vserver 1.9.5.x5
[linux-2.6.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,
430         unsigned long page)
431 {
432         struct flush_cache_page_args args;
433
434         /*
435          * If ownes no valid ASID yet, cannot possibly have gotten
436          * this page into the cache.
437          */
438         if (cpu_context(smp_processor_id(), vma->vm_mm) == 0)
439                 return;
440
441         args.vma = vma;
442         args.page = page;
443
444         on_each_cpu(local_r4k_flush_cache_page, &args, 1, 1);
445 }
446
447 static inline void local_r4k_flush_data_cache_page(void * addr)
448 {
449         r4k_blast_dcache_page((unsigned long) addr);
450 }
451
452 static void r4k_flush_data_cache_page(unsigned long addr)
453 {
454         on_each_cpu(local_r4k_flush_data_cache_page, (void *) addr, 1, 1);
455 }
456
457 struct flush_icache_range_args {
458         unsigned long start;
459         unsigned long end;
460 };
461
462 static inline void local_r4k_flush_icache_range(void *args)
463 {
464         struct flush_icache_range_args *fir_args = args;
465         unsigned long dc_lsize = current_cpu_data.dcache.linesz;
466         unsigned long ic_lsize = current_cpu_data.icache.linesz;
467         unsigned long sc_lsize = current_cpu_data.scache.linesz;
468         unsigned long start = fir_args->start;
469         unsigned long end = fir_args->end;
470         unsigned long addr, aend;
471
472         if (!cpu_has_ic_fills_f_dc) {
473                 if (end - start > dcache_size) {
474                         r4k_blast_dcache();
475                 } else {
476                         addr = start & ~(dc_lsize - 1);
477                         aend = (end - 1) & ~(dc_lsize - 1);
478
479                         while (1) {
480                                 /* Hit_Writeback_Inv_D */
481                                 protected_writeback_dcache_line(addr);
482                                 if (addr == aend)
483                                         break;
484                                 addr += dc_lsize;
485                         }
486                 }
487
488                 if (!cpu_icache_snoops_remote_store) {
489                         if (end - start > scache_size) {
490                                 r4k_blast_scache();
491                         } else {
492                                 addr = start & ~(sc_lsize - 1);
493                                 aend = (end - 1) & ~(sc_lsize - 1);
494
495                                 while (1) {
496                                         /* Hit_Writeback_Inv_D */
497                                         protected_writeback_scache_line(addr);
498                                         if (addr == aend)
499                                                 break;
500                                         addr += sc_lsize;
501                                 }
502                         }
503                 }
504         }
505
506         if (end - start > icache_size)
507                 r4k_blast_icache();
508         else {
509                 addr = start & ~(ic_lsize - 1);
510                 aend = (end - 1) & ~(ic_lsize - 1);
511                 while (1) {
512                         /* Hit_Invalidate_I */
513                         protected_flush_icache_line(addr);
514                         if (addr == aend)
515                                 break;
516                         addr += ic_lsize;
517                 }
518         }
519 }
520
521 static void r4k_flush_icache_range(unsigned long start, unsigned long end)
522 {
523         struct flush_icache_range_args args;
524
525         args.start = start;
526         args.end = end;
527
528         on_each_cpu(local_r4k_flush_icache_range, &args, 1, 1);
529 }
530
531 /*
532  * Ok, this seriously sucks.  We use them to flush a user page but don't
533  * know the virtual address, so we have to blast away the whole icache
534  * which is significantly more expensive than the real thing.  Otoh we at
535  * least know the kernel address of the page so we can flush it
536  * selectivly.
537  */
538
539 struct flush_icache_page_args {
540         struct vm_area_struct *vma;
541         struct page *page;
542 };
543
544 static inline void local_r4k_flush_icache_page(void *args)
545 {
546         struct flush_icache_page_args *fip_args = args;
547         struct vm_area_struct *vma = fip_args->vma;
548         struct page *page = fip_args->page;
549
550         /*
551          * Tricky ...  Because we don't know the virtual address we've got the
552          * choice of either invalidating the entire primary and secondary
553          * caches or invalidating the secondary caches also.  With the subset
554          * enforcment on R4000SC, R4400SC, R10000 and R12000 invalidating the
555          * secondary cache will result in any entries in the primary caches
556          * also getting invalidated which hopefully is a bit more economical.
