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/sched.h>
9 #include <linux/interrupt.h>
10 #include <linux/seq_file.h>
11 #include <linux/debugfs.h>
12 #include <linux/pfn.h>
13 #include <linux/percpu.h>
14 #include <linux/gfp.h>
15 #include <linux/pci.h>
16 #include <linux/vmalloc.h>
19 #include <asm/processor.h>
20 #include <asm/tlbflush.h>
21 #include <asm/sections.h>
22 #include <asm/setup.h>
23 #include <asm/uaccess.h>
24 #include <asm/pgalloc.h>
25 #include <asm/proto.h>
29 * The current flushing context - we pass it instead of 5 arguments:
36 unsigned long numpages;
39 unsigned force_split : 1;
45 * Serialize cpa() (for !DEBUG_PAGEALLOC which uses large identity mappings)
46 * using cpa_lock. So that we don't allow any other cpu, with stale large tlb
47 * entries change the page attribute in parallel to some other cpu
48 * splitting a large page entry along with changing the attribute.
50 static DEFINE_SPINLOCK(cpa_lock);
52 #define CPA_FLUSHTLB 1
54 #define CPA_PAGES_ARRAY 4
55 #define CPA_FREE_PAGETABLES 8
58 static unsigned long direct_pages_count[PG_LEVEL_NUM];
60 void update_page_count(int level, unsigned long pages)
62 /* Protect against CPA */
64 direct_pages_count[level] += pages;
65 spin_unlock(&pgd_lock);
68 static void split_page_count(int level)
70 direct_pages_count[level]--;
71 direct_pages_count[level - 1] += PTRS_PER_PTE;
74 void arch_report_meminfo(struct seq_file *m)
76 seq_printf(m, "DirectMap4k: %8lu kB\n",
77 direct_pages_count[PG_LEVEL_4K] << 2);
78 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
79 seq_printf(m, "DirectMap2M: %8lu kB\n",
80 direct_pages_count[PG_LEVEL_2M] << 11);
82 seq_printf(m, "DirectMap4M: %8lu kB\n",
83 direct_pages_count[PG_LEVEL_2M] << 12);
86 seq_printf(m, "DirectMap1G: %8lu kB\n",
87 direct_pages_count[PG_LEVEL_1G] << 20);
90 static inline void split_page_count(int level) { }
95 static inline unsigned long highmap_start_pfn(void)
97 return __pa_symbol(_text) >> PAGE_SHIFT;
100 static inline unsigned long highmap_end_pfn(void)
102 return __pa_symbol(roundup(_brk_end, PMD_SIZE)) >> PAGE_SHIFT;
107 #ifdef CONFIG_DEBUG_PAGEALLOC
108 # define debug_pagealloc 1
110 # define debug_pagealloc 0
114 within(unsigned long addr, unsigned long start, unsigned long end)
116 return addr >= start && addr < end;
124 * clflush_cache_range - flush a cache range with clflush
125 * @vaddr: virtual start address
126 * @size: number of bytes to flush
128 * clflushopt is an unordered instruction which needs fencing with mfence or
129 * sfence to avoid ordering issues.
131 void clflush_cache_range(void *vaddr, unsigned int size)
133 unsigned long clflush_mask = boot_cpu_data.x86_clflush_size - 1;
134 void *vend = vaddr + size;
139 for (p = (void *)((unsigned long)vaddr & ~clflush_mask);
140 p < vend; p += boot_cpu_data.x86_clflush_size)
145 EXPORT_SYMBOL_GPL(clflush_cache_range);
147 static void __cpa_flush_all(void *arg)
149 unsigned long cache = (unsigned long)arg;
152 * Flush all to work around Errata in early athlons regarding
153 * large page flushing.
157 if (cache && boot_cpu_data.x86 >= 4)
161 static void cpa_flush_all(unsigned long cache)
163 BUG_ON(irqs_disabled());
165 on_each_cpu(__cpa_flush_all, (void *) cache, 1);
168 static void __cpa_flush_range(void *arg)
171 * We could optimize that further and do individual per page
172 * tlb invalidates for a low number of pages. Caveat: we must
173 * flush the high aliases on 64bit as well.
178 static void cpa_flush_range(unsigned long start, int numpages, int cache)
180 unsigned int i, level;
183 BUG_ON(irqs_disabled());
184 WARN_ON(PAGE_ALIGN(start) != start);
186 on_each_cpu(__cpa_flush_range, NULL, 1);
192 * We only need to flush on one CPU,
193 * clflush is a MESI-coherent instruction that
194 * will cause all other CPUs to flush the same
197 for (i = 0, addr = start; i < numpages; i++, addr += PAGE_SIZE) {
198 pte_t *pte = lookup_address(addr, &level);
201 * Only flush present addresses:
203 if (pte && (pte_val(*pte) & _PAGE_PRESENT))
204 clflush_cache_range((void *) addr, PAGE_SIZE);
208 static void cpa_flush_array(unsigned long *start, int numpages, int cache,
209 int in_flags, struct page **pages)
211 unsigned int i, level;
212 unsigned long do_wbinvd = cache && numpages >= 1024; /* 4M threshold */
214 BUG_ON(irqs_disabled());
216 on_each_cpu(__cpa_flush_all, (void *) do_wbinvd, 1);
218 if (!cache || do_wbinvd)
222 * We only need to flush on one CPU,
223 * clflush is a MESI-coherent instruction that
224 * will cause all other CPUs to flush the same
227 for (i = 0; i < numpages; i++) {
231 if (in_flags & CPA_PAGES_ARRAY)
232 addr = (unsigned long)page_address(pages[i]);
236 pte = lookup_address(addr, &level);
239 * Only flush present addresses:
241 if (pte && (pte_val(*pte) & _PAGE_PRESENT))
242 clflush_cache_range((void *)addr, PAGE_SIZE);
247 * Certain areas of memory on x86 require very specific protection flags,
248 * for example the BIOS area or kernel text. Callers don't always get this
249 * right (again, ioremap() on BIOS memory is not uncommon) so this function
250 * checks and fixes these known static required protection bits.
252 static inline pgprot_t static_protections(pgprot_t prot, unsigned long address,
255 pgprot_t forbidden = __pgprot(0);
258 * The BIOS area between 640k and 1Mb needs to be executable for
259 * PCI BIOS based config access (CONFIG_PCI_GOBIOS) support.
261 #ifdef CONFIG_PCI_BIOS
262 if (pcibios_enabled && within(pfn, BIOS_BEGIN >> PAGE_SHIFT, BIOS_END >> PAGE_SHIFT))
263 pgprot_val(forbidden) |= _PAGE_NX;
267 * The kernel text needs to be executable for obvious reasons
268 * Does not cover __inittext since that is gone later on. On
269 * 64bit we do not enforce !NX on the low mapping
271 if (within(address, (unsigned long)_text, (unsigned long)_etext))
272 pgprot_val(forbidden) |= _PAGE_NX;
275 * The .rodata section needs to be read-only. Using the pfn
276 * catches all aliases.
