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("ic_conf = 0x%lx alias %d blk %d line %d shift %d\n",
176 *(unsigned long *) (&cache_info.ic_conf),
177 cache_info.ic_conf.cc_alias,
178 cache_info.ic_conf.cc_block,
179 cache_info.ic_conf.cc_line,
180 cache_info.ic_conf.cc_shift);
181 printk(" wt %d sh %d cst %d hv %d\n",
182 cache_info.ic_conf.cc_wt,
183 cache_info.ic_conf.cc_sh,
184 cache_info.ic_conf.cc_cst,
185 cache_info.ic_conf.cc_hv);
187 printk("D-TLB conf: sh %d page %d cst %d aid %d pad1 %d\n",
188 cache_info.dt_conf.tc_sh,
189 cache_info.dt_conf.tc_page,
190 cache_info.dt_conf.tc_cst,
191 cache_info.dt_conf.tc_aid,
192 cache_info.dt_conf.tc_pad1);
194 printk("I-TLB conf: sh %d page %d cst %d aid %d pad1 %d\n",
195 cache_info.it_conf.tc_sh,
196 cache_info.it_conf.tc_page,
197 cache_info.it_conf.tc_cst,
198 cache_info.it_conf.tc_aid,
199 cache_info.it_conf.tc_pad1);
203 if (cache_info.dt_conf.tc_sh == 0 || cache_info.dt_conf.tc_sh == 2) {
204 if (cache_info.dt_conf.tc_sh == 2)
205 printk(KERN_WARNING "Unexpected TLB configuration. "
206 "Will flush I/D separately (could be optimized).\n");
211 /* "New and Improved" version from Jim Hull
212 * (1 << (cc_block-1)) * (cc_line << (4 + cnf.cc_shift))
213 * The following CAFL_STRIDE is an optimized version, see
214 * http://lists.parisc-linux.org/pipermail/parisc-linux/2004-June/023625.html
215 * http://lists.parisc-linux.org/pipermail/parisc-linux/2004-June/023671.html
217 #define CAFL_STRIDE(cnf) (cnf.cc_line << (3 + cnf.cc_block + cnf.cc_shift))
218 dcache_stride = CAFL_STRIDE(cache_info.dc_conf);
219 icache_stride = CAFL_STRIDE(cache_info.ic_conf);
223 if (pdc_btlb_info(&btlb_info) < 0) {
224 memset(&btlb_info, 0, sizeof btlb_info);
228 if ((boot_cpu_data.pdc.capabilities & PDC_MODEL_NVA_MASK) ==
229 PDC_MODEL_NVA_UNSUPPORTED) {
230 printk(KERN_WARNING "parisc_cache_init: Only equivalent aliasing supported!\n");
232 panic("SMP kernel required to avoid non-equivalent aliasing");
237 void disable_sr_hashing(void)
239 int srhash_type, retval;
240 unsigned long space_bits;
242 switch (boot_cpu_data.cpu_type) {
243 case pcx: /* We shouldn't get this far. setup.c should prevent it. */
250 srhash_type = SRHASH_PCXST;
254 srhash_type = SRHASH_PCXL;
257 case pcxl2: /* pcxl2 doesn't support space register hashing */
260 default: /* Currently all PA2.0 machines use the same ins. sequence */
261 srhash_type = SRHASH_PA20;
265 disable_sr_hashing_asm(srhash_type);
267 retval = pdc_spaceid_bits(&space_bits);
268 /* If this procedure isn't implemented, don't panic. */
269 if (retval < 0 && retval != PDC_BAD_OPTION)
270 panic("pdc_spaceid_bits call failed.\n");
272 panic("SpaceID hashing is still on!\n");
276 __flush_cache_page(struct vm_area_struct *vma, unsigned long vmaddr,
277 unsigned long physaddr)
280 flush_dcache_page_asm(physaddr, vmaddr);
281 if (vma->vm_flags & VM_EXEC)
282 flush_icache_page_asm(physaddr, vmaddr);
286 void flush_dcache_page(struct page *page)
288 struct address_space *mapping = page_mapping(page);
289 struct vm_area_struct *mpnt;
290 unsigned long offset;
291 unsigned long addr, old_addr = 0;
294 if (mapping && !mapping_mapped(mapping)) {
295 set_bit(PG_dcache_dirty, &page->flags);
299 flush_kernel_dcache_page(page);
304 pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
306 /* We have carefully arranged in arch_get_unmapped_area() that
307 * *any* mappings of a file are always congruently mapped (whether
308 * declared as MAP_PRIVATE or MAP_SHARED), so we only need
309 * to flush one address here for them all to become coherent */
311 flush_dcache_mmap_lock(mapping);
312 vma_interval_tree_foreach(mpnt, &mapping->i_mmap, pgoff, pgoff) {
313 offset = (pgoff - mpnt->vm_pgoff) << PAGE_SHIFT;
314 addr = mpnt->vm_start + offset;
316 /* The TLB is the engine of coherence on parisc: The
317 * CPU is entitled to speculate any page with a TLB
318 * mapping, so here we kill the mapping then flush the
319 * page along a special flush only alias mapping.
