2 * Copyright 2002 Andi Kleen, SuSE Labs.
3 * Thanks to Ben LaHaise for precious feedback.
5 #include <linux/highmem.h>
6 #include <linux/bootmem.h>
7 #include <linux/module.h>
8 #include <linux/sched.h>
9 #include <linux/slab.h>
11 #include <linux/interrupt.h>
12 #include <linux/seq_file.h>
13 #include <linux/debugfs.h>
16 #include <asm/processor.h>
17 #include <asm/tlbflush.h>
18 #include <asm/sections.h>
19 #include <asm/uaccess.h>
20 #include <asm/pgalloc.h>
21 #include <asm/proto.h>
25 * The current flushing context - we pass it instead of 5 arguments:
34 unsigned force_split : 1;
38 #define CPA_FLUSHTLB 1
42 static unsigned long direct_pages_count[PG_LEVEL_NUM];
44 void update_page_count(int level, unsigned long pages)
48 /* Protect against CPA */
49 spin_lock_irqsave(&pgd_lock, flags);
50 direct_pages_count[level] += pages;
51 spin_unlock_irqrestore(&pgd_lock, flags);
54 static void split_page_count(int level)
56 direct_pages_count[level]--;
57 direct_pages_count[level - 1] += PTRS_PER_PTE;
60 int arch_report_meminfo(char *page)
62 int n = sprintf(page, "DirectMap4k: %8lu kB\n",
63 direct_pages_count[PG_LEVEL_4K] << 2);
64 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
65 n += sprintf(page + n, "DirectMap2M: %8lu kB\n",
66 direct_pages_count[PG_LEVEL_2M] << 11);
68 n += sprintf(page + n, "DirectMap4M: %8lu kB\n",
69 direct_pages_count[PG_LEVEL_2M] << 12);
73 n += sprintf(page + n, "DirectMap1G: %8lu kB\n",
74 direct_pages_count[PG_LEVEL_1G] << 20);
79 static inline void split_page_count(int level) { }
84 static inline unsigned long highmap_start_pfn(void)
86 return __pa(_text) >> PAGE_SHIFT;
89 static inline unsigned long highmap_end_pfn(void)
91 return __pa(round_up((unsigned long)_end, PMD_SIZE)) >> PAGE_SHIFT;
96 #ifdef CONFIG_DEBUG_PAGEALLOC
97 # define debug_pagealloc 1
99 # define debug_pagealloc 0
103 within(unsigned long addr, unsigned long start, unsigned long end)
105 return addr >= start && addr < end;
113 * clflush_cache_range - flush a cache range with clflush
114 * @addr: virtual start address
115 * @size: number of bytes to flush
117 * clflush is an unordered instruction which needs fencing with mfence
118 * to avoid ordering issues.
120 void clflush_cache_range(void *vaddr, unsigned int size)
122 void *vend = vaddr + size - 1;
126 for (; vaddr < vend; vaddr += boot_cpu_data.x86_clflush_size)
129 * Flush any possible final partial cacheline:
136 static void __cpa_flush_all(void *arg)
138 unsigned long cache = (unsigned long)arg;
141 * Flush all to work around Errata in early athlons regarding
142 * large page flushing.
146 if (cache && boot_cpu_data.x86_model >= 4)
150 static void cpa_flush_all(unsigned long cache)
152 BUG_ON(irqs_disabled());
154 on_each_cpu(__cpa_flush_all, (void *) cache, 1);
157 static void __cpa_flush_range(void *arg)
160 * We could optimize that further and do individual per page
161 * tlb invalidates for a low number of pages. Caveat: we must
162 * flush the high aliases on 64bit as well.
