03100b8a45d98cb4ed560f1916ed09fdc04948cc
[firefly-linux-kernel-4.4.55.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         pud_t *pudp;
376         pmd_t *pmdp;
377         pte_t *ptep;
378
379         /*
380          * If ownes no valid ASID yet, cannot possibly have gotten
381          * this page into the cache.
382          */
383         if (cpu_context(smp_processor_id(), mm) == 0)
384                 return;
385
386         page &= PAGE_MASK;
387         pgdp = pgd_offset(mm, page);
388         pudp = pud_offset(pgdp, page);
389         pmdp = pmd_offset(pudp, page);
390         ptep = pte_offset(pmdp, page);
391
392         /*
393          * If the page isn't marked valid, the page cannot possibly be
394          * in the cache.
395          */
396         if (!(pte_val(*ptep) & _PAGE_PRESENT))
397                 return;
398
399         /*
400          * Doing flushes for another ASID than the current one is
401          * too difficult since stupid R4k caches do a TLB translation
402          * for every cache flush operation.  So we do indexed flushes
403          * in that case, which doesn't overly flush the cache too much.
404          */
405         if ((mm == current->active_mm) && (pte_val(*ptep) & _PAGE_VALID)) {
406                 if (cpu_has_dc_aliases || (exec && !cpu_has_ic_fills_f_dc)) {
407                         r4k_blast_dcache_page(page);
408                         if (exec && !cpu_icache_snoops_remote_store)
409                                 r4k_blast_scache_page(page);
410                 }
411                 if (exec)
412                         r4k_blast_icache_page(page);
413
414                 return;
415         }
416
417         /*
418          * Do indexed flush, too much work to get the (possible) TLB refills
419          * to work correctly.
420          */
421         page = INDEX_BASE + (page & (dcache_size - 1));
422         if (cpu_has_dc_aliases || (exec && !cpu_has_ic_fills_f_dc)) {
423                 r4k_blast_dcache_page_indexed(page);
424                 if (exec && !cpu_icache_snoops_remote_store)
425                         r4k_blast_scache_page_indexed(page);
426         }
427         if (exec) {
428                 if (cpu_has_vtag_icache) {
429                         int cpu = smp_processor_id();
430
431                         if (cpu_context(cpu, mm) != 0)
432                                 drop_mmu_context(mm, cpu);
433                 } else
434                         r4k_blast_icache_page_indexed(page);
435         }
436 }
437
438 static void r4k_flush_cache_page(struct vm_area_struct *vma, unsigned long page, unsigned long pfn)
439 {
440         struct flush_cache_page_args args;
441
442         args.vma = vma;
443         args.page = page;
444
445         on_each_cpu(local_r4k_flush_cache_page, &args, 1, 1);
446 }
447
448 static inline void local_r4k_flush_data_cache_page(void * addr)
449 {
450         r4k_blast_dcache_page((unsigned long) addr);
451 }
452
453 static void r4k_flush_data_cache_page(unsigned long addr)
454 {
455         on_each_cpu(local_r4k_flush_data_cache_page, (void *) addr, 1, 1);
456 }
457
458 struct flush_icache_range_args {
459         unsigned long start;
460         unsigned long end;
461 };
462
463 static inline void local_r4k_flush_icache_range(void *args)
464 {
465         struct flush_icache_range_args *fir_args = args;
466         unsigned long dc_lsize = current_cpu_data.dcache.linesz;
467         unsigned long ic_lsize = current_cpu_data.icache.linesz;
468         unsigned long sc_lsize = current_cpu_data.scache.linesz;
469         unsigned long start = fir_args->start;
470         unsigned long end = fir_args->end;
471         unsigned long addr, aend;
472
473         if (!cpu_has_ic_fills_f_dc) {
474                 if (end - start > dcache_size) {
475                         r4k_blast_dcache();
476                 } else {
477                         addr = start & ~(dc_lsize - 1);
478                         aend = (end - 1) & ~(dc_lsize - 1);
479
480                         while (1) {
481                                 /* Hit_Writeback_Inv_D */
482                                 protected_writeback_dcache_line(addr);
483                                 if (addr == aend)
484                                         break;
