2 * Copyright 2002 Andi Kleen, SuSE Labs.
3 * Thanks to Ben LaHaise for precious feedback.
6 #include <linux/highmem.h>
7 #include <linux/module.h>
8 #include <linux/sched.h>
9 #include <linux/slab.h>
12 #include <asm/processor.h>
13 #include <asm/tlbflush.h>
14 #include <asm/sections.h>
15 #include <asm/uaccess.h>
16 #include <asm/pgalloc.h>
18 pte_t *lookup_address(unsigned long address, int *level)
20 pgd_t *pgd = pgd_offset_k(address);
26 pud = pud_offset(pgd, address);
29 pmd = pmd_offset(pud, address);
37 return pte_offset_kernel(pmd, address);
40 static void set_pmd_pte(pte_t *kpte, unsigned long address, pte_t pte)
46 set_pte_atomic(kpte, pte);
47 if (SHARED_KERNEL_PMD)
50 spin_lock_irqsave(&pgd_lock, flags);
51 for (page = pgd_list; page; page = (struct page *)page->index) {
56 pgd = (pgd_t *)page_address(page) + pgd_index(address);
57 pud = pud_offset(pgd, address);
58 pmd = pmd_offset(pud, address);
59 set_pte_atomic((pte_t *)pmd, pte);
61 spin_unlock_irqrestore(&pgd_lock, flags);
65 split_large_page(pte_t *kpte, unsigned long address, pgprot_t ref_prot)
72 base = alloc_pages(GFP_KERNEL, 0);
76 down_write(&init_mm.mmap_sem);
78 * Check for races, another CPU might have split this page
81 tmp = lookup_address(address, &level);
85 address = __pa(address);
86 addr = address & LARGE_PAGE_MASK;
87 pbase = (pte_t *)page_address(base);
88 paravirt_alloc_pt(&init_mm, page_to_pfn(base));
90 for (i = 0; i < PTRS_PER_PTE; i++, addr += PAGE_SIZE)
91 set_pte(&pbase[i], pfn_pte(addr >> PAGE_SHIFT, ref_prot));
94 * Install the new, split up pagetable:
96 set_pmd_pte(kpte, address, mk_pte(base, ref_prot));
100 up_write(&init_mm.mmap_sem);
103 __free_pages(base, 0);
108 static int __change_page_attr(struct page *page, pgprot_t prot)
110 pgprot_t ref_prot = PAGE_KERNEL;
111 struct page *kpte_page;
112 unsigned long address;
117 BUG_ON(PageHighMem(page));
118 address = (unsigned long)page_address(page);
121 kpte = lookup_address(address, &level);
125 oldprot = pte_pgprot(*kpte);
126 kpte_page = virt_to_page(kpte);
127 BUG_ON(PageLRU(kpte_page));
128 BUG_ON(PageCompound(kpte_page));
131 * Better fail early if someone sets the kernel text to NX.
132 * Does not cover __inittext
134 BUG_ON(address >= (unsigned long)&_text &&
135 address < (unsigned long)&_etext &&
136 (pgprot_val(prot) & _PAGE_NX));
138 if ((address & LARGE_PAGE_MASK) < (unsigned long)&_etext)
139 ref_prot = PAGE_KERNEL_EXEC;
141 ref_prot = canon_pgprot(ref_prot);
142 prot = canon_pgprot(prot);
145 set_pte_atomic(kpte, mk_pte(page, prot));
147 err = split_large_page(kpte, address, ref_prot);
155 * Change the page attributes of an page in the linear mapping.
157 * This should be used when a page is mapped with a different caching policy
158 * than write-back somewhere - some CPUs do not like it when mappings with
159 * different caching policies exist. This changes the page attributes of the
160 * in kernel linear mapping too.
162 * The caller needs to ensure that there are no conflicting mappings elsewhere.
163 * This function only deals with the kernel linear map.
165 * Caller must call global_flush_tlb() after this.
167 int change_page_attr(struct page *page, int numpages, pgprot_t prot)
171 for (i = 0; i < numpages; i++, page++) {
172 err = __change_page_attr(page, prot);
179 EXPORT_SYMBOL(change_page_attr);
181 int change_page_attr_addr(unsigned long addr, int numpages, pgprot_t prot)
184 unsigned long pfn = (addr >> PAGE_SHIFT);
186 for (i = 0; i < numpages; i++) {
187 if (!pfn_valid(pfn + i)) {
191 pte_t *pte = lookup_address(addr + i*PAGE_SIZE, &level);
192 BUG_ON(pte && !pte_none(*pte));
195 return change_page_attr(virt_to_page(addr), i, prot);
198 static void flush_kernel_map(void *arg)
201 * Flush all to work around Errata in early athlons regarding
202 * large page flushing.
206 if (boot_cpu_data.x86_model >= 4)
210 void global_flush_tlb(void)
212 BUG_ON(irqs_disabled());
214 on_each_cpu(flush_kernel_map, NULL, 1, 1);
216 EXPORT_SYMBOL(global_flush_tlb);
218 #ifdef CONFIG_DEBUG_PAGEALLOC
219 void kernel_map_pages(struct page *page, int numpages, int enable)
221 if (PageHighMem(page))
224 debug_check_no_locks_freed(page_address(page),
225 numpages * PAGE_SIZE);
229 * the return value is ignored - the calls cannot fail,
230 * large pages are disabled at boot time.
232 change_page_attr(page, numpages, enable ? PAGE_KERNEL : __pgprot(0));
235 * we should perform an IPI and flush all tlbs,
236 * but that can deadlock->flush only current cpu.