Merge branch 'wip-mips-pm' of https://github.com/paulburton/linux into mips-for-linux...
[pandora-kernel.git] / arch / mips / mm / c-r4k.c
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/cpu_pm.h>
11 #include <linux/hardirq.h>
12 #include <linux/init.h>
13 #include <linux/highmem.h>
14 #include <linux/kernel.h>
15 #include <linux/linkage.h>
16 #include <linux/preempt.h>
17 #include <linux/sched.h>
18 #include <linux/smp.h>
19 #include <linux/mm.h>
20 #include <linux/module.h>
21 #include <linux/bitops.h>
22
23 #include <asm/bcache.h>
24 #include <asm/bootinfo.h>
25 #include <asm/cache.h>
26 #include <asm/cacheops.h>
27 #include <asm/cpu.h>
28 #include <asm/cpu-features.h>
29 #include <asm/cpu-type.h>
30 #include <asm/io.h>
31 #include <asm/page.h>
32 #include <asm/pgtable.h>
33 #include <asm/r4kcache.h>
34 #include <asm/sections.h>
35 #include <asm/mmu_context.h>
36 #include <asm/war.h>
37 #include <asm/cacheflush.h> /* for run_uncached() */
38 #include <asm/traps.h>
39 #include <asm/dma-coherence.h>
40
41 /*
42  * Special Variant of smp_call_function for use by cache functions:
43  *
44  *  o No return value
45  *  o collapses to normal function call on UP kernels
46  *  o collapses to normal function call on systems with a single shared
47  *    primary cache.
48  *  o doesn't disable interrupts on the local CPU
49  */
50 static inline void r4k_on_each_cpu(void (*func) (void *info), void *info)
51 {
52         preempt_disable();
53
54 #ifndef CONFIG_MIPS_MT_SMP
55         smp_call_function(func, info, 1);
56 #endif
57         func(info);
58         preempt_enable();
59 }
60
61 #if defined(CONFIG_MIPS_CMP) || defined(CONFIG_MIPS_CPS)
62 #define cpu_has_safe_index_cacheops 0
63 #else
64 #define cpu_has_safe_index_cacheops 1
65 #endif
66
67 /*
68  * Must die.
69  */
70 static unsigned long icache_size __read_mostly;
71 static unsigned long dcache_size __read_mostly;
72 static unsigned long scache_size __read_mostly;
73
74 /*
75  * Dummy cache handling routines for machines without boardcaches
76  */
77 static void cache_noop(void) {}
78
79 static struct bcache_ops no_sc_ops = {
80         .bc_enable = (void *)cache_noop,
81         .bc_disable = (void *)cache_noop,
82         .bc_wback_inv = (void *)cache_noop,
83         .bc_inv = (void *)cache_noop
84 };
85
86 struct bcache_ops *bcops = &no_sc_ops;
87
88 #define cpu_is_r4600_v1_x()     ((read_c0_prid() & 0xfffffff0) == 0x00002010)
89 #define cpu_is_r4600_v2_x()     ((read_c0_prid() & 0xfffffff0) == 0x00002020)
90
91 #define R4600_HIT_CACHEOP_WAR_IMPL                                      \
92 do {                                                                    \
93         if (R4600_V2_HIT_CACHEOP_WAR && cpu_is_r4600_v2_x())            \
94                 *(volatile unsigned long *)CKSEG1;                      \
95         if (R4600_V1_HIT_CACHEOP_WAR)                                   \
96                 __asm__ __volatile__("nop;nop;nop;nop");                \
97 } while (0)
98
99 static void (*r4k_blast_dcache_page)(unsigned long addr);
100
101 static inline void r4k_blast_dcache_page_dc32(unsigned long addr)
102 {
103         R4600_HIT_CACHEOP_WAR_IMPL;
104         blast_dcache32_page(addr);
105 }
106
107 static inline void r4k_blast_dcache_page_dc64(unsigned long addr)
108 {
109         blast_dcache64_page(addr);
110 }
111
112 static void r4k_blast_dcache_page_setup(void)
113 {
114         unsigned long  dc_lsize = cpu_dcache_line_size();
115
116         if (dc_lsize == 0)
117                 r4k_blast_dcache_page = (void *)cache_noop;
118         else if (dc_lsize == 16)
119                 r4k_blast_dcache_page = blast_dcache16_page;
120         else if (dc_lsize == 32)
121                 r4k_blast_dcache_page = r4k_blast_dcache_page_dc32;
122         else if (dc_lsize == 64)
123                 r4k_blast_dcache_page = r4k_blast_dcache_page_dc64;
124 }
125
126 #ifndef CONFIG_EVA
127 #define r4k_blast_dcache_user_page  r4k_blast_dcache_page
128 #else
129
130 static void (*r4k_blast_dcache_user_page)(unsigned long addr);
131
132 static void r4k_blast_dcache_user_page_setup(void)
133 {
134         unsigned long  dc_lsize = cpu_dcache_line_size();
135
136         if (dc_lsize == 0)
137                 r4k_blast_dcache_user_page = (void *)cache_noop;
138         else if (dc_lsize == 16)
139                 r4k_blast_dcache_user_page = blast_dcache16_user_page;
140         else if (dc_lsize == 32)
141                 r4k_blast_dcache_user_page = blast_dcache32_user_page;
142         else if (dc_lsize == 64)
143                 r4k_blast_dcache_user_page = blast_dcache64_user_page;
144 }
145
146 #endif
147
148 static void (* r4k_blast_dcache_page_indexed)(unsigned long addr);
149
150 static void r4k_blast_dcache_page_indexed_setup(void)
151 {
152         unsigned long dc_lsize = cpu_dcache_line_size();
153
154         if (dc_lsize == 0)
155                 r4k_blast_dcache_page_indexed = (void *)cache_noop;
156         else if (dc_lsize == 16)
157                 r4k_blast_dcache_page_indexed = blast_dcache16_page_indexed;
158         else if (dc_lsize == 32)
159                 r4k_blast_dcache_page_indexed = blast_dcache32_page_indexed;
160         else if (dc_lsize == 64)
161                 r4k_blast_dcache_page_indexed = blast_dcache64_page_indexed;
162 }
163
164 void (* r4k_blast_dcache)(void);
165 EXPORT_SYMBOL(r4k_blast_dcache);
166
167 static void r4k_blast_dcache_setup(void)
168 {
169         unsigned long dc_lsize = cpu_dcache_line_size();
170
171         if (dc_lsize == 0)
172                 r4k_blast_dcache = (void *)cache_noop;
173         else if (dc_lsize == 16)
174                 r4k_blast_dcache = blast_dcache16;
175         else if (dc_lsize == 32)
176                 r4k_blast_dcache = blast_dcache32;
177         else if (dc_lsize == 64)
178                 r4k_blast_dcache = blast_dcache64;
179 }
180
181 /* force code alignment (used for TX49XX_ICACHE_INDEX_INV_WAR) */
182 #define JUMP_TO_ALIGN(order) \
183         __asm__ __volatile__( \
184                 "b\t1f\n\t" \
185                 ".align\t" #order "\n\t" \
186                 "1:\n\t" \
187                 )
188 #define CACHE32_UNROLL32_ALIGN  JUMP_TO_ALIGN(10) /* 32 * 32 = 1024 */
189 #define CACHE32_UNROLL32_ALIGN2 JUMP_TO_ALIGN(11)
190
191 static inline void blast_r4600_v1_icache32(void)
192 {
193         unsigned long flags;
194
195         local_irq_save(flags);
196         blast_icache32();
197         local_irq_restore(flags);
198 }
199
200 static inline void tx49_blast_icache32(void)
201 {
202         unsigned long start = INDEX_BASE;
203         unsigned long end = start + current_cpu_data.icache.waysize;
204         unsigned long ws_inc = 1UL << current_cpu_data.icache.waybit;
205         unsigned long ws_end = current_cpu_data.icache.ways <<
206                                current_cpu_data.icache.waybit;
207         unsigned long ws, addr;
208
209         CACHE32_UNROLL32_ALIGN2;
210         /* I'm in even chunk.  blast odd chunks */
211         for (ws = 0; ws < ws_end; ws += ws_inc)
212                 for (addr = start + 0x400; addr < end; addr += 0x400 * 2)
213                         cache32_unroll32(addr|ws, Index_Invalidate_I);
214         CACHE32_UNROLL32_ALIGN;
215         /* I'm in odd chunk.  blast even chunks */
216         for (ws = 0; ws < ws_end; ws += ws_inc)
217                 for (addr = start; addr < end; addr += 0x400 * 2)
218                         cache32_unroll32(addr|ws, Index_Invalidate_I);
219 }
220
221 static inline void blast_icache32_r4600_v1_page_indexed(unsigned long page)
222 {
223         unsigned long flags;
224
225         local_irq_save(flags);
226         blast_icache32_page_indexed(page);
227         local_irq_restore(flags);
