%2, 0x2000(%0)\n\t"
742 			"cache\t%2, 0x3000(%0)\n\t"
743 			"cache\t%1, 0(%0)\n\t"
744 			"cache\t%1, 0x1000(%0)\n\t"
745 			"cache\t%1, 0x2000(%0)\n\t"
746 			"cache\t%1, 0x3000(%0)\n\t"
747 			".set pop\n"
748 			:
749 			: "r" (addr), "i" (Index_Store_Tag_I), "i" (Fill));
750 	}
751 }
752 
753 static char *way_string[] __cpuinitdata = { NULL, "direct mapped", "2-way",
754 	"3-way", "4-way", "5-way", "6-way", "7-way", "8-way"
755 };
756 
757 static void __cpuinit probe_pcache(void)
758 {
759 	struct cpuinfo_mips *c = &current_cpu_data;
760 	unsigned int config = read_c0_config();
761 	unsigned int prid = read_c0_prid();
762 	unsigned long config1;
763 	unsigned int lsize;
764 
765 	switch (c->cputype) {
766 	case CPU_R4600:			/* QED style two way caches? */
767 	case CPU_R4700:
768 	case CPU_R5000:
769 	case CPU_NEVADA:
770 		icache_size = 1 << (12 + ((config & CONF_IC) >> 9));
771 		c->icache.linesz = 16 << ((config & CONF_IB) >> 5);
772 		c->icache.ways = 2;
773 		c->icache.waybit = __ffs(icache_size/2);
774 
775 		dcache_size = 1 << (12 + ((config & CONF_DC) >> 6));
776 		c->dcache.linesz = 16 << ((config & CONF_DB) >> 4);
777 		c->dcache.ways = 2;
778 		c->dcache.waybit= __ffs(dcache_size/2);
779 
780 		c->options |= MIPS_CPU_CACHE_CDEX_P;
781 		break;
782 
783 	case CPU_R5432:
784 	case CPU_R5500:
785 		icache_size = 1 << (12 + ((config & CONF_IC) >> 9));
786 		c->icache.linesz = 16 << ((config & CONF_IB) >> 5);
787 		c->icache.ways = 2;
788 		c->icache.waybit= 0;
789 
790 		dcache_size = 1 << (12 + ((config & CONF_DC) >> 6));
791 		c->dcache.linesz = 16 << ((config & CONF_DB) >> 4);
792 		c->dcache.ways = 2;
793 		c->dcache.waybit = 0;
794 
795 		c->options |= MIPS_CPU_CACHE_CDEX_P | MIPS_CPU_PREFETCH;
796 		break;
797 
798 	case CPU_TX49XX:
799 		icache_size = 1 << (12 + ((config & CONF_IC) >> 9));
800 		c->icache.linesz = 16 << ((config & CONF_IB) >> 5);
801 		c->icache.ways = 4;
802 		c->icache.waybit= 0;
803 
804 		dcache_size = 1 << (12 + ((config & CONF_DC) >> 6));
805 		c->dcache.linesz = 16 << ((config & CONF_DB) >> 4);
806 		c->dcache.ways = 4;
807 		c->dcache.waybit = 0;
808 
809 		c->options |= MIPS_CPU_CACHE_CDEX_P;
810 		c->options |= MIPS_CPU_PREFETCH;
811 		break;
812 
813 	case CPU_R4000PC:
814 	case CPU_R4000SC:
815 	case CPU_R4000MC:
816 	case CPU_R4400PC:
817 	case CPU_R4400SC:
818 	case CPU_R4400MC:
819 	case CPU_R4300:
820 		icache_size = 1 << (12 + ((config & CONF_IC) >> 9));
821 		c->icache.linesz = 16 << ((config & CONF_IB) >> 5);
822 		c->icache.ways = 1;
823 		c->icache.waybit = 0; 	/* doesn't matter */
824 
825 		dcache_size = 1 << (12 + ((config & CONF_DC) >> 6));
826 		c->dcache.linesz = 16 << ((config & CONF_DB) >> 4);
827 		c->dcache.ways = 1;
828 		c->dcache.waybit = 0;	/* does not matter */
829 
830 		c->options |= MIPS_CPU_CACHE_CDEX_P;
831 		break;
832 
833 	case CPU_R10000:
834 	case CPU_R12000:
835 	case CPU_R14000:
836 		icache_size = 1 << (12 + ((config & R10K_CONF_IC) >> 29));
837 		c->icache.linesz = 64;
838 		c->icache.ways = 2;
839 		c->icache.waybit = 0;
