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)
43 set_pte_atomic(kpte, pte);
45 if (SHARED_KERNEL_PMD)
50 for (page = pgd_list; page; page = (struct page *)page->index) {
55 pgd = (pgd_t *)page_address(page) + pgd_index(address);
56 pud = pud_offset(pgd, address);
57 pmd = pmd_offset(pud, address);
58 set_pte_atomic((pte_t *)pmd, pte);
64 static int split_large_page(pte_t *kpte, unsigned long address)
66 pgprot_t ref_prot = pte_pgprot(pte_clrhuge(*kpte));
67 gfp_t gfp_flags = GFP_KERNEL;
75 #ifdef CONFIG_DEBUG_PAGEALLOC
76 gfp_flags = GFP_ATOMIC;
78 base = alloc_pages(gfp_flags, 0);
82 spin_lock_irqsave(&pgd_lock, flags);
84 * Check for races, another CPU might have split this page
87 tmp = lookup_address(address, &level);
93 address = __pa(address);
94 addr = address & LARGE_PAGE_MASK;
95 pbase = (pte_t *)page_address(base);
97 paravirt_alloc_pt(&init_mm, page_to_pfn(base));
100 for (i = 0; i < PTRS_PER_PTE; i++, addr += PAGE_SIZE)
101 set_pte(&pbase[i], pfn_pte(addr >> PAGE_SHIFT, ref_prot));
104 * Install the new, split up pagetable:
106 __set_pmd_pte(kpte, address, mk_pte(base, ref_prot));
110 spin_unlock_irqrestore(&pgd_lock, flags);
113 __free_pages(base, 0);
119 __change_page_attr(unsigned long address, struct page *page, pgprot_t prot)
121 struct page *kpte_page;
125 BUG_ON(PageHighMem(page));
128 kpte = lookup_address(address, &level);
132 kpte_page = virt_to_page(kpte);
133 BUG_ON(PageLRU(kpte_page));
134 BUG_ON(PageCompound(kpte_page));
137 * Better fail early if someone sets the kernel text to NX.
138 * Does not cover __inittext
140 BUG_ON(address >= (unsigned long)&_text &&
141 address < (unsigned long)&_etext &&
142 (pgprot_val(prot) & _PAGE_NX));
145 set_pte_atomic(kpte, mk_pte(page, canon_pgprot(prot)));
147 err = split_large_page(kpte, address);
155 * change_page_attr_addr - Change page table attributes in linear mapping
156 * @address: Virtual address in linear mapping.
157 * @numpages: Number of pages to change
158 * @prot: New page table attribute (PAGE_*)
160 * Change page attributes of a page in the direct mapping. This is a variant
161 * of change_page_attr() that also works on memory holes that do not have
162 * mem_map entry (pfn_valid() is false).
164 * See change_page_attr() documentation for more details.
167 int change_page_attr_addr(unsigned long address, int numpages, pgprot_t prot)
169 int err = 0, kernel_map = 0, i;
172 if (address >= __START_KERNEL_map &&
173 address < __START_KERNEL_map + KERNEL_TEXT_SIZE) {
175 address = (unsigned long)__va(__pa(address));
180 for (i = 0; i < numpages; i++, address += PAGE_SIZE) {
181 unsigned long pfn = __pa(address) >> PAGE_SHIFT;
183 if (!kernel_map || pte_present(pfn_pte(0, prot))) {
184 err = __change_page_attr(address, pfn_to_page(pfn), prot);
190 * Handle kernel mapping too which aliases part of
193 if (__pa(address) < KERNEL_TEXT_SIZE) {
197 addr2 = __START_KERNEL_map + __pa(address);
198 /* Make sure the kernel mappings stay executable */
199 prot2 = pte_pgprot(pte_mkexec(pfn_pte(0, prot)));
200 err = __change_page_attr(addr2, pfn_to_page(pfn), prot2);
209 * change_page_attr - Change page table attributes in the linear mapping.
210 * @page: First page to change
211 * @numpages: Number of pages to change
212 * @prot: New protection/caching type (PAGE_*)
214 * Returns 0 on success, otherwise a negated errno.
216 * This should be used when a page is mapped with a different caching policy
217 * than write-back somewhere - some CPUs do not like it when mappings with
218 * different caching policies exist. This changes the page attributes of the
219 * in kernel linear mapping too.
221 * Caller must call global_flush_tlb() later to make the changes active.
223 * The caller needs to ensure that there are no conflicting mappings elsewhere
224 * (e.g. in user space) * This function only deals with the kernel linear map.
226 * For MMIO areas without mem_map use change_page_attr_addr() instead.
228 int change_page_attr(struct page *page, int numpages, pgprot_t prot)
230 unsigned long addr = (unsigned long)page_address(page);
232 return change_page_attr_addr(addr, numpages, prot);
234 EXPORT_SYMBOL(change_page_attr);
236 static void flush_kernel_map(void *arg)
239 * Flush all to work around Errata in early athlons regarding
240 * large page flushing.
244 if (boot_cpu_data.x86_model >= 4)
248 void global_flush_tlb(void)
250 BUG_ON(irqs_disabled());
252 on_each_cpu(flush_kernel_map, NULL, 1, 1);
254 EXPORT_SYMBOL(global_flush_tlb);
256 #ifdef CONFIG_DEBUG_PAGEALLOC
257 void kernel_map_pages(struct page *page, int numpages, int enable)
259 if (PageHighMem(page))
262 debug_check_no_locks_freed(page_address(page),
263 numpages * PAGE_SIZE);
267 * If page allocator is not up yet then do not call c_p_a():
269 if (!debug_pagealloc_enabled)
273 * the return value is ignored - the calls cannot fail,
274 * large pages are disabled at boot time.
276 change_page_attr(page, numpages, enable ? PAGE_KERNEL : __pgprot(0));
279 * we should perform an IPI and flush all tlbs,
280 * but that can deadlock->flush only current cpu.