aboutsummaryrefslogtreecommitdiffstats
path: root/arch/sh/mm/cache-sh4.c
diff options
context:
space:
mode:
authorAndrea Bastoni <bastoni@cs.unc.edu>2010-05-30 19:16:45 -0400
committerAndrea Bastoni <bastoni@cs.unc.edu>2010-05-30 19:16:45 -0400
commitada47b5fe13d89735805b566185f4885f5a3f750 (patch)
tree644b88f8a71896307d71438e9b3af49126ffb22b /arch/sh/mm/cache-sh4.c
parent43e98717ad40a4ae64545b5ba047c7b86aa44f4f (diff)
parent3280f21d43ee541f97f8cda5792150d2dbec20d5 (diff)
Merge branch 'wip-2.6.34' into old-private-masterarchived-private-master
Diffstat (limited to 'arch/sh/mm/cache-sh4.c')
-rw-r--r--arch/sh/mm/cache-sh4.c525
1 files changed, 85 insertions, 440 deletions
diff --git a/arch/sh/mm/cache-sh4.c b/arch/sh/mm/cache-sh4.c
index b7f235c74d66..2cfae81914aa 100644
--- a/arch/sh/mm/cache-sh4.c
+++ b/arch/sh/mm/cache-sh4.c
@@ -2,7 +2,7 @@
2 * arch/sh/mm/cache-sh4.c 2 * arch/sh/mm/cache-sh4.c
3 * 3 *
4 * Copyright (C) 1999, 2000, 2002 Niibe Yutaka 4 * Copyright (C) 1999, 2000, 2002 Niibe Yutaka
5 * Copyright (C) 2001 - 2007 Paul Mundt 5 * Copyright (C) 2001 - 2009 Paul Mundt
6 * Copyright (C) 2003 Richard Curnow 6 * Copyright (C) 2003 Richard Curnow
7 * Copyright (c) 2007 STMicroelectronics (R&D) Ltd. 7 * Copyright (c) 2007 STMicroelectronics (R&D) Ltd.
8 * 8 *
@@ -15,6 +15,8 @@
15#include <linux/io.h> 15#include <linux/io.h>
16#include <linux/mutex.h> 16#include <linux/mutex.h>
17#include <linux/fs.h> 17#include <linux/fs.h>
18#include <linux/highmem.h>
19#include <asm/pgtable.h>
18#include <asm/mmu_context.h> 20#include <asm/mmu_context.h>
19#include <asm/cacheflush.h> 21#include <asm/cacheflush.h>
20 22
@@ -23,27 +25,18 @@
23 * flushing. Anything exceeding this will simply flush the dcache in its 25 * flushing. Anything exceeding this will simply flush the dcache in its
24 * entirety. 26 * entirety.
25 */ 27 */
26#define MAX_DCACHE_PAGES 64 /* XXX: Tune for ways */
27#define MAX_ICACHE_PAGES 32 28#define MAX_ICACHE_PAGES 32
28 29
29static void __flush_cache_one(unsigned long addr, unsigned long phys, 30static void __flush_cache_one(unsigned long addr, unsigned long phys,
30 unsigned long exec_offset); 31 unsigned long exec_offset);
31 32
32/* 33/*
33 * This is initialised here to ensure that it is not placed in the BSS. If
34 * that were to happen, note that cache_init gets called before the BSS is
35 * cleared, so this would get nulled out which would be hopeless.
36 */
37static void (*__flush_dcache_segment_fn)(unsigned long, unsigned long) =
38 (void (*)(unsigned long, unsigned long))0xdeadbeef;
39
40/*
41 * Write back the range of D-cache, and purge the I-cache. 34 * Write back the range of D-cache, and purge the I-cache.