167 static void cpa_flush_range(unsigned long start, int numpages, int cache)
169 unsigned int i, level;
172 BUG_ON(irqs_disabled());
173 WARN_ON(PAGE_ALIGN(start) != start);
175 on_each_cpu(__cpa_flush_range, NULL, 1);
181 * We only need to flush on one CPU,
182 * clflush is a MESI-coherent instruction that
183 * will cause all other CPUs to flush the same
186 for (i = 0, addr = start; i < numpages; i++, addr += PAGE_SIZE) {
187 pte_t *pte = lookup_address(addr, &level);
190 * Only flush present addresses:
192 if (pte && (pte_val(*pte) & _PAGE_PRESENT))
193 clflush_cache_range((void *) addr, PAGE_SIZE);
197 static void cpa_flush_array(unsigned long *start, int numpages, int cache)
199 unsigned int i, level;
202 BUG_ON(irqs_disabled());
204 on_each_cpu(__cpa_flush_range, NULL, 1);
210 if (numpages >= 1024) {
211 if (boot_cpu_data.x86_model >= 4)
216 * We only need to flush on one CPU,
217 * clflush is a MESI-coherent instruction that
218 * will cause all other CPUs to flush the same
221 for (i = 0, addr = start; i < numpages; i++, addr++) {
222 pte_t *pte = lookup_address(*addr, &level);
225 * Only flush present addresses:
227 if (pte && (pte_val(*pte) & _PAGE_PRESENT))
228 clflush_cache_range((void *) *addr, PAGE_SIZE);
233 * Certain areas of memory on x86 require very specific protection flags,
234 * for example the BIOS area or kernel text. Callers don't always get this
235 * right (again, ioremap() on BIOS memory is not uncommon) so this function
236 * checks and fixes these known static required protection bits.
238 static inline pgprot_t static_protections(pgprot_t prot, unsigned long address,
241 pgprot_t forbidden = __pgprot(0);
244 * The BIOS area between 640k and 1Mb needs to be executable for
245 * PCI BIOS based config access (CONFIG_PCI_GOBIOS) support.
247 if (within(pfn, BIOS_BEGIN >> PAGE_SHIFT, BIOS_END >> PAGE_SHIFT))
248 pgprot_val(forbidden) |= _PAGE_NX;
251 * The kernel text needs to be executable for obvious reasons
252 * Does not cover __inittext since that is gone later on. On
253 * 64bit we do not enforce !NX on the low mapping
255 if (within(address, (unsigned long)_text, (unsigned long)_etext))
256 pgprot_val(forbidden) |= _PAGE_NX;
259 * The .rodata section needs to be read-only. Using the pfn
260 * catches all aliases.
262 if (within(pfn, __pa((unsigned long)__start_rodata) >> PAGE_SHIFT,
263 __pa((unsigned long)__end_rodata) >> PAGE_SHIFT))
264 pgprot_val(forbidden) |= _PAGE_RW;
266 prot = __pgprot(pgprot_val(prot) & ~pgprot_val(forbidden));
272 * Lookup the page table entry for a virtual address. Return a pointer
273 * to the entry and the level of the mapping.
275 * Note: We return pud and pmd either when the entry is marked large
276 * or when the present bit is not set. Otherwise we would return a
277 * pointer to a nonexisting mapping.