485                                 addr += dc_lsize;
486                         }
487                 }
488
489                 if (!cpu_icache_snoops_remote_store) {
490                         if (end - start > scache_size) {
491                                 r4k_blast_scache();
492                         } else {
493                                 addr = start & ~(sc_lsize - 1);
494                                 aend = (end - 1) & ~(sc_lsize - 1);
495
496                                 while (1) {
497                                         /* Hit_Writeback_Inv_D */
498                                         protected_writeback_scache_line(addr);
499                                         if (addr == aend)
500                                                 break;
501                                         addr += sc_lsize;
502                                 }
503                         }
504                 }
505         }
506
507         if (end - start > icache_size)
508                 r4k_blast_icache();
509         else {
510                 addr = start & ~(ic_lsize - 1);
511                 aend = (end - 1) & ~(ic_lsize - 1);
512                 while (1) {
513                         /* Hit_Invalidate_I */
514                         protected_flush_icache_line(addr);
515                         if (addr == aend)
516                                 break;
517                         addr += ic_lsize;
518                 }
519         }
520 }
521
522 static void r4k_flush_icache_range(unsigned long start, unsigned long end)
523 {
524         struct flush_icache_range_args args;
525
526         args.start = start;
527         args.end = end;
528
529         on_each_cpu(local_r4k_flush_icache_range, &args, 1, 1);
530 }
531
532 /*
533  * Ok, this seriously sucks.  We use them to flush a user page but don't
534  * know the virtual address, so we have to blast away the whole icache
535  * which is significantly more expensive than the real thing.  Otoh we at
536  * least know the kernel address of the page so we can flush it
537  * selectivly.
538  */
539
540 struct flush_icache_page_args {
541         struct vm_area_struct *vma;
542         struct page *page;
543 };
544
545 static inline void local_r4k_flush_icache_page(void *args)
546 {
547         struct flush_icache_page_args *fip_args = args;
548         struct vm_area_struct *vma = fip_args->vma;
549         struct page *page = fip_args->page;
550
551         /*
552          * Tricky ...  Because we don't know the virtual address we've got the
553          * choice of either invalidating the entire primary and secondary
554          * caches or invalidating the secondary caches also.  With the subset
555          * enforcment on R4000SC, R4400SC, R10000 and R12000 invalidating the
556          * secondary cache will result in any entries in the primary caches
557          * also getting invalidated which hopefully is a bit more economical.
558          */
559         if (cpu_has_subset_pcaches) {
560                 unsigned long addr = (unsigned long) page_address(page);
561
562                 r4k_blast_scache_page(addr);
563                 ClearPageDcacheDirty(page);
564
565                 return;
566         }
567
568         if (!cpu_has_ic_fills_f_dc) {
569                 unsigned long addr = (unsigned long) page_address(page);
570                 r4k_blast_dcache_page(addr);
571                 if (!cpu_icache_snoops_remote_store)
572                         r4k_blast_scache_page(addr);
573                 ClearPageDcacheDirty(page);
574         }
575
576         /*
577          * We're not sure of the virtual address(es) involved here, so
578          * we have to flush the entire I-cache.
579          */
580         if (cpu_has_vtag_icache) {
581                 int cpu = smp_processor_id();
582
583                 if (cpu_context(cpu, vma->vm_mm) != 0)
584                         drop_mmu_context(vma->vm_mm, cpu);
585         } else
586                 r4k_blast_icache();
587 }
588
589 static void r4k_flush_icache_page(struct vm_area_struct *vma,
590         struct page *page)
591 {
592         struct flush_icache_page_args args;
593
594         /*
595          * If there's no context yet, or the page isn't executable, no I-cache
596          * flush is needed.