840 
841 		dcache_size = 1 << (12 + ((config & R10K_CONF_DC) >> 26));
842 		c->dcache.linesz = 32;
843 		c->dcache.ways = 2;
844 		c->dcache.waybit = 0;
845 
846 		c->options |= MIPS_CPU_PREFETCH;
847 		break;
848 
849 	case CPU_VR4133:
850 		write_c0_config(config & ~VR41_CONF_P4K);
851 	case CPU_VR4131:
852 		/* Workaround for cache instruction bug of VR4131 */
853 		if (c->processor_id == 0x0c80U || c->processor_id == 0x0c81U ||
854 		    c->processor_id == 0x0c82U) {
855 			config |= 0x00400000U;
856 			if (c->processor_id == 0x0c80U)
857 				config |= VR41_CONF_BP;
858 			write_c0_config(config);
859 		} else
860 			c->options |= MIPS_CPU_CACHE_CDEX_P;
861 
862 		icache_size = 1 << (10 + ((config & CONF_IC) >> 9));
863 		c->icache.linesz = 16 << ((config & CONF_IB) >> 5);
864 		c->icache.ways = 2;
865 		c->icache.waybit = __ffs(icache_size/2);
866 
867 		dcache_size = 1 << (10 + ((config & CONF_DC) >> 6));
868 		c->dcache.linesz = 16 << ((config & CONF_DB) >> 4);
869 		c->dcache.ways = 2;
870 		c->dcache.waybit = __ffs(dcache_size/2);
871 		break;
872 
873 	case CPU_VR41XX:
874 	case CPU_VR4111:
875 	case CPU_VR4121:
876 	case CPU_VR4122:
877 	case CPU_VR4181:
878 	case CPU_VR4181A:
879 		icache_size = 1 << (10 + ((config & CONF_IC) >> 9));
880 		c->icache.linesz = 16 << ((config & CONF_IB) >> 5);
881 		c->icache.ways = 1;
882 		c->icache.waybit = 0; 	/* doesn't matter */
883 
884 		dcache_size = 1 << (10 + ((config & CONF_DC) >> 6));
885 		c->dcache.linesz = 16 << ((config & CONF_DB) >> 4);
886 		c->dcache.ways = 1;
887 		c->dcache.waybit = 0;	/* does not matter */
888 
889 		c->options |= MIPS_CPU_CACHE_CDEX_P;
890 		break;
891 
892 	case CPU_RM7000:
893 		rm7k_erratum31();
894 
895 	case CPU_RM9000:
896 		icache_size = 1 << (12 + ((config & CONF_IC) >> 9));
897 		c->icache.linesz = 16 << ((config & CONF_IB) >> 5);
898 		c->icache.ways = 4;
899 		c->icache.waybit = __ffs(icache_size / c->icache.ways);
900 
901 		dcache_size = 1 << (12 + ((config & CONF_DC) >> 6));
902 		c->dcache.linesz = 16 << ((config & CONF_DB) >> 4);
903 		c->dcache.ways = 4;
904 		c->dcache.waybit = __ffs(dcache_size / c->dcache.ways);
905 
906 #if !defined(CONFIG_SMP) || !defined(RM9000_CDEX_SMP_WAR)
907 		c->options |= MIPS_CPU_CACHE_CDEX_P;
908 #endif
909 		c->options |= MIPS_CPU_PREFETCH;
910 		break;
911 
912 	case CPU_LOONGSON2:
913 		icache_size = 1 << (12 + ((config & CONF_IC) >> 9));
914 		c->icache.linesz = 16 << ((config & CONF_IB) >> 5);
915 		if (prid & 0x3)
916 			c->icache.ways = 4;
917 		else
918 			c->icache.ways = 2;
919 		c->icache.waybit = 0;
920 
921 		dcache_size = 1 << (12 + ((config & CONF_DC) >> 6));
922 		c->dcache.linesz = 16 << ((config & CONF_DB) >> 4);
923 		if (prid & 0x3)
924 			c->dcache.ways = 4;
925 		else
926 			c->dcache.ways = 2;
927 		c->dcache.waybit = 0;
928 		break;
929 
930 	default:
931 		if (!(config & MIPS_CONF_M))
932 			panic("Don't know how to probe P-caches on this cpu.");
933 
934 		/*
935 		 * So we seem to be a MIPS32 or MIPS64 CPU
936 		 * So let's probe the I-cache ...
