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>
21 #include <linux/sched/mm.h>
23 #include <asm/cache.h>
24 #include <asm/cacheflush.h>
25 #include <asm/tlbflush.h>
27 #include <asm/pgalloc.h>
28 #include <asm/processor.h>
29 #include <asm/sections.h>
30 #include <asm/shmparam.h>
32 int split_tlb __ro_after_init
;
33 int dcache_stride __ro_after_init
;
34 int icache_stride __ro_after_init
;
35 EXPORT_SYMBOL(dcache_stride
);
37 void flush_dcache_page_asm(unsigned long phys_addr
, unsigned long vaddr
);
38 EXPORT_SYMBOL(flush_dcache_page_asm
);
39 void purge_dcache_page_asm(unsigned long phys_addr
, unsigned long vaddr
);
40 void flush_icache_page_asm(unsigned long phys_addr
, unsigned long vaddr
);
43 /* On some machines (i.e., ones with the Merced bus), there can be
44 * only a single PxTLB broadcast at a time; this must be guaranteed
45 * by software. We need a spinlock around all TLB flushes to ensure
48 DEFINE_SPINLOCK(pa_tlb_flush_lock
);
50 /* Swapper page setup lock. */
51 DEFINE_SPINLOCK(pa_swapper_pg_lock
);
53 #if defined(CONFIG_64BIT) && defined(CONFIG_SMP)
54 int pa_serialize_tlb_flushes __ro_after_init
;
57 struct pdc_cache_info cache_info __ro_after_init
;
59 static struct pdc_btlb_info btlb_info __ro_after_init
;
64 flush_data_cache(void)
66 on_each_cpu(flush_data_cache_local
, NULL
, 1);
69 flush_instruction_cache(void)
71 on_each_cpu(flush_instruction_cache_local
, NULL
, 1);
76 flush_cache_all_local(void)
78 flush_instruction_cache_local(NULL
);
79 flush_data_cache_local(NULL
);
81 EXPORT_SYMBOL(flush_cache_all_local
);
83 /* Virtual address of pfn. */
84 #define pfn_va(pfn) __va(PFN_PHYS(pfn))
87 update_mmu_cache(struct vm_area_struct
*vma
, unsigned long address
, pte_t
*ptep
)
89 unsigned long pfn
= pte_pfn(*ptep
);
92 /* We don't have pte special. As a result, we can be called with
93 an invalid pfn and we don't need to flush the kernel dcache page.
94 This occurs with FireGL card in C8000. */
98 page
= pfn_to_page(pfn
);
99 if (page_mapping_file(page
) &&
100 test_bit(PG_dcache_dirty
, &page
->flags
)) {
101 flush_kernel_dcache_page_addr(pfn_va(pfn
));
102 clear_bit(PG_dcache_dirty
, &page
->flags
);
103 } else if (parisc_requires_coherency())
104 flush_kernel_dcache_page_addr(pfn_va(pfn
));
108 show_cache_info(struct seq_file
*m
)
112 seq_printf(m
, "I-cache\t\t: %ld KB\n",
113 cache_info
.ic_size
/1024 );
114 if (cache_info
.dc_loop
!= 1)
115 snprintf(buf
, 32, "%lu-way associative", cache_info
.dc_loop
);
116 seq_printf(m
, "D-cache\t\t: %ld KB (%s%s, %s)\n",
117 cache_info
.dc_size
/1024,
118 (cache_info
.dc_conf
.cc_wt
? "WT":"WB"),
119 (cache_info
.dc_conf
.cc_sh
? ", shared I/D":""),
120 ((cache_info
.dc_loop
== 1) ? "direct mapped" : buf
));
121 seq_printf(m
, "ITLB entries\t: %ld\n" "DTLB entries\t: %ld%s\n",
124 cache_info
.dt_conf
.tc_sh
? " - shared with ITLB":""
128 /* BTLB - Block TLB */
129 if (btlb_info
.max_size
==0) {
130 seq_printf(m
, "BTLB\t\t: not supported\n" );
133 "BTLB fixed\t: max. %d pages, pagesize=%d (%dMB)\n"