597          */
598         if (!(vma->vm_flags & VM_EXEC))
599                 return;
600
601         args.vma = vma;
602         args.page = page;
603
604         on_each_cpu(local_r4k_flush_icache_page, &args, 1, 1);
605 }
606
607
608 #ifdef CONFIG_DMA_NONCOHERENT
609
610 static void r4k_dma_cache_wback_inv(unsigned long addr, unsigned long size)
611 {
612         unsigned long end, a;
613
614         /* Catch bad driver code */
615         BUG_ON(size == 0);
616
617         if (cpu_has_subset_pcaches) {
618                 unsigned long sc_lsize = current_cpu_data.scache.linesz;
619
620                 if (size >= scache_size) {
621                         r4k_blast_scache();
622                         return;
623                 }
624
625                 a = addr & ~(sc_lsize - 1);
626                 end = (addr + size - 1) & ~(sc_lsize - 1);
627                 while (1) {
628                         flush_scache_line(a);   /* Hit_Writeback_Inv_SD */
629                         if (a == end)
630                                 break;
631                         a += sc_lsize;
632                 }
633                 return;
634         }
635
636         /*
637          * Either no secondary cache or the available caches don't have the
638          * subset property so we have to flush the primary caches
639          * explicitly
640          */
641         if (size >= dcache_size) {
642                 r4k_blast_dcache();
643         } else {
644                 unsigned long dc_lsize = current_cpu_data.dcache.linesz;
645
646                 R4600_HIT_CACHEOP_WAR_IMPL;
647                 a = addr & ~(dc_lsize - 1);
648                 end = (addr + size - 1) & ~(dc_lsize - 1);
649                 while (1) {
650                         flush_dcache_line(a);   /* Hit_Writeback_Inv_D */
651                         if (a == end)
652                                 break;
653                         a += dc_lsize;
654                 }
655         }
656
657         bc_wback_inv(addr, size);
658 }
659
660 static void r4k_dma_cache_inv(unsigned long addr, unsigned long size)
661 {
662         unsigned long end, a;
663
664         /* Catch bad driver code */
665         BUG_ON(size == 0);
666
667         if (cpu_has_subset_pcaches) {
668                 unsigned long sc_lsize = current_cpu_data.scache.linesz;
669
670                 if (size >= scache_size) {
671                         r4k_blast_scache();
672                         return;
673                 }
674
675                 a = addr & ~(sc_lsize - 1);
676                 end = (addr + size - 1) & ~(sc_lsize - 1);
677                 while (1) {
678                         flush_scache_line(a);   /* Hit_Writeback_Inv_SD */
679                         if (a == end)
680                                 break;
681                         a += sc_lsize;
682                 }
683                 return;
684         }
685
686         if (size >= dcache_size) {
687                 r4k_blast_dcache();
688         } else {
689                 unsigned long dc_lsize = current_cpu_data.dcache.linesz;
690
691                 R4600_HIT_CACHEOP_WAR_IMPL;
692                 a = addr & ~(dc_lsize - 1);
693                 end = (addr + size - 1) & ~(dc_lsize - 1);
694                 while (1) {
695                         flush_dcache_line(a);   /* Hit_Writeback_Inv_D */
696                         if (a == end)
697                                 break;
698                         a += dc_lsize;
699                 }
700         }
701
702         bc_inv(addr, size);
703 }
704 #endif /* CONFIG_DMA_NONCOHERENT */
705
706 /*
707  * While we're protected against bad userland addresses we don't care
708  * very much about what happens in that case.  Usually a segmentation
709  * fault will dump the process later on anyway ...
710  */
711 static void local_r4k_flush_cache_sigtramp(void * arg)
712 {
713         unsigned long ic_lsize = current_cpu_data.icache.linesz;
714         unsigned long dc_lsize = current_cpu_data.dcache.linesz;
715         unsigned long sc_lsize = current_cpu_data.scache.linesz;
716         unsigned long addr = (unsigned long) arg;
717
718         R4600_HIT_CACHEOP_WAR_IMPL;
719         protected_writeback_dcache_line(addr & ~(dc_lsize - 1));
720         if (!cpu_icache_snoops_remote_store)
721                 protected_writeback_scache_line(addr & ~(sc_lsize - 1));
722         protected_flush_icache_line(addr & ~(ic_lsize - 1));
723         if (MIPS4K_ICACHE_REFILL_WAR) {
724                 __asm__ __volatile__ (