937 		 */
938 		config1 = read_c0_config1();
939 
940 		if ((lsize = ((config1 >> 19) & 7)))
941 			c->icache.linesz = 2 << lsize;
942 		else
943 			c->icache.linesz = lsize;
944 		c->icache.sets = 64 << ((config1 >> 22) & 7);
945 		c->icache.ways = 1 + ((config1 >> 16) & 7);
946 
947 		icache_size = c->icache.sets *
948 		              c->icache.ways *
949 		              c->icache.linesz;
950 		c->icache.waybit = __ffs(icache_size/c->icache.ways);
951 
952 		if (config & 0x8)		/* VI bit */
953 			c->icache.flags |= MIPS_CACHE_VTAG;
954 
955 		/*
956 		 * Now probe the MIPS32 / MIPS64 data cache.
957 		 */
958 		c->dcache.flags = 0;
959 
960 		if ((lsize = ((config1 >> 10) & 7)))
961 			c->dcache.linesz = 2 << lsize;
962 		else
963 			c->dcache.linesz= lsize;
964 		c->dcache.sets = 64 << ((config1 >> 13) & 7);
965 		c->dcache.ways = 1 + ((config1 >> 7) & 7);
966 
967 		dcache_size = c->dcache.sets *
968 		              c->dcache.ways *
969 		              c->dcache.linesz;
970 		c->dcache.waybit = __ffs(dcache_size/c->dcache.ways);
971 
972 		c->options |= MIPS_CPU_PREFETCH;
973 		break;
974 	}
975 
976 	/*
977 	 * Processor configuration sanity check for the R4000SC erratum
978 	 * #5.  With page sizes larger than 32kB there is no possibility
979 	 * to get a VCE exception anymore so we don't care about this
980 	 * misconfiguration.  The case is rather theoretical anyway;
981 	 * presumably no vendor is shipping his hardware in the "bad"
982 	 * configuration.
983 	 */
984 	if ((prid & 0xff00) == PRID_IMP_R4000 && (prid & 0xff) < 0x40 &&
985 	    !(config & CONF_SC) && c->icache.linesz != 16 &&
986 	    PAGE_SIZE <= 0x8000)
987 		panic("Improper R4000SC processor configuration detected");
988 
989 	/* compute a couple of other cache variables */
990 	c->icache.waysize = icache_size / c->icache.ways;
991 	c->dcache.waysize = dcache_size / c->dcache.ways;
992 
993 	c->icache.sets = c->icache.linesz ?
994 		icache_size / (c->icache.linesz * c->icache.ways) : 0;
995 	c->dcache.sets = c->dcache.linesz ?
996 		dcache_size / (c->dcache.linesz * c->dcache.ways) : 0;
997 
998 	/*
999 	 * R10000 and R12000 P-caches are odd in a positive way.  They're 32kB
1000 	 * 2-way virtually indexed so normally would suffer from aliases.  So
1001 	 * normally they'd suffer from aliases but magic in the hardware deals
1002 	 * with that for us so we don't need to take care ourselves.
1003 	 */
1004 	switch (c->cputype) {
1005 	case CPU_20KC:
1006 	case CPU_25KF:
1007 	case CPU_SB1:
1008 	case CPU_SB1A:
1009 	case CPU_XLR:
1010 		c->dcache.flags |= MIPS_CACHE_PINDEX;
1011 		break;
1012 
1013 	case CPU_R10000:
1014 	case CPU_R12000:
1015 	case CPU_R14000:
1016 		break;
1017 
1018 	case CPU_24K:
1019 	case CPU_34K:
1020 	case CPU_74K:
1021 	case CPU_1004K:
1022 		if ((read_c0_config7() & (1 << 16))) {
1023 			/* effectively physically indexed dcache,
1024 			   thus no virtual aliases. */
1025 			c->dcache.flags |= MIPS_CACHE_PINDEX;
1026 			break;
1027 		}
1028 	default:
1029 		if (c->dcache.waysize > PAGE_SIZE)
1030 			c->dcache.flags |= MIPS_CACHE_ALIASES;
1031 	}
1032 
1033 	switch (c->cputype) {
1034 	case CPU_20KC:
1035 		/*
1036 		 * Some older 20Kc chips doesn't have the 'VI' bit in
1037 		 * the config register.
