2 * linux/arch/i386/mm/pgtable.c
5 #include <linux/sched.h>
6 #include <linux/kernel.h>
7 #include <linux/errno.h>
9 #include <linux/swap.h>
10 #include <linux/smp.h>
11 #include <linux/highmem.h>
12 #include <linux/slab.h>
13 #include <linux/pagemap.h>
14 #include <linux/spinlock.h>
15 #include <linux/module.h>
16 #include <linux/quicklist.h>
18 #include <asm/system.h>
19 #include <asm/pgtable.h>
20 #include <asm/pgalloc.h>
21 #include <asm/fixmap.h>
24 #include <asm/tlbflush.h>
28 int total = 0, reserved = 0;
29 int shared = 0, cached = 0;
36 printk(KERN_INFO "Mem-info:\n");
38 printk(KERN_INFO "Free swap: %6ldkB\n", nr_swap_pages<<(PAGE_SHIFT-10));
39 for_each_online_pgdat(pgdat) {
40 pgdat_resize_lock(pgdat, &flags);
41 for (i = 0; i < pgdat->node_spanned_pages; ++i) {
42 page = pgdat_page_nr(pgdat, i);
44 if (PageHighMem(page))
46 if (PageReserved(page))
48 else if (PageSwapCache(page))
50 else if (page_count(page))
51 shared += page_count(page) - 1;
53 pgdat_resize_unlock(pgdat, &flags);
55 printk(KERN_INFO "%d pages of RAM\n", total);
56 printk(KERN_INFO "%d pages of HIGHMEM\n", highmem);
57 printk(KERN_INFO "%d reserved pages\n", reserved);
58 printk(KERN_INFO "%d pages shared\n", shared);
59 printk(KERN_INFO "%d pages swap cached\n", cached);
61 printk(KERN_INFO "%lu pages dirty\n", global_page_state(NR_FILE_DIRTY));
62 printk(KERN_INFO "%lu pages writeback\n",
63 global_page_state(NR_WRITEBACK));
64 printk(KERN_INFO "%lu pages mapped\n", global_page_state(NR_FILE_MAPPED));
65 printk(KERN_INFO "%lu pages slab\n",
66 global_page_state(NR_SLAB_RECLAIMABLE) +
67 global_page_state(NR_SLAB_UNRECLAIMABLE));
68 printk(KERN_INFO "%lu pages pagetables\n",
69 global_page_state(NR_PAGETABLE));
73 * Associate a virtual page frame with a given physical page frame
74 * and protection flags for that frame.
76 static void set_pte_pfn(unsigned long vaddr, unsigned long pfn, pgprot_t flags)
83 pgd = swapper_pg_dir + pgd_index(vaddr);
88 pud = pud_offset(pgd, vaddr);
93 pmd = pmd_offset(pud, vaddr);
98 pte = pte_offset_kernel(pmd, vaddr);
99 if (pgprot_val(flags))
100 set_pte_present(&init_mm, vaddr, pte, pfn_pte(pfn, flags));
102 pte_clear(&init_mm, vaddr, pte);
105 * It's enough to flush this one mapping.
106 * (PGE mappings get flushed as well)
108 __flush_tlb_one(vaddr);
112 * Associate a large virtual page frame with a given physical page frame
113 * and protection flags for that frame. pfn is for the base of the page,
114 * vaddr is what the page gets mapped to - both must be properly aligned.
115 * The pmd must already be instantiated. Assumes PAE mode.
117 void set_pmd_pfn(unsigned long vaddr, unsigned long pfn, pgprot_t flags)
123 if (vaddr & (PMD_SIZE-1)) { /* vaddr is misaligned */
124 printk(KERN_WARNING "set_pmd_pfn: vaddr misaligned\n");
127 if (pfn & (PTRS_PER_PTE-1)) { /* pfn is misaligned */
128 printk(KERN_WARNING "set_pmd_pfn: pfn misaligned\n");
131 pgd = swapper_pg_dir + pgd_index(vaddr);
132 if (pgd_none(*pgd)) {
133 printk(KERN_WARNING "set_pmd_pfn: pgd_none\n");
136 pud = pud_offset(pgd, vaddr);
137 pmd = pmd_offset(pud, vaddr);
138 set_pmd(pmd, pfn_pmd(pfn, flags));
140 * It's enough to flush this one mapping.