557          */
558         if (cpu_has_subset_pcaches) {
559                 unsigned long addr = (unsigned long) page_address(page);
560
561                 r4k_blast_scache_page(addr);
562                 ClearPageDcacheDirty(page);
563
564                 return;
565         }
566
567         if (!cpu_has_ic_fills_f_dc) {
568                 unsigned long addr = (unsigned long) page_address(page);
569                 r4k_blast_dcache_page(addr);
570                 if (!cpu_icache_snoops_remote_store)
571                         r4k_blast_scache_page(addr);
572                 ClearPageDcacheDirty(page);
573         }
574
575         /*
576          * We're not sure of the virtual address(es) involved here, so
577          * we have to flush the entire I-cache.
578          */
579         if (cpu_has_vtag_icache) {
580                 int cpu = smp_processor_id();
581
582                 if (cpu_context(cpu, vma->vm_mm) != 0)
583                         drop_mmu_context(vma->vm_mm, cpu);
584         } else
585                 r4k_blast_icache();
586 }
587
588 static void r4k_flush_icache_page(struct vm_area_struct *vma,
589         struct page *page)
590 {
591         struct flush_icache_page_args args;
592
593         /*
594          * If there's no context yet, or the page isn't executable, no I-cache
595          * flush is needed.
596          */
597         if (!(vma->vm_flags & VM_EXEC))
598                 return;
599
600         args.vma = vma;
601         args.page = page;
602
603         on_each_cpu(local_r4k_flush_icache_page, &args, 1, 1);
604 }
605
606
607 #ifdef CONFIG_DMA_NONCOHERENT
608
609 static void r4k_dma_cache_wback_inv(unsigned long addr, unsigned long size)
610 {
611         unsigned long end, a;
612
613         /* Catch bad driver code */
614         BUG_ON(size == 0);
615
616         if (cpu_has_subset_pcaches) {
617                 unsigned long sc_lsize = current_cpu_data.scache.linesz;
618
619                 if (size >= scache_size) {
620                         r4k_blast_scache();
621                         return;
622                 }
623
624                 a = addr & ~(sc_lsize - 1);
625                 end = (addr + size - 1) & ~(sc_lsize - 1);
626                 while (1) {
627                         flush_scache_line(a);   /* Hit_Writeback_Inv_SD */
628                         if (a == end)
629                                 break;
630                         a += sc_lsize;
631                 }
632                 return;
633         }
634
635         /*
636          * Either no secondary cache or the available caches don't have the
637          * subset property so we have to flush the primary caches
638          * explicitly
639          */
640         if (size >= dcache_size) {
641                 r4k_blast_dcache();
642         } else {
643                 unsigned long dc_lsize = current_cpu_data.dcache.linesz;
644
645                 R4600_HIT_CACHEOP_WAR_IMPL;
646                 a = addr & ~(dc_lsize - 1);
647                 end = (addr + size - 1) & ~(dc_lsize - 1);
648                 while (1) {
649                         flush_dcache_line(a);   /* Hit_Writeback_Inv_D */
650                         if (a == end)
651                                 break;
652                         a += dc_lsize;
653                 }
654         }
655
656         bc_wback_inv(addr, size);
657 }
658
659 static void r4k_dma_cache_inv(unsigned long addr, unsigned long size)
660 {
661         unsigned long end, a;
662
663         /* Catch bad driver code */
664         BUG_ON(size == 0);
665
666         if (cpu_has_subset_pcaches) {
667                 unsigned long sc_lsize = current_cpu_data.scache.linesz;
668
669                 if (size >= scache_size) {
670                         r4k_blast_scache();
671                         return;
672                 }
673
674                 a = addr & ~(sc_lsize - 1);
675                 end = (addr + size - 1) & ~(sc_lsize - 1);
676                 while (1) {
677                         flush_scache_line(a);   /* Hit_Writeback_Inv_SD */
678                         if (a == end)
679                                 break;
680                         a += sc_lsize;
681                 }
682                 return;
683         }
684
685         if (size >= dcache_size) {
686                 r4k_blast_dcache();
687         } else {
688                 unsigned long dc_lsize = current_cpu_data.dcache.linesz;
689
690                 R4600_HIT_CACHEOP_WAR_IMPL;
691                 a = addr & ~(dc_lsize - 1);
692                 end = (addr + size - 1) & ~(dc_lsize - 1);
693                 while (1) {
694                         flush_dcache_line(a);   /* Hit_Writeback_Inv_D */
695                         if (a == end)
696                                 break;
697                         a += dc_lsize;
698                 }
699         }
700
701         bc_inv(addr, size);
702 }
703 #endif /* CONFIG_DMA_NONCOHERENT */
704
705 /*
706  * While we're protected against bad userland addresses we don't care
707  * very much about what happens in that case.  Usually a segmentation
708  * fault will dump the process later on anyway ...