320 * This guarantees that the page is no-longer in the
321 * cache for any process and nor may it be
322 * speculatively read in (until the user or kernel
323 * specifically accesses it, of course) */
325 flush_tlb_page(mpnt, addr);
326 if (old_addr == 0 || (old_addr & (SHMLBA - 1)) != (addr & (SHMLBA - 1))) {
327 __flush_cache_page(mpnt, addr, page_to_phys(page));
329 printk(KERN_ERR "INEQUIVALENT ALIASES 0x%lx and 0x%lx in file %s\n", old_addr, addr, mpnt->vm_file ? (char *)mpnt->vm_file->f_path.dentry->d_name.name : "(null)");
333 flush_dcache_mmap_unlock(mapping);
335 EXPORT_SYMBOL(flush_dcache_page);
337 /* Defined in arch/parisc/kernel/pacache.S */
338 EXPORT_SYMBOL(flush_kernel_dcache_range_asm);
339 EXPORT_SYMBOL(flush_kernel_dcache_page_asm);
340 EXPORT_SYMBOL(flush_data_cache_local);
341 EXPORT_SYMBOL(flush_kernel_icache_range_asm);
343 #define FLUSH_THRESHOLD 0x80000 /* 0.5MB */
344 int parisc_cache_flush_threshold __read_mostly = FLUSH_THRESHOLD;
346 void __init parisc_setup_cache_timing(void)
348 unsigned long rangetime, alltime;
353 alltime = mfctl(16) - alltime;
355 size = (unsigned long)(_end - _text);
356 rangetime = mfctl(16);
357 flush_kernel_dcache_range((unsigned long)_text, size);
358 rangetime = mfctl(16) - rangetime;
360 printk(KERN_DEBUG "Whole cache flush %lu cycles, flushing %lu bytes %lu cycles\n",
361 alltime, size, rangetime);
363 /* Racy, but if we see an intermediate value, it's ok too... */
364 parisc_cache_flush_threshold = size * alltime / rangetime;
366 parisc_cache_flush_threshold = (parisc_cache_flush_threshold + L1_CACHE_BYTES - 1) &~ (L1_CACHE_BYTES - 1);
367 if (!parisc_cache_flush_threshold)
368 parisc_cache_flush_threshold = FLUSH_THRESHOLD;
370 if (parisc_cache_flush_threshold > cache_info.dc_size)
371 parisc_cache_flush_threshold = cache_info.dc_size;
373 printk(KERN_INFO "Setting cache flush threshold to %x (%d CPUs online)\n", parisc_cache_flush_threshold, num_online_cpus());
376 extern void purge_kernel_dcache_page_asm(unsigned long);
377 extern void clear_user_page_asm(void *, unsigned long);
378 extern void copy_user_page_asm(void *, void *, unsigned long);
380 void flush_kernel_dcache_page_addr(void *addr)
384 flush_kernel_dcache_page_asm(addr);
385 purge_tlb_start(flags);
387 purge_tlb_end(flags);
389 EXPORT_SYMBOL(flush_kernel_dcache_page_addr);
391 void clear_user_page(void *vto, unsigned long vaddr, struct page *page)
394 if (!parisc_requires_coherency())
395 flush_kernel_dcache_page_asm(vto);
397 EXPORT_SYMBOL(clear_user_page);
399 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
402 /* Copy using kernel mapping. No coherency is needed
403 (all in kmap/kunmap) on machines that don't support
404 non-equivalent aliasing. However, the `from' page
405 needs to be flushed before it can be accessed through
406 the kernel mapping. */
408 flush_dcache_page_asm(__pa(vfrom), vaddr);
410 copy_page_asm(vto, vfrom);
411 if (!parisc_requires_coherency())
412 flush_kernel_dcache_page_asm(vto);
414 EXPORT_SYMBOL(copy_user_page);
418 void kunmap_parisc(void *addr)
420 if (parisc_requires_coherency())
421 flush_kernel_dcache_page_addr(addr);
423 EXPORT_SYMBOL(kunmap_parisc);
426 void purge_tlb_entries(struct mm_struct *mm, unsigned long addr)
430 /* Note: purge_tlb_entries can be called at startup with
433 purge_tlb_start(flags);
434 mtsp(mm->context, 1);
437 purge_tlb_end(flags);
439 EXPORT_SYMBOL(purge_tlb_entries);
441 void __flush_tlb_range(unsigned long sid, unsigned long start,
444 unsigned long npages;
446 npages = ((end - (start & PAGE_MASK)) + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
447 if (npages >= 512) /* 2MB of space: arbitrary, should be tuned */
452 purge_tlb_start(flags);
466 purge_tlb_end(flags);
470 static void cacheflush_h_tmp_function(void *dummy)
472 flush_cache_all_local();
475 void flush_cache_all(void)
477 on_each_cpu(cacheflush_h_tmp_function, NULL, 1);
480 static inline unsigned long mm_total_size(struct mm_struct *mm)
482 struct vm_area_struct *vma;
483 unsigned long usize = 0;