42 * 35 *
43 * Called from kernel/module.c:sys_init_module and routine for a.out format, 36 * Called from kernel/module.c:sys_init_module and routine for a.out format,
44 * signal handler code and kprobes code 37 * signal handler code and kprobes code
45 */ 38 */
46static void __uses_jump_to_uncached sh4_flush_icache_range(void *args) 39static void sh4_flush_icache_range(void *args)
47{ 40{
48 struct flusher_data *data = args; 41 struct flusher_data *data = args;
49 unsigned long start, end; 42 unsigned long start, end;
@@ -97,15 +90,15 @@ static inline void flush_cache_one(unsigned long start, unsigned long phys)
97 unsigned long flags, exec_offset = 0; 90 unsigned long flags, exec_offset = 0;
98 91
99 /* 92 /*
100 * All types of SH-4 require PC to be in P2 to operate on the I-cache. 93 * All types of SH-4 require PC to be uncached to operate on the I-cache.
101 * Some types of SH-4 require PC to be in P2 to operate on the D-cache. 94 * Some types of SH-4 require PC to be uncached to operate on the D-cache.
102 */ 95 */
103 if ((boot_cpu_data.flags & CPU_HAS_P2_FLUSH_BUG) || 96 if ((boot_cpu_data.flags & CPU_HAS_P2_FLUSH_BUG) ||
104 (start < CACHE_OC_ADDRESS_ARRAY)) 97 (start < CACHE_OC_ADDRESS_ARRAY))
105 exec_offset = 0x20000000; 98 exec_offset = cached_to_uncached;
106 99
107 local_irq_save(flags); 100 local_irq_save(flags);
108 __flush_cache_one(start | SH_CACHE_ASSOC, P1SEGADDR(phys), exec_offset); 101 __flush_cache_one(start, phys, exec_offset);
109 local_irq_restore(flags); 102 local_irq_restore(flags);
110} 103}
111 104
@@ -116,6 +109,7 @@ static inline void flush_cache_one(unsigned long start, unsigned long phys)
116static void sh4_flush_dcache_page(void *arg) 109static void sh4_flush_dcache_page(void *arg)
117{ 110{
118 struct page *page = arg; 111 struct page *page = arg;
112 unsigned long addr = (unsigned long)page_address(page);
119#ifndef CONFIG_SMP 113#ifndef CONFIG_SMP
120 struct address_space *mapping = page_mapping(page); 114 struct address_space *mapping = page_mapping(page);
121 115
@@ -123,22 +117,14 @@ static void sh4_flush_dcache_page(void *arg)
123 set_bit(PG_dcache_dirty, &page->flags); 117 set_bit(PG_dcache_dirty, &page->flags);
124 else 118 else
125#endif 119#endif
126 { 120 flush_cache_one(CACHE_OC_ADDRESS_ARRAY |
127 unsigned long phys = PHYSADDR(page_address(page)); 121 (addr & shm_align_mask), page_to_phys(page));
128 unsigned long addr = CACHE_OC_ADDRESS_ARRAY;
129 int i, n;
130
131 /* Loop all the D-cache */
132 n = boot_cpu_data.dcache.n_aliases;
133 for (i = 0; i < n; i++, addr += PAGE_SIZE)
134 flush_cache_one(addr, phys);
135 }
136 122
137 wmb(); 123 wmb();
138} 124}
139 125
140/* TODO: Selective icache invalidation through IC address array.. */ 126/* TODO: Selective icache invalidation through IC address array.. */
141static void __uses_jump_to_uncached flush_icache_all(void) 127static void flush_icache_all(void)
142{ 128{
143 unsigned long flags, ccr; 129 unsigned long flags, ccr;
144 130
@@ -146,9 +132,9 @@ static void __uses_jump_to_uncached flush_icache_all(void)
146 jump_to_uncached(); 132 jump_to_uncached();
147 133
148 /* Flush I-cache */ 134 /* Flush I-cache */