279 pte_t *lookup_address(unsigned long address, unsigned int *level)
281 pgd_t *pgd = pgd_offset_k(address);
285 *level = PG_LEVEL_NONE;
290 pud = pud_offset(pgd, address);
294 *level = PG_LEVEL_1G;
295 if (pud_large(*pud) || !pud_present(*pud))
298 pmd = pmd_offset(pud, address);
302 *level = PG_LEVEL_2M;
303 if (pmd_large(*pmd) || !pmd_present(*pmd))
306 *level = PG_LEVEL_4K;
308 return pte_offset_kernel(pmd, address);
310 EXPORT_SYMBOL_GPL(lookup_address);
313 * Set the new pmd in all the pgds we know about:
315 static void __set_pmd_pte(pte_t *kpte, unsigned long address, pte_t pte)
318 set_pte_atomic(kpte, pte);
320 if (!SHARED_KERNEL_PMD) {
323 list_for_each_entry(page, &pgd_list, lru) {
328 pgd = (pgd_t *)page_address(page) + pgd_index(address);
329 pud = pud_offset(pgd, address);
330 pmd = pmd_offset(pud, address);
331 set_pte_atomic((pte_t *)pmd, pte);
338 try_preserve_large_page(pte_t *kpte, unsigned long address,
339 struct cpa_data *cpa)
341 unsigned long nextpage_addr, numpages, pmask, psize, flags, addr, pfn;
342 pte_t new_pte, old_pte, *tmp;
343 pgprot_t old_prot, new_prot;
347 if (cpa->force_split)
350 spin_lock_irqsave(&pgd_lock, flags);
352 * Check for races, another CPU might have split this page
355 tmp = lookup_address(address, &level);
361 psize = PMD_PAGE_SIZE;
362 pmask = PMD_PAGE_MASK;
366 psize = PUD_PAGE_SIZE;
367 pmask = PUD_PAGE_MASK;
376 * Calculate the number of pages, which fit into this large
377 * page starting at address:
379 nextpage_addr = (address + psize) & pmask;
380 numpages = (nextpage_addr - address) >> PAGE_SHIFT;
381 if (numpages < cpa->numpages)
382 cpa->numpages = numpages;
385 * We are safe now. Check whether the new pgprot is the same:
388 old_prot = new_prot = pte_pgprot(old_pte);
390 pgprot_val(new_prot) &= ~pgprot_val(cpa->mask_clr);
391 pgprot_val(new_prot) |= pgprot_val(cpa->mask_set);
394 * old_pte points to the large page base address. So we need
395 * to add the offset of the virtual address:
397 pfn = pte_pfn(old_pte) + ((address & (psize - 1)) >> PAGE_SHIFT);
400 new_prot = static_protections(new_prot, address, pfn);
403 * We need to check the full range, whether
404 * static_protection() requires a different pgprot for one of
405 * the pages in the range we try to preserve:
407 addr = address + PAGE_SIZE;
409 for (i = 1; i < cpa->numpages; i++, addr += PAGE_SIZE, pfn++) {
410 pgprot_t chk_prot = static_protections(new_prot, addr, pfn);
412 if (pgprot_val(chk_prot) != pgprot_val(new_prot))
417 * If there are no changes, return. maxpages has been updated
420 if (pgprot_val(new_prot) == pgprot_val(old_prot)) {
426 * We need to change the attributes. Check, whether we can
427 * change the large page in one go. We request a split, when
428 * the address is not aligned and the number of pages is
429 * smaller than the number of pages in the large page. Note
430 * that we limited the number of possible pages already to
431 * the number of pages in the large page.
433 if (address == (nextpage_addr - psize) && cpa->numpages == numpages) {
435 * The address is aligned and the number of pages
436 * covers the full page.
438 new_pte = pfn_pte(pte_pfn(old_pte), canon_pgprot(new_prot));
439 __set_pmd_pte(kpte, address, new_pte);
440 cpa->flags |= CPA_FLUSHTLB;
445 spin_unlock_irqrestore(&pgd_lock, flags);
450 static LIST_HEAD(page_pool);
451 static unsigned long pool_size, pool_pages, pool_low;
452 static unsigned long pool_used, pool_failed;
454 static void cpa_fill_pool(struct page **ret)
456 gfp_t gfp = GFP_KERNEL;
461 * Avoid recursion (on debug-pagealloc) and also signal
462 * our priority to get to these pagetables:
464 if (current->flags & PF_MEMALLOC)
466 current->flags |= PF_MEMALLOC;
469 * Allocate atomically from atomic contexts:
471 if (in_atomic() || irqs_disabled() || debug_pagealloc)
472 gfp = GFP_ATOMIC | __GFP_NORETRY | __GFP_NOWARN;
474 while (pool_pages < pool_size || (ret && !*ret)) {
