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
6 * Copyright (C) 1999-2006 Helge Deller <deller@gmx.de> (07-13-1999)
7 * Copyright (C) 1999 SuSE GmbH Nuernberg
8 * Copyright (C) 2000 Philipp Rumpf (prumpf@tux.org)
10 * Cache and TLB management
14 #include <linux/init.h>
15 #include <linux/kernel.h>
17 #include <linux/module.h>
18 #include <linux/seq_file.h>
19 #include <linux/pagemap.h>
20 #include <linux/sched.h>
22 #include <asm/cache.h>
23 #include <asm/cacheflush.h>
24 #include <asm/tlbflush.h>
26 #include <asm/pgalloc.h>
27 #include <asm/processor.h>
28 #include <asm/sections.h>
29 #include <asm/shmparam.h>
31 int split_tlb __read_mostly
;
32 int dcache_stride __read_mostly
;
33 int icache_stride __read_mostly
;
34 EXPORT_SYMBOL(dcache_stride
);
36 void flush_dcache_page_asm(unsigned long phys_addr
, unsigned long vaddr
);
37 EXPORT_SYMBOL(flush_dcache_page_asm
);
38 void flush_icache_page_asm(unsigned long phys_addr
, unsigned long vaddr
);
41 /* On some machines (e.g. ones with the Merced bus), there can be
42 * only a single PxTLB broadcast at a time; this must be guaranteed
43 * by software. We put a spinlock around all TLB flushes to
46 DEFINE_SPINLOCK(pa_tlb_lock
);
48 struct pdc_cache_info cache_info __read_mostly
;
50 static struct pdc_btlb_info btlb_info __read_mostly
;
55 flush_data_cache(void)
57 on_each_cpu(flush_data_cache_local
, NULL
, 1);
60 flush_instruction_cache(void)
62 on_each_cpu(flush_instruction_cache_local
, NULL
, 1);
67 flush_cache_all_local(void)
69 flush_instruction_cache_local(NULL
);
70 flush_data_cache_local(NULL
);
72 EXPORT_SYMBOL(flush_cache_all_local
);
74 /* Virtual address of pfn. */
75 #define pfn_va(pfn) __va(PFN_PHYS(pfn))
78 update_mmu_cache(struct vm_area_struct
*vma
, unsigned long address
, pte_t
*ptep
)
80 unsigned long pfn
= pte_pfn(*ptep
);
83 /* We don't have pte special. As a result, we can be called with
84 an invalid pfn and we don't need to flush the kernel dcache page.
85 This occurs with FireGL card in C8000. */
89 page
= pfn_to_page(pfn
);
90 if (page_mapping(page
) && test_bit(PG_dcache_dirty
, &page
->flags
)) {
91 flush_kernel_dcache_page_addr(pfn_va(pfn
));
92 clear_bit(PG_dcache_dirty
, &page
->flags
);
93 } else if (parisc_requires_coherency())
94 flush_kernel_dcache_page_addr(pfn_va(pfn
));
98 show_cache_info(struct seq_file
*m
)
102 seq_printf(m
, "I-cache\t\t: %ld KB\n",
103 cache_info
.ic_size
/1024 );
104 if (cache_info
.dc_loop
!= 1)
105 snprintf(buf
, 32, "%lu-way associative", cache_info
.dc_loop
);
106 seq_printf(m
, "D-cache\t\t: %ld KB (%s%s, %s)\n",
107 cache_info
.dc_size
/1024,
108 (cache_info
.dc_conf
.cc_wt
? "WT":"WB"),
109 (cache_info
.dc_conf
.cc_sh
? ", shared I/D":""),
110 ((cache_info
.dc_loop
== 1) ? "direct mapped" : buf
));
111 seq_printf(m
, "ITLB entries\t: %ld\n" "DTLB entries\t: %ld%s\n",
114 cache_info
.dt_conf
.tc_sh
? " - shared with ITLB":""
118 /* BTLB - Block TLB */
119 if (btlb_info
.max_size
==0) {
120 seq_printf(m