725                         ".set push\n\t"
726                         ".set noat\n\t"
727                         ".set mips3\n\t"
728 #ifdef CONFIG_32BIT
729                         "la     $at,1f\n\t"
730 #endif
731 #ifdef CONFIG_64BIT
732                         "dla    $at,1f\n\t"
733 #endif
734                         "cache  %0,($at)\n\t"
735                         "nop; nop; nop\n"
736                         "1:\n\t"
737                         ".set pop"
738                         :
739                         : "i" (Hit_Invalidate_I));
740         }
741         if (MIPS_CACHE_SYNC_WAR)
742                 __asm__ __volatile__ ("sync");
743 }
744
745 static void r4k_flush_cache_sigtramp(unsigned long addr)
746 {
747         on_each_cpu(local_r4k_flush_cache_sigtramp, (void *) addr, 1, 1);
748 }
749
750 static void r4k_flush_icache_all(void)
751 {
752         if (cpu_has_vtag_icache)
753                 r4k_blast_icache();
754 }
755
756 static inline void rm7k_erratum31(void)
757 {
758         const unsigned long ic_lsize = 32;
759         unsigned long addr;
760
761         /* RM7000 erratum #31. The icache is screwed at startup. */
762         write_c0_taglo(0);
763         write_c0_taghi(0);
764
765         for (addr = INDEX_BASE; addr <= INDEX_BASE + 4096; addr += ic_lsize) {
766                 __asm__ __volatile__ (
767                         ".set noreorder\n\t"
768                         ".set mips3\n\t"
769                         "cache\t%1, 0(%0)\n\t"
770                         "cache\t%1, 0x1000(%0)\n\t"
771                         "cache\t%1, 0x2000(%0)\n\t"
772                         "cache\t%1, 0x3000(%0)\n\t"
773                         "cache\t%2, 0(%0)\n\t"
774                         "cache\t%2, 0x1000(%0)\n\t"
775                         "cache\t%2, 0x2000(%0)\n\t"
776                         "cache\t%2, 0x3000(%0)\n\t"
777                         "cache\t%1, 0(%0)\n\t"
778                         "cache\t%1, 0x1000(%0)\n\t"
779                         "cache\t%1, 0x2000(%0)\n\t"
780                         "cache\t%1, 0x3000(%0)\n\t"
781                         ".set\tmips0\n\t"
782                         ".set\treorder\n\t"
783                         :
784                         : "r" (addr), "i" (Index_Store_Tag_I), "i" (Fill));
785         }
786 }
787
788 static char *way_string[] __initdata = { NULL, "direct mapped", "2-way",
789         "3-way", "4-way", "5-way", "6-way", "7-way", "8-way"
790 };
791
792 static void __init probe_pcache(void)
793 {
794         struct cpuinfo_mips *c = &current_cpu_data;
795         unsigned int config = read_c0_config();
796         unsigned int prid = read_c0_prid();
797         unsigned long config1;
798         unsigned int lsize;
799
800         switch (c->cputype) {
801         case CPU_R4600:                 /* QED style two way caches? */
802         case CPU_R4700:
803         case CPU_R5000:
804         case CPU_NEVADA:
805                 icache_size = 1 << (12 + ((config & CONF_IC) >> 9));
806                 c->icache.linesz = 16 << ((config & CONF_IB) >> 5);
807                 c->icache.ways = 2;
808                 c->icache.waybit = ffs(icache_size/2) - 1;
809
810                 dcache_size = 1 << (12 + ((config & CONF_DC) >> 6));
811                 c->dcache.linesz = 16 << ((config & CONF_DB) >> 4);
812                 c->dcache.ways = 2;
813                 c->dcache.waybit= ffs(dcache_size/2) - 1;
814
815                 c->options |= MIPS_CPU_CACHE_CDEX_P;
816                 break;
817
818         case CPU_R5432:
819         case CPU_R5500:
820                 icache_size = 1 << (12 + ((config & CONF_IC) >> 9));
821                 c->icache.linesz = 16 << ((config & CONF_IB) >> 5);
822                 c->icache.ways = 2;
823                 c->icache.waybit= 0;
824
825                 dcache_size = 1 << (12 + ((config & CONF_DC) >> 6));
826                 c->dcache.linesz = 16 << ((config & CONF_DB) >> 4);
827                 c->dcache.ways = 2;
828                 c->dcache.waybit = 0;
829
830                 c->options |= MIPS_CPU_CACHE_CDEX_P;
831                 break;
832
833         case CPU_TX49XX:
834                 icache_size = 1 << (12 + ((config & CONF_IC) >> 9));
835                 c->icache.linesz = 16 << ((config & CONF_IB) >> 5);
836                 c->icache.ways = 4;
837                 c->icache.waybit= 0;
838
839                 dcache_size = 1 << (12 + ((config & CONF_DC) >> 6));
840                 c->dcache.linesz = 16 << ((config & CONF_DB) >> 4);