134 "BTLB fix-entr.\t: %d instruction, %d data (%d combined)\n"
135 "BTLB var-entr.\t: %d instruction, %d data (%d combined)\n",
136 btlb_info
.max_size
, (int)4096,
137 btlb_info
.max_size
>>8,
138 btlb_info
.fixed_range_info
.num_i
,
139 btlb_info
.fixed_range_info
.num_d
,
140 btlb_info
.fixed_range_info
.num_comb
,
141 btlb_info
.variable_range_info
.num_i
,
142 btlb_info
.variable_range_info
.num_d
,
143 btlb_info
.variable_range_info
.num_comb
150 parisc_cache_init(void)
152 if (pdc_cache_info(&cache_info
) < 0)
153 panic("parisc_cache_init: pdc_cache_info failed");
156 printk("ic_size %lx dc_size %lx it_size %lx\n",
161 printk("DC base 0x%lx stride 0x%lx count 0x%lx loop 0x%lx\n",
163 cache_info
.dc_stride
,
167 printk("dc_conf = 0x%lx alias %d blk %d line %d shift %d\n",
168 *(unsigned long *) (&cache_info
.dc_conf
),
169 cache_info
.dc_conf
.cc_alias
,
170 cache_info
.dc_conf
.cc_block
,
171 cache_info
.dc_conf
.cc_line
,
172 cache_info
.dc_conf
.cc_shift
);
173 printk(" wt %d sh %d cst %d hv %d\n",
174 cache_info
.dc_conf
.cc_wt
,
175 cache_info
.dc_conf
.cc_sh
,
176 cache_info
.dc_conf
.cc_cst
,
177 cache_info
.dc_conf
.cc_hv
);
179 printk("IC base 0x%lx stride 0x%lx count 0x%lx loop 0x%lx\n",
181 cache_info
.ic_stride
,
185 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",
186 cache_info
.it_sp_base
,
187 cache_info
.it_sp_stride
,
188 cache_info
.it_sp_count
,
190 cache_info
.it_off_base
,
191 cache_info
.it_off_stride
,
192 cache_info
.it_off_count
);
194 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",
195 cache_info
.dt_sp_base
,
196 cache_info
.dt_sp_stride
,
197 cache_info
.dt_sp_count
,
199 cache_info
.dt_off_base
,
200 cache_info
.dt_off_stride
,
201 cache_info
.dt_off_count
);
203 printk("ic_conf = 0x%lx alias %d blk %d line %d shift %d\n",
204 *(unsigned long *) (&cache_info
.ic_conf
),
205 cache_info
.ic_conf
.cc_alias
,
206 cache_info
.ic_conf
.cc_block
,
207 cache_info
.ic_conf
.cc_line
,
208 cache_info
.ic_conf
.cc_shift
);
209 printk(" wt %d sh %d cst %d hv %d\n",
210 cache_info
.ic_conf
.cc_wt
,
211 cache_info
.ic_conf
.cc_sh
,
212 cache_info
.ic_conf
.cc_cst
,
213 cache_info
.ic_conf
.cc_hv
);
215 printk("D-TLB conf: sh %d page %d cst %d aid %d sr %d\n",
216 cache_info
.dt_conf
.tc_sh
,
217 cache_info
.dt_conf
.tc_page
,
218 cache_info
.dt_conf
.tc_cst
,
219 cache_info
.dt_conf
.tc_aid
,
220 cache_info
.dt_conf
.tc_sr
);
222 printk("I-TLB conf: sh %d page %d cst %d aid %d sr %d\n",
223 cache_info
.it_conf
.tc_sh
,
224 cache_info
.it_conf
.tc_page
,
225 cache_info
.it_conf
.tc_cst
,
226 cache_info
.it_conf
.tc_aid
,
227 cache_info
.it_conf
.tc_sr
);
231 if (cache_info
.dt_conf
.tc_sh
== 0 || cache_info
.dt_conf
.tc_sh
== 2) {
232 if (cache_info
.dt_conf
.tc_sh
== 2)
233 printk(KERN_WARNING
"Unexpected TLB configuration. "
234 "Will flush I/D separately (could be optimized).\n");
239 /* "New and Improved" version from Jim Hull
240 * (1 << (cc_block-1)) * (cc_line << (4 + cnf.cc_shift))