278 if (within(pfn, __pa_symbol(__start_rodata) >> PAGE_SHIFT,
279 __pa_symbol(__end_rodata) >> PAGE_SHIFT))
280 pgprot_val(forbidden) |= _PAGE_RW;
282 #if defined(CONFIG_X86_64) && defined(CONFIG_DEBUG_RODATA)
284 * Once the kernel maps the text as RO (kernel_set_to_readonly is set),
285 * kernel text mappings for the large page aligned text, rodata sections
286 * will be always read-only. For the kernel identity mappings covering
287 * the holes caused by this alignment can be anything that user asks.
289 * This will preserve the large page mappings for kernel text/data
292 if (kernel_set_to_readonly &&
293 within(address, (unsigned long)_text,
294 (unsigned long)__end_rodata_hpage_align)) {
298 * Don't enforce the !RW mapping for the kernel text mapping,
299 * if the current mapping is already using small page mapping.
300 * No need to work hard to preserve large page mappings in this
303 * This also fixes the Linux Xen paravirt guest boot failure
304 * (because of unexpected read-only mappings for kernel identity
305 * mappings). In this paravirt guest case, the kernel text
306 * mapping and the kernel identity mapping share the same
307 * page-table pages. Thus we can't really use different
308 * protections for the kernel text and identity mappings. Also,
309 * these shared mappings are made of small page mappings.
310 * Thus this don't enforce !RW mapping for small page kernel
311 * text mapping logic will help Linux Xen parvirt guest boot
314 if (lookup_address(address, &level) && (level != PG_LEVEL_4K))
315 pgprot_val(forbidden) |= _PAGE_RW;
319 prot = __pgprot(pgprot_val(prot) & ~pgprot_val(forbidden));
325 * Lookup the page table entry for a virtual address in a specific pgd.
326 * Return a pointer to the entry and the level of the mapping.
328 pte_t *lookup_address_in_pgd(pgd_t *pgd, unsigned long address,
334 *level = PG_LEVEL_NONE;
339 pud = pud_offset(pgd, address);
343 *level = PG_LEVEL_1G;
344 if (pud_large(*pud) || !pud_present(*pud))
347 pmd = pmd_offset(pud, address);
351 *level = PG_LEVEL_2M;
352 if (pmd_large(*pmd) || !pmd_present(*pmd))
355 *level = PG_LEVEL_4K;
357 return pte_offset_kernel(pmd, address);
361 * Lookup the page table entry for a virtual address. Return a pointer
362 * to the entry and the level of the mapping.
364 * Note: We return pud and pmd either when the entry is marked large
365 * or when the present bit is not set. Otherwise we would return a
366 * pointer to a nonexisting mapping.
368 pte_t *lookup_address(unsigned long address, unsigned int *level)
370 return lookup_address_in_pgd(pgd_offset_k(address), address, level);
372 EXPORT_SYMBOL_GPL(lookup_address);
374 static pte_t *_lookup_address_cpa(struct cpa_data *cpa, unsigned long address,
378 return lookup_address_in_pgd(cpa->pgd + pgd_index(address),
381 return lookup_address(address, level);
385 * Lookup the PMD entry for a virtual address. Return a pointer to the entry
386 * or NULL if not present.
388 pmd_t *lookup_pmd_address(unsigned long address)
393 pgd = pgd_offset_k(address);
397 pud = pud_offset(pgd, address);
398 if (pud_none(*pud) || pud_large(*pud) || !pud_present(*pud))
401 return pmd_offset(pud, address);
405 * This is necessary because __pa() does not work on some
406 * kinds of memory, like vmalloc() or the alloc_remap()
407 * areas on 32-bit NUMA systems. The percpu areas can
408 * end up in this kind of memory, for instance.
410 * This could be optimized, but it is only intended to be
411 * used at inititalization time, and keeping it
412 * unoptimized should increase the testing coverage for
413 * the more obscure platforms.
415 phys_addr_t slow_virt_to_phys(void *__virt_addr)
417 unsigned long virt_addr = (unsigned long)__virt_addr;
418 phys_addr_t phys_addr;
419 unsigned long offset;
423 pte = lookup_address(virt_addr, &level);
427 * pXX_pfn() returns unsigned long, which must be cast to phys_addr_t
428 * before being left-shifted PAGE_SHIFT bits -- this trick is to
429 * make 32-PAE kernel work correctly.
433 phys_addr = (phys_addr_t)pud_pfn(*(pud_t *)pte) << PAGE_SHIFT;
434 offset = virt_addr & ~PUD_PAGE_MASK;
437 phys_addr = (phys_addr_t)pmd_pfn(*(pmd_t *)pte) << PAGE_SHIFT;
438 offset = virt_addr & ~PMD_PAGE_MASK;
441 phys_addr = (phys_addr_t)pte_pfn(*pte) << PAGE_SHIFT;
442 offset = virt_addr & ~PAGE_MASK;
445 return (phys_addr_t)(phys_addr | offset);
447 EXPORT_SYMBOL_GPL(slow_virt_to_phys);
450 * Set the new pmd in all the pgds we know about:
452 static void __set_pmd_pte(pte_t *kpte, unsigned long address, pte_t pte)
455 set_pte_atomic(kpte, pte);
457 if (!SHARED_KERNEL_PMD) {
460 list_for_each_entry(page, &pgd_list, lru) {
465 pgd = (pgd_t *)page_address(page) + pgd_index(address);
466 pud = pud_offset(pgd, address);
467 pmd = pmd_offset(pud, address);
468 set_pte_atomic((pte_t *)pmd, pte);
475 try_preserve_large_page(pte_t *kpte, unsigned long address,
476 struct cpa_data *cpa)
478 unsigned long nextpage_addr, numpages, pmask, psize, addr, pfn, old_pfn;
479 pte_t new_pte, old_pte, *tmp;
480 pgprot_t old_prot, new_prot, req_prot;
484 if (cpa->force_split)
487 spin_lock(&pgd_lock);
489 * Check for races, another CPU might have split this page
492 tmp = _lookup_address_cpa(cpa, address, &level);
498 old_prot = pmd_pgprot(*(pmd_t *)kpte);
499 old_pfn = pmd_pfn(*(pmd_t *)kpte);
502 old_prot = pud_pgprot(*(pud_t *)kpte);
503 old_pfn = pud_pfn(*(pud_t *)kpte);
510 psize = page_level_size(level);
511 pmask = page_level_mask(level);
514 * Calculate the number of pages, which fit into this large
515 * page starting at address:
517 nextpage_addr = (address + psize) & pmask;
518 numpages = (nextpage_addr - address) >> PAGE_SHIFT;
519 if (numpages < cpa->numpages)
520 cpa->numpages = numpages;
523 * We are safe now. Check whether the new pgprot is the same:
524 * Convert protection attributes to 4k-format, as cpa->mask* are set
528 req_prot = pgprot_large_2_4k(old_prot);
530 pgprot_val(req_prot) &= ~pgprot_val(cpa->mask_clr);
531 pgprot_val(req_prot) |= pgprot_val(cpa->mask_set);
534 * req_prot is in format of 4k pages. It must be converted to large
535 * page format: the caching mode includes the PAT bit located at
536 * different bit positions in the two formats.