228 }
229
230 static inline void tx49_blast_icache32_page_indexed(unsigned long page)
231 {
232         unsigned long indexmask = current_cpu_data.icache.waysize - 1;
233         unsigned long start = INDEX_BASE + (page & indexmask);
234         unsigned long end = start + PAGE_SIZE;
235         unsigned long ws_inc = 1UL << current_cpu_data.icache.waybit;
236         unsigned long ws_end = current_cpu_data.icache.ways <<
237                                current_cpu_data.icache.waybit;
238         unsigned long ws, addr;
239
240         CACHE32_UNROLL32_ALIGN2;
241         /* I'm in even chunk.  blast odd chunks */
242         for (ws = 0; ws < ws_end; ws += ws_inc)
243                 for (addr = start + 0x400; addr < end; addr += 0x400 * 2)
244                         cache32_unroll32(addr|ws, Index_Invalidate_I);
245         CACHE32_UNROLL32_ALIGN;
246         /* I'm in odd chunk.  blast even chunks */
247         for (ws = 0; ws < ws_end; ws += ws_inc)
248                 for (addr = start; addr < end; addr += 0x400 * 2)
249                         cache32_unroll32(addr|ws, Index_Invalidate_I);
250 }
251
252 static void (* r4k_blast_icache_page)(unsigned long addr);
253
254 static void r4k_blast_icache_page_setup(void)
255 {
256         unsigned long ic_lsize = cpu_icache_line_size();
257
258         if (ic_lsize == 0)
259                 r4k_blast_icache_page = (void *)cache_noop;
260         else if (ic_lsize == 16)
261                 r4k_blast_icache_page = blast_icache16_page;
262         else if (ic_lsize == 32 && current_cpu_type() == CPU_LOONGSON2)
263                 r4k_blast_icache_page = loongson2_blast_icache32_page;
264         else if (ic_lsize == 32)
265                 r4k_blast_icache_page = blast_icache32_page;
266         else if (ic_lsize == 64)
267                 r4k_blast_icache_page = blast_icache64_page;
268 }
269
270 #ifndef CONFIG_EVA
271 #define r4k_blast_icache_user_page  r4k_blast_icache_page
272 #else
273
274 static void (*r4k_blast_icache_user_page)(unsigned long addr);
275
276 static void __cpuinit r4k_blast_icache_user_page_setup(void)
277 {
278         unsigned long ic_lsize = cpu_icache_line_size();
279
280         if (ic_lsize == 0)
281                 r4k_blast_icache_user_page = (void *)cache_noop;
282         else if (ic_lsize == 16)
283                 r4k_blast_icache_user_page = blast_icache16_user_page;
284         else if (ic_lsize == 32)
285                 r4k_blast_icache_user_page = blast_icache32_user_page;
286         else if (ic_lsize == 64)
287                 r4k_blast_icache_user_page = blast_icache64_user_page;
288 }
289
290 #endif
291
292 static void (* r4k_blast_icache_page_indexed)(unsigned long addr);
293
294 static void r4k_blast_icache_page_indexed_setup(void)
295 {
296         unsigned long ic_lsize = cpu_icache_line_size();
297
298         if (ic_lsize == 0)
299                 r4k_blast_icache_page_indexed = (void *)cache_noop;
300         else if (ic_lsize == 16)
301                 r4k_blast_icache_page_indexed = blast_icache16_page_indexed;
302         else if (ic_lsize == 32) {
303                 if (R4600_V1_INDEX_ICACHEOP_WAR && cpu_is_r4600_v1_x())
304                         r4k_blast_icache_page_indexed =
305                                 blast_icache32_r4600_v1_page_indexed;
306                 else if (TX49XX_ICACHE_INDEX_INV_WAR)
307                         r4k_blast_icache_page_indexed =
308                                 tx49_blast_icache32_page_indexed;
309                 else if (current_cpu_type() == CPU_LOONGSON2)
310                         r4k_blast_icache_page_indexed =
311                                 loongson2_blast_icache32_page_indexed;
312                 else
313                         r4k_blast_icache_page_indexed =
314                                 blast_icache32_page_indexed;
315         } else if (ic_lsize == 64)
316                 r4k_blast_icache_page_indexed = blast_icache64_page_indexed;
317 }
318
319 void (* r4k_blast_icache)(void);
320 EXPORT_SYMBOL(r4k_blast_icache);
321
322 static void r4k_blast_icache_setup(void)
323 {
324         unsigned long ic_lsize = cpu_icache_line_size();
325
326         if (ic_lsize == 0)
327                 r4k_blast_icache = (void *)cache_noop;
328         else if (ic_lsize == 16)
329                 r4k_blast_icache = blast_icache16;
330         else if (ic_lsize == 32) {
331                 if (R4600_V1_INDEX_ICACHEOP_WAR && cpu_is_r4600_v1_x())
332                         r4k_blast_icache = blast_r4600_v1_icache32;
333                 else if (TX49XX_ICACHE_INDEX_INV_WAR)
334                         r4k_blast_icache = tx49_blast_icache32;
335                 else if (current_cpu_type() == CPU_LOONGSON2)
336                         r4k_blast_icache = loongson2_blast_icache32;
337                 else
338                         r4k_blast_icache = blast_icache32;
339         } else if (ic_lsize == 64)
340                 r4k_blast_icache = blast_icache64;
341 }
342
343 static void (* r4k_blast_scache_page)(unsigned long addr);
344
345 static void r4k_blast_scache_page_setup(void)
346 {
347         unsigned long sc_lsize = cpu_scache_line_size();
348
349         if (scache_size == 0)
350                 r4k_blast_scache_page = (void *)cache_noop;
351         else if (sc_lsize == 16)
352                 r4k_blast_scache_page = blast_scache16_page;
353         else if (sc_lsize == 32)
354                 r4k_blast_scache_page = blast_scache32_page;
355         else if (sc_lsize == 64)
356                 r4k_blast_scache_page = blast_scache64_page;
357         else if (sc_lsize == 128)
358                 r4k_blast_scache_page = blast_scache128_page;
359 }
360
361 static void (* r4k_blast_scache_page_indexed)(unsigned long addr);
362
363 static void r4k_blast_scache_page_indexed_setup(void)
364 {
365         unsigned long sc_lsize = cpu_scache_line_size();
366
367         if (scache_size == 0)
368                 r4k_blast_scache_page_indexed = (void *)cache_noop;
369         else if (sc_lsize == 16)
370                 r4k_blast_scache_page_indexed = blast_scache16_page_indexed;
371         else if (sc_lsize == 32)
372                 r4k_blast_scache_page_indexed = blast_scache32_page_indexed;
373         else if (sc_lsize == 64)
374                 r4k_blast_scache_page_indexed = blast_scache64_page_indexed;
375         else if (sc_lsize == 128)
376                 r4k_blast_scache_page_indexed = blast_scache128_page_indexed;
377 }
378
379 static void (* r4k_blast_scache)(void);
380
381 static void r4k_blast_scache_setup(void)
382 {
383         unsigned long sc_lsize = cpu_scache_line_size();
384
385         if (scache_size == 0)
386                 r4k_blast_scache = (void *)cache_noop;
387         else if (sc_lsize == 16)
388                 r4k_blast_scache = blast_scache16;
389         else if (sc_lsize == 32)
390                 r4k_blast_scache = blast_scache32;
391         else if (sc_lsize == 64)
392                 r4k_blast_scache = blast_scache64;
393         else if (sc_lsize == 128)
394                 r4k_blast_scache = blast_scache128;
395 }
396
397 static inline void local_r4k___flush_cache_all(void * args)
398 {
399         switch (current_cpu_type()) {
400         case CPU_LOONGSON2:
401         case CPU_LOONGSON3:
402         case CPU_R4000SC:
403         case CPU_R4000MC:
404         case CPU_R4400SC:
405         case CPU_R4400MC:
406         case CPU_R10000:
407         case CPU_R12000:
408         case CPU_R14000:
409                 /*
410                  * These caches are inclusive caches, that is, if something
411                  * is not cached in the S-cache, we know it also won't be
412                  * in one of the primary caches.