149 ccr = ctrl_inl(CCR); 135 ccr = __raw_readl(CCR);
150 ccr |= CCR_CACHE_ICI; 136 ccr |= CCR_CACHE_ICI;
151 ctrl_outl(ccr, CCR); 137 __raw_writel(ccr, CCR);
152 138
153 /* 139 /*
154 * back_to_cached() will take care of the barrier for us, don't add 140 * back_to_cached() will take care of the barrier for us, don't add
@@ -159,10 +145,27 @@ static void __uses_jump_to_uncached flush_icache_all(void)
159 local_irq_restore(flags); 145 local_irq_restore(flags);
160} 146}
161 147
162static inline void flush_dcache_all(void) 148static void flush_dcache_all(void)
163{ 149{
164 (*__flush_dcache_segment_fn)(0UL, boot_cpu_data.dcache.way_size); 150 unsigned long addr, end_addr, entry_offset;
165 wmb(); 151
152 end_addr = CACHE_OC_ADDRESS_ARRAY +
153 (current_cpu_data.dcache.sets <<
154 current_cpu_data.dcache.entry_shift) *
155 current_cpu_data.dcache.ways;
156
157 entry_offset = 1 << current_cpu_data.dcache.entry_shift;
158
159 for (addr = CACHE_OC_ADDRESS_ARRAY; addr < end_addr; ) {
160 __raw_writel(0, addr); addr += entry_offset;
161 __raw_writel(0, addr); addr += entry_offset;
162 __raw_writel(0, addr); addr += entry_offset;
163 __raw_writel(0, addr); addr += entry_offset;
164 __raw_writel(0, addr); addr += entry_offset;
165 __raw_writel(0, addr); addr += entry_offset;
166 __raw_writel(0, addr); addr += entry_offset;
167 __raw_writel(0, addr); addr += entry_offset;
168 }
166} 169}
167 170
168static void sh4_flush_cache_all(void *unused) 171static void sh4_flush_cache_all(void *unused)
@@ -171,89 +174,13 @@ static void sh4_flush_cache_all(void *unused)
171 flush_icache_all(); 174 flush_icache_all();
172} 175}
173 176
174static void __flush_cache_mm(struct mm_struct *mm, unsigned long start,
175 unsigned long end)
176{
177 unsigned long d = 0, p = start & PAGE_MASK;
178 unsigned long alias_mask = boot_cpu_data.dcache.alias_mask;
179 unsigned long n_aliases = boot_cpu_data.dcache.n_aliases;
180 unsigned long select_bit;
181 unsigned long all_aliases_mask;
182 unsigned long addr_offset;
183 pgd_t *dir;
184 pmd_t *pmd;
185 pud_t *pud;
186 pte_t *pte;
187 int i;
188
189 dir = pgd_offset(mm, p);
190 pud = pud_offset(dir, p);
191 pmd = pmd_offset(pud, p);
192 end = PAGE_ALIGN(end);
193
194 all_aliases_mask = (1 << n_aliases) - 1;
195
196 do {
197 if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd))) {
198 p &= PMD_MASK;
199 p += PMD_SIZE;
200 pmd++;
201
202 continue;
203 }
204
205 pte = pte_offset_kernel(pmd, p);
206
207 do {
208 unsigned long phys;
209 pte_t entry = *pte;
210
211 if (!(pte_val(entry) & _PAGE_PRESENT)) {
212 pte++;
213 p += PAGE_SIZE;
214 continue;
215 }
216
217 phys = pte_val(entry) & PTE_PHYS_MASK;
218
219 if ((p ^ phys) & alias_mask) {
220 d |= 1 << ((p & alias_mask) >> PAGE_SHIFT);
221 d |= 1 << ((phys & alias_mask) >> PAGE_SHIFT);
222
223 if (d == all_aliases_mask)
224 goto loop_exit;
225 }
226
227 pte++;
228 p += PAGE_SIZE;
229 } while (p < end && ((unsigned long)pte & ~PAGE_MASK));
230 pmd++;
231 } while (p < end);
232
233loop_exit:
234 addr_offset = 0;
235 select_bit = 1;
236
237 for (i = 0; i < n_aliases; i++) {
238 if (d & select_bit) {
239 (*__flush_dcache_segment_fn)(addr_offset, PAGE_SIZE);
