Linux-2.6.12-rc2
[linux-flexiantxendom0-natty.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 (dm@engr.sgi.com)
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/config.h>
11 #include <linux/init.h>
12 #include <linux/kernel.h>
13 #include <linux/sched.h>
14 #include <linux/mm.h>
15 #include <linux/bitops.h>
16
17 #include <asm/bcache.h>
18 #include <asm/bootinfo.h>
19 #include <asm/cacheops.h>
20 #include <asm/cpu.h>
21 #include <asm/cpu-features.h>
22 #include <asm/io.h>
23 #include <asm/page.h>
24 #include <asm/pgtable.h>
25 #include <asm/r4kcache.h>
26 #include <asm/system.h>
27 #include <asm/mmu_context.h>
28 #include <asm/war.h>
29
30 static unsigned long icache_size, dcache_size, scache_size;
31
32 /*
33  * Dummy cache handling routines for machines without boardcaches
34  */
35 static void no_sc_noop(void) {}
36
37 static struct bcache_ops no_sc_ops = {
38         .bc_enable = (void *)no_sc_noop,
39         .bc_disable = (void *)no_sc_noop,
40         .bc_wback_inv = (void *)no_sc_noop,
41         .bc_inv = (void *)no_sc_noop
42 };
43
44 struct bcache_ops *bcops = &no_sc_ops;
45
46 #define cpu_is_r4600_v1_x()     ((read_c0_prid() & 0xfffffff0) == 0x2010)
47 #define cpu_is_r4600_v2_x()     ((read_c0_prid() & 0xfffffff0) == 0x2020)
48
49 #define R4600_HIT_CACHEOP_WAR_IMPL                                      \
50 do {                                                                    \
51         if (R4600_V2_HIT_CACHEOP_WAR && cpu_is_r4600_v2_x())            \
52                 *(volatile unsigned long *)CKSEG1;                      \
53         if (R4600_V1_HIT_CACHEOP_WAR)                                   \
54                 __asm__ __volatile__("nop;nop;nop;nop");                \
55 } while (0)
56
57 static void (*r4k_blast_dcache_page)(unsigned long addr);
58
59 static inline void r4k_blast_dcache_page_dc32(unsigned long addr)
60 {
61         R4600_HIT_CACHEOP_WAR_IMPL;
62         blast_dcache32_page(addr);
63 }
64
65 static inline void r4k_blast_dcache_page_setup(void)
66 {
67         unsigned long  dc_lsize = cpu_dcache_line_size();
68
69         if (dc_lsize == 16)
70                 r4k_blast_dcache_page = blast_dcache16_page;
71         else if (dc_lsize == 32)
72                 r4k_blast_dcache_page = r4k_blast_dcache_page_dc32;
73 }
74
75 static void (* r4k_blast_dcache_page_indexed)(unsigned long addr);
76
77 static inline void r4k_blast_dcache_page_indexed_setup(void)
78 {
79         unsigned long dc_lsize = cpu_dcache_line_size();
80
81         if (dc_lsize == 16)
82                 r4k_blast_dcache_page_indexed = blast_dcache16_page_indexed;
83         else if (dc_lsize == 32)
84                 r4k_blast_dcache_page_indexed = blast_dcache32_page_indexed;
85 }
86
87 static void (* r4k_blast_dcache)(void);
88
89 static inline void r4k_blast_dcache_setup(void)
90 {
91         unsigned long dc_lsize = cpu_dcache_line_size();
92
93         if (dc_lsize == 16)
94                 r4k_blast_dcache = blast_dcache16;
95         else if (dc_lsize == 32)
96                 r4k_blast_dcache = blast_dcache32;
97 }
98
99 /* force code alignment (used for TX49XX_ICACHE_INDEX_INV_WAR) */
100 #define JUMP_TO_ALIGN(order) \
101         __asm__ __volatile__( \
102                 "b\t1f\n\t" \
103                 ".align\t" #order "\n\t" \
104                 "1:\n\t" \
105                 )
106 #define CACHE32_UNROLL32_ALIGN  JUMP_TO_ALIGN(10) /* 32 * 32 = 1024 */
107 #define CACHE32_UNROLL32_ALIGN2 JUMP_TO_ALIGN(11)
108
109 static inline void blast_r4600_v1_icache32(void)
110 {
111         unsigned long flags;
112
113         local_irq_save(flags);
114         blast_icache32();
115         local_irq_restore(flags);
116 }
117
118 static inline void tx49_blast_icache32(void)
119 {
120         unsigned long start = INDEX_BASE;
121         unsigned long end = start + current_cpu_data.icache.waysize;
122         unsigned long ws_inc = 1UL << current_cpu_data.icache.waybit;
123         unsigned long ws_end = current_cpu_data.icache.ways <<
124                                current_cpu_data.icache.waybit;
125         unsigned long ws, addr;
126
127         CACHE32_UNROLL32_ALIGN2;
128         /* I'm in even chunk.  blast odd chunks */
129         for (ws = 0; ws < ws_end; ws += ws_inc) 
130                 for (addr = start + 0x400; addr < end; addr += 0x400 * 2) 
131                         cache32_unroll32(addr|ws,Index_Invalidate_I);
132         CACHE32_UNROLL32_ALIGN;
133         /* I'm in odd chunk.  blast even chunks */
134         for (ws = 0; ws < ws_end; ws += ws_inc) 
135                 for (addr = start; addr < end; addr += 0x400 * 2) 
136                         cache32_unroll32(addr|ws,Index_Invalidate_I);
137 }
138
139 static inline void blast_icache32_r4600_v1_page_indexed(unsigned long page)
140 {
141         unsigned long flags;
142
143         local_irq_save(flags);
144         blast_icache32_page_indexed(page);
145         local_irq_restore(flags);
146 }
147
148 static inline void tx49_blast_icache32_page_indexed(unsigned long page)
149 {
150         unsigned long start = page;
151         unsigned long end = start + PAGE_SIZE;