, "BTLB\t\t: not supported\n" );
123 "BTLB fixed\t: max. %d pages, pagesize=%d (%dMB)\n"
124 "BTLB fix-entr.\t: %d instruction, %d data (%d combined)\n"
125 "BTLB var-entr.\t: %d instruction, %d data (%d combined)\n",
126 btlb_info
.max_size
, (int)4096,
127 btlb_info
.max_size
>>8,
128 btlb_info
.fixed_range_info
.num_i
,
129 btlb_info
.fixed_range_info
.num_d
,
130 btlb_info
.fixed_range_info
.num_comb
,
131 btlb_info
.variable_range_info
.num_i
,
132 btlb_info
.variable_range_info
.num_d
,
133 btlb_info
.variable_range_info
.num_comb
140 parisc_cache_init(void)
142 if (pdc_cache_info(&cache_info
) < 0)
143 panic("parisc_cache_init: pdc_cache_info failed");
146 printk("ic_size %lx dc_size %lx it_size %lx\n",
151 printk("DC base 0x%lx stride 0x%lx count 0x%lx loop 0x%lx\n",
153 cache_info
.dc_stride
,
157 printk("dc_conf = 0x%lx alias %d blk %d line %d shift %d\n",
158 *(unsigned long *) (&cache_info
.dc_conf
),
159 cache_info
.dc_conf
.cc_alias
,
160 cache_info
.dc_conf
.cc_block
,
161 cache_info
.dc_conf
.cc_line
,
162 cache_info
.dc_conf
.cc_shift
);
163 printk(" wt %d sh %d cst %d hv %d\n",
164 cache_info
.dc_conf
.cc_wt
,
165 cache_info
.dc_conf
.cc_sh
,
166 cache_info
.dc_conf
.cc_cst
,
167 cache_info
.dc_conf
.cc_hv
);
169 printk("IC base 0x%lx stride 0x%lx count 0x%lx loop 0x%lx\n",
171 cache_info
.ic_stride
,
175 printk("IT base 0x%lx stride 0x%lx count 0x%lx loop 0x%lx off_base 0x%lx off_stride 0x%lx off_count 0x%lx\n",
176 cache_info
.it_sp_base
,
177 cache_info
.it_sp_stride
,
178 cache_info
.it_sp_count
,
180 cache_info
.it_off_base
,
181 cache_info
.it_off_stride
,
182 cache_info
.it_off_count
);
184 printk("DT base 0x%lx stride 0x%lx count 0x%lx loop 0x%lx off_base 0x%lx off_stride 0x%lx off_count 0x%lx\n",
185 cache_info
.dt_sp_base
,
186 cache_info
.dt_sp_stride
,
187 cache_info
.dt_sp_count
,
189 cache_info
.dt_off_base
,
190 cache_info
.dt_off_stride
,
191 cache_info
.dt_off_count
);
193 printk("ic_conf = 0x%lx alias %d blk %d line %d shift %d\n",
194 *(unsigned long *) (&cache_info
.ic_conf
),
195 cache_info
.ic_conf
.cc_alias
,
196 cache_info
.ic_conf
.cc_block
,
197 cache_info
.ic_conf
.cc_line
,
198 cache_info
.ic_conf
.cc_shift
);
199 printk(" wt %d sh %d cst %d hv %d\n",
200 cache_info
.ic_conf
.cc_wt
,
201 cache_info
.ic_conf
.cc_sh
,
202 cache_info
.ic_conf
.cc_cst
,
203 cache_info
.ic_conf
.cc_hv
);
205 printk("D-TLB conf: sh %d page %d cst %d aid %d sr %d\n",
206 cache_info
.dt_conf
.tc_sh
,
207 cache_info
.dt_conf
.tc_page
,
208 cache_info
.dt_conf
.tc_cst
,
209 cache_info
.dt_conf
.tc_aid
,
210 cache_info
.dt_conf
.tc_sr
);
212 printk("I-TLB conf: sh %d page %d cst %d aid %d sr %d\n",
213 cache_info
.it_conf
.tc_sh
,
214 cache_info
.it_conf
.tc_page
,
215 cache_info
.it_conf
.tc_cst
,
216 cache_info
.it_conf
.tc_aid
,
217 cache_info
.it_conf
.tc_sr
);
221 if (cache_info
.dt_conf
.tc_sh
== 0 || cache_info
.dt_conf
.tc_sh
== 2) {
222 if (cache_info
.dt_conf
.tc_sh
== 2)