538 req_prot = pgprot_4k_2_large(req_prot);
541 * Set the PSE and GLOBAL flags only if the PRESENT flag is
542 * set otherwise pmd_present/pmd_huge will return true even on
543 * a non present pmd. The canon_pgprot will clear _PAGE_GLOBAL
544 * for the ancient hardware that doesn't support it.
546 if (pgprot_val(req_prot) & _PAGE_PRESENT)
547 pgprot_val(req_prot) |= _PAGE_PSE | _PAGE_GLOBAL;
549 pgprot_val(req_prot) &= ~(_PAGE_PSE | _PAGE_GLOBAL);
551 req_prot = canon_pgprot(req_prot);
554 * old_pfn points to the large page base pfn. So we need
555 * to add the offset of the virtual address:
557 pfn = old_pfn + ((address & (psize - 1)) >> PAGE_SHIFT);
560 new_prot = static_protections(req_prot, address, pfn);
563 * We need to check the full range, whether
564 * static_protection() requires a different pgprot for one of
565 * the pages in the range we try to preserve:
567 addr = address & pmask;
569 for (i = 0; i < (psize >> PAGE_SHIFT); i++, addr += PAGE_SIZE, pfn++) {
570 pgprot_t chk_prot = static_protections(req_prot, addr, pfn);
572 if (pgprot_val(chk_prot) != pgprot_val(new_prot))
577 * If there are no changes, return. maxpages has been updated
580 if (pgprot_val(new_prot) == pgprot_val(old_prot)) {
586 * We need to change the attributes. Check, whether we can
587 * change the large page in one go. We request a split, when
588 * the address is not aligned and the number of pages is
589 * smaller than the number of pages in the large page. Note
590 * that we limited the number of possible pages already to
591 * the number of pages in the large page.
593 if (address == (address & pmask) && cpa->numpages == (psize >> PAGE_SHIFT)) {
595 * The address is aligned and the number of pages
596 * covers the full page.
598 new_pte = pfn_pte(old_pfn, new_prot);
599 __set_pmd_pte(kpte, address, new_pte);
600 cpa->flags |= CPA_FLUSHTLB;
605 spin_unlock(&pgd_lock);
611 __split_large_page(struct cpa_data *cpa, pte_t *kpte, unsigned long address,
614 pte_t *pbase = (pte_t *)page_address(base);
615 unsigned long ref_pfn, pfn, pfninc = 1;
616 unsigned int i, level;
620 spin_lock(&pgd_lock);
622 * Check for races, another CPU might have split this page
625 tmp = _lookup_address_cpa(cpa, address, &level);
627 spin_unlock(&pgd_lock);
631 paravirt_alloc_pte(&init_mm, page_to_pfn(base));
635 ref_prot = pmd_pgprot(*(pmd_t *)kpte);
636 /* clear PSE and promote PAT bit to correct position */
637 ref_prot = pgprot_large_2_4k(ref_prot);
638 ref_pfn = pmd_pfn(*(pmd_t *)kpte);
642 ref_prot = pud_pgprot(*(pud_t *)kpte);
643 ref_pfn = pud_pfn(*(pud_t *)kpte);
644 pfninc = PMD_PAGE_SIZE >> PAGE_SHIFT;
647 * Clear the PSE flags if the PRESENT flag is not set
648 * otherwise pmd_present/pmd_huge will return true
649 * even on a non present pmd.
651 if (!(pgprot_val(ref_prot) & _PAGE_PRESENT))
652 pgprot_val(ref_prot) &= ~_PAGE_PSE;
656 spin_unlock(&pgd_lock);
661 * Set the GLOBAL flags only if the PRESENT flag is set
662 * otherwise pmd/pte_present will return true even on a non
663 * present pmd/pte. The canon_pgprot will clear _PAGE_GLOBAL
664 * for the ancient hardware that doesn't support it.
666 if (pgprot_val(ref_prot) & _PAGE_PRESENT)
667 pgprot_val(ref_prot) |= _PAGE_GLOBAL;
669 pgprot_val(ref_prot) &= ~_PAGE_GLOBAL;
672 * Get the target pfn from the original entry:
675 for (i = 0; i < PTRS_PER_PTE; i++, pfn += pfninc)
676 set_pte(&pbase[i], pfn_pte(pfn, canon_pgprot(ref_prot)));
678 if (virt_addr_valid(address)) {
679 unsigned long pfn = PFN_DOWN(__pa(address));
681 if (pfn_range_is_mapped(pfn, pfn + 1))
682 split_page_count(level);
686 * Install the new, split up pagetable.
688 * We use the standard kernel pagetable protections for the new
689 * pagetable protections, the actual ptes set above control the
690 * primary protection behavior:
692 __set_pmd_pte(kpte, address, mk_pte(base, __pgprot(_KERNPG_TABLE)));
695 * Intel Atom errata AAH41 workaround.
697 * The real fix should be in hw or in a microcode update, but
698 * we also probabilistically try to reduce the window of having
699 * a large TLB mixed with 4K TLBs while instruction fetches are
703 spin_unlock(&pgd_lock);
708 static int split_large_page(struct cpa_data *cpa, pte_t *kpte,
709 unsigned long address)
713 if (!debug_pagealloc)
714 spin_unlock(&cpa_lock);
715 base = alloc_pages(GFP_KERNEL | __GFP_NOTRACK, 0);
716 if (!debug_pagealloc)
717 spin_lock(&cpa_lock);
721 if (__split_large_page(cpa, kpte, address, base))
727 static bool try_to_free_pte_page(struct cpa_data *cpa, pte_t *pte)
731 if (!(cpa->flags & CPA_FREE_PAGETABLES))
734 for (i = 0; i < PTRS_PER_PTE; i++)
735 if (!pte_none(pte[i]))
738 free_page((unsigned long)pte);
742 static bool try_to_free_pmd_page(struct cpa_data *cpa, pmd_t *pmd)
746 if (!(cpa->flags & CPA_FREE_PAGETABLES))
749 for (i = 0; i < PTRS_PER_PMD; i++)
750 if (!pmd_none(pmd[i]))
753 free_page((unsigned long)pmd);
757 static bool try_to_free_pud_page(pud_t *pud)
761 for (i = 0; i < PTRS_PER_PUD; i++)
762 if (!pud_none(pud[i]))
765 free_page((unsigned long)pud);
769 static bool unmap_pte_range(struct cpa_data *cpa, pmd_t *pmd,
773 pte_t *pte = pte_offset_kernel(pmd, start);
775 while (start < end) {
776 set_pte(pte, __pte(0));
782 if (try_to_free_pte_page(cpa, (pte_t *)pmd_page_vaddr(*pmd))) {
789 static void __unmap_pmd_range(struct cpa_data *cpa, pud_t *pud, pmd_t *pmd,
790 unsigned long start, unsigned long end)
792 if (unmap_pte_range(cpa, pmd, start, end))
793 if (try_to_free_pmd_page(cpa, (pmd_t *)pud_page_vaddr(*pud)))
797 static void unmap_pmd_range(struct cpa_data *cpa, pud_t *pud,
798 unsigned long start, unsigned long end)
800 pmd_t *pmd = pmd_offset(pud, start);
803 * Not on a 2MB page boundary?