152         unsigned long ws_inc = 1UL << current_cpu_data.icache.waybit;
153         unsigned long ws_end = current_cpu_data.icache.ways <<
154                                current_cpu_data.icache.waybit;
155         unsigned long ws, addr;
156
157         CACHE32_UNROLL32_ALIGN2;
158         /* I'm in even chunk.  blast odd chunks */
159         for (ws = 0; ws < ws_end; ws += ws_inc) 
160                 for (addr = start + 0x400; addr < end; addr += 0x400 * 2) 
161                         cache32_unroll32(addr|ws,Index_Invalidate_I);
162         CACHE32_UNROLL32_ALIGN;
163         /* I'm in odd chunk.  blast even chunks */
164         for (ws = 0; ws < ws_end; ws += ws_inc) 
165                 for (addr = start; addr < end; addr += 0x400 * 2) 
166                         cache32_unroll32(addr|ws,Index_Invalidate_I);
167 }
168
169 static void (* r4k_blast_icache_page)(unsigned long addr);
170
171 static inline void r4k_blast_icache_page_setup(void)
172 {
173         unsigned long ic_lsize = cpu_icache_line_size();
174
175         if (ic_lsize == 16)
176                 r4k_blast_icache_page = blast_icache16_page;
177         else if (ic_lsize == 32)
178                 r4k_blast_icache_page = blast_icache32_page;
179         else if (ic_lsize == 64)
180                 r4k_blast_icache_page = blast_icache64_page;
181 }
182
183
184 static void (* r4k_blast_icache_page_indexed)(unsigned long addr);
185
186 static inline void r4k_blast_icache_page_indexed_setup(void)
187 {
188         unsigned long ic_lsize = cpu_icache_line_size();
189
190         if (ic_lsize == 16)
191                 r4k_blast_icache_page_indexed = blast_icache16_page_indexed;
192         else if (ic_lsize == 32) {
193                 if (TX49XX_ICACHE_INDEX_INV_WAR)
194                         r4k_blast_icache_page_indexed =
195                                 tx49_blast_icache32_page_indexed;
196                 else if (R4600_V1_INDEX_ICACHEOP_WAR && cpu_is_r4600_v1_x())
197                         r4k_blast_icache_page_indexed =
198                                 blast_icache32_r4600_v1_page_indexed;
199                 else
200                         r4k_blast_icache_page_indexed =
201                                 blast_icache32_page_indexed;
202         } else if (ic_lsize == 64)
203                 r4k_blast_icache_page_indexed = blast_icache64_page_indexed;
204 }
205
206 static void (* r4k_blast_icache)(void);
207
208 static inline void r4k_blast_icache_setup(void)
209 {
210         unsigned long ic_lsize = cpu_icache_line_size();
211
212         if (ic_lsize == 16)
213                 r4k_blast_icache = blast_icache16;
214         else if (ic_lsize == 32) {
215                 if (R4600_V1_INDEX_ICACHEOP_WAR && cpu_is_r4600_v1_x())
216                         r4k_blast_icache = blast_r4600_v1_icache32;
217                 else if (TX49XX_ICACHE_INDEX_INV_WAR)
218                         r4k_blast_icache = tx49_blast_icache32;
219                 else
220                         r4k_blast_icache = blast_icache32;
221         } else if (ic_lsize == 64)
222                 r4k_blast_icache = blast_icache64;
223 }
224
225 static void (* r4k_blast_scache_page)(unsigned long addr);
226
227 static inline void r4k_blast_scache_page_setup(void)
228 {
229         unsigned long sc_lsize = cpu_scache_line_size();
230
231         if (sc_lsize == 16)
232                 r4k_blast_scache_page = blast_scache16_page;
233         else if (sc_lsize == 32)
234                 r4k_blast_scache_page = blast_scache32_page;
235         else if (sc_lsize == 64)
236                 r4k_blast_scache_page = blast_scache64_page;
237         else if (sc_lsize == 128)
238                 r4k_blast_scache_page = blast_scache128_page;
239 }
240
241 static void (* r4k_blast_scache_page_indexed)(unsigned long addr);
242
243 static inline void r4k_blast_scache_page_indexed_setup(void)
244 {
245         unsigned long sc_lsize = cpu_scache_line_size();
246
247         if (sc_lsize == 16)
248                 r4k_blast_scache_page_indexed = blast_scache16_page_indexed;
249         else if (sc_lsize == 32)
250                 r4k_blast_scache_page_indexed = blast_scache32_page_indexed;
251         else if (sc_lsize == 64)
252                 r4k_blast_scache_page_indexed = blast_scache64_page_indexed;
253         else if (sc_lsize == 128)
254                 r4k_blast_scache_page_indexed = blast_scache128_page_indexed;
255 }
256
257 static void (* r4k_blast_scache)(void);
258
259 static inline void r4k_blast_scache_setup(void)
260 {
261         unsigned long sc_lsize = cpu_scache_line_size();
262
263         if (sc_lsize == 16)
264                 r4k_blast_scache = blast_scache16;
265         else if (sc_lsize == 32)
266                 r4k_blast_scache = blast_scache32;
267         else if (sc_lsize == 64)
268                 r4k_blast_scache = blast_scache64;
269         else if (sc_lsize == 128)
270                 r4k_blast_scache = blast_scache128;
271 }
272
273 /*
274  * This is former mm's flush_cache_all() which really should be
275  * flush_cache_vunmap these days ...