240 wmb();
241 }
242
243 select_bit <<= 1;
244 addr_offset += PAGE_SIZE;
245 }
246}
247
248/* 177/*
249 * Note : (RPC) since the caches are physically tagged, the only point 178 * Note : (RPC) since the caches are physically tagged, the only point
250 * of flush_cache_mm for SH-4 is to get rid of aliases from the 179 * of flush_cache_mm for SH-4 is to get rid of aliases from the
251 * D-cache. The assumption elsewhere, e.g. flush_cache_range, is that 180 * D-cache. The assumption elsewhere, e.g. flush_cache_range, is that
252 * lines can stay resident so long as the virtual address they were 181 * lines can stay resident so long as the virtual address they were
253 * accessed with (hence cache set) is in accord with the physical 182 * accessed with (hence cache set) is in accord with the physical
254 * address (i.e. tag). It's no different here. So I reckon we don't 183 * address (i.e. tag). It's no different here.
255 * need to flush the I-cache, since aliases don't matter for that. We
256 * should try that.
257 * 184 *
258 * Caller takes mm->mmap_sem. 185 * Caller takes mm->mmap_sem.
259 */ 186 */
@@ -264,33 +191,7 @@ static void sh4_flush_cache_mm(void *arg)
264 if (cpu_context(smp_processor_id(), mm) == NO_CONTEXT) 191 if (cpu_context(smp_processor_id(), mm) == NO_CONTEXT)
265 return; 192 return;
266 193
267 /* 194 flush_dcache_all();
268 * If cache is only 4k-per-way, there are never any 'aliases'. Since
269 * the cache is physically tagged, the data can just be left in there.
270 */
271 if (boot_cpu_data.dcache.n_aliases == 0)
272 return;
273
274 /*
275 * Don't bother groveling around the dcache for the VMA ranges
276 * if there are too many PTEs to make it worthwhile.
277 */
278 if (mm->nr_ptes >= MAX_DCACHE_PAGES)
279 flush_dcache_all();
280 else {
281 struct vm_area_struct *vma;
282
283 /*
284 * In this case there are reasonably sized ranges to flush,
285 * iterate through the VMA list and take care of any aliases.
286 */
287 for (vma = mm->mmap; vma; vma = vma->vm_next)
288 __flush_cache_mm(mm, vma->vm_start, vma->vm_end);
289 }
290
291 /* Only touch the icache if one of the VMAs has VM_EXEC set. */
292 if (mm->exec_vm)
293 flush_icache_all();
294} 195}
295 196
296/* 197/*
@@ -303,44 +204,62 @@ static void sh4_flush_cache_page(void *args)
303{ 204{
304 struct flusher_data *data = args; 205 struct flusher_data *data = args;
305 struct vm_area_struct *vma; 206 struct vm_area_struct *vma;
207 struct page *page;
306 unsigned long address, pfn, phys; 208 unsigned long address, pfn, phys;
307 unsigned int alias_mask; 209 int map_coherent = 0;
210 pgd_t *pgd;
211 pud_t *pud;
212 pmd_t *pmd;
213 pte_t *pte;
214 void *vaddr;
308 215
309 vma = data->vma; 216 vma = data->vma;
310 address = data->addr1; 217 address = data->addr1 & PAGE_MASK;
311 pfn = data->addr2; 218 pfn = data->addr2;
312 phys = pfn << PAGE_SHIFT; 219 phys = pfn << PAGE_SHIFT;
220 page = pfn_to_page(pfn);
313 221
314 if (cpu_context(smp_processor_id(), vma->vm_mm) == NO_CONTEXT) 222 if (cpu_context(smp_processor_id(), vma->vm_mm) == NO_CONTEXT)
315 return; 223 return;
316 224
317 alias_mask = boot_cpu_data.dcache.alias_mask; 225 pgd = pgd_offset(vma->vm_mm, address);