475 p = alloc_pages(gfp, 0);
481 * If the call site needs a page right now, provide it:
487 spin_lock_irqsave(&pgd_lock, flags);
488 list_add(&p->lru, &page_pool);
490 spin_unlock_irqrestore(&pgd_lock, flags);
493 current->flags &= ~PF_MEMALLOC;
496 #define SHIFT_MB (20 - PAGE_SHIFT)
497 #define ROUND_MB_GB ((1 << 10) - 1)
498 #define SHIFT_MB_GB 10
499 #define POOL_PAGES_PER_GB 16
501 void __init cpa_init(void)
508 * Calculate the number of pool pages:
510 * Convert totalram (nr of pages) to MiB and round to the next
511 * GiB. Shift MiB to Gib and multiply the result by
514 if (debug_pagealloc) {
515 gb = ((si.totalram >> SHIFT_MB) + ROUND_MB_GB) >> SHIFT_MB_GB;
516 pool_size = POOL_PAGES_PER_GB * gb;
520 pool_low = pool_size;
524 "CPA: page pool initialized %lu of %lu pages preallocated\n",
525 pool_pages, pool_size);
528 static int split_large_page(pte_t *kpte, unsigned long address)
530 unsigned long flags, pfn, pfninc = 1;
531 unsigned int i, level;
537 * Get a page from the pool. The pool list is protected by the
538 * pgd_lock, which we have to take anyway for the split
541 spin_lock_irqsave(&pgd_lock, flags);
542 if (list_empty(&page_pool)) {
543 spin_unlock_irqrestore(&pgd_lock, flags);
545 cpa_fill_pool(&base);
548 spin_lock_irqsave(&pgd_lock, flags);
550 base = list_first_entry(&page_pool, struct page, lru);
551 list_del(&base->lru);
554 if (pool_pages < pool_low)
555 pool_low = pool_pages;
559 * Check for races, another CPU might have split this page
562 tmp = lookup_address(address, &level);
566 pbase = (pte_t *)page_address(base);
567 paravirt_alloc_pte(&init_mm, page_to_pfn(base));
568 ref_prot = pte_pgprot(pte_clrhuge(*kpte));
571 if (level == PG_LEVEL_1G) {
572 pfninc = PMD_PAGE_SIZE >> PAGE_SHIFT;
573 pgprot_val(ref_prot) |= _PAGE_PSE;
578 * Get the target pfn from the original entry:
580 pfn = pte_pfn(*kpte);
581 for (i = 0; i < PTRS_PER_PTE; i++, pfn += pfninc)
582 set_pte(&pbase[i], pfn_pte(pfn, ref_prot));
584 if (address >= (unsigned long)__va(0) &&
585 address < (unsigned long)__va(max_low_pfn_mapped << PAGE_SHIFT))
586 split_page_count(level);
589 if (address >= (unsigned long)__va(1UL<<32) &&
590 address < (unsigned long)__va(max_pfn_mapped << PAGE_SHIFT))
591 split_page_count(level);
595 * Install the new, split up pagetable. Important details here:
597 * On Intel the NX bit of all levels must be cleared to make a
598 * page executable. See section 4.13.2 of Intel 64 and IA-32
599 * Architectures Software Developer's Manual).
601 * Mark the entry present. The current mapping might be
602 * set to not present, which we preserved above.
604 ref_prot = pte_pgprot(pte_mkexec(pte_clrhuge(*kpte)));
605 pgprot_val(ref_prot) |= _PAGE_PRESENT;
606 __set_pmd_pte(kpte, address, mk_pte(base, ref_prot));
611 * If we dropped out via the lookup_address check under
612 * pgd_lock then stick the page back into the pool:
615 list_add(&base->lru, &page_pool);
619 spin_unlock_irqrestore(&pgd_lock, flags);
624 static int __change_page_attr(struct cpa_data *cpa, int primary)
626 unsigned long address;
629 pte_t *kpte, old_pte;
631 if (cpa->flags & CPA_ARRAY)
632 address = cpa->vaddr[cpa->curpage];
634 address = *cpa->vaddr;
637 kpte = lookup_address(address, &level);
642 if (!pte_val(old_pte)) {
645 WARN(1, KERN_WARNING "CPA: called for zero pte. "
646 "vaddr = %lx cpa->vaddr = %lx\n", address,
651 if (level == PG_LEVEL_4K) {
653 pgprot_t new_prot = pte_pgprot(old_pte);
654 unsigned long pfn = pte_pfn(old_pte);
656 pgprot_val(new_prot) &= ~pgprot_val(cpa->mask_clr);
657 pgprot_val(new_prot) |= pgprot_val(cpa->mask_set);
659 new_prot = static_protections(new_prot, address, pfn);
662 * We need to keep the pfn from the existing PTE,
663 * after all we're only going to change it's attributes
664 * not the memory it points to
666 new_pte = pfn_pte(pfn, canon_pgprot(new_prot));
669 * Do we really change anything ?