241 * The following CAFL_STRIDE is an optimized version, see
242 * http://lists.parisc-linux.org/pipermail/parisc-linux/2004-June/023625.html
243 * http://lists.parisc-linux.org/pipermail/parisc-linux/2004-June/023671.html
245 #define CAFL_STRIDE(cnf) (cnf.cc_line << (3 + cnf.cc_block + cnf.cc_shift))
246 dcache_stride
= CAFL_STRIDE(cache_info
.dc_conf
);
247 icache_stride
= CAFL_STRIDE(cache_info
.ic_conf
);
251 if (pdc_btlb_info(&btlb_info
) < 0) {
252 memset(&btlb_info
, 0, sizeof btlb_info
);
256 if ((boot_cpu_data
.pdc
.capabilities
& PDC_MODEL_NVA_MASK
) ==
257 PDC_MODEL_NVA_UNSUPPORTED
) {
258 printk(KERN_WARNING
"parisc_cache_init: Only equivalent aliasing supported!\n");
260 panic("SMP kernel required to avoid non-equivalent aliasing");
265 void __init
disable_sr_hashing(void)
267 int srhash_type
, retval
;
268 unsigned long space_bits
;
270 switch (boot_cpu_data
.cpu_type
) {
271 case pcx
: /* We shouldn't get this far. setup.c should prevent it. */
278 srhash_type
= SRHASH_PCXST
;
282 srhash_type
= SRHASH_PCXL
;
285 case pcxl2
: /* pcxl2 doesn't support space register hashing */
288 default: /* Currently all PA2.0 machines use the same ins. sequence */
289 srhash_type
= SRHASH_PA20
;
293 disable_sr_hashing_asm(srhash_type
);
295 retval
= pdc_spaceid_bits(&space_bits
);
296 /* If this procedure isn't implemented, don't panic. */
297 if (retval
< 0 && retval
!= PDC_BAD_OPTION
)
298 panic("pdc_spaceid_bits call failed.\n");
300 panic("SpaceID hashing is still on!\n");
304 __flush_cache_page(struct vm_area_struct
*vma
, unsigned long vmaddr
,
305 unsigned long physaddr
)
308 flush_dcache_page_asm(physaddr
, vmaddr
);
309 if (vma
->vm_flags
& VM_EXEC
)
310 flush_icache_page_asm(physaddr
, vmaddr
);
315 __purge_cache_page(struct vm_area_struct
*vma
, unsigned long vmaddr
,
316 unsigned long physaddr
)
319 purge_dcache_page_asm(physaddr
, vmaddr
);
320 if (vma
->vm_flags
& VM_EXEC
)
321 flush_icache_page_asm(physaddr
, vmaddr
);
325 void flush_dcache_page(struct page
*page
)
327 struct address_space
*mapping
= page_mapping_file(page
);
328 struct vm_area_struct
*mpnt
;
329 unsigned long offset
;
330 unsigned long addr
, old_addr
= 0;
333 if (mapping
&& !mapping_mapped(mapping
)) {
334 set_bit(PG_dcache_dirty
, &page
->flags
);
338 flush_kernel_dcache_page(page
);
345 /* We have carefully arranged in arch_get_unmapped_area() that
346 * *any* mappings of a file are always congruently mapped (whether
347 * declared as MAP_PRIVATE or MAP_SHARED), so we only need
348 * to flush one address here for them all to become coherent */
350 flush_dcache_mmap_lock(mapping
);
351 vma_interval_tree_foreach(mpnt
, &mapping
->i_mmap
, pgoff
, pgoff
) {
352 offset
= (pgoff
- mpnt
->vm_pgoff
) << PAGE_SHIFT
;
353 addr
= mpnt
->vm_start
+ offset
;
355 /* The TLB is the engine of coherence on parisc: The
356 * CPU is entitled to speculate any page with a TLB
357 * mapping, so here we kill the mapping then flush the
358 * page along a special flush only alias mapping.