805 if (start & (PMD_SIZE - 1)) {
806 unsigned long next_page = (start + PMD_SIZE) & PMD_MASK;
807 unsigned long pre_end = min_t(unsigned long, end, next_page);
809 __unmap_pmd_range(cpa, pud, pmd, start, pre_end);
816 * Try to unmap in 2M chunks.
818 while (end - start >= PMD_SIZE) {
822 __unmap_pmd_range(cpa, pud, pmd,
823 start, start + PMD_SIZE);
833 return __unmap_pmd_range(cpa, pud, pmd, start, end);
836 * Try again to free the PMD page if haven't succeeded above.
839 if (try_to_free_pmd_page(cpa, (pmd_t *)pud_page_vaddr(*pud)))
843 static void __unmap_pud_range(struct cpa_data *cpa, pgd_t *pgd,
847 pud_t *pud = pud_offset(pgd, start);
850 * Not on a GB page boundary?
852 if (start & (PUD_SIZE - 1)) {
853 unsigned long next_page = (start + PUD_SIZE) & PUD_MASK;
854 unsigned long pre_end = min_t(unsigned long, end, next_page);
856 unmap_pmd_range(cpa, pud, start, pre_end);
863 * Try to unmap in 1G chunks?
865 while (end - start >= PUD_SIZE) {
870 unmap_pmd_range(cpa, pud, start, start + PUD_SIZE);
880 unmap_pmd_range(cpa, pud, start, end);
883 * No need to try to free the PUD page because we'll free it in
884 * populate_pgd's error path
888 static void unmap_pud_range(pgd_t *pgd, unsigned long start, unsigned long end)
890 struct cpa_data cpa = {
891 .flags = CPA_FREE_PAGETABLES,
894 __unmap_pud_range(&cpa, pgd, start, end);
897 void unmap_pud_range_nofree(pgd_t *pgd, unsigned long start, unsigned long end)
899 struct cpa_data cpa = {
903 __unmap_pud_range(&cpa, pgd, start, end);
906 static void unmap_pgd_range(pgd_t *root, unsigned long addr, unsigned long end)
908 pgd_t *pgd_entry = root + pgd_index(addr);
910 unmap_pud_range(pgd_entry, addr, end);
912 if (try_to_free_pud_page((pud_t *)pgd_page_vaddr(*pgd_entry)))
913 pgd_clear(pgd_entry);
916 static int alloc_pte_page(pmd_t *pmd)
918 pte_t *pte = (pte_t *)get_zeroed_page(GFP_KERNEL | __GFP_NOTRACK);
922 set_pmd(pmd, __pmd(__pa(pte) | _KERNPG_TABLE));
926 static int alloc_pmd_page(pud_t *pud)
928 pmd_t *pmd = (pmd_t *)get_zeroed_page(GFP_KERNEL | __GFP_NOTRACK);
932 set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE));
936 static void populate_pte(struct cpa_data *cpa,
937 unsigned long start, unsigned long end,
938 unsigned num_pages, pmd_t *pmd, pgprot_t pgprot)
942 pte = pte_offset_kernel(pmd, start);
944 while (num_pages-- && start < end) {
946 /* deal with the NX bit */
947 if (!(pgprot_val(pgprot) & _PAGE_NX))
948 cpa->pfn &= ~_PAGE_NX;
950 set_pte(pte, pfn_pte(cpa->pfn >> PAGE_SHIFT, pgprot));
953 cpa->pfn += PAGE_SIZE;
958 static int populate_pmd(struct cpa_data *cpa,
959 unsigned long start, unsigned long end,
960 unsigned num_pages, pud_t *pud, pgprot_t pgprot)
962 unsigned int cur_pages = 0;
967 * Not on a 2M boundary?
969 if (start & (PMD_SIZE - 1)) {
970 unsigned long pre_end = start + (num_pages << PAGE_SHIFT);
971 unsigned long next_page = (start + PMD_SIZE) & PMD_MASK;
973 pre_end = min_t(unsigned long, pre_end, next_page);
974 cur_pages = (pre_end - start) >> PAGE_SHIFT;
975 cur_pages = min_t(unsigned int, num_pages, cur_pages);
980 pmd = pmd_offset(pud, start);
982 if (alloc_pte_page(pmd))
985 populate_pte(cpa, start, pre_end, cur_pages, pmd, pgprot);
991 * We mapped them all?
993 if (num_pages == cur_pages)
996 pmd_pgprot = pgprot_4k_2_large(pgprot);
998 while (end - start >= PMD_SIZE) {
1001 * We cannot use a 1G page so allocate a PMD page if needed.
1004 if (alloc_pmd_page(pud))
1007 pmd = pmd_offset(pud, start);
1009 set_pmd(pmd, __pmd(cpa->pfn | _PAGE_PSE |
1010 massage_pgprot(pmd_pgprot)));
1013 cpa->pfn += PMD_SIZE;
1014 cur_pages += PMD_SIZE >> PAGE_SHIFT;
1018 * Map trailing 4K pages.