671 if (pte_val(old_pte) != pte_val(new_pte)) {
672 set_pte_atomic(kpte, new_pte);
673 cpa->flags |= CPA_FLUSHTLB;
680 * Check, whether we can keep the large page intact
681 * and just change the pte:
683 do_split = try_preserve_large_page(kpte, address, cpa);
685 * When the range fits into the existing large page,
686 * return. cp->numpages and cpa->tlbflush have been updated in
693 * We have to split the large page:
695 err = split_large_page(kpte, address);
697 cpa->flags |= CPA_FLUSHTLB;
704 static int __change_page_attr_set_clr(struct cpa_data *cpa, int checkalias);
706 static int cpa_process_alias(struct cpa_data *cpa)
708 struct cpa_data alias_cpa;
710 unsigned long temp_cpa_vaddr, vaddr;
712 if (cpa->pfn >= max_pfn_mapped)
716 if (cpa->pfn >= max_low_pfn_mapped && cpa->pfn < (1UL<<(32-PAGE_SHIFT)))
720 * No need to redo, when the primary call touched the direct
723 if (cpa->flags & CPA_ARRAY)
724 vaddr = cpa->vaddr[cpa->curpage];
728 if (!(within(vaddr, PAGE_OFFSET,
729 PAGE_OFFSET + (max_low_pfn_mapped << PAGE_SHIFT))
731 || within(vaddr, PAGE_OFFSET + (1UL<<32),
732 PAGE_OFFSET + (max_pfn_mapped << PAGE_SHIFT))
737 temp_cpa_vaddr = (unsigned long) __va(cpa->pfn << PAGE_SHIFT);
738 alias_cpa.vaddr = &temp_cpa_vaddr;
739 alias_cpa.flags &= ~CPA_ARRAY;
742 ret = __change_page_attr_set_clr(&alias_cpa, 0);
749 * No need to redo, when the primary call touched the high
752 if (within(vaddr, (unsigned long) _text, (unsigned long) _end))
756 * If the physical address is inside the kernel map, we need
757 * to touch the high mapped kernel as well:
759 if (!within(cpa->pfn, highmap_start_pfn(), highmap_end_pfn()))
763 temp_cpa_vaddr = (cpa->pfn << PAGE_SHIFT) + __START_KERNEL_map - phys_base;
764 alias_cpa.vaddr = &temp_cpa_vaddr;
765 alias_cpa.flags &= ~CPA_ARRAY;
768 * The high mapping range is imprecise, so ignore the return value.
770 __change_page_attr_set_clr(&alias_cpa, 0);
775 static int __change_page_attr_set_clr(struct cpa_data *cpa, int checkalias)
777 int ret, numpages = cpa->numpages;
781 * Store the remaining nr of pages for the large page
782 * preservation check.
784 cpa->numpages = numpages;
785 /* for array changes, we can't use large page */
786 if (cpa->flags & CPA_ARRAY)
789 ret = __change_page_attr(cpa, checkalias);
794 ret = cpa_process_alias(cpa);
800 * Adjust the number of pages with the result of the
801 * CPA operation. Either a large page has been
802 * preserved or a single page update happened.