709  */
710 static void local_r4k_flush_cache_sigtramp(void * arg)
711 {
712         unsigned long ic_lsize = current_cpu_data.icache.linesz;
713         unsigned long dc_lsize = current_cpu_data.dcache.linesz;
714         unsigned long sc_lsize = current_cpu_data.scache.linesz;
715         unsigned long addr = (unsigned long) arg;
716
717         R4600_HIT_CACHEOP_WAR_IMPL;
718         protected_writeback_dcache_line(addr & ~(dc_lsize - 1));
719         if (!cpu_icache_snoops_remote_store)
720                 protected_writeback_scache_line(addr & ~(sc_lsize - 1));
721         protected_flush_icache_line(addr & ~(ic_lsize - 1));
722         if (MIPS4K_ICACHE_REFILL_WAR) {
723                 __asm__ __volatile__ (
724                         ".set push\n\t"
725                         ".set noat\n\t"
726                         ".set mips3\n\t"
727 #ifdef CONFIG_MIPS32
728                         "la     $at,1f\n\t"
729 #endif
730 #ifdef CONFIG_MIPS64
731                         "dla    $at,1f\n\t"
732 #endif
733                         "cache  %0,($at)\n\t"
734                         "nop; nop; nop\n"
735                         "1:\n\t"
736                         ".set pop"
737                         :
738                         : "i" (Hit_Invalidate_I));
739         }
740         if (MIPS_CACHE_SYNC_WAR)
741                 __asm__ __volatile__ ("sync");
742 }
743
744 static void r4k_flush_cache_sigtramp(unsigned long addr)
745 {
746         on_each_cpu(local_r4k_flush_cache_sigtramp, (void *) addr, 1, 1);
747 }
748
749 static void r4k_flush_icache_all(void)
750 {
751         if (cpu_has_vtag_icache)
752                 r4k_blast_icache();
753 }
754
755 static inline void rm7k_erratum31(void)
756 {
757         const unsigned long ic_lsize = 32;
758         unsigned long addr;
759
760         /* RM7000 erratum #31. The icache is screwed at startup. */
761         write_c0_taglo(0);
762         write_c0_taghi(0);
763
764         for (addr = INDEX_BASE; addr <= INDEX_BASE + 4096; addr += ic_lsize) {
765                 __asm__ __volatile__ (
766                         ".set noreorder\n\t"
767                         ".set mips3\n\t"
768                         "cache\t%1, 0(%0)\n\t"
769                         "cache\t%1, 0x1000(%0)\n\t"
770                         "cache\t%1, 0x2000(%0)\n\t"
771                         "cache\t%1, 0x3000(%0)\n\t"
772                         "cache\t%2, 0(%0)\n\t"
773                         "cache\t%2, 0x1000(%0)\n\t"
774                         "cache\t%2, 0x2000(%0)\n\t"
775                         "cache\t%2, 0x3000(%0)\n\t"
776                         "cache\t%1, 0(%0)\n\t"
777                         "cache\t%1, 0x1000(%0)\n\t"
778                         "cache\t%1, 0x2000(%0)\n\t"
779                         "cache\t%1, 0x3000(%0)\n\t"
780                         ".set\tmips0\n\t"
781                         ".set\treorder\n\t"
782                         :
783                         : "r" (addr), "i" (Index_Store_Tag_I), "i" (Fill));
784         }
785 }
786
787 static char *way_string[] __initdata = { NULL, "direct mapped", "2-way",
788         "3-way", "4-way", "5-way", "6-way", "7-way", "8-way"
789 };
790
791 static void __init probe_pcache(void)
792 {
793         struct cpuinfo_mips *c = &current_cpu_data;
794         unsigned int config = read_c0_config();
795         unsigned int prid = read_c0_prid();
796         unsigned long config1;
797         unsigned int lsize;
798
799         switch (c->cputype) {
800         case CPU_R4600:                 /* QED style two way caches? */