413                  */
414                 r4k_blast_scache();
415                 break;
416
417         default:
418                 r4k_blast_dcache();
419                 r4k_blast_icache();
420                 break;
421         }
422 }
423
424 static void r4k___flush_cache_all(void)
425 {
426         r4k_on_each_cpu(local_r4k___flush_cache_all, NULL);
427 }
428
429 static inline int has_valid_asid(const struct mm_struct *mm)
430 {
431 #ifdef CONFIG_MIPS_MT_SMP
432         int i;
433
434         for_each_online_cpu(i)
435                 if (cpu_context(i, mm))
436                         return 1;
437
438         return 0;
439 #else
440         return cpu_context(smp_processor_id(), mm);
441 #endif
442 }
443
444 static void r4k__flush_cache_vmap(void)
445 {
446         r4k_blast_dcache();
447 }
448
449 static void r4k__flush_cache_vunmap(void)
450 {
451         r4k_blast_dcache();
452 }
453
454 static inline void local_r4k_flush_cache_range(void * args)
455 {
456         struct vm_area_struct *vma = args;
457         int exec = vma->vm_flags & VM_EXEC;
458
459         if (!(has_valid_asid(vma->vm_mm)))
460                 return;
461
462         r4k_blast_dcache();
463         if (exec)
464                 r4k_blast_icache();
465 }
466
467 static void r4k_flush_cache_range(struct vm_area_struct *vma,
468         unsigned long start, unsigned long end)
469 {
470         int exec = vma->vm_flags & VM_EXEC;
471
472         if (cpu_has_dc_aliases || (exec && !cpu_has_ic_fills_f_dc))
473                 r4k_on_each_cpu(local_r4k_flush_cache_range, vma);
474 }
475
476 static inline void local_r4k_flush_cache_mm(void * args)
477 {
478         struct mm_struct *mm = args;
479
480         if (!has_valid_asid(mm))
481                 return;
482
483         /*
484          * Kludge alert.  For obscure reasons R4000SC and R4400SC go nuts if we
485          * only flush the primary caches but R10000 and R12000 behave sane ...
486          * R4000SC and R4400SC indexed S-cache ops also invalidate primary
487          * caches, so we can bail out early.
488          */
489         if (current_cpu_type() == CPU_R4000SC ||
490             current_cpu_type() == CPU_R4000MC ||
491             current_cpu_type() == CPU_R4400SC ||
492             current_cpu_type() == CPU_R4400MC) {
493                 r4k_blast_scache();
494                 return;
495         }
496
497         r4k_blast_dcache();
498 }
499
500 static void r4k_flush_cache_mm(struct mm_struct *mm)
501 {
502         if (!cpu_has_dc_aliases)
503                 return;
504
505         r4k_on_each_cpu(local_r4k_flush_cache_mm, mm);
506 }
507
508 struct flush_cache_page_args {
509         struct vm_area_struct *vma;
510         unsigned long addr;
511         unsigned long pfn;
512 };
513
514 static inline void local_r4k_flush_cache_page(void *args)
515 {
516         struct flush_cache_page_args *fcp_args = args;
517         struct vm_area_struct *vma = fcp_args->vma;
518         unsigned long addr = fcp_args->addr;
519         struct page *page = pfn_to_page(fcp_args->pfn);
520         int exec = vma->vm_flags & VM_EXEC;
521         struct mm_struct *mm = vma->vm_mm;
522         int map_coherent = 0;
523         pgd_t *pgdp;
524         pud_t *pudp;
525         pmd_t *pmdp;
526         pte_t *ptep;
527         void *vaddr;
528
529         /*
530          * If ownes no valid ASID yet, cannot possibly have gotten
531          * this page into the cache.
532          */
533         if (!has_valid_asid(mm))
534                 return;
535
536         addr &= PAGE_MASK;
537         pgdp = pgd_offset(mm, addr);
538         pudp = pud_offset(pgdp, addr);
539         pmdp = pmd_offset(pudp, addr);
540         ptep = pte_offset(pmdp, addr);
541
542         /*
543          * If the page isn't marked valid, the page cannot possibly be
544          * in the cache.
545          */
546         if (!(pte_present(*ptep)))
547                 return;
548
549         if ((mm == current->active_mm) && (pte_val(*ptep) & _PAGE_VALID))
550                 vaddr = NULL;
551         else {
552                 /*
553                  * Use kmap_coherent or kmap_atomic to do flushes for
554                  * another ASID than the current one.
555                  */
556                 map_coherent = (cpu_has_dc_aliases &&
557                                 page_mapped(page) && !Page_dcache_dirty(page));
558                 if (map_coherent)
559                         vaddr = kmap_coherent(page, addr);
560                 else
561                         vaddr = kmap_atomic(page);
562                 addr = (unsigned long)vaddr;
563         }
564
565         if (cpu_has_dc_aliases || (exec && !cpu_has_ic_fills_f_dc)) {
566                 vaddr ? r4k_blast_dcache_page(addr) :
567                         r4k_blast_dcache_user_page(addr);
568                 if (exec && !cpu_icache_snoops_remote_store)
569                         r4k_blast_scache_page(addr);
570         }
571         if (exec) {
572                 if (vaddr && cpu_has_vtag_icache && mm == current->active_mm) {
573                         int cpu = smp_processor_id();
574
575                         if (cpu_context(cpu, mm) != 0)
576                                 drop_mmu_context(mm, cpu);
577                 } else
578                         vaddr ? r4k_blast_icache_page(addr) :
579                                 r4k_blast_icache_user_page(addr);
580         }
581
582         if (vaddr) {
583                 if (map_coherent)
584                         kunmap_coherent();
585                 else
586                         kunmap_atomic(vaddr);
587         }
588 }
589
590 static void r4k_flush_cache_page(struct vm_area_struct *vma,
591         unsigned long addr, unsigned long pfn)
592 {
593         struct flush_cache_page_args args;
594
595         args.vma = vma;
596         args.addr = addr;
597         args.pfn = pfn;
598
599         r4k_on_each_cpu(local_r4k_flush_cache_page, &args);
600 }
601
602 static inline void local_r4k_flush_data_cache_page(void * addr)
603 {
604         r4k_blast_dcache_page((unsigned long) addr);
605 }
606
607 static void r4k_flush_data_cache_page(unsigned long addr)
608 {
609         if (in_atomic())
610                 local_r4k_flush_data_cache_page((void *)addr);
611         else
612                 r4k_on_each_cpu(local_r4k_flush_data_cache_page, (void *) addr);
613 }
614
615 struct flush_icache_range_args {
616         unsigned long start;
617         unsigned long end;
618 };
619
620 static inline void local_r4k_flush_icache_range(unsigned long start, unsigned long end)
621 {
622         if (!cpu_has_ic_fills_f_dc) {
623                 if (end - start >= dcache_size) {
624                         r4k_blast_dcache();
625                 } else {
626                         R4600_HIT_CACHEOP_WAR_IMPL;
627                         protected_blast_dcache_range(start, end);
628                 }
629         }
630
631         if (end - start > icache_size)
632                 r4k_blast_icache();
633         else {
634                 switch (boot_cpu_type()) {
635                 case CPU_LOONGSON2:
636                         protected_loongson2_blast_icache_range(start, end);
637                         break;
638
639                 default:
640                         protected_blast_icache_range(start, end);
641                         break;
642                 }
643         }
644 #ifdef CONFIG_EVA
645         /*
646          * Due to all possible segment mappings, there might cache aliases
647          * caused by the bootloader being in non-EVA mode, and the CPU switching
648          * to EVA during early kernel init. It's best to flush the scache
649          * to avoid having secondary cores fetching stale data and lead to
650          * kernel crashes.