485 for (vma = mm->mmap; vma; vma = vma->vm_next)
486 usize += vma->vm_end - vma->vm_start;
490 static inline pte_t *get_ptep(pgd_t *pgd, unsigned long addr)
494 if (!pgd_none(*pgd)) {
495 pud_t *pud = pud_offset(pgd, addr);
496 if (!pud_none(*pud)) {
497 pmd_t *pmd = pmd_offset(pud, addr);
499 ptep = pte_offset_map(pmd, addr);
505 void flush_cache_mm(struct mm_struct *mm)
507 struct vm_area_struct *vma;
510 /* Flushing the whole cache on each cpu takes forever on
511 rp3440, etc. So, avoid it if the mm isn't too big. */
512 if (mm_total_size(mm) >= parisc_cache_flush_threshold) {
517 if (mm->context == mfsp(3)) {
518 for (vma = mm->mmap; vma; vma = vma->vm_next) {
519 flush_user_dcache_range_asm(vma->vm_start, vma->vm_end);
520 if ((vma->vm_flags & VM_EXEC) == 0)
522 flush_user_icache_range_asm(vma->vm_start, vma->vm_end);
528 for (vma = mm->mmap; vma; vma = vma->vm_next) {
531 for (addr = vma->vm_start; addr < vma->vm_end;
534 pte_t *ptep = get_ptep(pgd, addr);
537 pfn = pte_pfn(*ptep);
540 __flush_cache_page(vma, addr, PFN_PHYS(pfn));
546 flush_user_dcache_range(unsigned long start, unsigned long end)
548 if ((end - start) < parisc_cache_flush_threshold)
549 flush_user_dcache_range_asm(start,end);
555 flush_user_icache_range(unsigned long start, unsigned long end)
557 if ((end - start) < parisc_cache_flush_threshold)
558 flush_user_icache_range_asm(start,end);
560 flush_instruction_cache();
563 void flush_cache_range(struct vm_area_struct *vma,
564 unsigned long start, unsigned long end)
569 BUG_ON(!vma->vm_mm->context);
571 if ((end - start) >= parisc_cache_flush_threshold) {
576 if (vma->vm_mm->context == mfsp(3)) {
577 flush_user_dcache_range_asm(start, end);
578 if (vma->vm_flags & VM_EXEC)
579 flush_user_icache_range_asm(start, end);
583 pgd = vma->vm_mm->pgd;
584 for (addr = start & PAGE_MASK; addr < end; addr += PAGE_SIZE) {
586 pte_t *ptep = get_ptep(pgd, addr);
589 pfn = pte_pfn(*ptep);
591 __flush_cache_page(vma, addr, PFN_PHYS(pfn));
596 flush_cache_page(struct vm_area_struct *vma, unsigned long vmaddr, unsigned long pfn)
598 BUG_ON(!vma->vm_mm->context);
600 if (pfn_valid(pfn)) {
601 flush_tlb_page(vma, vmaddr);
602 __flush_cache_page(vma, vmaddr, PFN_PHYS(pfn));
606 #ifdef CONFIG_PARISC_TMPALIAS
608 void clear_user_highpage(struct page *page, unsigned long vaddr)
613 /* Clear using TMPALIAS region. The page doesn't need to
614 be flushed but the kernel mapping needs to be purged. */
616 vto = kmap_atomic(page);
618 /* The PA-RISC 2.0 Architecture book states on page F-6:
619 "Before a write-capable translation is enabled, *all*
620 non-equivalently-aliased translations must be removed
621 from the page table and purged from the TLB. (Note
622 that the caches are not required to be flushed at this
623 time.) Before any non-equivalent aliased translation
624 is re-enabled, the virtual address range for the writeable
625 page (the entire page) must be flushed from the cache,
626 and the write-capable translation removed from the page
627 table and purged from the TLB." */
629 purge_kernel_dcache_page_asm((unsigned long)vto);
630 purge_tlb_start(flags);
632 purge_tlb_end(flags);
634 clear_user_page_asm(vto, vaddr);
637 pagefault_enable(); /* kunmap_atomic(addr, KM_USER0); */
640 void copy_user_highpage(struct page *to, struct page *from,
641 unsigned long vaddr, struct vm_area_struct *vma)
646 /* Copy using TMPALIAS region. This has the advantage
647 that the `from' page doesn't need to be flushed. However,
648 the `to' page must be flushed in copy_user_page_asm since
649 it can be used to bring in executable code. */
651 vfrom = kmap_atomic(from);
652 vto = kmap_atomic(to);
654 purge_kernel_dcache_page_asm((unsigned long)vto);
655 purge_tlb_start(flags);
658 purge_tlb_end(flags);
660 copy_user_page_asm(vto, vfrom, vaddr);
661 flush_dcache_page_asm(__pa(vto), vaddr);
664 pagefault_enable(); /* kunmap_atomic(addr, KM_USER1); */
665 pagefault_enable(); /* kunmap_atomic(addr, KM_USER0); */
668 #endif /* CONFIG_PARISC_TMPALIAS */