801         case CPU_R4700:
802         case CPU_R5000:
803         case CPU_NEVADA:
804                 icache_size = 1 << (12 + ((config & CONF_IC) >> 9));
805                 c->icache.linesz = 16 << ((config & CONF_IB) >> 5);
806                 c->icache.ways = 2;
807                 c->icache.waybit = ffs(icache_size/2) - 1;
808
809                 dcache_size = 1 << (12 + ((config & CONF_DC) >> 6));
810                 c->dcache.linesz = 16 << ((config & CONF_DB) >> 4);
811                 c->dcache.ways = 2;
812                 c->dcache.waybit= ffs(dcache_size/2) - 1;
813
814                 c->options |= MIPS_CPU_CACHE_CDEX_P;
815                 break;
816
817         case CPU_R5432:
818         case CPU_R5500:
819                 icache_size = 1 << (12 + ((config & CONF_IC) >> 9));
820                 c->icache.linesz = 16 << ((config & CONF_IB) >> 5);
821                 c->icache.ways = 2;
822                 c->icache.waybit= 0;
823
824                 dcache_size = 1 << (12 + ((config & CONF_DC) >> 6));
825                 c->dcache.linesz = 16 << ((config & CONF_DB) >> 4);
826                 c->dcache.ways = 2;
827                 c->dcache.waybit = 0;
828
829                 c->options |= MIPS_CPU_CACHE_CDEX_P;
830                 break;
831
832         case CPU_TX49XX:
833                 icache_size = 1 << (12 + ((config & CONF_IC) >> 9));
834                 c->icache.linesz = 16 << ((config & CONF_IB) >> 5);
835                 c->icache.ways = 4;
836                 c->icache.waybit= 0;
837
838                 dcache_size = 1 << (12 + ((config & CONF_DC) >> 6));
839                 c->dcache.linesz = 16 << ((config & CONF_DB) >> 4);
840                 c->dcache.ways = 4;
841                 c->dcache.waybit = 0;
842
843                 c->options |= MIPS_CPU_CACHE_CDEX_P;
844                 break;
845
846         case CPU_R4000PC:
847         case CPU_R4000SC:
848         case CPU_R4000MC:
849         case CPU_R4400PC:
850         case CPU_R4400SC:
851         case CPU_R4400MC:
852         case CPU_R4300:
853                 icache_size = 1 << (12 + ((config & CONF_IC) >> 9));
854                 c->icache.linesz = 16 << ((config & CONF_IB) >> 5);
855                 c->icache.ways = 1;
856                 c->icache.waybit = 0;   /* doesn't matter */
857
858                 dcache_size = 1 << (12 + ((config & CONF_DC) >> 6));
859                 c->dcache.linesz = 16 << ((config & CONF_DB) >> 4);
860                 c->dcache.ways = 1;
861                 c->dcache.waybit = 0;   /* does not matter */
862
863                 c->options |= MIPS_CPU_CACHE_CDEX_P;
864                 break;
865
866         case CPU_R10000:
867         case CPU_R12000:
868                 icache_size = 1 << (12 + ((config & R10K_CONF_IC) >> 29));
869                 c->icache.linesz = 64;
870                 c->icache.ways = 2;
871                 c->icache.waybit = 0;
872
873                 dcache_size = 1 << (12 + ((config & R10K_CONF_DC) >> 26));
874                 c->dcache.linesz = 32;
875                 c->dcache.ways = 2;
876                 c->dcache.waybit = 0;
877
878                 c->options |= MIPS_CPU_PREFETCH;
879                 break;
880
881         case CPU_VR4133:
882                 write_c0_config(config & ~CONF_EB);
883         case CPU_VR4131:
884                 /* Workaround for cache instruction bug of VR4131 */
885                 if (c->processor_id == 0x0c80U || c->processor_id == 0x0c81U ||
886                     c->processor_id == 0x0c82U) {