276  */
277 static inline void local_r4k_flush_cache_all(void * args)
278 {
279         r4k_blast_dcache();
280         r4k_blast_icache();
281 }
282
283 static void r4k_flush_cache_all(void)
284 {
285         if (!cpu_has_dc_aliases)
286                 return;
287
288         on_each_cpu(local_r4k_flush_cache_all, NULL, 1, 1);
289 }
290
291 static inline void local_r4k___flush_cache_all(void * args)
292 {
293         r4k_blast_dcache();
294         r4k_blast_icache();
295
296         switch (current_cpu_data.cputype) {
297         case CPU_R4000SC:
298         case CPU_R4000MC:
299         case CPU_R4400SC:
300         case CPU_R4400MC:
301         case CPU_R10000:
302         case CPU_R12000:
303                 r4k_blast_scache();
304         }
305 }
306
307 static void r4k___flush_cache_all(void)
308 {
309         on_each_cpu(local_r4k___flush_cache_all, NULL, 1, 1);
310 }
311
312 static inline void local_r4k_flush_cache_range(void * args)
313 {
314         struct vm_area_struct *vma = args;
315         int exec;
316
317         if (!(cpu_context(smp_processor_id(), vma->vm_mm)))
318                 return;
319
320         exec = vma->vm_flags & VM_EXEC;
321         if (cpu_has_dc_aliases || exec)
322                 r4k_blast_dcache();
323         if (exec)
324                 r4k_blast_icache();
325 }
326
327 static void r4k_flush_cache_range(struct vm_area_struct *vma,
328         unsigned long start, unsigned long end)
329 {
330         on_each_cpu(local_r4k_flush_cache_range, vma, 1, 1);
331 }
332
333 static inline void local_r4k_flush_cache_mm(void * args)
334 {
335         struct mm_struct *mm = args;
336
337         if (!cpu_context(smp_processor_id(), mm))
338                 return;
339
340         r4k_blast_dcache();
341         r4k_blast_icache();
342
343         /*
344          * Kludge alert.  For obscure reasons R4000SC and R4400SC go nuts if we
345          * only flush the primary caches but R10000 and R12000 behave sane ...
346          */
347         if (current_cpu_data.cputype == CPU_R4000SC ||
348             current_cpu_data.cputype == CPU_R4000MC ||
349             current_cpu_data.cputype == CPU_R4400SC ||
350             current_cpu_data.cputype == CPU_R4400MC)
351                 r4k_blast_scache();
352 }
353
354 static void r4k_flush_cache_mm(struct mm_struct *mm)
355 {
356         if (!cpu_has_dc_aliases)
357                 return;
358
359         on_each_cpu(local_r4k_flush_cache_mm, mm, 1, 1);
360 }
361
362 struct flush_cache_page_args {
363         struct vm_area_struct *vma;
364         unsigned long page;
365 };
366
367 static inline void local_r4k_flush_cache_page(void *args)
368 {
369         struct flush_cache_page_args *fcp_args = args;
370         struct vm_area_struct *vma = fcp_args->vma;
371         unsigned long page = fcp_args->page;
372         int exec = vma->vm_flags & VM_EXEC;
373         struct mm_struct *mm = vma->vm_mm;
374         pgd_t *pgdp;
375         pmd_t *pmdp;
376         pte_t *ptep;
377
378         page &= PAGE_MASK;
379         pgdp = pgd_offset(mm, page);
380         pmdp = pmd_offset(pgdp, page);
381         ptep = pte_offset(pmdp, page);
382
383         /*
384          * If the page isn't marked valid, the page cannot possibly be
385          * in the cache.
386          */
387         if (!(pte_val(*ptep) & _PAGE_PRESENT))
388                 return;
389
390         /*
391          * Doing flushes for another ASID than the current one is
392          * too difficult since stupid R4k caches do a TLB translation
393          * for every cache flush operation.  So we do indexed flushes
394          * in that case, which doesn't overly flush the cache too much.
395          */
396         if ((mm == current->active_mm) && (pte_val(*ptep) & _PAGE_VALID)) {
397                 if (cpu_has_dc_aliases || (exec && !cpu_has_ic_fills_f_dc)) {
398                         r4k_blast_dcache_page(page);
399                         if (exec && !cpu_icache_snoops_remote_store)
400                                 r4k_blast_scache_page(page);
401                 }
402                 if (exec)
403                         r4k_blast_icache_page(page);
404
405                 return;
406         }
407
408         /*
409          * Do indexed flush, too much work to get the (possible) TLB refills
410          * to work correctly.
411          */
412         page = INDEX_BASE + (page & (dcache_size - 1));
413         if (cpu_has_dc_aliases || (exec && !cpu_has_ic_fills_f_dc)) {
414                 r4k_blast_dcache_page_indexed(page);
415                 if (exec && !cpu_icache_snoops_remote_store)
416                         r4k_blast_scache_page_indexed(page);
417         }
418         if (exec) {
419                 if (cpu_has_vtag_icache) {
420                         int cpu = smp_processor_id();
421
422                         if (cpu_context(cpu, vma->vm_mm) != 0)
423                                 drop_mmu_context(vma->vm_mm, cpu);
424                 } else
425                         r4k_blast_icache_page_indexed(page);
426         }
427 }
428
429 static void r4k_flush_cache_page(struct vm_area_struct *vma, unsigned long page, unsigned long pfn)
430 {
431         struct flush_cache_page_args args;
432
433         /*
434          * If ownes no valid ASID yet, cannot possibly have gotten
435          * this page into the cache.
436          */
437         if (cpu_context(smp_processor_id(), vma->vm_mm) == 0)
438                 return;
439
440         args.vma = vma;
441         args.page = page;
442
443         on_each_cpu(local_r4k_flush_cache_page, &args, 1, 1);
444 }
445
446 static inline void local_r4k_flush_data_cache_page(void * addr)
447 {
448         r4k_blast_dcache_page((unsigned long) addr);
449 }
450
451 static void r4k_flush_data_cache_page(unsigned long addr)
452 {
453         on_each_cpu(local_r4k_flush_data_cache_page, (void *) addr, 1, 1);
454 }
455
456 struct flush_icache_range_args {
457         unsigned long start;
458         unsigned long end;
459 };
460
461 static inline void local_r4k_flush_icache_range(void *args)
462 {
463         struct flush_icache_range_args *fir_args = args;
464         unsigned long dc_lsize = current_cpu_data.dcache.linesz;
465         unsigned long ic_lsize = current_cpu_data.icache.linesz;
466         unsigned long sc_lsize = current_cpu_data.scache.linesz;
467         unsigned long start = fir_args->start;
468         unsigned long end = fir_args->end;
469         unsigned long addr, aend;
470
471         if (!cpu_has_ic_fills_f_dc) {
472                 if (end - start > dcache_size) {
473                         r4k_blast_dcache();
474                 } else {
475                         addr = start & ~(dc_lsize - 1);
476                         aend = (end - 1) & ~(dc_lsize - 1);
477
478                         while (1) {
479                                 /* Hit_Writeback_Inv_D */
480                                 protected_writeback_dcache_line(addr);
481                                 if (addr == aend)
482                                         break;
483                                 addr += dc_lsize;
484                         }
485                 }
486
487                 if (!cpu_icache_snoops_remote_store) {
488                         if (end - start > scache_size) {
489                                 r4k_blast_scache();
490                         } else {
491                                 addr = start & ~(sc_lsize - 1);
492                                 aend = (end - 1) & ~(sc_lsize - 1);
493
494                                 while (1) {
495                                         /* Hit_Writeback_Inv_D */
496                                         protected_writeback_scache_line(addr);
497                                         if (addr == aend)
498                                                 break;
499                                         addr += sc_lsize;
500                                 }
501                         }
502                 }
503         }
504
505         if (end - start > icache_size)
506                 r4k_blast_icache();
507         else {
508                 addr = start & ~(ic_lsize - 1);
509                 aend = (end - 1) & ~(ic_lsize - 1);
510                 while (1) {
511                         /* Hit_Invalidate_I */
512                         protected_flush_icache_line(addr);
513                         if (addr == aend)
514                                 break;
515                         addr += ic_lsize;
516                 }
517         }
518 }
519
520 static void r4k_flush_icache_range(unsigned long start, unsigned long end)
521 {
522         struct flush_icache_range_args args;
523
524         args.start = start;
525         args.end = end;
526
527         on_each_cpu(local_r4k_flush_icache_range, &args, 1, 1);
528 }
529
530 /*
531  * Ok, this seriously sucks.  We use them to flush a user page but don't
532  * know the virtual address, so we have to blast away the whole icache
533  * which is significantly more expensive than the real thing.  Otoh we at
534  * least know the kernel address of the page so we can flush it
535  * selectivly.