359 * This guarantees that the page is no-longer in the
360 * cache for any process and nor may it be
361 * speculatively read in (until the user or kernel
362 * specifically accesses it, of course) */
364 flush_tlb_page(mpnt
, addr
);
365 if (old_addr
== 0 || (old_addr
& (SHM_COLOUR
- 1))
366 != (addr
& (SHM_COLOUR
- 1))) {
367 __flush_cache_page(mpnt
, addr
, page_to_phys(page
));
368 if (parisc_requires_coherency() && old_addr
)
369 printk(KERN_ERR
"INEQUIVALENT ALIASES 0x%lx and 0x%lx in file %pD\n", old_addr
, addr
, mpnt
->vm_file
);
373 flush_dcache_mmap_unlock(mapping
);
375 EXPORT_SYMBOL(flush_dcache_page
);
377 /* Defined in arch/parisc/kernel/pacache.S */
378 EXPORT_SYMBOL(flush_kernel_dcache_range_asm
);
379 EXPORT_SYMBOL(flush_kernel_dcache_page_asm
);
380 EXPORT_SYMBOL(flush_data_cache_local
);
381 EXPORT_SYMBOL(flush_kernel_icache_range_asm
);
383 #define FLUSH_THRESHOLD 0x80000 /* 0.5MB */
384 static unsigned long parisc_cache_flush_threshold __ro_after_init
= FLUSH_THRESHOLD
;
386 #define FLUSH_TLB_THRESHOLD (16*1024) /* 16 KiB minimum TLB threshold */
387 static unsigned long parisc_tlb_flush_threshold __ro_after_init
= FLUSH_TLB_THRESHOLD
;
389 void __init
parisc_setup_cache_timing(void)
391 unsigned long rangetime
, alltime
;
392 unsigned long size
, start
;
393 unsigned long threshold
;
397 alltime
= mfctl(16) - alltime
;
399 size
= (unsigned long)(_end
- _text
);
400 rangetime
= mfctl(16);
401 flush_kernel_dcache_range((unsigned long)_text
, size
);
402 rangetime
= mfctl(16) - rangetime
;
404 printk(KERN_DEBUG
"Whole cache flush %lu cycles, flushing %lu bytes %lu cycles\n",
405 alltime
, size
, rangetime
);
407 threshold
= L1_CACHE_ALIGN(size
* alltime
/ rangetime
);
408 if (threshold
> cache_info
.dc_size
)
409 threshold
= cache_info
.dc_size
;
411 parisc_cache_flush_threshold
= threshold
;
412 printk(KERN_INFO
"Cache flush threshold set to %lu KiB\n",
413 parisc_cache_flush_threshold
/1024);
415 /* calculate TLB flush threshold */
417 /* On SMP machines, skip the TLB measure of kernel text which
418 * has been mapped as huge pages. */
419 if (num_online_cpus() > 1 && !parisc_requires_coherency()) {
420 threshold
= max(cache_info
.it_size
, cache_info
.dt_size
);
421 threshold
*= PAGE_SIZE
;
422 threshold
/= num_online_cpus();
423 goto set_tlb_threshold
;
427 start
= (unsigned long) _text
;
428 rangetime
= mfctl(16);
429 while (start
< (unsigned long) _end
) {
430 flush_tlb_kernel_range(start
, start
+ PAGE_SIZE
);
434 rangetime
= mfctl(16) - rangetime
;
438 alltime
= mfctl(16) - alltime
;
440 printk(KERN_INFO
"Whole TLB flush %lu cycles, Range flush %lu bytes %lu cycles\n",
441 alltime
, size
, rangetime
);
443 threshold
= PAGE_ALIGN((num_online_cpus() * size