887                         config &= ~0x00000030U;
888                         config |= 0x00410000U;
889                         write_c0_config(config);
890                 }
891                 icache_size = 1 << (10 + ((config & CONF_IC) >> 9));
892                 c->icache.linesz = 16 << ((config & CONF_IB) >> 5);
893                 c->icache.ways = 2;
894                 c->icache.waybit = ffs(icache_size/2) - 1;
895
896                 dcache_size = 1 << (10 + ((config & CONF_DC) >> 6));
897                 c->dcache.linesz = 16 << ((config & CONF_DB) >> 4);
898                 c->dcache.ways = 2;
899                 c->dcache.waybit = ffs(dcache_size/2) - 1;
900
901                 c->options |= MIPS_CPU_CACHE_CDEX_P;
902                 break;
903
904         case CPU_VR41XX:
905         case CPU_VR4111:
906         case CPU_VR4121:
907         case CPU_VR4122:
908         case CPU_VR4181:
909         case CPU_VR4181A:
910                 icache_size = 1 << (10 + ((config & CONF_IC) >> 9));
911                 c->icache.linesz = 16 << ((config & CONF_IB) >> 5);
912                 c->icache.ways = 1;
913                 c->icache.waybit = 0;   /* doesn't matter */
914
915                 dcache_size = 1 << (10 + ((config & CONF_DC) >> 6));
916                 c->dcache.linesz = 16 << ((config & CONF_DB) >> 4);
917                 c->dcache.ways = 1;
918                 c->dcache.waybit = 0;   /* does not matter */
919
920                 c->options |= MIPS_CPU_CACHE_CDEX_P;
921                 break;
922
923         case CPU_RM7000:
924                 rm7k_erratum31();
925
926         case CPU_RM9000:
927                 icache_size = 1 << (12 + ((config & CONF_IC) >> 9));
928                 c->icache.linesz = 16 << ((config & CONF_IB) >> 5);
929                 c->icache.ways = 4;
930                 c->icache.waybit = ffs(icache_size / c->icache.ways) - 1;
931
932                 dcache_size = 1 << (12 + ((config & CONF_DC) >> 6));
933                 c->dcache.linesz = 16 << ((config & CONF_DB) >> 4);
934                 c->dcache.ways = 4;
935                 c->dcache.waybit = ffs(dcache_size / c->dcache.ways) - 1;
936
937 #if !defined(CONFIG_SMP) || !defined(RM9000_CDEX_SMP_WAR)
938                 c->options |= MIPS_CPU_CACHE_CDEX_P;
939 #endif
940                 c->options |= MIPS_CPU_PREFETCH;
941                 break;
942
943         default:
944                 if (!(config & MIPS_CONF_M))
945                         panic("Don't know how to probe P-caches on this cpu.");
946
947                 /*
948                  * So we seem to be a MIPS32 or MIPS64 CPU
949                  * So let's probe the I-cache ...
950                  */
951                 config1 = read_c0_config1();
952
953                 if ((lsize = ((config1 >> 19) & 7)))
954                         c->icache.linesz = 2 << lsize;
955                 else
956                         c->icache.linesz = lsize;
957                 c->icache.sets = 64 << ((config1 >> 22) & 7);
958                 c->icache.ways = 1 + ((config1 >> 16) & 7);
959
960                 icache_size = c->icache.sets *
961                               c->icache.ways *
962                               c->icache.linesz;
963                 c->icache.waybit = ffs(icache_size/c->icache.ways) - 1;
964
965                 if (config & 0x8)               /* VI bit */
966                         c->icache.flags |= MIPS_CACHE_VTAG;
967
968                 /*
969                  * Now probe the MIPS32 / MIPS64 data cache.