536  */
537
538 struct flush_icache_page_args {
539         struct vm_area_struct *vma;
540         struct page *page;
541 };
542
543 static inline void local_r4k_flush_icache_page(void *args)
544 {
545         struct flush_icache_page_args *fip_args = args;
546         struct vm_area_struct *vma = fip_args->vma;
547         struct page *page = fip_args->page;
548
549         /*
550          * Tricky ...  Because we don't know the virtual address we've got the
551          * choice of either invalidating the entire primary and secondary
552          * caches or invalidating the secondary caches also.  With the subset
553          * enforcment on R4000SC, R4400SC, R10000 and R12000 invalidating the
554          * secondary cache will result in any entries in the primary caches
555          * also getting invalidated which hopefully is a bit more economical.
556          */
557         if (cpu_has_subset_pcaches) {
558                 unsigned long addr = (unsigned long) page_address(page);
559
560                 r4k_blast_scache_page(addr);
561                 ClearPageDcacheDirty(page);
562
563                 return;
564         }
565
566         if (!cpu_has_ic_fills_f_dc) {
567                 unsigned long addr = (unsigned long) page_address(page);
568                 r4k_blast_dcache_page(addr);
569                 if (!cpu_icache_snoops_remote_store)
570                         r4k_blast_scache_page(addr);
571                 ClearPageDcacheDirty(page);
572         }
573
574         /*
575          * We're not sure of the virtual address(es) involved here, so
576          * we have to flush the entire I-cache.
577          */
578         if (cpu_has_vtag_icache) {
579                 int cpu = smp_processor_id();
580
581                 if (cpu_context(cpu, vma->vm_mm) != 0)
582                         drop_mmu_context(vma->vm_mm, cpu);
583         } else
584                 r4k_blast_icache();
585 }
586
587 static void r4k_flush_icache_page(struct vm_area_struct *vma,
588         struct page *page)
589 {
590         struct flush_icache_page_args args;
591
592         /*
593          * If there's no context yet, or the page isn't executable, no I-cache
594          * flush is needed.
595          */
596         if (!(vma->vm_flags & VM_EXEC))
597                 return;
598
599         args.vma = vma;
600         args.page = page;
601
602         on_each_cpu(local_r4k_flush_icache_page, &args, 1, 1);
603 }
604
605
606 #ifdef CONFIG_DMA_NONCOHERENT
607
608 static void r4k_dma_cache_wback_inv(unsigned long addr, unsigned long size)
609 {
610         unsigned long end, a;
611
612         /* Catch bad driver code */
613         BUG_ON(size == 0);
614
615         if (cpu_has_subset_pcaches) {
616                 unsigned long sc_lsize = current_cpu_data.scache.linesz;
617
618                 if (size >= scache_size) {
619                         r4k_blast_scache();
620                         return;
621                 }
622
623                 a = addr & ~(sc_lsize - 1);
624                 end = (addr + size - 1) & ~(sc_lsize - 1);
625                 while (1) {
626                         flush_scache_line(a);   /* Hit_Writeback_Inv_SD */
627                         if (a == end)
628                                 break;
629                         a += sc_lsize;
630                 }
631                 return;
632         }
633
634         /*
635          * Either no secondary cache or the available caches don't have the
636          * subset property so we have to flush the primary caches
637          * explicitly
638          */
639         if (size >= dcache_size) {
640                 r4k_blast_dcache();
641         } else {
642                 unsigned long dc_lsize = current_cpu_data.dcache.linesz;
643
644                 R4600_HIT_CACHEOP_WAR_IMPL;
645                 a = addr & ~(dc_lsize - 1);
646                 end = (addr + size - 1) & ~(dc_lsize - 1);
647                 while (1) {
648                         flush_dcache_line(a);   /* Hit_Writeback_Inv_D */
649                         if (a == end)
650                                 break;
651                         a += dc_lsize;
652                 }
653         }
654
655         bc_wback_inv(addr, size);
656 }
657
658 static void r4k_dma_cache_inv(unsigned long addr, unsigned long size)
659 {
660         unsigned long end, a;
661
662         /* Catch bad driver code */
663         BUG_ON(size == 0);
664
665         if (cpu_has_subset_pcaches) {
666                 unsigned long sc_lsize = current_cpu_data.scache.linesz;
667
668                 if (size >= scache_size) {
669                         r4k_blast_scache();
670                         return;
671                 }
672
673                 a = addr & ~(sc_lsize - 1);
674                 end = (addr + size - 1) & ~(sc_lsize - 1);
675                 while (1) {
676                         flush_scache_line(a);   /* Hit_Writeback_Inv_SD */
677                         if (a == end)
678                                 break;
679                         a += sc_lsize;
680                 }
681                 return;
682         }
683
684         if (size >= dcache_size) {
685                 r4k_blast_dcache();
686         } else {
687                 unsigned long dc_lsize = current_cpu_data.dcache.linesz;
688
689                 R4600_HIT_CACHEOP_WAR_IMPL;
690                 a = addr & ~(dc_lsize - 1);
691                 end = (addr + size - 1) & ~(dc_lsize - 1);
692                 while (1) {
693                         flush_dcache_line(a);   /* Hit_Writeback_Inv_D */
694                         if (a == end)
695                                 break;
696                         a += dc_lsize;
697                 }
698         }
699
700         bc_inv(addr, size);
701 }
702 #endif /* CONFIG_DMA_NONCOHERENT */
703
704 /*
705  * While we're protected against bad userland addresses we don't care
706  * very much about what happens in that case.  Usually a segmentation
707  * fault will dump the process later on anyway ...