318 226 pud = pud_offset(pgd, address);
319 /* We only need to flush D-cache when we have alias */ 227 pmd = pmd_offset(pud, address);
320 if ((address^phys) & alias_mask) { 228 pte = pte_offset_kernel(pmd, address);
321 /* Loop 4K of the D-cache */ 229
322 flush_cache_one( 230 /* If the page isn't present, there is nothing to do here. */
323 CACHE_OC_ADDRESS_ARRAY | (address & alias_mask), 231 if (!(pte_val(*pte) & _PAGE_PRESENT))
324 phys); 232 return;
325 /* Loop another 4K of the D-cache */
326 flush_cache_one(
327 CACHE_OC_ADDRESS_ARRAY | (phys & alias_mask),
328 phys);
329 }
330 233
331 alias_mask = boot_cpu_data.icache.alias_mask; 234 if ((vma->vm_mm == current->active_mm))
332 if (vma->vm_flags & VM_EXEC) { 235 vaddr = NULL;
236 else {
333 /* 237 /*
334 * Evict entries from the portion of the cache from which code 238 * Use kmap_coherent or kmap_atomic to do flushes for
335 * may have been executed at this address (virtual). There's 239 * another ASID than the current one.
336 * no need to evict from the portion corresponding to the
337 * physical address as for the D-cache, because we know the
338 * kernel has never executed the code through its identity
339 * translation.
340 */ 240 */
341 flush_cache_one( 241 map_coherent = (current_cpu_data.dcache.n_aliases &&
342 CACHE_IC_ADDRESS_ARRAY | (address & alias_mask), 242 !test_bit(PG_dcache_dirty, &page->flags) &&
343 phys); 243 page_mapped(page));
244 if (map_coherent)
245 vaddr = kmap_coherent(page, address);
246 else
247 vaddr = kmap_atomic(page, KM_USER0);
248
249 address = (unsigned long)vaddr;
250 }
251
252 flush_cache_one(CACHE_OC_ADDRESS_ARRAY |
253 (address & shm_align_mask), phys);
254
255 if (vma->vm_flags & VM_EXEC)
256 flush_icache_all();
257
258 if (vaddr) {
259 if (map_coherent)
260 kunmap_coherent(vaddr);
261 else
262 kunmap_atomic(vaddr, KM_USER0);
344 } 263 }
345} 264}
346 265
@@ -373,24 +292,10 @@ static void sh4_flush_cache_range(void *args)
373 if (boot_cpu_data.dcache.n_aliases == 0) 292 if (boot_cpu_data.dcache.n_aliases == 0)
374 return; 293 return;
375 294
376 /* 295 flush_dcache_all();
377 * Don't bother with the lookup and alias check if we have a
378 * wide range to cover, just blow away the dcache in its
379 * entirety instead. -- PFM.
380 */
381 if (((end - start) >> PAGE_SHIFT) >= MAX_DCACHE_PAGES)
382 flush_dcache_all();
383 else
384 __flush_cache_mm(vma->vm_mm, start, end);
385 296
386 if (vma->vm_flags & VM_EXEC) { 297 if (vma->vm_flags & VM_EXEC)
387 /*
388 * TODO: Is this required??? Need to look at how I-cache
389 * coherency is assured when new programs are loaded to see if
390 * this matters.
391 */
392 flush_icache_all(); 298 flush_icache_all();
393 }
394} 299}
395 300
396/** 301/**
@@ -464,245 +369,6 @@ static void __flush_cache_one(unsigned long addr, unsigned long phys,
464 } while (--way_count != 0); 369 } while (--way_count != 0);
465} 370}
466 371
467/*
468 * Break the 1, 2 and 4 way variants of this out into separate functions to
469 * avoid nearly all the overhead of having the conditional stuff in the function
470 * bodies (+ the 1 and 2 way cases avoid saving any registers too).