970                  */
971                 c->dcache.flags = 0;
972
973                 if ((lsize = ((config1 >> 10) & 7)))
974                         c->dcache.linesz = 2 << lsize;
975                 else
976                         c->dcache.linesz= lsize;
977                 c->dcache.sets = 64 << ((config1 >> 13) & 7);
978                 c->dcache.ways = 1 + ((config1 >> 7) & 7);
979
980                 dcache_size = c->dcache.sets *
981                               c->dcache.ways *
982                               c->dcache.linesz;
983                 c->dcache.waybit = ffs(dcache_size/c->dcache.ways) - 1;
984
985                 c->options |= MIPS_CPU_PREFETCH;
986                 break;
987         }
988
989         /*
990          * Processor configuration sanity check for the R4000SC erratum
991          * #5.  With page sizes larger than 32kB there is no possibility
992          * to get a VCE exception anymore so we don't care about this
993          * misconfiguration.  The case is rather theoretical anyway;
994          * presumably no vendor is shipping his hardware in the "bad"
995          * configuration.
996          */
997         if ((prid & 0xff00) == PRID_IMP_R4000 && (prid & 0xff) < 0x40 &&
998             !(config & CONF_SC) && c->icache.linesz != 16 &&
999             PAGE_SIZE <= 0x8000)
1000                 panic("Improper R4000SC processor configuration detected");
1001
1002         /* compute a couple of other cache variables */
1003         c->icache.waysize = icache_size / c->icache.ways;
1004         c->dcache.waysize = dcache_size / c->dcache.ways;
1005
1006         c->icache.sets = icache_size / (c->icache.linesz * c->icache.ways);
1007         c->dcache.sets = dcache_size / (c->dcache.linesz * c->dcache.ways);
1008
1009         /*
1010          * R10000 and R12000 P-caches are odd in a positive way.  They're 32kB
1011          * 2-way virtually indexed so normally would suffer from aliases.  So
1012          * normally they'd suffer from aliases but magic in the hardware deals
1013          * with that for us so we don't need to take care ourselves.
1014          */
1015         if (c->cputype != CPU_R10000 && c->cputype != CPU_R12000)
1016                 if (c->dcache.waysize > PAGE_SIZE)
1017                         c->dcache.flags |= MIPS_CACHE_ALIASES;
1018
1019         switch (c->cputype) {
1020         case CPU_20KC:
1021                 /*
1022                  * Some older 20Kc chips doesn't have the 'VI' bit in
1023                  * the config register.
1024                  */
1025                 c->icache.flags |= MIPS_CACHE_VTAG;
1026                 break;
1027
1028         case CPU_AU1500:
1029                 c->icache.flags |= MIPS_CACHE_IC_F_DC;
1030                 break;
1031         }
1032
1033         printk("Primary instruction cache %ldkB, %s, %s, linesize %d bytes.\n",
1034                icache_size >> 10,
1035                cpu_has_vtag_icache ? "virtually tagged" : "physically tagged",
1036                way_string[c->icache.ways], c->icache.linesz);
1037
1038         printk("Primary data cache %ldkB, %s, linesize %d bytes.\n",
1039                dcache_size >> 10, way_string[c->dcache.ways], c->dcache.linesz);
1040 }
1041
1042 /*
1043  * If you even _breathe_ on this function, look at the gcc output and make sure
1044  * it does not pop things on and off the stack for the cache sizing loop that
1045  * executes in KSEG1 space or else you will crash and burn badly.  You have
1046  * been warned.
1047  */
1048 static int __init probe_scache(void)
1049 {
1050         extern unsigned long stext;
1051         unsigned long flags, addr, begin, end, pow2;
1052         unsigned int config = read_c0_config();
1053         struct cpuinfo_mips *c = &current_cpu_data;
1054         int tmp;
1055
1056         if (config & CONF_SC)
1057                 return 0;
1058
1059         begin = (unsigned long) &stext;
1060         begin &= ~((4 * 1024 * 1024) - 1);
1061         end = begin + (4 * 1024 * 1024);
1062
1063         /*
1064          * This is such a bitch, you'd think they would make it easy to do
1065          * this.  Away you daemons of stupidity!