841                 c->dcache.ways = 4;
842                 c->dcache.waybit = 0;
843
844                 c->options |= MIPS_CPU_CACHE_CDEX_P;
845                 break;
846
847         case CPU_R4000PC:
848         case CPU_R4000SC:
849         case CPU_R4000MC:
850         case CPU_R4400PC:
851         case CPU_R4400SC:
852         case CPU_R4400MC:
853         case CPU_R4300:
854                 icache_size = 1 << (12 + ((config & CONF_IC) >> 9));
855                 c->icache.linesz = 16 << ((config & CONF_IB) >> 5);
856                 c->icache.ways = 1;
857                 c->icache.waybit = 0;   /* doesn't matter */
858
859                 dcache_size = 1 << (12 + ((config & CONF_DC) >> 6));
860                 c->dcache.linesz = 16 << ((config & CONF_DB) >> 4);
861                 c->dcache.ways = 1;
862                 c->dcache.waybit = 0;   /* does not matter */
863
864                 c->options |= MIPS_CPU_CACHE_CDEX_P;
865                 break;
866
867         case CPU_R10000:
868         case CPU_R12000:
869                 icache_size = 1 << (12 + ((config & R10K_CONF_IC) >> 29));
870                 c->icache.linesz = 64;
871                 c->icache.ways = 2;
872                 c->icache.waybit = 0;
873
874                 dcache_size = 1 << (12 + ((config & R10K_CONF_DC) >> 26));
875                 c->dcache.linesz = 32;
876                 c->dcache.ways = 2;
877                 c->dcache.waybit = 0;
878
879                 c->options |= MIPS_CPU_PREFETCH;
880                 break;
881
882         case CPU_VR4133:
883                 write_c0_config(config & ~CONF_EB);
884         case CPU_VR4131:
885                 /* Workaround for cache instruction bug of VR4131 */
886                 if (c->processor_id == 0x0c80U || c->processor_id == 0x0c81U ||
887                     c->processor_id == 0x0c82U) {
888                         config &= ~0x00000030U;
889                         config |= 0x00410000U;
890                         write_c0_config(config);
891                 }
892                 icache_size = 1 << (10 + ((config & CONF_IC) >> 9));
893                 c->icache.linesz = 16 << ((config & CONF_IB) >> 5);
894                 c->icache.ways = 2;
895                 c->icache.waybit = ffs(icache_size/2) - 1;
896
897                 dcache_size = 1 << (10 + ((config & CONF_DC) >> 6));
898                 c->dcache.linesz = 16 << ((config & CONF_DB) >> 4);
899                 c->dcache.ways = 2;
900                 c->dcache.waybit = ffs(dcache_size/2) - 1;
901
902                 c->options |= MIPS_CPU_CACHE_CDEX_P;
903                 break;
904
905         case CPU_VR41XX:
906         case CPU_VR4111:
907         case CPU_VR4121:
908         case CPU_VR4122:
909         case CPU_VR4181:
910         case CPU_VR4181A:
911                 icache_size = 1 << (10 + ((config & CONF_IC) >> 9));
912                 c->icache.linesz = 16 << ((config & CONF_IB) >> 5);
913                 c->icache.ways = 1;
914                 c->icache.waybit = 0;   /* doesn't matter */
915
916                 dcache_size = 1 << (10 + ((config & CONF_DC) >> 6));
917                 c->dcache.linesz = 16 << ((config & CONF_DB) >> 4);
918                 c->dcache.ways = 1;
919                 c->dcache.waybit = 0;   /* does not matter */
920
921                 c->options |= MIPS_CPU_CACHE_CDEX_P;
922                 break;
923
924         case CPU_RM7000:
925                 rm7k_erratum31();
926
927         case CPU_RM9000:
928                 icache_size = 1 << (12 + ((config & CONF_IC) >> 9));
929                 c->icache.linesz = 16 << ((config & CONF_IB) >> 5);
930                 c->icache.ways = 4;
931                 c->icache.waybit = ffs(icache_size / c->icache.ways) - 1;
932
933                 dcache_size = 1 << (12 + ((config & CONF_DC) >> 6));
934                 c->dcache.linesz = 16 << ((config & CONF_DB) >> 4);
935                 c->dcache.ways = 4;
936                 c->dcache.waybit = ffs(dcache_size / c->dcache.ways) - 1;
937
938 #if !defined(CONFIG_SMP) || !defined(RM9000_CDEX_SMP_WAR)
939                 c->options |= MIPS_CPU_CACHE_CDEX_P;
940 #endif
941                 c->options |= MIPS_CPU_PREFETCH;
942                 break;
943
944         default:
945                 if (!(config & MIPS_CONF_M))
946                         panic("Don't know how to probe P-caches on this cpu.");
947
948                 /*
949                  * So we seem to be a MIPS32 or MIPS64 CPU
950                  * So let's probe the I-cache ...