471 *
472 * We want to eliminate unnecessary bus transactions, so this code uses
473 * a non-obvious technique.
474 *
475 * Loop over a cache way sized block of, one cache line at a time. For each
476 * line, use movca.a to cause the current cache line contents to be written
477 * back, but without reading anything from main memory. However this has the
478 * side effect that the cache is now caching that memory location. So follow
479 * this with a cache invalidate to mark the cache line invalid. And do all
480 * this with interrupts disabled, to avoid the cache line being accidently
481 * evicted while it is holding garbage.
482 *
483 * This also breaks in a number of circumstances:
484 * - if there are modifications to the region of memory just above
485 * empty_zero_page (for example because a breakpoint has been placed
486 * there), then these can be lost.
487 *
488 * This is because the the memory address which the cache temporarily
489 * caches in the above description is empty_zero_page. So the
490 * movca.l hits the cache (it is assumed that it misses, or at least
491 * isn't dirty), modifies the line and then invalidates it, losing the
492 * required change.
493 *
494 * - If caches are disabled or configured in write-through mode, then
495 * the movca.l writes garbage directly into memory.
496 */
497static void __flush_dcache_segment_writethrough(unsigned long start,
498 unsigned long extent_per_way)
499{
500 unsigned long addr;
501 int i;
502
503 addr = CACHE_OC_ADDRESS_ARRAY | (start & cpu_data->dcache.entry_mask);
504
505 while (extent_per_way) {
506 for (i = 0; i < cpu_data->dcache.ways; i++)
507 __raw_writel(0, addr + cpu_data->dcache.way_incr * i);
508
509 addr += cpu_data->dcache.linesz;
510 extent_per_way -= cpu_data->dcache.linesz;
511 }
512}
513
514static void __flush_dcache_segment_1way(unsigned long start,
515 unsigned long extent_per_way)
516{
517 unsigned long orig_sr, sr_with_bl;
518 unsigned long base_addr;
519 unsigned long way_incr, linesz, way_size;
520 struct cache_info *dcache;
521 register unsigned long a0, a0e;
522
523 asm volatile("stc sr, %0" : "=r" (orig_sr));
524 sr_with_bl = orig_sr | (1<<28);
525 base_addr = ((unsigned long)&empty_zero_page[0]);
526
527 /*
528 * The previous code aligned base_addr to 16k, i.e. the way_size of all
529 * existing SH-4 D-caches. Whilst I don't see a need to have this
530 * aligned to any better than the cache line size (which it will be
531 * anyway by construction), let's align it to at least the way_size of
532 * any existing or conceivable SH-4 D-cache. -- RPC
533 */
534 base_addr = ((base_addr >> 16) << 16);
535 base_addr |= start;
536
537 dcache = &boot_cpu_data.dcache;
538 linesz = dcache->linesz;
539 way_incr = dcache->way_incr;
540 way_size = dcache->way_size;
541
542 a0 = base_addr;