1066          */
1067         local_irq_save(flags);
1068
1069         /* Fill each size-multiple cache line with a valid tag. */
1070         pow2 = (64 * 1024);
1071         for (addr = begin; addr < end; addr = (begin + pow2)) {
1072                 unsigned long *p = (unsigned long *) addr;
1073                 __asm__ __volatile__("nop" : : "r" (*p)); /* whee... */
1074                 pow2 <<= 1;
1075         }
1076
1077         /* Load first line with zero (therefore invalid) tag. */
1078         write_c0_taglo(0);
1079         write_c0_taghi(0);
1080         __asm__ __volatile__("nop; nop; nop; nop;"); /* avoid the hazard */
1081         cache_op(Index_Store_Tag_I, begin);
1082         cache_op(Index_Store_Tag_D, begin);
1083         cache_op(Index_Store_Tag_SD, begin);
1084
1085         /* Now search for the wrap around point. */
1086         pow2 = (128 * 1024);
1087         tmp = 0;
1088         for (addr = begin + (128 * 1024); addr < end; addr = begin + pow2) {
1089                 cache_op(Index_Load_Tag_SD, addr);
1090                 __asm__ __volatile__("nop; nop; nop; nop;"); /* hazard... */
1091                 if (!read_c0_taglo())
1092                         break;
1093                 pow2 <<= 1;
1094         }
1095         local_irq_restore(flags);
1096         addr -= begin;
1097
1098         scache_size = addr;
1099         c->scache.linesz = 16 << ((config & R4K_CONF_SB) >> 22);
1100         c->scache.ways = 1;
1101         c->dcache.waybit = 0;           /* does not matter */
1102
1103         return 1;
1104 }
1105
1106 typedef int (*probe_func_t)(unsigned long);
1107 extern int r5k_sc_init(void);
1108 extern int rm7k_sc_init(void);
1109
1110 static void __init setup_scache(void)
1111 {
1112         struct cpuinfo_mips *c = &current_cpu_data;
1113         unsigned int config = read_c0_config();
1114         probe_func_t probe_scache_kseg1;
1115         int sc_present = 0;
1116
1117         /*
1118          * Do the probing thing on R4000SC and R4400SC processors.  Other
1119          * processors don't have a S-cache that would be relevant to the
1120          * Linux memory managment.
1121          */
1122         switch (c->cputype) {
1123         case CPU_R4000SC:
1124         case CPU_R4000MC:
1125         case CPU_R4400SC:
1126         case CPU_R4400MC:
1127                 probe_scache_kseg1 = (probe_func_t) (CKSEG1ADDR(&probe_scache));
1128                 sc_present = probe_scache_kseg1(config);
1129                 if (sc_present)
1130                         c->options |= MIPS_CPU_CACHE_CDEX_S;
1131                 break;
1132
1133         case CPU_R10000:
1134         case CPU_R12000:
1135                 scache_size = 0x80000 << ((config & R10K_CONF_SS) >> 16);
1136                 c->scache.linesz = 64 << ((config >> 13) & 1);
1137                 c->scache.ways = 2;
1138                 c->scache.waybit= 0;
1139                 sc_present = 1;
1140                 break;
1141
1142         case CPU_R5000:
1143         case CPU_NEVADA:
1144 #ifdef CONFIG_R5000_CPU_SCACHE
1145                 r5k_sc_init();
1146 #endif
1147                 return;
1148
1149         case CPU_RM7000:
1150         case CPU_RM9000:
1151 #ifdef CONFIG_RM7000_CPU_SCACHE
1152                 rm7k_sc_init();
1153 #endif
1154                 return;
1155
1156         default:
1157                 sc_present = 0;
1158         }
1159
1160         if (!sc_present)
1161                 return;
1162
1163         if ((c->isa_level == MIPS_CPU_ISA_M32 ||
1164              c->isa_level == MIPS_CPU_ISA_M64) &&
1165             !(c->scache.flags & MIPS_CACHE_NOT_PRESENT))
1166                 panic("Dunno how to handle MIPS32 / MIPS64 second level cache");
1167
1168         /* compute a couple of other cache variables */
1169         c->scache.waysize = scache_size / c->scache.ways;
1170
1171         c->scache.sets = scache_size / (c->scache.linesz * c->scache.ways);
1172
1173         printk("Unified secondary cache %ldkB %s, linesize %d bytes.\n",
1174                scache_size >> 10, way_string[c->scache.ways], c->scache.linesz);
1175
1176         c->options |= MIPS_CPU_SUBSET_CACHES;
1177 }
1178
1179 static inline void coherency_setup(void)
1180 {
1181         change_c0_config(CONF_CM_CMASK, CONF_CM_DEFAULT);
1182
1183         /*
1184          * c0_status.cu=0 specifies that updates by the sc instruction use
1185          * the coherency mode specified by the TLB; 1 means cachable
1186          * coherent update on write will be used.  Not all processors have
1187          * this bit and; some wire it to zero, others like Toshiba had the
1188          * silly idea of putting something else there ...