651          */
652         bc_wback_inv(start, (end - start));
653         __sync();
654 #endif
655 }
656
657 static inline void local_r4k_flush_icache_range_ipi(void *args)
658 {
659         struct flush_icache_range_args *fir_args = args;
660         unsigned long start = fir_args->start;
661         unsigned long end = fir_args->end;
662
663         local_r4k_flush_icache_range(start, end);
664 }
665
666 static void r4k_flush_icache_range(unsigned long start, unsigned long end)
667 {
668         struct flush_icache_range_args args;
669
670         args.start = start;
671         args.end = end;
672
673         r4k_on_each_cpu(local_r4k_flush_icache_range_ipi, &args);
674         instruction_hazard();
675 }
676
677 #if defined(CONFIG_DMA_NONCOHERENT) || defined(CONFIG_DMA_MAYBE_COHERENT)
678
679 static void r4k_dma_cache_wback_inv(unsigned long addr, unsigned long size)
680 {
681         /* Catch bad driver code */
682         BUG_ON(size == 0);
683
684         preempt_disable();
685         if (cpu_has_inclusive_pcaches) {
686                 if (size >= scache_size)
687                         r4k_blast_scache();
688                 else
689                         blast_scache_range(addr, addr + size);
690                 preempt_enable();
691                 __sync();
692                 return;
693         }
694
695         /*
696          * Either no secondary cache or the available caches don't have the
697          * subset property so we have to flush the primary caches
698          * explicitly
699          */
700         if (cpu_has_safe_index_cacheops && size >= dcache_size) {
701                 r4k_blast_dcache();
702         } else {
703                 R4600_HIT_CACHEOP_WAR_IMPL;
704                 blast_dcache_range(addr, addr + size);
705         }
706         preempt_enable();
707
708         bc_wback_inv(addr, size);
709         __sync();
710 }
711
712 static void r4k_dma_cache_inv(unsigned long addr, unsigned long size)
713 {
714         /* Catch bad driver code */
715         BUG_ON(size == 0);
716
717         preempt_disable();
718         if (cpu_has_inclusive_pcaches) {
719                 if (size >= scache_size)
720                         r4k_blast_scache();
721                 else {
722                         /*
723                          * There is no clearly documented alignment requirement
724                          * for the cache instruction on MIPS processors and
725                          * some processors, among them the RM5200 and RM7000
726                          * QED processors will throw an address error for cache
727                          * hit ops with insufficient alignment.  Solved by
728                          * aligning the address to cache line size.
729                          */
730                         blast_inv_scache_range(addr, addr + size);
731                 }
732                 preempt_enable();
733                 __sync();
734                 return;
735         }
736
737         if (cpu_has_safe_index_cacheops && size >= dcache_size) {
738                 r4k_blast_dcache();
739         } else {
740                 R4600_HIT_CACHEOP_WAR_IMPL;
741                 blast_inv_dcache_range(addr, addr + size);
742         }
743         preempt_enable();
744
745         bc_inv(addr, size);
746         __sync();
747 }
748 #endif /* CONFIG_DMA_NONCOHERENT || CONFIG_DMA_MAYBE_COHERENT */
749
750 /*
751  * While we're protected against bad userland addresses we don't care
752  * very much about what happens in that case.  Usually a segmentation
753  * fault will dump the process later on anyway ...
754  */
755 static void local_r4k_flush_cache_sigtramp(void * arg)
756 {
757         unsigned long ic_lsize = cpu_icache_line_size();
758         unsigned long dc_lsize = cpu_dcache_line_size();
759         unsigned long sc_lsize = cpu_scache_line_size();
760         unsigned long addr = (unsigned long) arg;
761
762         R4600_HIT_CACHEOP_WAR_IMPL;
763         if (dc_lsize)
764                 protected_writeback_dcache_line(addr & ~(dc_lsize - 1));
765         if (!cpu_icache_snoops_remote_store && scache_size)
766                 protected_writeback_scache_line(addr & ~(sc_lsize - 1));
767         if (ic_lsize)
768                 protected_flush_icache_line(addr & ~(ic_lsize - 1));
769         if (MIPS4K_ICACHE_REFILL_WAR) {
770                 __asm__ __volatile__ (
771                         ".set push\n\t"
772                         ".set noat\n\t"
773                         ".set mips3\n\t"
774 #ifdef CONFIG_32BIT
775                         "la     $at,1f\n\t"
776 #endif
777 #ifdef CONFIG_64BIT
778                         "dla    $at,1f\n\t"
779 #endif
780                         "cache  %0,($at)\n\t"
781                         "nop; nop; nop\n"
782                         "1:\n\t"
783                         ".set pop"
784                         :
785                         : "i" (Hit_Invalidate_I));
786         }
787         if (MIPS_CACHE_SYNC_WAR)
788                 __asm__ __volatile__ ("sync");
789 }
790
791 static void r4k_flush_cache_sigtramp(unsigned long addr)
792 {
793         r4k_on_each_cpu(local_r4k_flush_cache_sigtramp, (void *) addr);
794 }
795
796 static void r4k_flush_icache_all(void)
797 {
798         if (cpu_has_vtag_icache)
799                 r4k_blast_icache();
800 }
801
802 struct flush_kernel_vmap_range_args {
803         unsigned long   vaddr;
804         int             size;
805 };
806
807 static inline void local_r4k_flush_kernel_vmap_range(void *args)
808 {
809         struct flush_kernel_vmap_range_args *vmra = args;
810         unsigned long vaddr = vmra->vaddr;
811         int size = vmra->size;
812
813         /*
814          * Aliases only affect the primary caches so don't bother with
815          * S-caches or T-caches.
816          */
817         if (cpu_has_safe_index_cacheops && size >= dcache_size)
818                 r4k_blast_dcache();
819         else {
820                 R4600_HIT_CACHEOP_WAR_IMPL;
821                 blast_dcache_range(vaddr, vaddr + size);
822         }
823 }
824
825 static void r4k_flush_kernel_vmap_range(unsigned long vaddr, int size)
826 {
827         struct flush_kernel_vmap_range_args args;
828
829         args.vaddr = (unsigned long) vaddr;
830         args.size = size;
831
832         r4k_on_each_cpu(local_r4k_flush_kernel_vmap_range, &args);
833 }
834
835 static inline void rm7k_erratum31(void)
836 {
837         const unsigned long ic_lsize = 32;
838         unsigned long addr;
839
840         /* RM7000 erratum #31. The icache is screwed at startup. */
841         write_c0_taglo(0);
842         write_c0_taghi(0);
843
844         for (addr = INDEX_BASE; addr <= INDEX_BASE + 4096; addr += ic_lsize) {
845                 __asm__ __volatile__ (
846                         ".set push\n\t"
847                         ".set noreorder\n\t"
848                         ".set mips3\n\t"
849                         "cache\t%1, 0(%0)\n\t"
850                         "cache\t%1, 0x1000(%0)\n\t"
851                         "cache\t%1, 0x2000(%0)\n\t"
852                         "cache\t%1, 0x3000(%0)\n\t"
853                         "cache\t%2, 0(%0)\n\t"
854                         "cache\t%2, 0x1000(%0)\n\t"
855                         "cache\t%2, 0x2000(%0)\n\t"
856                         "cache\t%2, 0x3000(%0)\n\t"
857                         "cache\t%1, 0(%0)\n\t"
858                         "cache\t%1, 0x1000(%0)\n\t"
859                         "cache\t%1, 0x2000(%0)\n\t"
860                         "cache\t%1, 0x3000(%0)\n\t"
861                         ".set pop\n"
862                         :
863                         : "r" (addr), "i" (Index_Store_Tag_I), "i" (Fill));
864         }
865 }
866
867 static inline void alias_74k_erratum(struct cpuinfo_mips *c)
868 {
869         unsigned int imp = c->processor_id & PRID_IMP_MASK;
870         unsigned int rev = c->processor_id & PRID_REV_MASK;
871
872         /*
873          * Early versions of the 74K do not update the cache tags on a
874          * vtag miss/ptag hit which can occur in the case of KSEG0/KUSEG
875          * aliases. In this case it is better to treat the cache as always
876          * having aliases.