* alltime
) / rangetime
);
444 printk(KERN_INFO
"Calculated TLB flush threshold %lu KiB\n",
448 if (threshold
> parisc_tlb_flush_threshold
)
449 parisc_tlb_flush_threshold
= threshold
;
450 printk(KERN_INFO
"TLB flush threshold set to %lu KiB\n",
451 parisc_tlb_flush_threshold
/1024);
454 extern void purge_kernel_dcache_page_asm(unsigned long);
455 extern void clear_user_page_asm(void *, unsigned long);
456 extern void copy_user_page_asm(void *, void *, unsigned long);
458 void flush_kernel_dcache_page_addr(void *addr
)
462 flush_kernel_dcache_page_asm(addr
);
463 purge_tlb_start(flags
);
465 purge_tlb_end(flags
);
467 EXPORT_SYMBOL(flush_kernel_dcache_page_addr
);
469 void copy_user_page(void *vto
, void *vfrom
, unsigned long vaddr
,
472 /* Copy using kernel mapping. No coherency is needed (all in
473 kunmap) for the `to' page. However, the `from' page needs to
474 be flushed through a mapping equivalent to the user mapping
475 before it can be accessed through the kernel mapping. */
477 flush_dcache_page_asm(__pa(vfrom
), vaddr
);
478 copy_page_asm(vto
, vfrom
);
481 EXPORT_SYMBOL(copy_user_page
);
483 /* __flush_tlb_range()
485 * returns 1 if all TLBs were flushed.
487 int __flush_tlb_range(unsigned long sid
, unsigned long start
,
492 if ((!IS_ENABLED(CONFIG_SMP
) || !arch_irqs_disabled()) &&
493 end
- start
>= parisc_tlb_flush_threshold
) {
498 /* Purge TLB entries for small ranges using the pdtlb and
499 pitlb instructions. These instructions execute locally
500 but cause a purge request to be broadcast to other TLBs. */
501 while (start
< end
) {
502 purge_tlb_start(flags
);
506 purge_tlb_end(flags
);
512 static void cacheflush_h_tmp_function(void *dummy
)
514 flush_cache_all_local();
517 void flush_cache_all(void)
519 on_each_cpu(cacheflush_h_tmp_function
, NULL
, 1);
522 static inline unsigned long mm_total_size(struct mm_struct
*mm
)
524 struct vm_area_struct
*vma
;
525 unsigned long usize
= 0;
527 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
)
528 usize
+= vma
->vm_end
- vma
->vm_start
;
532 static inline pte_t
*get_ptep(pgd_t
*pgd
, unsigned long addr
)
536 if (!pgd_none(*pgd
)) {
537 p4d_t
*p4d
= p4d_offset(pgd
, addr
);
538 if (!p4d_none(*p4d
)) {
539 pud_t
*pud
= pud_offset(p4d
, addr
);
540 if (!pud_none(*pud
)) {
541 pmd_t
*pmd
= pmd_offset(pud
, addr
);
543 ptep
= pte_offset_map(pmd
, addr
);
550 void flush_cache_mm(struct mm_struct
*mm
)
552 struct vm_area_struct
*vma
;
555 /* Flushing the whole cache on each cpu takes forever on
556 rp3440, etc. So, avoid it if the mm isn't too big. */
557 if ((!IS_ENABLED(CONFIG_SMP
) || !arch_irqs_disabled()) &&
558 mm_total_size(mm
) >= parisc_cache_flush_threshold
) {
565 if (mm
->context
== mfsp(3)) {