223 printk(KERN_WARNING
"Unexpected TLB configuration. "
224 "Will flush I/D separately (could be optimized).\n");
229 /* "New and Improved" version from Jim Hull
230 * (1 << (cc_block-1)) * (cc_line << (4 + cnf.cc_shift))
231 * The following CAFL_STRIDE is an optimized version, see
232 * http://lists.parisc-linux.org/pipermail/parisc-linux/2004-June/023625.html
233 * http://lists.parisc-linux.org/pipermail/parisc-linux/2004-June/023671.html
235 #define CAFL_STRIDE(cnf) (cnf.cc_line << (3 + cnf.cc_block + cnf.cc_shift))
236 dcache_stride
= CAFL_STRIDE(cache_info
.dc_conf
);
237 icache_stride
= CAFL_STRIDE(cache_info
.ic_conf
);
241 if (pdc_btlb_info(&btlb_info
) < 0) {
242 memset(&btlb_info
, 0, sizeof btlb_info
);
246 if ((boot_cpu_data
.pdc
.capabilities
& PDC_MODEL_NVA_MASK
) ==
247 PDC_MODEL_NVA_UNSUPPORTED
) {
248 printk(KERN_WARNING
"parisc_cache_init: Only equivalent aliasing supported!\n");
250 panic("SMP kernel required to avoid non-equivalent aliasing");
255 void disable_sr_hashing(void)
257 int srhash_type
, retval
;
258 unsigned long space_bits
;
260 switch (boot_cpu_data
.cpu_type
) {
261 case pcx
: /* We shouldn't get this far. setup.c should prevent it. */
268 srhash_type
= SRHASH_PCXST
;
272 srhash_type
= SRHASH_PCXL
;
275 case pcxl2
: /* pcxl2 doesn't support space register hashing */
278 default: /* Currently all PA2.0 machines use the same ins. sequence */
279 srhash_type
= SRHASH_PA20
;
283 disable_sr_hashing_asm(srhash_type
);
285 retval
= pdc_spaceid_bits(&space_bits
);
286 /* If this procedure isn't implemented, don't panic. */
287 if (retval
< 0 && retval
!= PDC_BAD_OPTION
)
288 panic("pdc_spaceid_bits call failed.\n");
290 panic("SpaceID hashing is still on!\n");
294 __flush_cache_page(struct vm_area_struct
*vma
, unsigned long vmaddr
,
295 unsigned long physaddr
)
298 flush_dcache_page_asm(physaddr
, vmaddr
);
299 if (vma
->vm_flags
& VM_EXEC
)
300 flush_icache_page_asm(physaddr
, vmaddr
);
304 void flush_dcache_page(struct page
*page
)
306 struct address_space
*mapping
= page_mapping(page
);
307 struct vm_area_struct
*mpnt
;
308 unsigned long offset
;
309 unsigned long addr
, old_addr
= 0;
312 if (mapping
&& !mapping_mapped(mapping
)) {
313 set_bit(PG_dcache_dirty
, &page
->flags
);
317 flush_kernel_dcache_page(page
);
324 /* We have carefully arranged in arch_get_unmapped_area() that
325 * *any* mappings of a file are always congruently mapped (whether
326 * declared as MAP_PRIVATE or MAP_SHARED), so we only need
327 * to flush one address here for them all to become coherent */
329 flush_dcache_mmap_lock(mapping
);
330 vma_interval_tree_foreach(mpnt
, &mapping
->i_mmap
, pgoff
, pgoff
) {
331 offset
= (pgoff
- mpnt
->vm_pgoff
) << PAGE_SHIFT
;
332 addr
= mpnt
->vm_start
+ offset
;
334 /* The TLB is the engine of coherence on parisc: The
335 * CPU is entitled to speculate any page with a TLB
336 * mapping, so here we kill the mapping then flush the
337 * page along a special flush only alias mapping.