877          */
878         switch (imp) {
879         case PRID_IMP_74K:
880                 if (rev <= PRID_REV_ENCODE_332(2, 4, 0))
881                         c->dcache.flags |= MIPS_CACHE_VTAG;
882                 if (rev == PRID_REV_ENCODE_332(2, 4, 0))
883                         write_c0_config6(read_c0_config6() | MIPS_CONF6_SYND);
884                 break;
885         case PRID_IMP_1074K:
886                 if (rev <= PRID_REV_ENCODE_332(1, 1, 0)) {
887                         c->dcache.flags |= MIPS_CACHE_VTAG;
888                         write_c0_config6(read_c0_config6() | MIPS_CONF6_SYND);
889                 }
890                 break;
891         default:
892                 BUG();
893         }
894 }
895
896 static char *way_string[] = { NULL, "direct mapped", "2-way",
897         "3-way", "4-way", "5-way", "6-way", "7-way", "8-way"
898 };
899
900 static void probe_pcache(void)
901 {
902         struct cpuinfo_mips *c = &current_cpu_data;
903         unsigned int config = read_c0_config();
904         unsigned int prid = read_c0_prid();
905         unsigned long config1;
906         unsigned int lsize;
907
908         switch (current_cpu_type()) {
909         case CPU_R4600:                 /* QED style two way caches? */
910         case CPU_R4700:
911         case CPU_R5000:
912         case CPU_NEVADA:
913                 icache_size = 1 << (12 + ((config & CONF_IC) >> 9));
914                 c->icache.linesz = 16 << ((config & CONF_IB) >> 5);
915                 c->icache.ways = 2;
916                 c->icache.waybit = __ffs(icache_size/2);
917
918                 dcache_size = 1 << (12 + ((config & CONF_DC) >> 6));
919                 c->dcache.linesz = 16 << ((config & CONF_DB) >> 4);
920                 c->dcache.ways = 2;
921                 c->dcache.waybit= __ffs(dcache_size/2);
922
923                 c->options |= MIPS_CPU_CACHE_CDEX_P;
924                 break;
925
926         case CPU_R5432:
927         case CPU_R5500:
928                 icache_size = 1 << (12 + ((config & CONF_IC) >> 9));
929                 c->icache.linesz = 16 << ((config & CONF_IB) >> 5);
930                 c->icache.ways = 2;
931                 c->icache.waybit= 0;
932
933                 dcache_size = 1 << (12 + ((config & CONF_DC) >> 6));
934                 c->dcache.linesz = 16 << ((config & CONF_DB) >> 4);
935                 c->dcache.ways = 2;
936                 c->dcache.waybit = 0;
937
938                 c->options |= MIPS_CPU_CACHE_CDEX_P | MIPS_CPU_PREFETCH;
939                 break;
940
941         case CPU_TX49XX:
942                 icache_size = 1 << (12 + ((config & CONF_IC) >> 9));
943                 c->icache.linesz = 16 << ((config & CONF_IB) >> 5);
944                 c->icache.ways = 4;
945                 c->icache.waybit= 0;
946
947                 dcache_size = 1 << (12 + ((config & CONF_DC) >> 6));
948                 c->dcache.linesz = 16 << ((config & CONF_DB) >> 4);
949                 c->dcache.ways = 4;
950                 c->dcache.waybit = 0;
951
952                 c->options |= MIPS_CPU_CACHE_CDEX_P;
953                 c->options |= MIPS_CPU_PREFETCH;
954                 break;
955
956         case CPU_R4000PC:
957         case CPU_R4000SC:
958         case CPU_R4000MC:
959         case CPU_R4400PC:
960         case CPU_R4400SC:
961         case CPU_R4400MC:
962         case CPU_R4300:
963                 icache_size = 1 << (12 + ((config & CONF_IC) >> 9));
964                 c->icache.linesz = 16 << ((config & CONF_IB) >> 5);
965                 c->icache.ways = 1;
966                 c->icache.waybit = 0;   /* doesn't matter */
967
968                 dcache_size = 1 << (12 + ((config & CONF_DC) >> 6));
969                 c->dcache.linesz = 16 << ((config & CONF_DB) >> 4);
970                 c->dcache.ways = 1;
971                 c->dcache.waybit = 0;   /* does not matter */
972
973                 c->options |= MIPS_CPU_CACHE_CDEX_P;
974                 break;
975
976         case CPU_R10000:
977         case CPU_R12000:
978         case CPU_R14000:
979                 icache_size = 1 << (12 + ((config & R10K_CONF_IC) >> 29));
980                 c->icache.linesz = 64;
981                 c->icache.ways = 2;
982                 c->icache.waybit = 0;
983
984                 dcache_size = 1 << (12 + ((config & R10K_CONF_DC) >> 26));
985                 c->dcache.linesz = 32;
986                 c->dcache.ways = 2;
987                 c->dcache.waybit = 0;
988
989                 c->options |= MIPS_CPU_PREFETCH;
990                 break;
991
992         case CPU_VR4133:
993                 write_c0_config(config & ~VR41_CONF_P4K);
994         case CPU_VR4131:
995                 /* Workaround for cache instruction bug of VR4131 */
996                 if (c->processor_id == 0x0c80U || c->processor_id == 0x0c81U ||
997                     c->processor_id == 0x0c82U) {
998                         config |= 0x00400000U;
999                         if (c->processor_id == 0x0c80U)
1000                                 config |= VR41_CONF_BP;
1001                         write_c0_config(config);
1002                 } else
1003                         c->options |= MIPS_CPU_CACHE_CDEX_P;
1004
1005                 icache_size = 1 << (10 + ((config & CONF_IC) >> 9));
1006                 c->icache.linesz = 16 << ((config & CONF_IB) >> 5);
1007                 c->icache.ways = 2;
1008                 c->icache.waybit = __ffs(icache_size/2);
1009
1010                 dcache_size = 1 << (10 + ((config & CONF_DC) >> 6));
1011                 c->dcache.linesz = 16 << ((config & CONF_DB) >> 4);
1012                 c->dcache.ways = 2;
1013                 c->dcache.waybit = __ffs(dcache_size/2);
1014                 break;
1015
1016         case CPU_VR41XX:
1017         case CPU_VR4111:
1018         case CPU_VR4121:
1019         case CPU_VR4122:
1020         case CPU_VR4181:
1021         case CPU_VR4181A:
1022                 icache_size = 1 << (10 + ((config & CONF_IC) >> 9));
1023                 c->icache.linesz = 16 << ((config & CONF_IB) >> 5);
1024                 c->icache.ways = 1;
1025                 c->icache.waybit = 0;   /* doesn't matter */
1026
1027                 dcache_size = 1 << (10 + ((config & CONF_DC) >> 6));
1028                 c->dcache.linesz = 16 << ((config & CONF_DB) >> 4);
1029                 c->dcache.ways = 1;
1030                 c->dcache.waybit = 0;   /* does not matter */
1031
1032                 c->options |= MIPS_CPU_CACHE_CDEX_P;
1033                 break;
1034
1035         case CPU_RM7000:
1036                 rm7k_erratum31();
1037
1038                 icache_size = 1 << (12 + ((config & CONF_IC) >> 9));
1039                 c->icache.linesz = 16 << ((config & CONF_IB) >> 5);
1040                 c->icache.ways = 4;
1041                 c->icache.waybit = __ffs(icache_size / c->icache.ways);
1042
1043                 dcache_size = 1 << (12 + ((config & CONF_DC) >> 6));
1044                 c->dcache.linesz = 16 << ((config & CONF_DB) >> 4);
1045                 c->dcache.ways = 4;
1046                 c->dcache.waybit = __ffs(dcache_size / c->dcache.ways);
1047
1048                 c->options |= MIPS_CPU_CACHE_CDEX_P;
1049                 c->options |= MIPS_CPU_PREFETCH;
1050                 break;
1051
1052         case CPU_LOONGSON2:
1053                 icache_size = 1 << (12 + ((config & CONF_IC) >> 9));
1054                 c->icache.linesz = 16 << ((config & CONF_IB) >> 5);
1055                 if (prid & 0x3)
1056                         c->icache.ways = 4;
1057                 else
1058                         c->icache.ways = 2;
1059                 c->icache.waybit = 0;
1060
1061                 dcache_size = 1 << (12 + ((config & CONF_DC) >> 6));
1062                 c->dcache.linesz = 16 << ((config & CONF_DB) >> 4);
1063                 if (prid & 0x3)
1064                         c->dcache.ways = 4;
1065                 else
1066                         c->dcache.ways = 2;
1067                 c->dcache.waybit = 0;
1068                 break;
1069
1070         case CPU_LOONGSON3:
1071                 config1 = read_c0_config1();
1072                 lsize = (config1 >> 19) & 7;
1073                 if (lsize)
1074                         c->icache.linesz = 2 << lsize;
1075                 else
1076                         c->icache.linesz = 0;
1077                 c->icache.sets = 64 << ((config1 >> 22) & 7);
1078                 c->icache.ways = 1 + ((config1 >> 16) & 7);
1079                 icache_size = c->icache.sets *
1080                                           c->icache.ways *
1081                                           c->icache.linesz;
1082                 c->icache.waybit = 0;
1083
1084                 lsize = (config1 >> 10) & 7;
1085                 if (lsize)
1086                         c->dcache.linesz = 2 << lsize;
1087                 else
1088                         c->dcache.linesz = 0;
1089                 c->dcache.sets = 64 << ((config1 >> 13) & 7);
1090                 c->dcache.ways = 1 + ((config1 >> 7) & 7);
1091                 dcache_size = c->dcache.sets *
1092                                           c->dcache.ways *
1093                                           c->dcache.linesz;
1094                 c->dcache.waybit = 0;
1095                 break;
1096
1097         default:
1098                 if (!(config & MIPS_CONF_M))
1099                         panic("Don't know how to probe P-caches on this cpu.");
1100
1101                 /*
1102                  * So we seem to be a MIPS32 or MIPS64 CPU
1103                  * So let's probe the I-cache ...