566 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
567 flush_user_dcache_range_asm(vma
->vm_start
, vma
->vm_end
);
568 if (vma
->vm_flags
& VM_EXEC
)
569 flush_user_icache_range_asm(vma
->vm_start
, vma
->vm_end
);
570 flush_tlb_range(vma
, vma
->vm_start
, vma
->vm_end
);
576 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
579 for (addr
= vma
->vm_start
; addr
< vma
->vm_end
;
582 pte_t
*ptep
= get_ptep(pgd
, addr
);
585 pfn
= pte_pfn(*ptep
);
588 if (unlikely(mm
->context
)) {
589 flush_tlb_page(vma
, addr
);
590 __flush_cache_page(vma
, addr
, PFN_PHYS(pfn
));
592 __purge_cache_page(vma
, addr
, PFN_PHYS(pfn
));
598 void flush_cache_range(struct vm_area_struct
*vma
,
599 unsigned long start
, unsigned long end
)
604 if ((!IS_ENABLED(CONFIG_SMP
) || !arch_irqs_disabled()) &&
605 end
- start
>= parisc_cache_flush_threshold
) {
606 if (vma
->vm_mm
->context
)
607 flush_tlb_range(vma
, start
, end
);
612 if (vma
->vm_mm
->context
== mfsp(3)) {
613 flush_user_dcache_range_asm(start
, end
);
614 if (vma
->vm_flags
& VM_EXEC
)
615 flush_user_icache_range_asm(start
, end
);
616 flush_tlb_range(vma
, start
, end
);
620 pgd
= vma
->vm_mm
->pgd
;
621 for (addr
= vma
->vm_start
; addr
< vma
->vm_end
; addr
+= PAGE_SIZE
) {
623 pte_t
*ptep
= get_ptep(pgd
, addr
);
626 pfn
= pte_pfn(*ptep
);
627 if (pfn_valid(pfn
)) {
628 if (unlikely(vma
->vm_mm
->context
)) {
629 flush_tlb_page(vma
, addr
);
630 __flush_cache_page(vma
, addr
, PFN_PHYS(pfn
));
632 __purge_cache_page(vma
, addr
, PFN_PHYS(pfn
));
639 flush_cache_page(struct vm_area_struct
*vma
, unsigned long vmaddr
, unsigned long pfn
)
641 if (pfn_valid(pfn
)) {
642 if (likely(vma
->vm_mm
->context
)) {
643 flush_tlb_page(vma
, vmaddr
);
644 __flush_cache_page(vma
, vmaddr
, PFN_PHYS(pfn
));
646 __purge_cache_page(vma
, vmaddr
, PFN_PHYS(pfn
));
651 void flush_kernel_vmap_range(void *vaddr
, int size
)
653 unsigned long start
= (unsigned long)vaddr
;
654 unsigned long end
= start
+ size
;
656 if ((!IS_ENABLED(CONFIG_SMP
) || !arch_irqs_disabled()) &&
657 (unsigned long)size
>= parisc_cache_flush_threshold
) {
658 flush_tlb_kernel_range(start
, end
);
663 flush_kernel_dcache_range_asm(start
, end
);
664 flush_tlb_kernel_range(start
, end
);
666 EXPORT_SYMBOL(flush_kernel_vmap_range
);
668 void invalidate_kernel_vmap_range(void *vaddr
, int size
)
670 unsigned long start
= (unsigned long)vaddr
;
671 unsigned long end
= start
+ size
;
673 if ((!IS_ENABLED(CONFIG_SMP
) || !arch_irqs_disabled()) &&
674 (unsigned long)size
>= parisc_cache_flush_threshold
) {
675 flush_tlb_kernel_range(start
, end
);
680 purge_kernel_dcache_range_asm(start
, end
);
681 flush_tlb_kernel_range(start
, end
);
683 EXPORT_SYMBOL(invalidate_kernel_vmap_range
);