543 a0e = base_addr + extent_per_way;
544 do {
545 asm volatile("ldc %0, sr" : : "r" (sr_with_bl));
546 asm volatile("movca.l r0, @%0\n\t"
547 "ocbi @%0" : : "r" (a0));
548 a0 += linesz;
549 asm volatile("movca.l r0, @%0\n\t"
550 "ocbi @%0" : : "r" (a0));
551 a0 += linesz;
552 asm volatile("movca.l r0, @%0\n\t"
553 "ocbi @%0" : : "r" (a0));
554 a0 += linesz;
555 asm volatile("movca.l r0, @%0\n\t"
556 "ocbi @%0" : : "r" (a0));
557 asm volatile("ldc %0, sr" : : "r" (orig_sr));
558 a0 += linesz;
559 } while (a0 < a0e);
560}
561
562static void __flush_dcache_segment_2way(unsigned long start,
563 unsigned long extent_per_way)
564{
565 unsigned long orig_sr, sr_with_bl;
566 unsigned long base_addr;
567 unsigned long way_incr, linesz, way_size;
568 struct cache_info *dcache;
569 register unsigned long a0, a1, a0e;
570
571 asm volatile("stc sr, %0" : "=r" (orig_sr));
572 sr_with_bl = orig_sr | (1<<28);
573 base_addr = ((unsigned long)&empty_zero_page[0]);
574
575 /* See comment under 1-way above */
576 base_addr = ((base_addr >> 16) << 16);
577 base_addr |= start;
578
579 dcache = &boot_cpu_data.dcache;
580 linesz = dcache->linesz;
581 way_incr = dcache->way_incr;
582 way_size = dcache->way_size;
583
584 a0 = base_addr;
585 a1 = a0 + way_incr;
586 a0e = base_addr + extent_per_way;
587 do {
588 asm volatile("ldc %0, sr" : : "r" (sr_with_bl));
589 asm volatile("movca.l r0, @%0\n\t"
590 "movca.l r0, @%1\n\t"
591 "ocbi @%0\n\t"
592 "ocbi @%1" : :
593 "r" (a0), "r" (a1));
594 a0 += linesz;
595 a1 += linesz;
596 asm volatile("movca.l r0, @%0\n\t"
597 "movca.l r0, @%1\n\t"
598 "ocbi @%0\n\t"
599 "ocbi @%1" : :
600 "r" (a0), "r" (a1));
601 a0 += linesz;
602 a1 += linesz;
603 asm volatile("movca.l r0, @%0\n\t"
604 "movca.l r0, @%1\n\t"
605 "ocbi @%0\n\t"
606 "ocbi @%1" : :
607 "r" (a0), "r" (a1));
608 a0 += linesz;
609 a1 += linesz;
610 asm volatile("movca.l r0, @%0\n\t"
611 "movca.l r0, @%1\n\t"
612 "ocbi @%0\n\t"
613 "ocbi @%1" : :
614 "r" (a0), "r" (a1));
615 asm volatile("ldc %0, sr" : : "r" (orig_sr));
616 a0 += linesz;
617 a1 += linesz;
618 } while (a0 < a0e);
619}
620
621static void __flush_dcache_segment_4way(unsigned long start,
622 unsigned long extent_per_way)
623{
624 unsigned long orig_sr, sr_with_bl;
625 unsigned long base_addr;
626 unsigned long way_incr, linesz, way_size;
627 struct cache_info *dcache;
628 register unsigned long a0, a1, a2, a3, a0e;
629
630 asm volatile("stc sr, %0" : "=r" (orig_sr));
631 sr_with_bl = orig_sr | (1<<28);
632 base_addr = ((unsigned long)&empty_zero_page[0]);
633
634 /* See comment under 1-way above */
635 base_addr = ((base_addr >> 16) << 16);
636 base_addr |= start;
637
638 dcache = &boot_cpu_data.dcache;
639 linesz = dcache->linesz;
640 way_incr = dcache->way_incr;
641 way_size = dcache->way_size;
642
643 a0 = base_addr;