1021 pmd = pmd_offset(pud, start);
1023 if (alloc_pte_page(pmd))
1026 populate_pte(cpa, start, end, num_pages - cur_pages,
1032 static int populate_pud(struct cpa_data *cpa, unsigned long start, pgd_t *pgd,
1038 pgprot_t pud_pgprot;
1040 end = start + (cpa->numpages << PAGE_SHIFT);
1043 * Not on a Gb page boundary? => map everything up to it with
1046 if (start & (PUD_SIZE - 1)) {
1047 unsigned long pre_end;
1048 unsigned long next_page = (start + PUD_SIZE) & PUD_MASK;
1050 pre_end = min_t(unsigned long, end, next_page);
1051 cur_pages = (pre_end - start) >> PAGE_SHIFT;
1052 cur_pages = min_t(int, (int)cpa->numpages, cur_pages);
1054 pud = pud_offset(pgd, start);
1060 if (alloc_pmd_page(pud))
1063 cur_pages = populate_pmd(cpa, start, pre_end, cur_pages,
1071 /* We mapped them all? */
1072 if (cpa->numpages == cur_pages)
1075 pud = pud_offset(pgd, start);
1076 pud_pgprot = pgprot_4k_2_large(pgprot);
1079 * Map everything starting from the Gb boundary, possibly with 1G pages
1081 while (end - start >= PUD_SIZE) {
1082 set_pud(pud, __pud(cpa->pfn | _PAGE_PSE |
1083 massage_pgprot(pud_pgprot)));
1086 cpa->pfn += PUD_SIZE;
1087 cur_pages += PUD_SIZE >> PAGE_SHIFT;
1091 /* Map trailing leftover */
1095 pud = pud_offset(pgd, start);
1097 if (alloc_pmd_page(pud))
1100 tmp = populate_pmd(cpa, start, end, cpa->numpages - cur_pages,
1111 * Restrictions for kernel page table do not necessarily apply when mapping in
1114 static int populate_pgd(struct cpa_data *cpa, unsigned long addr)
1116 pgprot_t pgprot = __pgprot(_KERNPG_TABLE);
1117 pud_t *pud = NULL; /* shut up gcc */
1121 pgd_entry = cpa->pgd + pgd_index(addr);
1124 * Allocate a PUD page and hand it down for mapping.
1126 if (pgd_none(*pgd_entry)) {
1127 pud = (pud_t *)get_zeroed_page(GFP_KERNEL | __GFP_NOTRACK);
1131 set_pgd(pgd_entry, __pgd(__pa(pud) | _KERNPG_TABLE));
1134 pgprot_val(pgprot) &= ~pgprot_val(cpa->mask_clr);
1135 pgprot_val(pgprot) |= pgprot_val(cpa->mask_set);
1137 ret = populate_pud(cpa, addr, pgd_entry, pgprot);
1139 unmap_pgd_range(cpa->pgd, addr,
1140 addr + (cpa->numpages << PAGE_SHIFT));
1144 cpa->numpages = ret;
1148 static int __cpa_process_fault(struct cpa_data *cpa, unsigned long vaddr,
1152 return populate_pgd(cpa, vaddr);
1155 * Ignore all non primary paths.
1161 * Ignore the NULL PTE for kernel identity mapping, as it is expected
1163 * Also set numpages to '1' indicating that we processed cpa req for
1164 * one virtual address page and its pfn. TBD: numpages can be set based
1165 * on the initial value and the level returned by lookup_address().
1167 if (within(vaddr, PAGE_OFFSET,
1168 PAGE_OFFSET + (max_pfn_mapped << PAGE_SHIFT))) {
1170 cpa->pfn = __pa(vaddr) >> PAGE_SHIFT;
1173 WARN(1, KERN_WARNING "CPA: called for zero pte. "
1174 "vaddr = %lx cpa->vaddr = %lx\n", vaddr,
1181 static int __change_page_attr(struct cpa_data *cpa, int primary)
1183 unsigned long address;
1186 pte_t *kpte, old_pte;
1188 if (cpa->flags & CPA_PAGES_ARRAY) {
1189 struct page *page = cpa->pages[cpa->curpage];
1190 if (unlikely(PageHighMem(page)))
1192 address = (unsigned long)page_address(page);
1193 } else if (cpa->flags & CPA_ARRAY)
1194 address = cpa->vaddr[cpa->curpage];
1196 address = *cpa->vaddr;
1198 kpte = _lookup_address_cpa(cpa, address, &level);
1200 return __cpa_process_fault(cpa, address, primary);
1203 if (!pte_val(old_pte))
1204 return __cpa_process_fault(cpa, address, primary);
1206 if (level == PG_LEVEL_4K) {
1208 pgprot_t new_prot = pte_pgprot(old_pte);
1209 unsigned long pfn = pte_pfn(old_pte);
1211 pgprot_val(new_prot) &= ~pgprot_val(cpa->mask_clr);
1212 pgprot_val(new_prot) |= pgprot_val(cpa->mask_set);
1214 new_prot = static_protections(new_prot, address, pfn);
1217 * Set the GLOBAL flags only if the PRESENT flag is
1218 * set otherwise pte_present will return true even on
1219 * a non present pte. The canon_pgprot will clear
1220 * _PAGE_GLOBAL for the ancient hardware that doesn't
1223 if (pgprot_val(new_prot) & _PAGE_PRESENT)
1224 pgprot_val(new_prot) |= _PAGE_GLOBAL;
1226 pgprot_val(new_prot) &= ~_PAGE_GLOBAL;
1229 * We need to keep the pfn from the existing PTE,
1230 * after all we're only going to change it's attributes
1231 * not the memory it points to
1233 new_pte = pfn_pte(pfn, canon_pgprot(new_prot));
1236 * Do we really change anything ?
1238 if (pte_val(old_pte) != pte_val(new_pte)) {
1239 set_pte_atomic(kpte, new_pte);
1240 cpa->flags |= CPA_FLUSHTLB;
1247 * Check, whether we can keep the large page intact
1248 * and just change the pte:
1250 do_split = try_preserve_large_page(kpte, address, cpa);
1252 * When the range fits into the existing large page,
1253 * return. cp->numpages and cpa->tlbflush have been updated in
1260 * We have to split the large page:
1262 err = split_large_page(cpa, kpte, address);
1265 * Do a global flush tlb after splitting the large page
1266 * and before we do the actual change page attribute in the PTE.
1268 * With out this, we violate the TLB application note, that says
1269 * "The TLBs may contain both ordinary and large-page
1270 * translations for a 4-KByte range of linear addresses. This
1271 * may occur if software modifies the paging structures so that
1272 * the page size used for the address range changes. If the two
1273 * translations differ with respect to page frame or attributes
1274 * (e.g., permissions), processor behavior is undefined and may
1275 * be implementation-specific."
1277 * We do this global tlb flush inside the cpa_lock, so that we
1278 * don't allow any other cpu, with stale tlb entries change the
1279 * page attribute in parallel, that also falls into the
1280 * just split large page entry.