1038 		 */
1039 		c->icache.flags |= MIPS_CACHE_VTAG;
1040 		break;
1041 
1042 	case CPU_ALCHEMY:
1043 		c->icache.flags |= MIPS_CACHE_IC_F_DC;
1044 		break;
1045 	}
1046 
1047 #ifdef  CONFIG_CPU_LOONGSON2
1048 	/*
1049 	 * LOONGSON2 has 4 way icache, but when using indexed cache op,
1050 	 * one op will act on all 4 ways
1051 	 */
1052 	c->icache.ways = 1;
1053 #endif
1054 
1055 	printk("Primary instruction cache %ldkB, %s, %s, linesize %d bytes.\n",
1056 	       icache_size >> 10,
1057 	       c->icache.flags & MIPS_CACHE_VTAG ? "VIVT" : "VIPT",
1058 	       way_string[c->icache.ways], c->icache.linesz);
1059 
1060 	printk("Primary data cache %ldkB, %s, %s, %s, linesize %d bytes\n",
1061 	       dcache_size >> 10, way_string[c->dcache.ways],
1062 	       (c->dcache.flags & MIPS_CACHE_PINDEX) ? "PIPT" : "VIPT",
1063 	       (c->dcache.flags & MIPS_CACHE_ALIASES) ?
1064 			"cache aliases" : "no aliases",
1065 	       c->dcache.linesz);
1066 }
1067 
1068 /*
1069  * If you even _breathe_ on this function, look at the gcc output and make sure
1070  * it does not pop things on and off the stack for the cache sizing loop that
1071  * executes in KSEG1 space or else you will crash and burn badly.  You have
1072  * been warned.
1073  */
1074 static int __cpuinit probe_scache(void)
1075 {
1076 	unsigned long flags, addr, begin, end, pow2;
1077 	unsigned int config = read_c0_config();
1078 	struct cpuinfo_mips *c = &current_cpu_data;
1079 
1080 	if (config & CONF_SC)
1081 		return 0;
1082 
1083 	begin = (unsigned long) &_stext;
1084 	begin &= ~((4 * 1024 * 1024) - 1);
1085 	end = begin + (4 * 1024 * 1024);
1086 
1087 	/*
1088 	 * This is such a bitch, you'd think they would make it easy to do
1089 	 * this.  Away you daemons of stupidity!
1090 	 */
1091 	local_irq_save(flags);
1092 
1093 	/* Fill each size-multiple cache line with a valid tag. */
1094 	pow2 = (64 * 1024);
1095 	for (addr = begin; addr < end; addr = (begin + pow2)) {
1096 		unsigned long *p = (unsigned long *) addr;
1097 		__asm__ __volatile__("nop" : : "r" (*p)); /* whee... */
1098 		pow2 <<= 1;
1099 	}
1100 
1101 	/* Load first line with zero (therefore invalid) tag. */
1102 	write_c0_taglo(0);
1103 	write_c0_taghi(0);
1104 	__asm__ __volatile__("nop; nop; nop; nop;"); /* avoid the hazard */
1105 	cache_op(Index_Store_Tag_I, begin);
1106 	cache_op(Index_Store_Tag_D, begin);
1107 	cache_op(Index_Store_Tag_SD, begin);
1108 
1109 	/* Now search for the wrap around point. */
1110 	pow2 = (128 * 1024);
1111 	for (addr = begin + (128 * 1024); addr < end; addr = begin + pow2) {
1112 		cache_op(Index_Load_Tag_SD, addr);
1113 		__asm__ __volatile__("nop; nop; nop; nop;"); /* hazard... */
1114 		if (!read_c0_taglo())
1115 			break;
1116 		pow2 <<= 1;
1117 	}
1118 	local_irq_restore(flags);
1119 	addr -= begin;
1120 
1121 	scache_size = addr;
1122 	c->scache.linesz = 16 << ((config & R4K_CONF_SB) >> 22);
1123 	c->scache.ways = 1;
1124 	c->dcache.waybit = 0;		/* does not matter */
1125 
1126 	return 1;
1127 }
1128 
1129 #if defined(CONFIG_CPU_LOONGSON2)
1130 static void __init loongson2_sc_init(void)
1131 {
1132 	struct cpuinfo_mips *c = &current_cpu_data;
1133 
1134 	scache_size = 512*1024;
1135 	c->scache.linesz = 32;
1136 	c->scache.ways = 4;
1137 	c->scache.waybit = 0;