708  */
709 static void local_r4k_flush_cache_sigtramp(void * arg)
710 {
711         unsigned long ic_lsize = current_cpu_data.icache.linesz;
712         unsigned long dc_lsize = current_cpu_data.dcache.linesz;
713         unsigned long sc_lsize = current_cpu_data.scache.linesz;
714         unsigned long addr = (unsigned long) arg;
715
716         R4600_HIT_CACHEOP_WAR_IMPL;
717         protected_writeback_dcache_line(addr & ~(dc_lsize - 1));
718         if (!cpu_icache_snoops_remote_store)
719                 protected_writeback_scache_line(addr & ~(sc_lsize - 1));
720         protected_flush_icache_line(addr & ~(ic_lsize - 1));
721         if (MIPS4K_ICACHE_REFILL_WAR) {
722                 __asm__ __volatile__ (
723                         ".set push\n\t"
724                         ".set noat\n\t"
725                         ".set mips3\n\t"
726 #ifdef CONFIG_MIPS32
727                         "la     $at,1f\n\t"
728 #endif
729 #ifdef CONFIG_MIPS64
730                         "dla    $at,1f\n\t"
731 #endif
732                         "cache  %0,($at)\n\t"
733                         "nop; nop; nop\n"
734                         "1:\n\t"
735                         ".set pop"
736                         :
737                         : "i" (Hit_Invalidate_I));
738         }
739         if (MIPS_CACHE_SYNC_WAR)
740                 __asm__ __volatile__ ("sync");
741 }
742
743 static void r4k_flush_cache_sigtramp(unsigned long addr)
744 {
745         on_each_cpu(local_r4k_flush_cache_sigtramp, (void *) addr, 1, 1);
746 }
747
748 static void r4k_flush_icache_all(void)
749 {
750         if (cpu_has_vtag_icache)
751                 r4k_blast_icache();
752 }
753
754 static inline void rm7k_erratum31(void)
755 {
756         const unsigned long ic_lsize = 32;
757         unsigned long addr;
758
759         /* RM7000 erratum #31. The icache is screwed at startup. */
760         write_c0_taglo(0);
761         write_c0_taghi(0);
762
763         for (addr = INDEX_BASE; addr <= INDEX_BASE + 4096; addr += ic_lsize) {
764                 __asm__ __volatile__ (
765                         ".set noreorder\n\t"
766                         ".set mips3\n\t"
767                         "cache\t%1, 0(%0)\n\t"
768                         "cache\t%1, 0x1000(%0)\n\t"
769                         "cache\t%1, 0x2000(%0)\n\t"
770                         "cache\t%1, 0x3000(%0)\n\t"
771                         "cache\t%2, 0(%0)\n\t"
772                         "cache\t%2, 0x1000(%0)\n\t"
773                         "cache\t%2, 0x2000(%0)\n\t"
774                         "cache\t%2, 0x3000(%0)\n\t"
775                         "cache\t%1, 0(%0)\n\t"
776                         "cache\t%1, 0x1000(%0)\n\t"
777                         "cache\t%1, 0x2000(%0)\n\t"
778                         "cache\t%1, 0x3000(%0)\n\t"
779                         ".set\tmips0\n\t"
780                         ".set\treorder\n\t"
781                         :
782                         : "r" (addr), "i" (Index_Store_Tag_I), "i" (Fill));
783         }
784 }
785
786 static char *way_string[] __initdata = { NULL, "direct mapped", "2-way",
787         "3-way", "4-way", "5-way", "6-way", "7-way", "8-way"
788 };
789
790 static void __init probe_pcache(void)
791 {
792         struct cpuinfo_mips *c = &current_cpu_data;
793         unsigned int config = read_c0_config();
794         unsigned int prid = read_c0_prid();
795         unsigned long config1;
796         unsigned int lsize;
797
798         switch (c->cputype) {
799         case CPU_R4600:                 /* QED style two way caches? */
800         case CPU_R4700:
801         case CPU_R5000:
802         case CPU_NEVADA:
803                 icache_size = 1 << (12 + ((config & CONF_IC) >> 9));
804                 c->icache.linesz = 16 << ((config & CONF_IB) >> 5);
805                 c->icache.ways = 2;
806                 c->icache.waybit = ffs(icache_size/2) - 1;
807
808                 dcache_size = 1 << (12 + ((config & CONF_DC) >> 6));
809                 c->dcache.linesz = 16 << ((config & CONF_DB) >> 4);
810                 c->dcache.ways = 2;
811                 c->dcache.waybit= ffs(dcache_size/2) - 1;
812
813                 c->options |= MIPS_CPU_CACHE_CDEX_P;
814                 break;
815
816         case CPU_R5432:
817         case CPU_R5500:
818                 icache_size = 1 << (12 + ((config & CONF_IC) >> 9));
819                 c->icache.linesz = 16 << ((config & CONF_IB) >> 5);
820                 c->icache.ways = 2;
821                 c->icache.waybit= 0;
822
823                 dcache_size = 1 << (12 + ((config & CONF_DC) >> 6));
824                 c->dcache.linesz = 16 << ((config & CONF_DB) >> 4);
825                 c->dcache.ways = 2;
826                 c->dcache.waybit = 0;
827
828                 c->options |= MIPS_CPU_CACHE_CDEX_P;
829                 break;
830
831         case CPU_TX49XX:
832                 icache_size = 1 << (12 + ((config & CONF_IC) >> 9));
833                 c->icache.linesz = 16 << ((config & CONF_IB) >> 5);
834                 c->icache.ways = 4;
835                 c->icache.waybit= 0;
836
837                 dcache_size = 1 << (12 + ((config & CONF_DC) >> 6));
838                 c->dcache.linesz = 16 << ((config & CONF_DB) >> 4);
839                 c->dcache.ways = 4;
840                 c->dcache.waybit = 0;