1289 static int __change_page_attr_set_clr(struct cpa_data *cpa, int checkalias);
1291 static int cpa_process_alias(struct cpa_data *cpa)
1293 struct cpa_data alias_cpa;
1294 unsigned long laddr = (unsigned long)__va(cpa->pfn << PAGE_SHIFT);
1295 unsigned long vaddr;
1298 if (!pfn_range_is_mapped(cpa->pfn, cpa->pfn + 1))
1302 * No need to redo, when the primary call touched the direct
1305 if (cpa->flags & CPA_PAGES_ARRAY) {
1306 struct page *page = cpa->pages[cpa->curpage];
1307 if (unlikely(PageHighMem(page)))
1309 vaddr = (unsigned long)page_address(page);
1310 } else if (cpa->flags & CPA_ARRAY)
1311 vaddr = cpa->vaddr[cpa->curpage];
1313 vaddr = *cpa->vaddr;
1315 if (!(within(vaddr, PAGE_OFFSET,
1316 PAGE_OFFSET + (max_pfn_mapped << PAGE_SHIFT)))) {
1319 alias_cpa.vaddr = &laddr;
1320 alias_cpa.flags &= ~(CPA_PAGES_ARRAY | CPA_ARRAY);
1322 ret = __change_page_attr_set_clr(&alias_cpa, 0);
1327 #ifdef CONFIG_X86_64
1329 * If the primary call didn't touch the high mapping already
1330 * and the physical address is inside the kernel map, we need
1331 * to touch the high mapped kernel as well:
1333 if (!within(vaddr, (unsigned long)_text, _brk_end) &&
1334 within(cpa->pfn, highmap_start_pfn(), highmap_end_pfn())) {
1335 unsigned long temp_cpa_vaddr = (cpa->pfn << PAGE_SHIFT) +
1336 __START_KERNEL_map - phys_base;
1338 alias_cpa.vaddr = &temp_cpa_vaddr;
1339 alias_cpa.flags &= ~(CPA_PAGES_ARRAY | CPA_ARRAY);
1342 * The high mapping range is imprecise, so ignore the
1345 __change_page_attr_set_clr(&alias_cpa, 0);
1352 static int __change_page_attr_set_clr(struct cpa_data *cpa, int checkalias)
1354 int ret, numpages = cpa->numpages;
1358 * Store the remaining nr of pages for the large page
1359 * preservation check.
1361 cpa->numpages = numpages;
1362 /* for array changes, we can't use large page */
1363 if (cpa->flags & (CPA_ARRAY | CPA_PAGES_ARRAY))
1366 if (!debug_pagealloc)
1367 spin_lock(&cpa_lock);
1368 ret = __change_page_attr(cpa, checkalias);
1369 if (!debug_pagealloc)
1370 spin_unlock(&cpa_lock);
1375 ret = cpa_process_alias(cpa);
1381 * Adjust the number of pages with the result of the
1382 * CPA operation. Either a large page has been
1383 * preserved or a single page update happened.
1385 BUG_ON(cpa->numpages > numpages || !cpa->numpages);
1386 numpages -= cpa->numpages;
1387 if (cpa->flags & (CPA_PAGES_ARRAY | CPA_ARRAY))
1390 *cpa->vaddr += cpa->numpages * PAGE_SIZE;
1396 static int change_page_attr_set_clr(unsigned long *addr, int numpages,
1397 pgprot_t mask_set, pgprot_t mask_clr,
1398 int force_split, int in_flag,
1399 struct page **pages)
1401 struct cpa_data cpa;
1402 int ret, cache, checkalias;
1403 unsigned long baddr = 0;
1405 memset(&cpa, 0, sizeof(cpa));
1408 * Check, if we are requested to change a not supported
1411 mask_set = canon_pgprot(mask_set);
1412 mask_clr = canon_pgprot(mask_clr);
1413 if (!pgprot_val(mask_set) && !pgprot_val(mask_clr) && !force_split)
1416 /* Ensure we are PAGE_SIZE aligned */
1417 if (in_flag & CPA_ARRAY) {
1419 for (i = 0; i < numpages; i++) {
1420 if (addr[i] & ~PAGE_MASK) {
1421 addr[i] &= PAGE_MASK;
1425 } else if (!(in_flag & CPA_PAGES_ARRAY)) {
1427 * in_flag of CPA_PAGES_ARRAY implies it is aligned.
1428 * No need to cehck in that case
1430 if (*addr & ~PAGE_MASK) {
1433 * People should not be passing in unaligned addresses:
1438 * Save address for cache flush. *addr is modified in the call
1439 * to __change_page_attr_set_clr() below.
1444 /* Must avoid aliasing mappings in the highmem code */
1445 kmap_flush_unused();
1451 cpa.numpages = numpages;
1452 cpa.mask_set = mask_set;
1453 cpa.mask_clr = mask_clr;
1456 cpa.force_split = force_split;
1458 if (in_flag & (CPA_ARRAY | CPA_PAGES_ARRAY))
1459 cpa.flags |= in_flag;
1461 /* No alias checking for _NX bit modifications */
1462 checkalias = (pgprot_val(mask_set) | pgprot_val(mask_clr)) != _PAGE_NX;
1464 ret = __change_page_attr_set_clr(&cpa, checkalias);
1467 * Check whether we really changed something:
1469 if (!(cpa.flags & CPA_FLUSHTLB))
1473 * No need to flush, when we did not set any of the caching
1476 cache = !!pgprot2cachemode(mask_set);
1479 * On success we use CLFLUSH, when the CPU supports it to
1480 * avoid the WBINVD. If the CPU does not support it and in the
1481 * error case we fall back to cpa_flush_all (which uses
1484 if (!ret && cpu_has_clflush) {
1485 if (cpa.flags & (CPA_PAGES_ARRAY | CPA_ARRAY)) {
1486 cpa_flush_array(addr, numpages, cache,
1489 cpa_flush_range(baddr, numpages, cache);
1491 cpa_flush_all(cache);
1497 static inline int change_page_attr_set(unsigned long *addr, int numpages,
1498 pgprot_t mask, int array)
1500 return change_page_attr_set_clr(addr, numpages, mask, __pgprot(0), 0,
1501 (array ? CPA_ARRAY : 0), NULL);
1504 static inline int change_page_attr_clear(unsigned long *addr, int numpages,
1505 pgprot_t mask, int array)
1507 return change_page_attr_set_clr(addr, numpages, __pgprot(0), mask, 0,
1508 (array ? CPA_ARRAY : 0), NULL);
1511 static inline int cpa_set_pages_array(struct page **pages, int numpages,
1514 return change_page_attr_set_clr(NULL, numpages, mask, __pgprot(0), 0,
1515 CPA_PAGES_ARRAY, pages);
1518 static inline int cpa_clear_pages_array(struct page **pages, int numpages,
1521 return change_page_attr_set_clr(NULL, numpages, __pgprot(0), mask, 0,
1522 CPA_PAGES_ARRAY, pages);
1525 int _set_memory_uc(unsigned long addr, int numpages)
1528 * for now UC MINUS. see comments in ioremap_nocache()
1529 * If you really need strong UC use ioremap_uc(), but note
1530 * that you cannot override IO areas with set_memory_*() as
1531 * these helpers cannot work with IO memory.