1138 	c->scache.waysize = scache_size / (c->scache.ways);
1139 	c->scache.sets = scache_size / (c->scache.linesz * c->scache.ways);
1140 	pr_info("Unified secondary cache %ldkB %s, linesize %d bytes.\n",
1141 	       scache_size >> 10, way_string[c->scache.ways], c->scache.linesz);
1142 
1143 	c->options |= MIPS_CPU_INCLUSIVE_CACHES;
1144 }
1145 #endif
1146 
1147 extern int r5k_sc_init(void);
1148 extern int rm7k_sc_init(void);
1149 extern int mips_sc_init(void);
1150 
1151 static void __cpuinit setup_scache(void)
1152 {
1153 	struct cpuinfo_mips *c = &current_cpu_data;
1154 	unsigned int config = read_c0_config();
1155 	int sc_present = 0;
1156 
1157 	/*
1158 	 * Do the probing thing on R4000SC and R4400SC processors.  Other
1159 	 * processors don't have a S-cache that would be relevant to the
1160 	 * Linux memory management.
1161 	 */
1162 	switch (c->cputype) {
1163 	case CPU_R4000SC:
1164 	case CPU_R4000MC:
1165 	case CPU_R4400SC:
1166 	case CPU_R4400MC:
1167 		sc_present = run_uncached(probe_scache);
1168 		if (sc_present)
1169 			c->options |= MIPS_CPU_CACHE_CDEX_S;
1170 		break;
1171 
1172 	case CPU_R10000:
1173 	case CPU_R12000:
1174 	case CPU_R14000:
1175 		scache_size = 0x80000 << ((config & R10K_CONF_SS) >> 16);
1176 		c->scache.linesz = 64 << ((config >> 13) & 1);
1177 		c->scache.ways = 2;
1178 		c->scache.waybit= 0;
1179 		sc_present = 1;
1180 		break;
1181 
1182 	case CPU_R5000:
1183 	case CPU_NEVADA:
1184 #ifdef CONFIG_R5000_CPU_SCACHE
1185 		r5k_sc_init();
1186 #endif
1187                 return;
1188 
1189 	case CPU_RM7000:
1190 	case CPU_RM9000:
1191 #ifdef CONFIG_RM7000_CPU_SCACHE
1192 		rm7k_sc_init();
1193 #endif
1194 		return;
1195 
1196 #if defined(CONFIG_CPU_LOONGSON2)
1197 	case CPU_LOONGSON2:
1198 		loongson2_sc_init();
1199 		return;
1200 #endif
1201 
1202 	default:
1203 		if (c->isa_level == MIPS_CPU_ISA_M32R1 ||
1204 		    c->isa_level == MIPS_CPU_ISA_M32R2 ||
1205 		    c->isa_level == MIPS_CPU_ISA_M64R1 ||
1206 		    c->isa_level == MIPS_CPU_ISA_M64R2) {
1207 #ifdef CONFIG_MIPS_CPU_SCACHE
1208 			if (mips_sc_init ()) {
1209 				scache_size = c->scache.ways * c->scache.sets * c->scache.linesz;
1210 				printk("MIPS secondary cache %ldkB, %s, linesize %d bytes.\n",
1211 				       scache_size >> 10,
1212 				       way_string[c->scache.ways], c->scache.linesz);
1213 			}
1214 #else
1215 			if (!(c->scache.flags & MIPS_CACHE_NOT_PRESENT))
1216 				panic("Dunno how to handle MIPS32 / MIPS64 second level cache");
1217 #endif
1218 			return;
1219 		}
1220 		sc_present = 0;
1221 	}
1222 
1223 	if (!sc_present)
1224 		return;
1225 
1226 	/* compute a couple of other cache variables */
1227 	c->scache.waysize = scache_size / c->scache.ways;
1228 
1229 	c->scache.sets = scache_size / (c->scache.linesz * c->scache.ways);
1230 
1231 	printk("Unified secondary cache %ldkB %s, linesize %d bytes.\n",
1232 	       scache_size >> 10, way_string[c->scache.ways], c->scache.linesz);
1233 
1234 	c->options |= MIPS_CPU_INCLUSIVE_CACHES;
1235 }
1236 
1237 void au1x00_fixup_config_od(void)
1238 {
1239 	/*
1240 	 * c0_config.od (bit 19) was write only (and read as 0)
1241 	 * on the early revisions of Alchemy SOCs.  It disables the bus
1242 	 * transaction overlapping and needs to be set to fix various errata.