338 * This guarantees that the page is no-longer in the
339 * cache for any process and nor may it be
340 * speculatively read in (until the user or kernel
341 * specifically accesses it, of course) */
343 flush_tlb_page(mpnt
, addr
);
344 if (old_addr
== 0 || (old_addr
& (SHM_COLOUR
- 1))
345 != (addr
& (SHM_COLOUR
- 1))) {
346 __flush_cache_page(mpnt
, addr
, page_to_phys(page
));
348 printk(KERN_ERR
"INEQUIVALENT ALIASES 0x%lx and 0x%lx in file %pD\n", old_addr
, addr
, mpnt
->vm_file
);
352 flush_dcache_mmap_unlock(mapping
);
354 EXPORT_SYMBOL(flush_dcache_page
);
356 /* Defined in arch/parisc/kernel/pacache.S */
357 EXPORT_SYMBOL(flush_kernel_dcache_range_asm
);
358 EXPORT_SYMBOL(flush_kernel_dcache_page_asm
);
359 EXPORT_SYMBOL(flush_data_cache_local
);
360 EXPORT_SYMBOL(flush_kernel_icache_range_asm
);
362 #define FLUSH_THRESHOLD 0x80000 /* 0.5MB */
363 static unsigned long parisc_cache_flush_threshold __read_mostly
= FLUSH_THRESHOLD
;
365 #define FLUSH_TLB_THRESHOLD (2*1024*1024) /* 2MB initial TLB threshold */
366 static unsigned long parisc_tlb_flush_threshold __read_mostly
= FLUSH_TLB_THRESHOLD
;
368 void __init
parisc_setup_cache_timing(void)
370 unsigned long rangetime
, alltime
;
371 unsigned long size
, start
;
372 unsigned long threshold
;
376 alltime
= mfctl(16) - alltime
;
378 size
= (unsigned long)(_end
- _text
);
379 rangetime
= mfctl(16);
380 flush_kernel_dcache_range((unsigned long)_text
, size
);
381 rangetime
= mfctl(16) - rangetime
;
383 printk(KERN_DEBUG
"Whole cache flush %lu cycles, flushing %lu bytes %lu cycles\n",
384 alltime
, size
, rangetime
);
386 threshold
= L1_CACHE_ALIGN(size
* alltime
/ rangetime
);
387 if (threshold
> cache_info
.dc_size
)
388 threshold
= cache_info
.dc_size
;
390 parisc_cache_flush_threshold
= threshold
;
391 printk(KERN_INFO
"Cache flush threshold set to %lu KiB\n",
392 parisc_cache_flush_threshold
/1024);
394 /* calculate TLB flush threshold */
396 /* On SMP machines, skip the TLB measure of kernel text which
397 * has been mapped as huge pages. */
398 if (num_online_cpus() > 1 && !parisc_requires_coherency()) {
399 threshold
= max(cache_info
.it_size
, cache_info
.dt_size
);
400 threshold
*= PAGE_SIZE
;
401 threshold
/= num_online_cpus();
402 goto set_tlb_threshold
;
407 alltime
= mfctl(16) - alltime
;
410 start
= (unsigned long) _text
;
411 rangetime
= mfctl(16);
412 while (start
< (unsigned long) _end
) {
413 flush_tlb_kernel_range(start
, start
+ PAGE_SIZE
);
417 rangetime
= mfctl(16) - rangetime
;
419 printk(KERN_DEBUG
"Whole TLB flush %lu cycles, flushing %lu bytes %lu cycles\n",
420 alltime
, size
, rangetime
);
422 threshold
= PAGE_ALIGN(num_online_cpus() * size
* alltime
/ rangetime
);
426 parisc_tlb_flush_threshold
= threshold
;
427 printk(KERN_INFO
"TLB flush threshold set to %lu KiB\n",
428 parisc_tlb_flush_threshold
/1024);
431 extern void purge_kernel_dcache_page_asm(unsigned long);
432 extern void clear_user_page_asm(void *, unsigned long);
433 extern void copy_user_page_asm(void *, void *, unsigned long);
435 void flush_kernel_dcache_page_addr(void *addr
)
439 flush_kernel_dcache_page_asm(addr
);
440 purge_tlb_start(flags
);
442 purge_tlb_end(flags
);
444 EXPORT_SYMBOL(flush_kernel_dcache_page_addr
);
446 void copy_user_page(void *vto
, void *vfrom
, unsigned long vaddr
,
449 /* Copy using kernel mapping. No coherency is needed (all in
450 kunmap) for the `to' page. However, the `from' page needs to
451 be flushed through a mapping equivalent to the user mapping
452 before it can be accessed through the kernel mapping. */
454 flush_dcache_page_asm(__pa(vfrom
), vaddr
);
456 copy_page_asm(vto
, vfrom
);
458 EXPORT_SYMBOL(copy_user_page
);
460 /* __flush_tlb_range()
462 * returns 1 if all TLBs were flushed.