951                  */
952                 config1 = read_c0_config1();
953
954                 if ((lsize = ((config1 >> 19) & 7)))
955                         c->icache.linesz = 2 << lsize;
956                 else
957                         c->icache.linesz = lsize;
958                 c->icache.sets = 64 << ((config1 >> 22) & 7);
959                 c->icache.ways = 1 + ((config1 >> 16) & 7);
960
961                 icache_size = c->icache.sets *
962                               c->icache.ways *
963                               c->icache.linesz;
964                 c->icache.waybit = ffs(icache_size/c->icache.ways) - 1;
965
966                 if (config & 0x8)               /* VI bit */
967                         c->icache.flags |= MIPS_CACHE_VTAG;
968
969                 /*
970                  * Now probe the MIPS32 / MIPS64 data cache.
971                  */
972                 c->dcache.flags = 0;
973
974                 if ((lsize = ((config1 >> 10) & 7)))
975                         c->dcache.linesz = 2 << lsize;
976                 else
977                         c->dcache.linesz= lsize;
978                 c->dcache.sets = 64 << ((config1 >> 13) & 7);
979                 c->dcache.ways = 1 + ((config1 >> 7) & 7);
980
981                 dcache_size = c->dcache.sets *
982                               c->dcache.ways *
983                               c->dcache.linesz;
984                 c->dcache.waybit = ffs(dcache_size/c->dcache.ways) - 1;
985
986                 c->options |= MIPS_CPU_PREFETCH;
987                 break;
988         }
989
990         /*
991          * Processor configuration sanity check for the R4000SC erratum
992          * #5.  With page sizes larger than 32kB there is no possibility
993          * to get a VCE exception anymore so we don't care about this
994          * misconfiguration.  The case is rather theoretical anyway;
995          * presumably no vendor is shipping his hardware in the "bad"
996          * configuration.
997          */
998         if ((prid & 0xff00) == PRID_IMP_R4000 && (prid & 0xff) < 0x40 &&
999             !(config & CONF_SC) && c->icache.linesz != 16 &&
1000             PAGE_SIZE <= 0x8000)
1001                 panic("Improper R4000SC processor configuration detected");
1002
1003         /* compute a couple of other cache variables */
1004         c->icache.waysize = icache_size / c->icache.ways;
1005         c->dcache.waysize = dcache_size / c->dcache.ways;
1006
1007         c->icache.sets = icache_size / (c->icache.linesz * c->icache.ways);
1008         c->dcache.sets = dcache_size / (c->dcache.linesz * c->dcache.ways);
1009
1010         /*
1011          * R10000 and R12000 P-caches are odd in a positive way.  They're 32kB
1012          * 2-way virtually indexed so normally would suffer from aliases.  So
1013          * normally they'd suffer from aliases but magic in the hardware deals
1014          * with that for us so we don't need to take care ourselves.
1015          */
1016         switch (c->cputype) {
1017         case CPU_20KC:
1018         case CPU_25KF:
1019         case CPU_R10000:
1020         case CPU_R12000:
1021         case CPU_SB1:
1022                 break;
1023         case CPU_24K:
1024                 if (!(read_c0_config7() & (1 << 16)))
1025         default:
1026                         if (c->dcache.waysize > PAGE_SIZE)
1027                                 c->dcache.flags |= MIPS_CACHE_ALIASES;
1028         }
1029
1030         switch (c->cputype) {
1031         case CPU_20KC:
1032                 /*
1033                  * Some older 20Kc chips doesn't have the 'VI' bit in
1034                  * the config register.
1035                  */
1036                 c->icache.flags |= MIPS_CACHE_VTAG;
1037                 break;
1038
1039         case CPU_AU1500:
1040                 c->icache.flags |= MIPS_CACHE_IC_F_DC;
1041                 break;
1042         }
1043
1044         printk("Primary instruction cache %ldkB, %s, %s, linesize %d bytes.\n",
1045                icache_size >> 10,
1046                cpu_has_vtag_icache ? "virtually tagged" : "physically tagged",
1047                way_string[c->icache.ways], c->icache.linesz);
1048
1049         printk("Primary data cache %ldkB, %s, linesize %d bytes.\n",
1050                dcache_size >> 10, way_string[c->dcache.ways], c->dcache.linesz);
1051 }
1052
1053 /*
1054  * If you even _breathe_ on this function, look at the gcc output and make sure
1055  * it does not pop things on and off the stack for the cache sizing loop that
1056  * executes in KSEG1 space or else you will crash and burn badly.  You have
1057  * been warned.