1243 	 */
1244 	switch (read_c0_prid()) {
1245 	case 0x00030100: /* Au1000 DA */
1246 	case 0x00030201: /* Au1000 HA */
1247 	case 0x00030202: /* Au1000 HB */
1248 	case 0x01030200: /* Au1500 AB */
1249 	/*
1250 	 * Au1100 errata actually keeps silence about this bit, so we set it
1251 	 * just in case for those revisions that require it to be set according
1252 	 * to the (now gone) cpu table.
1253 	 */
1254 	case 0x02030200: /* Au1100 AB */
1255 	case 0x02030201: /* Au1100 BA */
1256 	case 0x02030202: /* Au1100 BC */
1257 		set_c0_config(1 << 19);
1258 		break;
1259 	}
1260 }
1261 
1262 /* CP0 hazard avoidance. */
1263 #define NXP_BARRIER()							\
1264 	 __asm__ __volatile__(						\
1265 	".set noreorder\n\t"						\
1266 	"nop; nop; nop; nop; nop; nop;\n\t"				\
1267 	".set reorder\n\t")
1268 
1269 static void nxp_pr4450_fixup_config(void)
1270 {
1271 	unsigned long config0;
1272 
1273 	config0 = read_c0_config();
1274 
1275 	/* clear all three cache coherency fields */
1276 	config0 &= ~(0x7 | (7 << 25) | (7 << 28));
1277 	config0 |= (((_page_cachable_default >> _CACHE_SHIFT) <<  0) |
1278 		    ((_page_cachable_default >> _CACHE_SHIFT) << 25) |
1279 		    ((_page_cachable_default >> _CACHE_SHIFT) << 28));
1280 	write_c0_config(config0);
1281 	NXP_BARRIER();
1282 }
1283 
1284 static int __cpuinitdata cca = -1;
1285 
1286 static int __init cca_setup(char *str)
1287 {
1288 	get_option(&str, &cca);
1289 
1290 	return 1;
1291 }
1292 
1293 __setup("cca=", cca_setup);
1294 
1295 static void __cpuinit coherency_setup(void)
1296 {
1297 	if (cca < 0 || cca > 7)
1298 		cca = read_c0_config() & CONF_CM_CMASK;
1299 	_page_cachable_default = cca << _CACHE_SHIFT;
1300 
1301 	pr_debug("Using cache attribute %d\n", cca);
1302 	change_c0_config(CONF_CM_CMASK, cca);
1303 
1304 	/*
1305 	 * c0_status.cu=0 specifies that updates by the sc instruction use
1306 	 * the coherency mode specified by the TLB; 1 means cachable
1307 	 * coherent update on write will be used.  Not all processors have
1308 	 * this bit and; some wire it to zero, others like Toshiba had the
1309 	 * silly idea of putting something else there ...