1533 return change_page_attr_set(&addr, numpages,
1534 cachemode2pgprot(_PAGE_CACHE_MODE_UC_MINUS),
1538 int set_memory_uc(unsigned long addr, int numpages)
1543 * for now UC MINUS. see comments in ioremap_nocache()
1545 ret = reserve_memtype(__pa(addr), __pa(addr) + numpages * PAGE_SIZE,
1546 _PAGE_CACHE_MODE_UC_MINUS, NULL);
1550 ret = _set_memory_uc(addr, numpages);
1557 free_memtype(__pa(addr), __pa(addr) + numpages * PAGE_SIZE);
1561 EXPORT_SYMBOL(set_memory_uc);
1563 static int _set_memory_array(unsigned long *addr, int addrinarray,
1564 enum page_cache_mode new_type)
1566 enum page_cache_mode set_type;
1570 for (i = 0; i < addrinarray; i++) {
1571 ret = reserve_memtype(__pa(addr[i]), __pa(addr[i]) + PAGE_SIZE,
1577 /* If WC, set to UC- first and then WC */
1578 set_type = (new_type == _PAGE_CACHE_MODE_WC) ?
1579 _PAGE_CACHE_MODE_UC_MINUS : new_type;
1581 ret = change_page_attr_set(addr, addrinarray,
1582 cachemode2pgprot(set_type), 1);
1584 if (!ret && new_type == _PAGE_CACHE_MODE_WC)
1585 ret = change_page_attr_set_clr(addr, addrinarray,
1587 _PAGE_CACHE_MODE_WC),
1588 __pgprot(_PAGE_CACHE_MASK),
1589 0, CPA_ARRAY, NULL);
1596 for (j = 0; j < i; j++)
1597 free_memtype(__pa(addr[j]), __pa(addr[j]) + PAGE_SIZE);
1602 int set_memory_array_uc(unsigned long *addr, int addrinarray)
1604 return _set_memory_array(addr, addrinarray, _PAGE_CACHE_MODE_UC_MINUS);
1606 EXPORT_SYMBOL(set_memory_array_uc);
1608 int set_memory_array_wc(unsigned long *addr, int addrinarray)
1610 return _set_memory_array(addr, addrinarray, _PAGE_CACHE_MODE_WC);
1612 EXPORT_SYMBOL(set_memory_array_wc);
1614 int set_memory_array_wt(unsigned long *addr, int addrinarray)
1616 return _set_memory_array(addr, addrinarray, _PAGE_CACHE_MODE_WT);
1618 EXPORT_SYMBOL_GPL(set_memory_array_wt);
1620 int _set_memory_wc(unsigned long addr, int numpages)
1623 unsigned long addr_copy = addr;
1625 ret = change_page_attr_set(&addr, numpages,
1626 cachemode2pgprot(_PAGE_CACHE_MODE_UC_MINUS),
1629 ret = change_page_attr_set_clr(&addr_copy, numpages,
1631 _PAGE_CACHE_MODE_WC),
1632 __pgprot(_PAGE_CACHE_MASK),
1638 int set_memory_wc(unsigned long addr, int numpages)
1642 ret = reserve_memtype(__pa(addr), __pa(addr) + numpages * PAGE_SIZE,
1643 _PAGE_CACHE_MODE_WC, NULL);
1647 ret = _set_memory_wc(addr, numpages);
1649 free_memtype(__pa(addr), __pa(addr) + numpages * PAGE_SIZE);
1653 EXPORT_SYMBOL(set_memory_wc);
1655 int _set_memory_wt(unsigned long addr, int numpages)
1657 return change_page_attr_set(&addr, numpages,
1658 cachemode2pgprot(_PAGE_CACHE_MODE_WT), 0);
1661 int set_memory_wt(unsigned long addr, int numpages)
1665 ret = reserve_memtype(__pa(addr), __pa(addr) + numpages * PAGE_SIZE,
1666 _PAGE_CACHE_MODE_WT, NULL);
1670 ret = _set_memory_wt(addr, numpages);
1672 free_memtype(__pa(addr), __pa(addr) + numpages * PAGE_SIZE);
1676 EXPORT_SYMBOL_GPL(set_memory_wt);
1678 int _set_memory_wb(unsigned long addr, int numpages)
1680 /* WB cache mode is hard wired to all cache attribute bits being 0 */
1681 return change_page_attr_clear(&addr, numpages,
1682 __pgprot(_PAGE_CACHE_MASK), 0);
1685 int set_memory_wb(unsigned long addr, int numpages)
1689 ret = _set_memory_wb(addr, numpages);
1693 free_memtype(__pa(addr), __pa(addr) + numpages * PAGE_SIZE);
1696 EXPORT_SYMBOL(set_memory_wb);
1698 int set_memory_array_wb(unsigned long *addr, int addrinarray)
1703 /* WB cache mode is hard wired to all cache attribute bits being 0 */
1704 ret = change_page_attr_clear(addr, addrinarray,
1705 __pgprot(_PAGE_CACHE_MASK), 1);
1709 for (i = 0; i < addrinarray; i++)
1710 free_memtype(__pa(addr[i]), __pa(addr[i]) + PAGE_SIZE);
1714 EXPORT_SYMBOL(set_memory_array_wb);
1716 int set_memory_x(unsigned long addr, int numpages)
1718 if (!(__supported_pte_mask & _PAGE_NX))
1721 return change_page_attr_clear(&addr, numpages, __pgprot(_PAGE_NX), 0);
1723 EXPORT_SYMBOL(set_memory_x);
1725 int set_memory_nx(unsigned long addr, int numpages)
1727 if (!(__supported_pte_mask & _PAGE_NX))
1730 return change_page_attr_set(&addr, numpages, __pgprot(_PAGE_NX), 0);
1732 EXPORT_SYMBOL(set_memory_nx);
1734 int set_memory_ro(unsigned long addr, int numpages)
1736 return change_page_attr_clear(&addr, numpages, __pgprot(_PAGE_RW), 0);
1739 int set_memory_rw(unsigned long addr, int numpages)
1741 return change_page_attr_set(&addr, numpages, __pgprot(_PAGE_RW), 0);
1744 int set_memory_np(unsigned long addr, int numpages)
1746 return change_page_attr_clear(&addr, numpages, __pgprot(_PAGE_PRESENT), 0);
1749 int set_memory_4k(unsigned long addr, int numpages)
1751 return change_page_attr_set_clr(&addr, numpages, __pgprot(0),
1752 __pgprot(0), 1, 0, NULL);
1755 int set_pages_uc(struct page *page, int numpages)
1757 unsigned long addr = (unsigned long)page_address(page);
1759 return set_memory_uc(addr, numpages);
1761 EXPORT_SYMBOL(set_pages_uc);
1763 static int _set_pages_array(struct page **pages, int addrinarray,
1764 enum page_cache_mode new_type)
1766 unsigned long start;
1768 enum page_cache_mode set_type;
1773 for (i = 0; i < addrinarray; i++) {
1774 if (PageHighMem(pages[i]))
1776 start = page_to_pfn(pages[i]) << PAGE_SHIFT;
1777 end = start + PAGE_SIZE;
1778 if (reserve_memtype(start, end, new_type, NULL))
1782 /* If WC, set to UC- first and then WC */
1783 set_type = (new_type == _PAGE_CACHE_MODE_WC) ?