804 BUG_ON(cpa->numpages > numpages);
805 numpages -= cpa->numpages;
806 if (cpa->flags & CPA_ARRAY)
809 *cpa->vaddr += cpa->numpages * PAGE_SIZE;
815 static inline int cache_attr(pgprot_t attr)
817 return pgprot_val(attr) &
818 (_PAGE_PAT | _PAGE_PAT_LARGE | _PAGE_PWT | _PAGE_PCD);
821 static int change_page_attr_set_clr(unsigned long *addr, int numpages,
822 pgprot_t mask_set, pgprot_t mask_clr,
823 int force_split, int array)
826 int ret, cache, checkalias;
829 * Check, if we are requested to change a not supported
832 mask_set = canon_pgprot(mask_set);
833 mask_clr = canon_pgprot(mask_clr);
834 if (!pgprot_val(mask_set) && !pgprot_val(mask_clr) && !force_split)
837 /* Ensure we are PAGE_SIZE aligned */
839 if (*addr & ~PAGE_MASK) {
842 * People should not be passing in unaligned addresses:
848 for (i = 0; i < numpages; i++) {
849 if (addr[i] & ~PAGE_MASK) {
850 addr[i] &= PAGE_MASK;
856 /* Must avoid aliasing mappings in the highmem code */
860 cpa.numpages = numpages;
861 cpa.mask_set = mask_set;
862 cpa.mask_clr = mask_clr;
865 cpa.force_split = force_split;
868 cpa.flags |= CPA_ARRAY;
870 /* No alias checking for _NX bit modifications */
871 checkalias = (pgprot_val(mask_set) | pgprot_val(mask_clr)) != _PAGE_NX;
873 ret = __change_page_attr_set_clr(&cpa, checkalias);
876 * Check whether we really changed something:
878 if (!(cpa.flags & CPA_FLUSHTLB))
882 * No need to flush, when we did not set any of the caching
885 cache = cache_attr(mask_set);
888 * On success we use clflush, when the CPU supports it to
889 * avoid the wbindv. If the CPU does not support it and in the
890 * error case we fall back to cpa_flush_all (which uses
893 if (!ret && cpu_has_clflush) {
894 if (cpa.flags & CPA_ARRAY)
895 cpa_flush_array(addr, numpages, cache);
897 cpa_flush_range(*addr, numpages, cache);
899 cpa_flush_all(cache);
907 static inline int change_page_attr_set(unsigned long *addr, int numpages,
908 pgprot_t mask, int array)
910 return change_page_attr_set_clr(addr, numpages, mask, __pgprot(0), 0,
914 static inline int change_page_attr_clear(unsigned long *addr, int numpages,
915 pgprot_t mask, int array)
917 return change_page_attr_set_clr(addr, numpages, __pgprot(0), mask, 0,
921 int _set_memory_uc(unsigned long addr, int numpages)
924 * for now UC MINUS. see comments in ioremap_nocache()
926 return change_page_attr_set(&addr, numpages,
927 __pgprot(_PAGE_CACHE_UC_MINUS), 0);
930 int set_memory_uc(unsigned long addr, int numpages)
933 * for now UC MINUS. see comments in ioremap_nocache()
935 if (reserve_memtype(__pa(addr), __pa(addr) + numpages * PAGE_SIZE,
936 _PAGE_CACHE_UC_MINUS, NULL))
939 return _set_memory_uc(addr, numpages);
941 EXPORT_SYMBOL(set_memory_uc);
943 int set_memory_array_uc(unsigned long *addr, int addrinarray)
949 * for now UC MINUS. see comments in ioremap_nocache()
951 for (i = 0; i < addrinarray; i++) {
952 start = __pa(addr[i]);
953 for (end = start + PAGE_SIZE; i < addrinarray - 1; end += PAGE_SIZE) {
954 if (end != __pa(addr[i + 1]))
958 if (reserve_memtype(start, end, _PAGE_CACHE_UC_MINUS, NULL))
962 return change_page_attr_set(addr, addrinarray,
963 __pgprot(_PAGE_CACHE_UC_MINUS), 1);