1058  */
1059 static int __init probe_scache(void)
1060 {
1061         extern unsigned long stext;
1062         unsigned long flags, addr, begin, end, pow2;
1063         unsigned int config = read_c0_config();
1064         struct cpuinfo_mips *c = &current_cpu_data;
1065         int tmp;
1066
1067         if (config & CONF_SC)
1068                 return 0;
1069
1070         begin = (unsigned long) &stext;
1071         begin &= ~((4 * 1024 * 1024) - 1);
1072         end = begin + (4 * 1024 * 1024);
1073
1074         /*
1075          * This is such a bitch, you'd think they would make it easy to do
1076          * this.  Away you daemons of stupidity!
1077          */
1078         local_irq_save(flags);
1079
1080         /* Fill each size-multiple cache line with a valid tag. */
1081         pow2 = (64 * 1024);
1082         for (addr = begin; addr < end; addr = (begin + pow2)) {
1083                 unsigned long *p = (unsigned long *) addr;
1084                 __asm__ __volatile__("nop" : : "r" (*p)); /* whee... */
1085                 pow2 <<= 1;
1086         }
1087
1088         /* Load first line with zero (therefore invalid) tag. */
1089         write_c0_taglo(0);
1090         write_c0_taghi(0);
1091         __asm__ __volatile__("nop; nop; nop; nop;"); /* avoid the hazard */
1092         cache_op(Index_Store_Tag_I, begin);
1093         cache_op(Index_Store_Tag_D, begin);
1094         cache_op(Index_Store_Tag_SD, begin);
1095
1096         /* Now search for the wrap around point. */
1097         pow2 = (128 * 1024);
1098         tmp = 0;
1099         for (addr = begin + (128 * 1024); addr < end; addr = begin + pow2) {
1100                 cache_op(Index_Load_Tag_SD, addr);
1101                 __asm__ __volatile__("nop; nop; nop; nop;"); /* hazard... */
1102                 if (!read_c0_taglo())
1103                         break;
1104                 pow2 <<= 1;
1105         }
1106         local_irq_restore(flags);
1107         addr -= begin;
1108
1109         scache_size = addr;
1110         c->scache.linesz = 16 << ((config & R4K_CONF_SB) >> 22);
1111         c->scache.ways = 1;
1112         c->dcache.waybit = 0;           /* does not matter */
1113
1114         return 1;
1115 }
1116
1117 typedef int (*probe_func_t)(unsigned long);
1118 extern int r5k_sc_init(void);
1119 extern int rm7k_sc_init(void);
1120
1121 static void __init setup_scache(void)
1122 {
1123         struct cpuinfo_mips *c = &current_cpu_data;
1124         unsigned int config = read_c0_config();
1125         probe_func_t probe_scache_kseg1;
1126         int sc_present = 0;
1127
1128         /*
1129          * Do the probing thing on R4000SC and R4400SC processors.  Other
1130          * processors don't have a S-cache that would be relevant to the
1131          * Linux memory managment.