644 a1 = a0 + way_incr;
645 a2 = a1 + way_incr;
646 a3 = a2 + way_incr;
647 a0e = base_addr + extent_per_way;
648 do {
649 asm volatile("ldc %0, sr" : : "r" (sr_with_bl));
650 asm volatile("movca.l r0, @%0\n\t"
651 "movca.l r0, @%1\n\t"
652 "movca.l r0, @%2\n\t"
653 "movca.l r0, @%3\n\t"
654 "ocbi @%0\n\t"
655 "ocbi @%1\n\t"
656 "ocbi @%2\n\t"
657 "ocbi @%3\n\t" : :
658 "r" (a0), "r" (a1), "r" (a2), "r" (a3));
659 a0 += linesz;
660 a1 += linesz;
661 a2 += linesz;
662 a3 += linesz;
663 asm volatile("movca.l r0, @%0\n\t"
664 "movca.l r0, @%1\n\t"
665 "movca.l r0, @%2\n\t"
666 "movca.l r0, @%3\n\t"
667 "ocbi @%0\n\t"
668 "ocbi @%1\n\t"
669 "ocbi @%2\n\t"
670 "ocbi @%3\n\t" : :
671 "r" (a0), "r" (a1), "r" (a2), "r" (a3));
672 a0 += linesz;
673 a1 += linesz;
674 a2 += linesz;
675 a3 += linesz;
676 asm volatile("movca.l r0, @%0\n\t"
677 "movca.l r0, @%1\n\t"
678 "movca.l r0, @%2\n\t"
679 "movca.l r0, @%3\n\t"
680 "ocbi @%0\n\t"
681 "ocbi @%1\n\t"
682 "ocbi @%2\n\t"
683 "ocbi @%3\n\t" : :
684 "r" (a0), "r" (a1), "r" (a2), "r" (a3));
685 a0 += linesz;
686 a1 += linesz;
687 a2 += linesz;
688 a3 += linesz;
689 asm volatile("movca.l r0, @%0\n\t"
690 "movca.l r0, @%1\n\t"
691 "movca.l r0, @%2\n\t"
692 "movca.l r0, @%3\n\t"
693 "ocbi @%0\n\t"
694 "ocbi @%1\n\t"
695 "ocbi @%2\n\t"
696 "ocbi @%3\n\t" : :
697 "r" (a0), "r" (a1), "r" (a2), "r" (a3));
698 asm volatile("ldc %0, sr" : : "r" (orig_sr));
699 a0 += linesz;
700 a1 += linesz;
701 a2 += linesz;
702 a3 += linesz;
703 } while (a0 < a0e);
704}
705
706extern void __weak sh4__flush_region_init(void); 372extern void __weak sh4__flush_region_init(void);
707 373
708/* 374/*
@@ -710,31 +376,10 @@ extern void __weak sh4__flush_region_init(void);
710 */ 376 */
711void __init sh4_cache_init(void) 377void __init sh4_cache_init(void)
712{ 378{
713 unsigned int wt_enabled = !!(__raw_readl(CCR) & CCR_CACHE_WT);
714
715 printk("PVR=%08x CVR=%08x PRR=%08x\n", 379 printk("PVR=%08x CVR=%08x PRR=%08x\n",
716 ctrl_inl(CCN_PVR), 380 __raw_readl(CCN_PVR),
717 ctrl_inl(CCN_CVR), 381 __raw_readl(CCN_CVR),
718 ctrl_inl(CCN_PRR)); 382 __raw_readl(CCN_PRR));
719
720 if (wt_enabled)
721 __flush_dcache_segment_fn = __flush_dcache_segment_writethrough;
722 else {
723 switch (boot_cpu_data.dcache.ways) {
724 case 1:
725 __flush_dcache_segment_fn = __flush_dcache_segment_1way;
726 break;
727 case 2:
728 __flush_dcache_segment_fn = __flush_dcache_segment_2way;
729 break;
730 case 4:
731 __flush_dcache_segment_fn = __flush_dcache_segment_4way;
732 break;
733 default:
734 panic("unknown number of cache ways\n");
735 break;
736 }
737 }
738 383
739 local_flush_icache_range = sh4_flush_icache_range; 384 local_flush_icache_range = sh4_flush_icache_range;
740 local_flush_dcache_page = sh4_flush_dcache_page; 385 local_flush_dcache_page = sh4_flush_dcache_page;