965 for (i = 0; i < addrinarray; i++) {
966 unsigned long tmp = __pa(addr[i]);
970 for (end = start + PAGE_SIZE; i < addrinarray - 1; end += PAGE_SIZE) {
971 if (end != __pa(addr[i + 1]))
975 free_memtype(tmp, end);
979 EXPORT_SYMBOL(set_memory_array_uc);
981 int _set_memory_wc(unsigned long addr, int numpages)
983 return change_page_attr_set(&addr, numpages,
984 __pgprot(_PAGE_CACHE_WC), 0);
987 int set_memory_wc(unsigned long addr, int numpages)
990 return set_memory_uc(addr, numpages);
992 if (reserve_memtype(__pa(addr), __pa(addr) + numpages * PAGE_SIZE,
993 _PAGE_CACHE_WC, NULL))
996 return _set_memory_wc(addr, numpages);
998 EXPORT_SYMBOL(set_memory_wc);
1000 int _set_memory_wb(unsigned long addr, int numpages)
1002 return change_page_attr_clear(&addr, numpages,
1003 __pgprot(_PAGE_CACHE_MASK), 0);
1006 int set_memory_wb(unsigned long addr, int numpages)
1008 free_memtype(__pa(addr), __pa(addr) + numpages * PAGE_SIZE);
1010 return _set_memory_wb(addr, numpages);
1012 EXPORT_SYMBOL(set_memory_wb);
1014 int set_memory_array_wb(unsigned long *addr, int addrinarray)
1018 for (i = 0; i < addrinarray; i++) {
1019 unsigned long start = __pa(addr[i]);
1022 for (end = start + PAGE_SIZE; i < addrinarray - 1; end += PAGE_SIZE) {
1023 if (end != __pa(addr[i + 1]))
1027 free_memtype(start, end);
1029 return change_page_attr_clear(addr, addrinarray,
1030 __pgprot(_PAGE_CACHE_MASK), 1);
1032 EXPORT_SYMBOL(set_memory_array_wb);
1034 int set_memory_x(unsigned long addr, int numpages)
1036 return change_page_attr_clear(&addr, numpages, __pgprot(_PAGE_NX), 0);
1038 EXPORT_SYMBOL(set_memory_x);
1040 int set_memory_nx(unsigned long addr, int numpages)
1042 return change_page_attr_set(&addr, numpages, __pgprot(_PAGE_NX), 0);
1044 EXPORT_SYMBOL(set_memory_nx);
1046 int set_memory_ro(unsigned long addr, int numpages)
1048 return change_page_attr_clear(&addr, numpages, __pgprot(_PAGE_RW), 0);
1051 int set_memory_rw(unsigned long addr, int numpages)
1053 return change_page_attr_set(&addr, numpages, __pgprot(_PAGE_RW), 0);
1056 int set_memory_np(unsigned long addr, int numpages)
1058 return change_page_attr_clear(&addr, numpages, __pgprot(_PAGE_PRESENT), 0);
1061 int set_memory_4k(unsigned long addr, int numpages)
1063 return change_page_attr_set_clr(&addr, numpages, __pgprot(0),
1067 int set_pages_uc(struct page *page, int numpages)
1069 unsigned long addr = (unsigned long)page_address(page);
1071 return set_memory_uc(addr, numpages);
1073 EXPORT_SYMBOL(set_pages_uc);
1075 int set_pages_wb(struct page *page, int numpages)
1077 unsigned long addr = (unsigned long)page_address(page);
1079 return set_memory_wb(addr, numpages);
1081 EXPORT_SYMBOL(set_pages_wb);
1083 int set_pages_x(struct page *page, int numpages)
1085 unsigned long addr = (unsigned long)page_address(page);
1087 return set_memory_x(addr, numpages);
1089 EXPORT_SYMBOL(set_pages_x);
1091 int set_pages_nx(struct page *page, int numpages)
1093 unsigned long addr = (unsigned long)page_address(page);
1095 return set_memory_nx(addr, numpages);
1097 EXPORT_SYMBOL(set_pages_nx);
1099 int set_pages_ro(struct page *page, int numpages)