1784 _PAGE_CACHE_MODE_UC_MINUS : new_type;
1786 ret = cpa_set_pages_array(pages, addrinarray,
1787 cachemode2pgprot(set_type));
1788 if (!ret && new_type == _PAGE_CACHE_MODE_WC)
1789 ret = change_page_attr_set_clr(NULL, addrinarray,
1791 _PAGE_CACHE_MODE_WC),
1792 __pgprot(_PAGE_CACHE_MASK),
1793 0, CPA_PAGES_ARRAY, pages);
1796 return 0; /* Success */
1799 for (i = 0; i < free_idx; i++) {
1800 if (PageHighMem(pages[i]))
1802 start = page_to_pfn(pages[i]) << PAGE_SHIFT;
1803 end = start + PAGE_SIZE;
1804 free_memtype(start, end);
1809 int set_pages_array_uc(struct page **pages, int addrinarray)
1811 return _set_pages_array(pages, addrinarray, _PAGE_CACHE_MODE_UC_MINUS);
1813 EXPORT_SYMBOL(set_pages_array_uc);
1815 int set_pages_array_wc(struct page **pages, int addrinarray)
1817 return _set_pages_array(pages, addrinarray, _PAGE_CACHE_MODE_WC);
1819 EXPORT_SYMBOL(set_pages_array_wc);
1821 int set_pages_array_wt(struct page **pages, int addrinarray)
1823 return _set_pages_array(pages, addrinarray, _PAGE_CACHE_MODE_WT);
1825 EXPORT_SYMBOL_GPL(set_pages_array_wt);
1827 int set_pages_wb(struct page *page, int numpages)
1829 unsigned long addr = (unsigned long)page_address(page);
1831 return set_memory_wb(addr, numpages);
1833 EXPORT_SYMBOL(set_pages_wb);
1835 int set_pages_array_wb(struct page **pages, int addrinarray)
1838 unsigned long start;
1842 /* WB cache mode is hard wired to all cache attribute bits being 0 */
1843 retval = cpa_clear_pages_array(pages, addrinarray,
1844 __pgprot(_PAGE_CACHE_MASK));
1848 for (i = 0; i < addrinarray; i++) {
1849 if (PageHighMem(pages[i]))
1851 start = page_to_pfn(pages[i]) << PAGE_SHIFT;
1852 end = start + PAGE_SIZE;
1853 free_memtype(start, end);
1858 EXPORT_SYMBOL(set_pages_array_wb);
1860 int set_pages_x(struct page *page, int numpages)
1862 unsigned long addr = (unsigned long)page_address(page);
1864 return set_memory_x(addr, numpages);
1866 EXPORT_SYMBOL(set_pages_x);
1868 int set_pages_nx(struct page *page, int numpages)
1870 unsigned long addr = (unsigned long)page_address(page);
1872 return set_memory_nx(addr, numpages);
1874 EXPORT_SYMBOL(set_pages_nx);
1876 int set_pages_ro(struct page *page, int numpages)
1878 unsigned long addr = (unsigned long)page_address(page);
1880 return set_memory_ro(addr, numpages);
1883 int set_pages_rw(struct page *page, int numpages)
1885 unsigned long addr = (unsigned long)page_address(page);
1887 return set_memory_rw(addr, numpages);
1890 #ifdef CONFIG_DEBUG_PAGEALLOC
1892 static int __set_pages_p(struct page *page, int numpages)
1894 unsigned long tempaddr = (unsigned long) page_address(page);
1895 struct cpa_data cpa = { .vaddr = &tempaddr,
1897 .numpages = numpages,
1898 .mask_set = __pgprot(_PAGE_PRESENT | _PAGE_RW),
1899 .mask_clr = __pgprot(0),
1903 * No alias checking needed for setting present flag. otherwise,
1904 * we may need to break large pages for 64-bit kernel text
1905 * mappings (this adds to complexity if we want to do this from
1906 * atomic context especially). Let's keep it simple!
1908 return __change_page_attr_set_clr(&cpa, 0);
1911 static int __set_pages_np(struct page *page, int numpages)
1913 unsigned long tempaddr = (unsigned long) page_address(page);
1914 struct cpa_data cpa = { .vaddr = &tempaddr,
1916 .numpages = numpages,
1917 .mask_set = __pgprot(0),
1918 .mask_clr = __pgprot(_PAGE_PRESENT | _PAGE_RW),
1922 * No alias checking needed for setting not present flag. otherwise,
1923 * we may need to break large pages for 64-bit kernel text
1924 * mappings (this adds to complexity if we want to do this from
1925 * atomic context especially). Let's keep it simple!
1927 return __change_page_attr_set_clr(&cpa, 0);
1930 void __kernel_map_pages(struct page *page, int numpages, int enable)
1932 if (PageHighMem(page))
1935 debug_check_no_locks_freed(page_address(page),
1936 numpages * PAGE_SIZE);
1940 * The return value is ignored as the calls cannot fail.
1941 * Large pages for identity mappings are not used at boot time
1942 * and hence no memory allocations during large page split.
1945 __set_pages_p(page, numpages);
1947 __set_pages_np(page, numpages);
1950 * We should perform an IPI and flush all tlbs,
1951 * but that can deadlock->flush only current cpu:
1955 arch_flush_lazy_mmu_mode();
1958 #ifdef CONFIG_HIBERNATION
1960 bool kernel_page_present(struct page *page)
1965 if (PageHighMem(page))
1968 pte = lookup_address((unsigned long)page_address(page), &level);
1969 return (pte_val(*pte) & _PAGE_PRESENT);
1972 #endif /* CONFIG_HIBERNATION */
1974 #endif /* CONFIG_DEBUG_PAGEALLOC */
1976 int kernel_map_pages_in_pgd(pgd_t *pgd, u64 pfn, unsigned long address,
1977 unsigned numpages, unsigned long page_flags)
1979 int retval = -EINVAL;
1981 struct cpa_data cpa = {
1985 .numpages = numpages,
1986 .mask_set = __pgprot(0),
1987 .mask_clr = __pgprot(0),
1991 if (!(__supported_pte_mask & _PAGE_NX))
1994 if (!(page_flags & _PAGE_NX))
1995 cpa.mask_clr = __pgprot(_PAGE_NX);
1997 cpa.mask_set = __pgprot(_PAGE_PRESENT | page_flags);
1999 retval = __change_page_attr_set_clr(&cpa, 0);
2006 void kernel_unmap_pages_in_pgd(pgd_t *root, unsigned long address,
2009 unmap_pgd_range(root, address, address + (numpages << PAGE_SHIFT));
2013 * The testcases use internal knowledge of the implementation that shouldn't
2014 * be exposed to the rest of the kernel. Include these directly here.
2016 #ifdef CONFIG_CPA_DEBUG
2017 #include "pageattr-test.c"