1189          */
1190         switch (current_cpu_data.cputype) {
1191         case CPU_R4000PC:
1192         case CPU_R4000SC:
1193         case CPU_R4000MC:
1194         case CPU_R4400PC:
1195         case CPU_R4400SC:
1196         case CPU_R4400MC:
1197                 clear_c0_config(CONF_CU);
1198                 break;
1199         }
1200 }
1201
1202 void __init ld_mmu_r4xx0(void)
1203 {
1204         extern void build_clear_page(void);
1205         extern void build_copy_page(void);
1206         extern char except_vec2_generic;
1207         struct cpuinfo_mips *c = &current_cpu_data;
1208
1209         /* Default cache error handler for R4000 and R5000 family */
1210         memcpy((void *)(CAC_BASE   + 0x100), &except_vec2_generic, 0x80);
1211         memcpy((void *)(UNCAC_BASE + 0x100), &except_vec2_generic, 0x80);
1212
1213         probe_pcache();
1214         setup_scache();
1215
1216         if (c->dcache.sets * c->dcache.ways > PAGE_SIZE)
1217                 c->dcache.flags |= MIPS_CACHE_ALIASES;
1218
1219         r4k_blast_dcache_page_setup();
1220         r4k_blast_dcache_page_indexed_setup();
1221         r4k_blast_dcache_setup();
1222         r4k_blast_icache_page_setup();
1223         r4k_blast_icache_page_indexed_setup();
1224         r4k_blast_icache_setup();
1225         r4k_blast_scache_page_setup();
1226         r4k_blast_scache_page_indexed_setup();
1227         r4k_blast_scache_setup();
1228
1229         /*
1230          * Some MIPS32 and MIPS64 processors have physically indexed caches.
1231          * This code supports virtually indexed processors and will be
1232          * unnecessarily inefficient on physically indexed processors.
1233          */
1234         shm_align_mask = max_t( unsigned long,
1235                                 c->dcache.sets * c->dcache.linesz - 1,
1236                                 PAGE_SIZE - 1);
1237
1238         flush_cache_all         = r4k_flush_cache_all;
1239         __flush_cache_all       = r4k___flush_cache_all;
1240         flush_cache_mm          = r4k_flush_cache_mm;
1241         flush_cache_page        = r4k_flush_cache_page;
1242         flush_icache_page       = r4k_flush_icache_page;
1243         flush_cache_range       = r4k_flush_cache_range;
1244
1245         flush_cache_sigtramp    = r4k_flush_cache_sigtramp;
1246         flush_icache_all        = r4k_flush_icache_all;
1247         flush_data_cache_page   = r4k_flush_data_cache_page;
1248         flush_icache_range      = r4k_flush_icache_range;
1249
1250 #ifdef CONFIG_DMA_NONCOHERENT
1251         _dma_cache_wback_inv    = r4k_dma_cache_wback_inv;
1252         _dma_cache_wback        = r4k_dma_cache_wback_inv;
1253         _dma_cache_inv          = r4k_dma_cache_inv;
1254 #endif
1255
1256         __flush_cache_all();
1257         coherency_setup();
1258
1259         build_clear_page();
1260         build_copy_page();
1261 }