841
842                 c->options |= MIPS_CPU_CACHE_CDEX_P;
843                 break;
844
845         case CPU_R4000PC:
846         case CPU_R4000SC:
847         case CPU_R4000MC:
848         case CPU_R4400PC:
849         case CPU_R4400SC:
850         case CPU_R4400MC:
851         case CPU_R4300:
852                 icache_size = 1 << (12 + ((config & CONF_IC) >> 9));
853                 c->icache.linesz = 16 << ((config & CONF_IB) >> 5);
854                 c->icache.ways = 1;
855                 c->icache.waybit = 0;   /* doesn't matter */
856
857                 dcache_size = 1 << (12 + ((config & CONF_DC) >> 6));
858                 c->dcache.linesz = 16 << ((config & CONF_DB) >> 4);
859                 c->dcache.ways = 1;
860                 c->dcache.waybit = 0;   /* does not matter */
861
862                 c->options |= MIPS_CPU_CACHE_CDEX_P;
863                 break;
864
865         case CPU_R10000:
866         case CPU_R12000:
867                 icache_size = 1 << (12 + ((config & R10K_CONF_IC) >> 29));
868                 c->icache.linesz = 64;
869                 c->icache.ways = 2;
870                 c->icache.waybit = 0;
871
872                 dcache_size = 1 << (12 + ((config & R10K_CONF_DC) >> 26));
873                 c->dcache.linesz = 32;
874                 c->dcache.ways = 2;
875                 c->dcache.waybit = 0;
876
877                 c->options |= MIPS_CPU_PREFETCH;
878                 break;
879
880         case CPU_VR4133:
881                 write_c0_config(config & ~CONF_EB);
882         case CPU_VR4131:
883                 /* Workaround for cache instruction bug of VR4131 */
884                 if (c->processor_id == 0x0c80U || c->processor_id == 0x0c81U ||
885                     c->processor_id == 0x0c82U) {
886                         config &= ~0x00000030U;
887                         config |= 0x00410000U;
888                         write_c0_config(config);
889                 }
890                 icache_size = 1 << (10 + ((config & CONF_IC) >> 9));
891                 c->icache.linesz = 16 << ((config & CONF_IB) >> 5);
892                 c->icache.ways = 2;
893                 c->icache.waybit = ffs(icache_size/2) - 1;
894
895                 dcache_size = 1 << (10 + ((config & CONF_DC) >> 6));
896                 c->dcache.linesz = 16 << ((config & CONF_DB) >> 4);
897                 c->dcache.ways = 2;
898                 c->dcache.waybit = ffs(dcache_size/2) - 1;
899
900                 c->options |= MIPS_CPU_CACHE_CDEX_P;
901                 break;
902
903         case CPU_VR41XX:
904         case CPU_VR4111:
905         case CPU_VR4121:
906         case CPU_VR4122:
907         case CPU_VR4181:
908         case CPU_VR4181A:
909                 icache_size = 1 << (10 + ((config & CONF_IC) >> 9));
910                 c->icache.linesz = 16 << ((config & CONF_IB) >> 5);
911                 c->icache.ways = 1;
912                 c->icache.waybit = 0;   /* doesn't matter */
913
914                 dcache_size = 1 << (10 + ((config & CONF_DC) >> 6));
915                 c->dcache.linesz = 16 << ((config & CONF_DB) >> 4);
916                 c->dcache.ways = 1;
917                 c->dcache.waybit = 0;   /* does not matter */
918
919                 c->options |= MIPS_CPU_CACHE_CDEX_P;
920                 break;
921
922         case CPU_RM7000:
923                 rm7k_erratum31();
924
925         case CPU_RM9000:
926                 icache_size = 1 << (12 + ((config & CONF_IC) >> 9));
927                 c->icache.linesz = 16 << ((config & CONF_IB) >> 5);
928                 c->icache.ways = 4;
929                 c->icache.waybit = ffs(icache_size / c->icache.ways) - 1;
930
931                 dcache_size = 1 << (12 + ((config & CONF_DC) >> 6));
932                 c->dcache.linesz = 16 << ((config & CONF_DB) >> 4);
933                 c->dcache.ways = 4;
934                 c->dcache.waybit = ffs(dcache_size / c->dcache.ways) - 1;
935
936 #if !defined(CONFIG_SMP) || !defined(RM9000_CDEX_SMP_WAR)
937                 c->options |= MIPS_CPU_CACHE_CDEX_P;
938 #endif
939                 c->options |= MIPS_CPU_PREFETCH;
940                 break;
941
942         default:
943                 if (!(config & MIPS_CONF_M))
944                         panic("Don't know how to probe P-caches on this cpu.");
945
946                 /*
947                  * So we seem to be a MIPS32 or MIPS64 CPU
948                  * So let's probe the I-cache ...
949                  */
950                 config1 = read_c0_config1();
951
952                 if ((lsize = ((config1 >> 19) & 7)))
953                         c->icache.linesz = 2 << lsize;
954                 else
955                         c->icache.linesz = lsize;
956                 c->icache.sets = 64 << ((config1 >> 22) & 7);
957                 c->icache.ways = 1 + ((config1 >> 16) & 7);
958
959                 icache_size = c->icache.sets *
960                               c->icache.ways *
961                               c->icache.linesz;
962                 c->icache.waybit = ffs(icache_size/c->icache.ways) - 1;
963
964                 if (config & 0x8)               /* VI bit */
965                         c->icache.flags |= MIPS_CACHE_VTAG;
966
967                 /*
968                  * Now probe the MIPS32 / MIPS64 data cache.