1104                  */
1105                 config1 = read_c0_config1();
1106
1107                 lsize = (config1 >> 19) & 7;
1108
1109                 /* IL == 7 is reserved */
1110                 if (lsize == 7)
1111                         panic("Invalid icache line size");
1112
1113                 c->icache.linesz = lsize ? 2 << lsize : 0;
1114
1115                 c->icache.sets = 32 << (((config1 >> 22) + 1) & 7);
1116                 c->icache.ways = 1 + ((config1 >> 16) & 7);
1117
1118                 icache_size = c->icache.sets *
1119                               c->icache.ways *
1120                               c->icache.linesz;
1121                 c->icache.waybit = __ffs(icache_size/c->icache.ways);
1122
1123                 if (config & 0x8)               /* VI bit */
1124                         c->icache.flags |= MIPS_CACHE_VTAG;
1125
1126                 /*
1127                  * Now probe the MIPS32 / MIPS64 data cache.
1128                  */
1129                 c->dcache.flags = 0;
1130
1131                 lsize = (config1 >> 10) & 7;
1132
1133                 /* DL == 7 is reserved */
1134                 if (lsize == 7)
1135                         panic("Invalid dcache line size");
1136
1137                 c->dcache.linesz = lsize ? 2 << lsize : 0;
1138
1139                 c->dcache.sets = 32 << (((config1 >> 13) + 1) & 7);
1140                 c->dcache.ways = 1 + ((config1 >> 7) & 7);
1141
1142                 dcache_size = c->dcache.sets *
1143                               c->dcache.ways *
1144                               c->dcache.linesz;
1145                 c->dcache.waybit = __ffs(dcache_size/c->dcache.ways);
1146
1147                 c->options |= MIPS_CPU_PREFETCH;
1148                 break;
1149         }
1150
1151         /*
1152          * Processor configuration sanity check for the R4000SC erratum
1153          * #5.  With page sizes larger than 32kB there is no possibility
1154          * to get a VCE exception anymore so we don't care about this
1155          * misconfiguration.  The case is rather theoretical anyway;
1156          * presumably no vendor is shipping his hardware in the "bad"
1157          * configuration.
1158          */
1159         if ((prid & PRID_IMP_MASK) == PRID_IMP_R4000 &&
1160             (prid & PRID_REV_MASK) < PRID_REV_R4400 &&
1161             !(config & CONF_SC) && c->icache.linesz != 16 &&
1162             PAGE_SIZE <= 0x8000)
1163                 panic("Improper R4000SC processor configuration detected");
1164
1165         /* compute a couple of other cache variables */
1166         c->icache.waysize = icache_size / c->icache.ways;
1167         c->dcache.waysize = dcache_size / c->dcache.ways;
1168
1169         c->icache.sets = c->icache.linesz ?
1170                 icache_size / (c->icache.linesz * c->icache.ways) : 0;
1171         c->dcache.sets = c->dcache.linesz ?
1172                 dcache_size / (c->dcache.linesz * c->dcache.ways) : 0;
1173
1174         /*
1175          * R10000 and R12000 P-caches are odd in a positive way.  They're 32kB
1176          * 2-way virtually indexed so normally would suffer from aliases.  So
1177          * normally they'd suffer from aliases but magic in the hardware deals
1178          * with that for us so we don't need to take care ourselves.
1179          */
1180         switch (current_cpu_type()) {
1181         case CPU_20KC:
1182         case CPU_25KF:
1183         case CPU_SB1:
1184         case CPU_SB1A:
1185         case CPU_XLR:
1186                 c->dcache.flags |= MIPS_CACHE_PINDEX;
1187                 break;
1188
1189         case CPU_R10000:
1190         case CPU_R12000:
1191         case CPU_R14000:
1192                 break;
1193
1194         case CPU_M14KC:
1195         case CPU_M14KEC:
1196         case CPU_24K:
1197         case CPU_34K:
1198         case CPU_74K:
1199         case CPU_1004K:
1200         case CPU_1074K:
1201         case CPU_INTERAPTIV:
1202         case CPU_P5600:
1203         case CPU_PROAPTIV:
1204         case CPU_M5150:
1205                 if ((c->cputype == CPU_74K) || (c->cputype == CPU_1074K))
1206                         alias_74k_erratum(c);
1207                 if (!(read_c0_config7() & MIPS_CONF7_IAR) &&
1208                     (c->icache.waysize > PAGE_SIZE))
1209                         c->icache.flags |= MIPS_CACHE_ALIASES;
1210                 if (read_c0_config7() & MIPS_CONF7_AR) {
1211                         /*
1212                          * Effectively physically indexed dcache,
1213                          * thus no virtual aliases.
1214                         */
1215                         c->dcache.flags |= MIPS_CACHE_PINDEX;
1216                         break;
1217                 }
1218         default:
1219                 if (c->dcache.waysize > PAGE_SIZE)
1220                         c->dcache.flags |= MIPS_CACHE_ALIASES;
1221         }
1222
1223         switch (current_cpu_type()) {
1224         case CPU_20KC:
1225                 /*
1226                  * Some older 20Kc chips doesn't have the 'VI' bit in
1227                  * the config register.
1228                  */
1229                 c->icache.flags |= MIPS_CACHE_VTAG;
1230                 break;
1231
1232         case CPU_ALCHEMY:
1233                 c->icache.flags |= MIPS_CACHE_IC_F_DC;
1234                 break;
1235
1236         case CPU_LOONGSON2:
1237                 /*
1238                  * LOONGSON2 has 4 way icache, but when using indexed cache op,
1239                  * one op will act on all 4 ways
1240                  */
1241                 c->icache.ways = 1;
1242         }
1243
1244         printk("Primary instruction cache %ldkB, %s, %s, linesize %d bytes.\n",
1245                icache_size >> 10,
1246                c->icache.flags & MIPS_CACHE_VTAG ? "VIVT" : "VIPT",
1247                way_string[c->icache.ways], c->icache.linesz);
1248
1249         printk("Primary data cache %ldkB, %s, %s, %s, linesize %d bytes\n",
1250                dcache_size >> 10, way_string[c->dcache.ways],
1251                (c->dcache.flags & MIPS_CACHE_PINDEX) ? "PIPT" : "VIPT",
1252                (c->dcache.flags & MIPS_CACHE_ALIASES) ?