1310 	 */
1311 	switch (current_cpu_type()) {
1312 	case CPU_R4000PC:
1313 	case CPU_R4000SC:
1314 	case CPU_R4000MC:
1315 	case CPU_R4400PC:
1316 	case CPU_R4400SC:
1317 	case CPU_R4400MC:
1318 		clear_c0_config(CONF_CU);
1319 		break;
1320 	/*
1321 	 * We need to catch the early Alchemy SOCs with
1322 	 * the write-only co_config.od bit and set it back to one on:
1323 	 * Au1000 rev DA, HA, HB;  Au1100 AB, BA, BC, Au1500 AB
1324 	 */
1325 	case CPU_ALCHEMY:
1326 		au1x00_fixup_config_od();
1327 		break;
1328 
1329 	case PRID_IMP_PR4450:
1330 		nxp_pr4450_fixup_config();
1331 		break;
1332 	}
1333 }
1334 
1335 #if defined(CONFIG_DMA_NONCOHERENT)
1336 
1337 static int __cpuinitdata coherentio;
1338 
1339 static int __init setcoherentio(char *str)
1340 {
1341 	coherentio = 1;
1342 
1343 	return 1;
1344 }
1345 
1346 __setup("coherentio", setcoherentio);
1347 #endif
1348 
1349 void __cpuinit r4k_cache_init(void)
1350 {
1351 	extern void build_clear_page(void);
1352 	extern void build_copy_page(void);
1353 	extern char __weak except_vec2_generic;
1354 	extern char __weak except_vec2_sb1;
1355 	struct cpuinfo_mips *c = &current_cpu_data;
1356 
1357 	switch (c->cputype) {
1358 	case CPU_SB1:
1359 	case CPU_SB1A:
1360 		set_uncached_handler(0x100, &except_vec2_sb1, 0x80);
1361 		break;
1362 
1363 	default:
1364 		set_uncached_handler(0x100, &except_vec2_generic, 0x80);
1365 		break;
1366 	}
1367 
1368 	probe_pcache();
1369 	setup_scache();
1370 
1371 	r4k_blast_dcache_page_setup();
1372 	r4k_blast_dcache_page_indexed_setup();
1373 	r4k_blast_dcache_setup();
1374 	r4k_blast_icache_page_setup();
1375 	r4k_blast_icache_page_indexed_setup();
1376 	r4k_blast_icache_setup();
1377 	r4k_blast_scache_page_setup();
1378 	r4k_blast_scache_page_indexed_setup();
1379 	r4k_blast_scache_setup();
1380 
1381 	/*
1382 	 * Some MIPS32 and MIPS64 processors have physically indexed caches.
1383 	 * This code supports virtually indexed processors and will be
1384 	 * unnecessarily inefficient on physically indexed processors.
1385 	 */
1386 	if (c->dcache.linesz)
1387 		shm_align_mask = max_t( unsigned long,
1388 					c->dcache.sets * c->dcache.linesz - 1,
1389 					PAGE_SIZE - 1);
1390 	else
1391 		shm_align_mask = PAGE_SIZE-1;
1392 
1393 	__flush_cache_vmap	= r4k__flush_cache_vmap;
1394 	__flush_cache_vunmap	= r4k__flush_cache_vunmap;
1395 
1396 	flush_cache_all		= cache_noop;
1397 	__flush_cache_all	= r4k___flush_cache_all;
1398 	flush_cache_mm		= r4k_flush_cache_mm;
1399 	flush_cache_page	= r4k_flush_cache_page;
1400 	flush_cache_range	= r4k_flush_cache_range;
1401 
1402 	flush_cache_sigtramp	= r4k_flush_cache_sigtramp;
1403 	flush_icache_all	= r4k_flush_icache_all;
1404 	local_flush_data_cache_page	= local_r4k_flush_data_cache_page;
1405 	flush_data_cache_page	= r4k_flush_data_cache_page;
1406 	flush_icache_range	= r4k_flush_icache_range;
1407 	local_flush_icache_range	= local_r4k_flush_icache_range;
1408 
1409 #if defined(CONFIG_DMA_NONCOHERENT)
1410 	if (coherentio) {
1411 		_dma_cache_wback_inv	= (void *)cache_noop;
1412 		_dma_cache_wback	= (void *)cache_noop;
1413 		_dma_cache_inv		= (void *)cache_noop;
1414 	} else {
1415 		_dma_cache_wback_inv	= r4k_dma_cache_wback_inv;
1416 		_dma_cache_wback	= r4k_dma_cache_wback_inv;
1417 		_dma_cache_inv		= r4k_dma_cache_inv;
1418 	}
1419 #endif
1420 
1421 	build_clear_page();
1422 	build_copy_page();
1423 #if !defined(CONFIG_MIPS_CMP)
1424 	local_r4k___flush_cache_all(NULL);
1425 #endif
1426 	coherency_setup();
1427 }
1428