1132          */
1133         switch (c->cputype) {
1134         case CPU_R4000SC:
1135         case CPU_R4000MC:
1136         case CPU_R4400SC:
1137         case CPU_R4400MC:
1138                 probe_scache_kseg1 = (probe_func_t) (CKSEG1ADDR(&probe_scache));
1139                 sc_present = probe_scache_kseg1(config);
1140                 if (sc_present)
1141                         c->options |= MIPS_CPU_CACHE_CDEX_S;
1142                 break;
1143
1144         case CPU_R10000:
1145         case CPU_R12000:
1146                 scache_size = 0x80000 << ((config & R10K_CONF_SS) >> 16);
1147                 c->scache.linesz = 64 << ((config >> 13) & 1);
1148                 c->scache.ways = 2;
1149                 c->scache.waybit= 0;
1150                 sc_present = 1;
1151                 break;
1152
1153         case CPU_R5000:
1154         case CPU_NEVADA:
1155 #ifdef CONFIG_R5000_CPU_SCACHE
1156                 r5k_sc_init();
1157 #endif
1158                 return;
1159
1160         case CPU_RM7000:
1161         case CPU_RM9000:
1162 #ifdef CONFIG_RM7000_CPU_SCACHE
1163                 rm7k_sc_init();
1164 #endif
1165                 return;
1166
1167         default:
1168                 sc_present = 0;
1169         }
1170
1171         if (!sc_present)
1172                 return;
1173
1174         if ((c->isa_level == MIPS_CPU_ISA_M32 ||
1175              c->isa_level == MIPS_CPU_ISA_M64) &&
1176             !(c->scache.flags & MIPS_CACHE_NOT_PRESENT))
1177                 panic("Dunno how to handle MIPS32 / MIPS64 second level cache");
1178
1179         /* compute a couple of other cache variables */
1180         c->scache.waysize = scache_size / c->scache.ways;
1181
1182         c->scache.sets = scache_size / (c->scache.linesz * c->scache.ways);
1183
1184         printk("Unified secondary cache %ldkB %s, linesize %d bytes.\n",
1185                scache_size >> 10, way_string[c->scache.ways], c->scache.linesz);
1186
1187         c->options |= MIPS_CPU_SUBSET_CACHES;
1188 }
1189
1190 static inline void coherency_setup(void)
1191 {
1192         change_c0_config(CONF_CM_CMASK, CONF_CM_DEFAULT);
1193
1194         /*
1195          * c0_status.cu=0 specifies that updates by the sc instruction use
1196          * the coherency mode specified by the TLB; 1 means cachable
1197          * coherent update on write will be used.  Not all processors have
1198          * this bit and; some wire it to zero, others like Toshiba had the
1199          * silly idea of putting something else there ...
1200          */
1201         switch (current_cpu_data.cputype) {
1202         case CPU_R4000PC:
1203         case CPU_R4000SC:
1204         case CPU_R4000MC:
1205         case CPU_R4400PC:
1206         case CPU_R4400SC:
1207         case CPU_R4400MC:
1208                 clear_c0_config(CONF_CU);
1209                 break;
1210         }
1211 }
1212
1213 void __init ld_mmu_r4xx0(void)
1214 {
1215         extern void build_clear_page(void);
1216         extern void build_copy_page(void);
1217         extern char except_vec2_generic;
1218         struct cpuinfo_mips *c = &current_cpu_data;
1219
1220         /* Default cache error handler for R4000 and R5000 family */
1221         memcpy((void *)(CAC_BASE   + 0x100), &except_vec2_generic, 0x80);
1222         memcpy((void *)(UNCAC_BASE + 0x100), &except_vec2_generic, 0x80);
1223
1224         probe_pcache();
1225         setup_scache();
1226
1227         r4k_blast_dcache_page_setup();
1228         r4k_blast_dcache_page_indexed_setup();
1229         r4k_blast_dcache_setup();
1230         r4k_blast_icache_page_setup();
1231         r4k_blast_icache_page_indexed_setup();
1232         r4k_blast_icache_setup();
1233         r4k_blast_scache_page_setup();
1234         r4k_blast_scache_page_indexed_setup();
1235         r4k_blast_scache_setup();
1236
1237         /*
1238          * Some MIPS32 and MIPS64 processors have physically indexed caches.
1239          * This code supports virtually indexed processors and will be
1240          * unnecessarily inefficient on physically indexed processors.
1241          */
1242         shm_align_mask = max_t( unsigned long,
1243                                 c->dcache.sets * c->dcache.linesz - 1,
1244                                 PAGE_SIZE - 1);
1245
1246         flush_cache_all         = r4k_flush_cache_all;
1247         __flush_cache_all       = r4k___flush_cache_all;
1248         flush_cache_mm          = r4k_flush_cache_mm;
1249         flush_cache_page        = r4k_flush_cache_page;
1250         flush_icache_page       = r4k_flush_icache_page;
1251         flush_cache_range       = r4k_flush_cache_range;
1252
1253         flush_cache_sigtramp    = r4k_flush_cache_sigtramp;
1254         flush_icache_all        = r4k_flush_icache_all;
1255         flush_data_cache_page   = r4k_flush_data_cache_page;
1256         flush_icache_range      = r4k_flush_icache_range;
1257
1258 #ifdef CONFIG_DMA_NONCOHERENT
1259         _dma_cache_wback_inv    = r4k_dma_cache_wback_inv;
1260         _dma_cache_wback        = r4k_dma_cache_wback_inv;
1261         _dma_cache_inv          = r4k_dma_cache_inv;
1262 #endif
1263
1264         __flush_cache_all();
1265         coherency_setup();
1266
1267         build_clear_page();
1268         build_copy_page();
1269 }