1253                         "cache aliases" : "no aliases",
1254                c->dcache.linesz);
1255 }
1256
1257 /*
1258  * If you even _breathe_ on this function, look at the gcc output and make sure
1259  * it does not pop things on and off the stack for the cache sizing loop that
1260  * executes in KSEG1 space or else you will crash and burn badly.  You have
1261  * been warned.
1262  */
1263 static int probe_scache(void)
1264 {
1265         unsigned long flags, addr, begin, end, pow2;
1266         unsigned int config = read_c0_config();
1267         struct cpuinfo_mips *c = &current_cpu_data;
1268
1269         if (config & CONF_SC)
1270                 return 0;
1271
1272         begin = (unsigned long) &_stext;
1273         begin &= ~((4 * 1024 * 1024) - 1);
1274         end = begin + (4 * 1024 * 1024);
1275
1276         /*
1277          * This is such a bitch, you'd think they would make it easy to do
1278          * this.  Away you daemons of stupidity!
1279          */
1280         local_irq_save(flags);
1281
1282         /* Fill each size-multiple cache line with a valid tag. */
1283         pow2 = (64 * 1024);
1284         for (addr = begin; addr < end; addr = (begin + pow2)) {
1285                 unsigned long *p = (unsigned long *) addr;
1286                 __asm__ __volatile__("nop" : : "r" (*p)); /* whee... */
1287                 pow2 <<= 1;
1288         }
1289
1290         /* Load first line with zero (therefore invalid) tag. */
1291         write_c0_taglo(0);
1292         write_c0_taghi(0);
1293         __asm__ __volatile__("nop; nop; nop; nop;"); /* avoid the hazard */
1294         cache_op(Index_Store_Tag_I, begin);
1295         cache_op(Index_Store_Tag_D, begin);
1296         cache_op(Index_Store_Tag_SD, begin);
1297
1298         /* Now search for the wrap around point. */
1299         pow2 = (128 * 1024);
1300         for (addr = begin + (128 * 1024); addr < end; addr = begin + pow2) {
1301                 cache_op(Index_Load_Tag_SD, addr);
1302                 __asm__ __volatile__("nop; nop; nop; nop;"); /* hazard... */
1303                 if (!read_c0_taglo())
1304                         break;
1305                 pow2 <<= 1;
1306         }
1307         local_irq_restore(flags);
1308         addr -= begin;
1309
1310         scache_size = addr;
1311         c->scache.linesz = 16 << ((config & R4K_CONF_SB) >> 22);
1312         c->scache.ways = 1;
1313         c->dcache.waybit = 0;           /* does not matter */
1314
1315         return 1;
1316 }
1317
1318 static void __init loongson2_sc_init(void)
1319 {
1320         struct cpuinfo_mips *c = &current_cpu_data;
1321
1322         scache_size = 512*1024;
1323         c->scache.linesz = 32;
1324         c->scache.ways = 4;
1325         c->scache.waybit = 0;
1326         c->scache.waysize = scache_size / (c->scache.ways);
1327         c->scache.sets = scache_size / (c->scache.linesz * c->scache.ways);
1328         pr_info("Unified secondary cache %ldkB %s, linesize %d bytes.\n",
1329                scache_size >> 10, way_string[c->scache.ways], c->scache.linesz);
1330
1331         c->options |= MIPS_CPU_INCLUSIVE_CACHES;
1332 }
1333
1334 static void __init loongson3_sc_init(void)
1335 {
1336         struct cpuinfo_mips *c = &current_cpu_data;
1337         unsigned int config2, lsize;
1338
1339         config2 = read_c0_config2();
1340         lsize = (config2 >> 4) & 15;
1341         if (lsize)
1342                 c->scache.linesz = 2 << lsize;
1343         else
1344                 c->scache.linesz = 0;
1345         c->scache.sets = 64 << ((config2 >> 8) & 15);
1346         c->scache.ways = 1 + (config2 & 15);
1347
1348         scache_size = c->scache.sets *
1349                                   c->scache.ways *
1350                                   c->scache.linesz;
1351         /* Loongson-3 has 4 cores, 1MB scache for each. scaches are shared */
1352         scache_size *= 4;
1353         c->scache.waybit = 0;
1354         pr_info("Unified secondary cache %ldkB %s, linesize %d bytes.\n",
1355                scache_size >> 10, way_string[c->scache.ways], c->scache.linesz);
1356         if (scache_size)
1357                 c->options |= MIPS_CPU_INCLUSIVE_CACHES;
1358         return;
1359 }
1360
1361 extern int r5k_sc_init(void);
1362 extern int rm7k_sc_init(void);
1363 extern int mips_sc_init(void);
1364
1365 static void setup_scache(void)
1366 {
1367         struct cpuinfo_mips *c = &current_cpu_data;
1368         unsigned int config = read_c0_config();
1369         int sc_present = 0;
1370
1371         /*
1372          * Do the probing thing on R4000SC and R4400SC processors.  Other
1373          * processors don't have a S-cache that would be relevant to the
1374          * Linux memory management.
1375          */
1376         switch (current_cpu_type()) {
1377         case CPU_R4000SC:
1378         case CPU_R4000MC:
1379         case CPU_R4400SC:
1380         case CPU_R4400MC:
1381                 sc_present = run_uncached(probe_scache);
1382                 if (sc_present)
1383                         c->options |= MIPS_CPU_CACHE_CDEX_S;
1384                 break;
1385
1386         case CPU_R10000:
1387         case CPU_R12000:
1388         case CPU_R14000:
1389                 scache_size = 0x80000 << ((config & R10K_CONF_SS) >> 16);
1390                 c->scache.linesz = 64 << ((config >> 13) & 1);
1391                 c->scache.ways = 2;
1392                 c->scache.waybit= 0;
1393                 sc_present = 1;
1394                 break;
1395
1396         case CPU_R5000:
1397         case CPU_NEVADA:
1398 #ifdef CONFIG_R5000_CPU_SCACHE
1399                 r5k_sc_init();
1400 #endif
1401                 return;
1402
1403         case CPU_RM7000:
1404 #ifdef CONFIG_RM7000_CPU_SCACHE
1405                 rm7k_sc_init();
1406 #endif
1407                 return;
1408
1409         case CPU_LOONGSON2:
1410                 loongson2_sc_init();
1411                 return;
1412
1413         case CPU_LOONGSON3:
1414                 loongson3_sc_init();
1415                 return;
1416
1417         case CPU_XLP:
1418                 /* don't need to worry about L2, fully coherent */
1419                 return;
1420
1421         default:
1422                 if (c->isa_level & (MIPS_CPU_ISA_M32R1 | MIPS_CPU_ISA_M32R2 |
1423                                     MIPS_CPU_ISA_M64R1 | MIPS_CPU_ISA_M64R2)) {
1424 #ifdef CONFIG_MIPS_CPU_SCACHE
1425                         if (mips_sc_init ()) {
1426                                 scache_size = c->scache.ways * c->scache.sets * c->scache.linesz;
1427                                 printk("MIPS secondary cache %ldkB, %s, linesize %d bytes.\n",
1428                                        scache_size >> 10,
1429                                        way_string[c->scache.ways], c->scache.linesz);
1430                         }
1431 #else
1432                         if (!(c->scache.flags & MIPS_CACHE_NOT_PRESENT))
1433                                 panic("Dunno how to handle MIPS32 / MIPS64 second level cache");
1434 #endif
1435                         return;
1436                 }
1437                 sc_present = 0;
1438         }
1439
1440         if (!sc_present)
1441                 return;
1442
1443         /* compute a couple of other cache variables */
1444         c->scache.waysize = scache_size / c->scache.ways;
1445
1446         c->scache.sets = scache_size / (c->scache.linesz * c->scache.ways);
1447
1448         printk("Unified secondary cache %ldkB %s, linesize %d bytes.\n",
1449                scache_size >> 10, way_string[c->scache.ways], c->scache.linesz);
1450
1451         c->options |= MIPS_CPU_INCLUSIVE_CACHES;
1452 }
1453
1454 void au1x00_fixup_config_od(void)
1455 {
1456         /*
1457          * c0_config.od (bit 19) was write only (and read as 0)
1458          * on the early revisions of Alchemy SOCs.  It disables the bus
1459          * transaction overlapping and needs to be set to fix various errata.