969                  */
970                 c->dcache.flags = 0;
971
972                 if ((lsize = ((config1 >> 10) & 7)))
973                         c->dcache.linesz = 2 << lsize;
974                 else
975                         c->dcache.linesz= lsize;
976                 c->dcache.sets = 64 << ((config1 >> 13) & 7);
977                 c->dcache.ways = 1 + ((config1 >> 7) & 7);
978
979                 dcache_size = c->dcache.sets *
980                               c->dcache.ways *
981                               c->dcache.linesz;
982                 c->dcache.waybit = ffs(dcache_size/c->dcache.ways) - 1;
983
984                 c->options |= MIPS_CPU_PREFETCH;
985                 break;
986         }
987
988         /*
989          * Processor configuration sanity check for the R4000SC erratum
990          * #5.  With page sizes larger than 32kB there is no possibility
991          * to get a VCE exception anymore so we don't care about this
992          * misconfiguration.  The case is rather theoretical anyway;
993          * presumably no vendor is shipping his hardware in the "bad"
994          * configuration.
995          */
996         if ((prid & 0xff00) == PRID_IMP_R4000 && (prid & 0xff) < 0x40 &&
997             !(config & CONF_SC) && c->icache.linesz != 16 &&
998             PAGE_SIZE <= 0x8000)
999                 panic("Improper R4000SC processor configuration detected");
1000
1001         /* compute a couple of other cache variables */
1002         c->icache.waysize = icache_size / c->icache.ways;
1003         c->dcache.waysize = dcache_size / c->dcache.ways;
1004
1005         c->icache.sets = icache_size / (c->icache.linesz * c->icache.ways);
1006         c->dcache.sets = dcache_size / (c->dcache.linesz * c->dcache.ways);
1007
1008         /*
1009          * R10000 and R12000 P-caches are odd in a positive way.  They're 32kB
1010          * 2-way virtually indexed so normally would suffer from aliases.  So
1011          * normally they'd suffer from aliases but magic in the hardware deals
1012          * with that for us so we don't need to take care ourselves.
1013          */
1014         if (c->cputype != CPU_R10000 && c->cputype != CPU_R12000)
1015                 if (c->dcache.waysize > PAGE_SIZE)
1016                         c->dcache.flags |= MIPS_CACHE_ALIASES;
1017
1018         switch (c->cputype) {
1019         case CPU_20KC:
1020                 /*
1021                  * Some older 20Kc chips doesn't have the 'VI' bit in
1022                  * the config register.
1023                  */
1024                 c->icache.flags |= MIPS_CACHE_VTAG;
1025                 break;
1026
1027         case CPU_AU1500:
1028                 c->icache.flags |= MIPS_CACHE_IC_F_DC;
1029                 break;
1030         }
1031
1032         printk("Primary instruction cache %ldkB, %s, %s, linesize %d bytes.\n",
1033                icache_size >> 10,
1034                cpu_has_vtag_icache ? "virtually tagged" : "physically tagged",
1035                way_string[c->icache.ways], c->icache.linesz);
1036
1037         printk("Primary data cache %ldkB, %s, linesize %d bytes.\n",
1038                dcache_size >> 10, way_string[c->dcache.ways], c->dcache.linesz);
1039 }
1040
1041 /*
1042  * If you even _breathe_ on this function, look at the gcc output and make sure
1043  * it does not pop things on and off the stack for the cache sizing loop that
1044  * executes in KSEG1 space or else you will crash and burn badly.  You have
1045  * been warned.
1046  */
1047 static int __init probe_scache(void)
1048 {
1049         extern unsigned long stext;
1050         unsigned long flags, addr, begin, end, pow2;
1051         unsigned int config = read_c0_config();
1052         struct cpuinfo_mips *c = &current_cpu_data;
1053         int tmp;
1054
1055         if (config & CONF_SC)
1056                 return 0;
1057
1058         begin = (unsigned long) &stext;
1059         begin &= ~((4 * 1024 * 1024) - 1);
1060         end = begin + (4 * 1024 * 1024);
1061
1062         /*
1063          * This is such a bitch, you'd think they would make it easy to do
1064          * this.  Away you daemons of stupidity!
1065          */
1066         local_irq_save(flags);
1067
1068         /* Fill each size-multiple cache line with a valid tag. */
1069         pow2 = (64 * 1024);
1070         for (addr = begin; addr < end; addr = (begin + pow2)) {
1071                 unsigned long *p = (unsigned long *) addr;
1072                 __asm__ __volatile__("nop" : : "r" (*p)); /* whee... */
1073                 pow2 <<= 1;
1074         }
1075
1076         /* Load first line with zero (therefore invalid) tag. */
1077         write_c0_taglo(0);
1078         write_c0_taghi(0);
1079         __asm__ __volatile__("nop; nop; nop; nop;"); /* avoid the hazard */
1080         cache_op(Index_Store_Tag_I, begin);
1081         cache_op(Index_Store_Tag_D, begin);
1082         cache_op(Index_Store_Tag_SD, begin);
1083
1084         /* Now search for the wrap around point. */
1085         pow2 = (128 * 1024);
1086         tmp = 0;
1087         for (addr = begin + (128 * 1024); addr < end; addr = begin + pow2) {
1088                 cache_op(Index_Load_Tag_SD, addr);
1089                 __asm__ __volatile__("nop; nop; nop; nop;"); /* hazard... */
1090                 if (!read_c0_taglo())
1091                         break;
1092                 pow2 <<= 1;
1093         }
1094         local_irq_restore(flags);
1095         addr -= begin;
1096
1097         scache_size = addr;
1098         c->scache.linesz = 16 << ((config & R4K_CONF_SB) >> 22);
1099         c->scache.ways = 1;
1100         c->dcache.waybit = 0;           /* does not matter */
1101
1102         return 1;
1103 }
1104
1105 typedef int (*probe_func_t)(unsigned long);
1106 extern int r5k_sc_init(void);
1107 extern int rm7k_sc_init(void);
1108
1109 static void __init setup_scache(void)
1110 {
1111         struct cpuinfo_mips *c = &current_cpu_data;
1112         unsigned int config = read_c0_config();
1113         probe_func_t probe_scache_kseg1;
1114         int sc_present = 0;
1115
1116         /*
1117          * Do the probing thing on R4000SC and R4400SC processors.  Other
1118          * processors don't have a S-cache that would be relevant to the
1119          * Linux memory managment.