464 int __flush_tlb_range(unsigned long sid
, unsigned long start
,
467 unsigned long flags
, size
;
469 size
= (end
- start
);
470 if (size
>= parisc_tlb_flush_threshold
) {
475 /* Purge TLB entries for small ranges using the pdtlb and
476 pitlb instructions. These instructions execute locally
477 but cause a purge request to be broadcast to other TLBs. */
478 if (likely(!split_tlb
)) {
479 while (start
< end
) {
480 purge_tlb_start(flags
);
483 purge_tlb_end(flags
);
490 while (start
< end
) {
491 purge_tlb_start(flags
);
495 purge_tlb_end(flags
);
501 static void cacheflush_h_tmp_function(void *dummy
)
503 flush_cache_all_local();
506 void flush_cache_all(void)
508 on_each_cpu(cacheflush_h_tmp_function
, NULL
, 1);
511 static inline unsigned long mm_total_size(struct mm_struct
*mm
)
513 struct vm_area_struct
*vma
;
514 unsigned long usize
= 0;
516 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
)
517 usize
+= vma
->vm_end
- vma
->vm_start
;
521 static inline pte_t
*get_ptep(pgd_t
*pgd
, unsigned long addr
)
525 if (!pgd_none(*pgd
)) {
526 pud_t
*pud
= pud_offset(pgd
, addr
);
527 if (!pud_none(*pud
)) {
528 pmd_t
*pmd
= pmd_offset(pud
, addr
);
530 ptep
= pte_offset_map(pmd
, addr
);
536 void flush_cache_mm(struct mm_struct
*mm
)
538 struct vm_area_struct
*vma
;
541 /* Flushing the whole cache on each cpu takes forever on
542 rp3440, etc. So, avoid it if the mm isn't too big. */
543 if (mm_total_size(mm
) >= parisc_cache_flush_threshold
) {
548 if (mm
->context
== mfsp(3)) {
549 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
550 flush_user_dcache_range_asm(vma
->vm_start
, vma
->vm_end
);
551 if ((vma
->vm_flags
& VM_EXEC
) == 0)
553 flush_user_icache_range_asm(vma
->vm_start
, vma
->vm_end
);
559 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
562 for (addr
= vma
->vm_start
; addr
< vma
->vm_end
;
565 pte_t
*ptep
= get_ptep(pgd
, addr
);
568 pfn
= pte_pfn(*ptep
);
571 __flush_cache_page(vma
, addr
, PFN_PHYS(pfn
));
577 flush_user_dcache_range(unsigned long start
, unsigned long end
)
579 if ((end
- start
) < parisc_cache_flush_threshold
)
580 flush_user_dcache_range_asm(start
,end
);
586 flush_user_icache_range(unsigned long start
, unsigned long end
)
588 if ((end
- start
) < parisc_cache_flush_threshold
)
589 flush_user_icache_range_asm(start
,end
);
591 flush_instruction_cache();
594 void flush_cache_range(struct vm_area_struct
*vma
,
595 unsigned long start
, unsigned long end
)
600 BUG_ON(!vma
->vm_mm
->context
);
602 if ((end
- start
) >= parisc_cache_flush_threshold
) {
607 if (vma
->vm_mm
->context
== mfsp(3)) {
608 flush_user_dcache_range_asm(start
, end
);
609 if (vma
->vm_flags
& VM_EXEC
)
610 flush_user_icache_range_asm(start
, end
);
614 pgd
= vma
->vm_mm
->pgd
;
615 for (addr
= start
& PAGE_MASK
; addr
< end
; addr
+= PAGE_SIZE
) {
617 pte_t
*ptep
= get_ptep(pgd
, addr
);
620 pfn
= pte_pfn(*ptep
);
622 __flush_cache_page(vma
, addr
, PFN_PHYS(pfn
));
627 flush_cache_page(struct vm_area_struct
*vma
, unsigned long vmaddr
, unsigned long pfn
)
629 BUG_ON(!vma
->vm_mm
->context
);
631 if (pfn_valid(pfn
)) {
632 flush_tlb_page(vma
, vmaddr
);
633 __flush_cache_page(vma
, vmaddr
, PFN_PHYS(pfn
));