1101 unsigned long addr = (unsigned long)page_address(page);
1103 return set_memory_ro(addr, numpages);
1106 int set_pages_rw(struct page *page, int numpages)
1108 unsigned long addr = (unsigned long)page_address(page);
1110 return set_memory_rw(addr, numpages);
1113 #ifdef CONFIG_DEBUG_PAGEALLOC
1115 static int __set_pages_p(struct page *page, int numpages)
1117 unsigned long tempaddr = (unsigned long) page_address(page);
1118 struct cpa_data cpa = { .vaddr = &tempaddr,
1119 .numpages = numpages,
1120 .mask_set = __pgprot(_PAGE_PRESENT | _PAGE_RW),
1121 .mask_clr = __pgprot(0),
1124 return __change_page_attr_set_clr(&cpa, 1);
1127 static int __set_pages_np(struct page *page, int numpages)
1129 unsigned long tempaddr = (unsigned long) page_address(page);
1130 struct cpa_data cpa = { .vaddr = &tempaddr,
1131 .numpages = numpages,
1132 .mask_set = __pgprot(0),
1133 .mask_clr = __pgprot(_PAGE_PRESENT | _PAGE_RW),
1136 return __change_page_attr_set_clr(&cpa, 1);
1139 void kernel_map_pages(struct page *page, int numpages, int enable)
1141 if (PageHighMem(page))
1144 debug_check_no_locks_freed(page_address(page),
1145 numpages * PAGE_SIZE);
1149 * If page allocator is not up yet then do not call c_p_a():
1151 if (!debug_pagealloc_enabled)
1155 * The return value is ignored as the calls cannot fail.
1156 * Large pages are kept enabled at boot time, and are
1157 * split up quickly with DEBUG_PAGEALLOC. If a splitup
1158 * fails here (due to temporary memory shortage) no damage
1159 * is done because we just keep the largepage intact up
1160 * to the next attempt when it will likely be split up:
1163 __set_pages_p(page, numpages);
1165 __set_pages_np(page, numpages);
1168 * We should perform an IPI and flush all tlbs,
1169 * but that can deadlock->flush only current cpu:
1174 * Try to refill the page pool here. We can do this only after
1177 cpa_fill_pool(NULL);
1180 #ifdef CONFIG_DEBUG_FS
1181 static int dpa_show(struct seq_file *m, void *v)
1183 seq_puts(m, "DEBUG_PAGEALLOC\n");
1184 seq_printf(m, "pool_size : %lu\n", pool_size);
1185 seq_printf(m, "pool_pages : %lu\n", pool_pages);
1186 seq_printf(m, "pool_low : %lu\n", pool_low);
1187 seq_printf(m, "pool_used : %lu\n", pool_used);
1188 seq_printf(m, "pool_failed : %lu\n", pool_failed);
1193 static int dpa_open(struct inode *inode, struct file *filp)
1195 return single_open(filp, dpa_show, NULL);
1198 static const struct file_operations dpa_fops = {
1201 .llseek = seq_lseek,
1202 .release = single_release,
1205 static int __init debug_pagealloc_proc_init(void)
1209 de = debugfs_create_file("debug_pagealloc", 0600, NULL, NULL,
1216 __initcall(debug_pagealloc_proc_init);
1219 #ifdef CONFIG_HIBERNATION
1221 bool kernel_page_present(struct page *page)
1226 if (PageHighMem(page))
1229 pte = lookup_address((unsigned long)page_address(page), &level);
1230 return (pte_val(*pte) & _PAGE_PRESENT);
1233 #endif /* CONFIG_HIBERNATION */
1235 #endif /* CONFIG_DEBUG_PAGEALLOC */
1238 * The testcases use internal knowledge of the implementation that shouldn't
1239 * be exposed to the rest of the kernel. Include these directly here.
1241 #ifdef CONFIG_CPA_DEBUG
1242 #include "pageattr-test.c"