1120          */
1121         switch (c->cputype) {
1122         case CPU_R4000SC:
1123         case CPU_R4000MC:
1124         case CPU_R4400SC:
1125         case CPU_R4400MC:
1126                 probe_scache_kseg1 = (probe_func_t) (CKSEG1ADDR(&probe_scache));
1127                 sc_present = probe_scache_kseg1(config);
1128                 if (sc_present)
1129                         c->options |= MIPS_CPU_CACHE_CDEX_S;
1130                 break;
1131
1132         case CPU_R10000:
1133         case CPU_R12000:
1134                 scache_size = 0x80000 << ((config & R10K_CONF_SS) >> 16);
1135                 c->scache.linesz = 64 << ((config >> 13) & 1);
1136                 c->scache.ways = 2;
1137                 c->scache.waybit= 0;
1138                 sc_present = 1;
1139                 break;
1140
1141         case CPU_R5000:
1142         case CPU_NEVADA:
1143 #ifdef CONFIG_R5000_CPU_SCACHE
1144                 r5k_sc_init();
1145 #endif
1146                 return;
1147
1148         case CPU_RM7000:
1149         case CPU_RM9000:
1150 #ifdef CONFIG_RM7000_CPU_SCACHE
1151                 rm7k_sc_init();
1152 #endif
1153                 return;
1154
1155         default:
1156                 sc_present = 0;
1157         }
1158
1159         if (!sc_present)
1160                 return;
1161
1162         if ((c->isa_level == MIPS_CPU_ISA_M32 ||
1163              c->isa_level == MIPS_CPU_ISA_M64) &&
1164             !(c->scache.flags & MIPS_CACHE_NOT_PRESENT))
1165                 panic("Dunno how to handle MIPS32 / MIPS64 second level cache");
1166
1167         /* compute a couple of other cache variables */
1168         c->scache.waysize = scache_size / c->scache.ways;
1169
1170         c->scache.sets = scache_size / (c->scache.linesz * c->scache.ways);
1171
1172         printk("Unified secondary cache %ldkB %s, linesize %d bytes.\n",
1173                scache_size >> 10, way_string[c->scache.ways], c->scache.linesz);
1174
1175         c->options |= MIPS_CPU_SUBSET_CACHES;
1176 }
1177
1178 static inline void coherency_setup(void)
1179 {
1180         change_c0_config(CONF_CM_CMASK, CONF_CM_DEFAULT);
1181
1182         /*
1183          * c0_status.cu=0 specifies that updates by the sc instruction use
1184          * the coherency mode specified by the TLB; 1 means cachable
1185          * coherent update on write will be used.  Not all processors have
1186          * this bit and; some wire it to zero, others like Toshiba had the
1187          * silly idea of putting something else there ...
1188          */
1189         switch (current_cpu_data.cputype) {
1190         case CPU_R4000PC:
1191         case CPU_R4000SC:
1192         case CPU_R4000MC:
1193         case CPU_R4400PC:
1194         case CPU_R4400SC:
1195         case CPU_R4400MC:
1196                 clear_c0_config(CONF_CU);
1197                 break;
1198         }
1199 }
1200
1201 void __init ld_mmu_r4xx0(void)
1202 {
1203         extern void build_clear_page(void);
1204         extern void build_copy_page(void);
1205         extern char except_vec2_generic;
1206         struct cpuinfo_mips *c = &current_cpu_data;
1207
1208         /* Default cache error handler for R4000 and R5000 family */
1209         memcpy((void *)(CAC_BASE   + 0x100), &except_vec2_generic, 0x80);
1210         memcpy((void *)(UNCAC_BASE + 0x100), &except_vec2_generic, 0x80);
1211
1212         probe_pcache();
1213         setup_scache();
1214
1215         if (c->dcache.sets * c->dcache.ways > PAGE_SIZE)
1216                 c->dcache.flags |= MIPS_CACHE_ALIASES;
1217
1218         r4k_blast_dcache_page_setup();
1219         r4k_blast_dcache_page_indexed_setup();
1220         r4k_blast_dcache_setup();
1221         r4k_blast_icache_page_setup();
1222         r4k_blast_icache_page_indexed_setup();
1223         r4k_blast_icache_setup();
1224         r4k_blast_scache_page_setup();
1225         r4k_blast_scache_page_indexed_setup();
1226         r4k_blast_scache_setup();
1227
1228         /*
1229          * Some MIPS32 and MIPS64 processors have physically indexed caches.
1230          * This code supports virtually indexed processors and will be
1231          * unnecessarily inefficient on physically indexed processors.
1232          */
1233         shm_align_mask = max_t( unsigned long,
1234                                 c->dcache.sets * c->dcache.linesz - 1,
1235                                 PAGE_SIZE - 1);
1236
1237         flush_cache_all         = r4k_flush_cache_all;
1238         __flush_cache_all       = r4k___flush_cache_all;
1239         flush_cache_mm          = r4k_flush_cache_mm;
1240         flush_cache_page        = r4k_flush_cache_page;
1241         flush_icache_page       = r4k_flush_icache_page;
1242         flush_cache_range       = r4k_flush_cache_range;
1243
1244         flush_cache_sigtramp    = r4k_flush_cache_sigtramp;
1245         flush_icache_all        = r4k_flush_icache_all;
1246         flush_data_cache_page   = r4k_flush_data_cache_page;
1247         flush_icache_range      = r4k_flush_icache_range;
1248
1249 #ifdef CONFIG_DMA_NONCOHERENT
1250         _dma_cache_wback_inv    = r4k_dma_cache_wback_inv;
1251         _dma_cache_wback        = r4k_dma_cache_wback_inv;
1252         _dma_cache_inv          = r4k_dma_cache_inv;
1253 #endif
1254
1255         __flush_cache_all();
1256         coherency_setup();
1257
1258         build_clear_page();
1259         build_copy_page();
1260 }