1460          */
1461         switch (read_c0_prid()) {
1462         case 0x00030100: /* Au1000 DA */
1463         case 0x00030201: /* Au1000 HA */
1464         case 0x00030202: /* Au1000 HB */
1465         case 0x01030200: /* Au1500 AB */
1466         /*
1467          * Au1100 errata actually keeps silence about this bit, so we set it
1468          * just in case for those revisions that require it to be set according
1469          * to the (now gone) cpu table.
1470          */
1471         case 0x02030200: /* Au1100 AB */
1472         case 0x02030201: /* Au1100 BA */
1473         case 0x02030202: /* Au1100 BC */
1474                 set_c0_config(1 << 19);
1475                 break;
1476         }
1477 }
1478
1479 /* CP0 hazard avoidance. */
1480 #define NXP_BARRIER()                                                   \
1481          __asm__ __volatile__(                                          \
1482         ".set noreorder\n\t"                                            \
1483         "nop; nop; nop; nop; nop; nop;\n\t"                             \
1484         ".set reorder\n\t")
1485
1486 static void nxp_pr4450_fixup_config(void)
1487 {
1488         unsigned long config0;
1489
1490         config0 = read_c0_config();
1491
1492         /* clear all three cache coherency fields */
1493         config0 &= ~(0x7 | (7 << 25) | (7 << 28));
1494         config0 |= (((_page_cachable_default >> _CACHE_SHIFT) <<  0) |
1495                     ((_page_cachable_default >> _CACHE_SHIFT) << 25) |
1496                     ((_page_cachable_default >> _CACHE_SHIFT) << 28));
1497         write_c0_config(config0);
1498         NXP_BARRIER();
1499 }
1500
1501 static int cca = -1;
1502
1503 static int __init cca_setup(char *str)
1504 {
1505         get_option(&str, &cca);
1506
1507         return 0;
1508 }
1509
1510 early_param("cca", cca_setup);
1511
1512 static void coherency_setup(void)
1513 {
1514         if (cca < 0 || cca > 7)
1515                 cca = read_c0_config() & CONF_CM_CMASK;
1516         _page_cachable_default = cca << _CACHE_SHIFT;
1517
1518         pr_debug("Using cache attribute %d\n", cca);
1519         change_c0_config(CONF_CM_CMASK, cca);
1520
1521         /*
1522          * c0_status.cu=0 specifies that updates by the sc instruction use
1523          * the coherency mode specified by the TLB; 1 means cachable
1524          * coherent update on write will be used.  Not all processors have
1525          * this bit and; some wire it to zero, others like Toshiba had the
1526          * silly idea of putting something else there ...
1527          */
1528         switch (current_cpu_type()) {
1529         case CPU_R4000PC:
1530         case CPU_R4000SC:
1531         case CPU_R4000MC:
1532         case CPU_R4400PC:
1533         case CPU_R4400SC:
1534         case CPU_R4400MC:
1535                 clear_c0_config(CONF_CU);
1536                 break;
1537         /*
1538          * We need to catch the early Alchemy SOCs with
1539          * the write-only co_config.od bit and set it back to one on:
1540          * Au1000 rev DA, HA, HB;  Au1100 AB, BA, BC, Au1500 AB
1541          */
1542         case CPU_ALCHEMY:
1543                 au1x00_fixup_config_od();
1544                 break;
1545
1546         case PRID_IMP_PR4450:
1547                 nxp_pr4450_fixup_config();
1548                 break;
1549         }
1550 }
1551
1552 static void r4k_cache_error_setup(void)
1553 {
1554         extern char __weak except_vec2_generic;
1555         extern char __weak except_vec2_sb1;
1556
1557         switch (current_cpu_type()) {
1558         case CPU_SB1:
1559         case CPU_SB1A:
1560                 set_uncached_handler(0x100, &except_vec2_sb1, 0x80);
1561                 break;
1562
1563         default:
1564                 set_uncached_handler(0x100, &except_vec2_generic, 0x80);
1565                 break;
1566         }
1567 }
1568
1569 void r4k_cache_init(void)
1570 {
1571         extern void build_clear_page(void);
1572         extern void build_copy_page(void);
1573         struct cpuinfo_mips *c = &current_cpu_data;
1574
1575         probe_pcache();
1576         setup_scache();
1577
1578         r4k_blast_dcache_page_setup();
1579         r4k_blast_dcache_page_indexed_setup();
1580         r4k_blast_dcache_setup();
1581         r4k_blast_icache_page_setup();
1582         r4k_blast_icache_page_indexed_setup();
1583         r4k_blast_icache_setup();
1584         r4k_blast_scache_page_setup();
1585         r4k_blast_scache_page_indexed_setup();
1586         r4k_blast_scache_setup();
1587 #ifdef CONFIG_EVA
1588         r4k_blast_dcache_user_page_setup();
1589         r4k_blast_icache_user_page_setup();
1590 #endif
1591
1592         /*
1593          * Some MIPS32 and MIPS64 processors have physically indexed caches.
1594          * This code supports virtually indexed processors and will be
1595          * unnecessarily inefficient on physically indexed processors.
1596          */
1597         if (c->dcache.linesz)
1598                 shm_align_mask = max_t( unsigned long,
1599                                         c->dcache.sets * c->dcache.linesz - 1,
1600                                         PAGE_SIZE - 1);
1601         else
1602                 shm_align_mask = PAGE_SIZE-1;
1603
1604         __flush_cache_vmap      = r4k__flush_cache_vmap;
1605         __flush_cache_vunmap    = r4k__flush_cache_vunmap;
1606
1607         flush_cache_all         = cache_noop;
1608         __flush_cache_all       = r4k___flush_cache_all;
1609         flush_cache_mm          = r4k_flush_cache_mm;
1610         flush_cache_page        = r4k_flush_cache_page;
1611         flush_cache_range       = r4k_flush_cache_range;
1612
1613         __flush_kernel_vmap_range = r4k_flush_kernel_vmap_range;
1614
1615         flush_cache_sigtramp    = r4k_flush_cache_sigtramp;
1616         flush_icache_all        = r4k_flush_icache_all;
1617         local_flush_data_cache_page     = local_r4k_flush_data_cache_page;
1618         flush_data_cache_page   = r4k_flush_data_cache_page;
1619         flush_icache_range      = r4k_flush_icache_range;
1620         local_flush_icache_range        = local_r4k_flush_icache_range;
1621
1622 #if defined(CONFIG_DMA_NONCOHERENT) || defined(CONFIG_DMA_MAYBE_COHERENT)
1623         if (coherentio) {
1624                 _dma_cache_wback_inv    = (void *)cache_noop;
1625                 _dma_cache_wback        = (void *)cache_noop;
1626                 _dma_cache_inv          = (void *)cache_noop;
1627         } else {
1628                 _dma_cache_wback_inv    = r4k_dma_cache_wback_inv;
1629                 _dma_cache_wback        = r4k_dma_cache_wback_inv;
1630                 _dma_cache_inv          = r4k_dma_cache_inv;
1631         }
1632 #endif
1633
1634         build_clear_page();
1635         build_copy_page();
1636
1637         /*
1638          * We want to run CMP kernels on core with and without coherent
1639          * caches. Therefore, do not use CONFIG_MIPS_CMP to decide whether
1640          * or not to flush caches.
1641          */
1642         local_r4k___flush_cache_all(NULL);
1643
1644         coherency_setup();
1645         board_cache_error_setup = r4k_cache_error_setup;
1646 }
1647
1648 static int r4k_cache_pm_notifier(struct notifier_block *self, unsigned long cmd,
1649                                void *v)
1650 {
1651         switch (cmd) {
1652         case CPU_PM_ENTER_FAILED:
1653         case CPU_PM_EXIT:
1654                 coherency_setup();
1655                 break;
1656         }
1657
1658         return NOTIFY_OK;
1659 }
1660
1661 static struct notifier_block r4k_cache_pm_notifier_block = {
1662         .notifier_call = r4k_cache_pm_notifier,
1663 };
1664
1665 int __init r4k_cache_init_pm(void)
1666 {
1667         return cpu_pm_register_notifier(&r4k_cache_pm_notifier_block);
1668 }
1669 arch_initcall(r4k_cache_init_pm);