141 * (PGE mappings get flushed as well)
143 __flush_tlb_one(vaddr);
147 unsigned long __FIXADDR_TOP = 0xfffff000;
148 EXPORT_SYMBOL(__FIXADDR_TOP);
150 void __set_fixmap (enum fixed_addresses idx, unsigned long phys, pgprot_t flags)
152 unsigned long address = __fix_to_virt(idx);
154 if (idx >= __end_of_fixed_addresses) {
158 set_pte_pfn(address, phys >> PAGE_SHIFT, flags);
163 * reserve_top_address - reserves a hole in the top of kernel address space
164 * @reserve - size of hole to reserve
166 * Can be used to relocate the fixmap area and poke a hole in the top
167 * of kernel address space to make room for a hypervisor.
169 void reserve_top_address(unsigned long reserve)
172 printk(KERN_INFO "Reserving virtual address space above 0x%08x\n",
174 __FIXADDR_TOP = -reserve - PAGE_SIZE;
175 __VMALLOC_RESERVE += reserve;
178 pte_t *pte_alloc_one_kernel(struct mm_struct *mm, unsigned long address)
180 return (pte_t *)__get_free_page(GFP_KERNEL|__GFP_REPEAT|__GFP_ZERO);
183 struct page *pte_alloc_one(struct mm_struct *mm, unsigned long address)
187 #ifdef CONFIG_HIGHPTE
188 pte = alloc_pages(GFP_KERNEL|__GFP_HIGHMEM|__GFP_REPEAT|__GFP_ZERO, 0);
190 pte = alloc_pages(GFP_KERNEL|__GFP_REPEAT|__GFP_ZERO, 0);
195 void pmd_ctor(struct kmem_cache *cache, void *pmd)
197 memset(pmd, 0, PTRS_PER_PMD*sizeof(pmd_t));
201 * List of all pgd's needed for non-PAE so it can invalidate entries
202 * in both cached and uncached pgd's; not needed for PAE since the
203 * kernel pmd is shared. If PAE were not to share the pmd a similar
204 * tactic would be needed. This is essentially codepath-based locking
205 * against pageattr.c; it is the unique case in which a valid change
206 * of kernel pagetables can't be lazily synchronized by vmalloc faults.
207 * vmalloc faults work because attached pagetables are never freed.
210 DEFINE_SPINLOCK(pgd_lock);
211 struct page *pgd_list;
213 static inline void pgd_list_add(pgd_t *pgd)
215 struct page *page = virt_to_page(pgd);
216 page->index = (unsigned long)pgd_list;
218 set_page_private(pgd_list, (unsigned long)&page->index);
220 set_page_private(page, (unsigned long)&pgd_list);
223 static inline void pgd_list_del(pgd_t *pgd)
225 struct page *next, **pprev, *page = virt_to_page(pgd);
226 next = (struct page *)page->index;
227 pprev = (struct page **)page_private(page);
230 set_page_private(next, (unsigned long)pprev);
235 #if (PTRS_PER_PMD == 1)
236 /* Non-PAE pgd constructor */
237 static void pgd_ctor(void *pgd)
241 /* !PAE, no pagetable sharing */
242 memset(pgd, 0, USER_PTRS_PER_PGD*sizeof(pgd_t));
244 spin_lock_irqsave(&pgd_lock, flags);
246 /* must happen under lock */
247 clone_pgd_range((pgd_t *)pgd + USER_PTRS_PER_PGD,
248 swapper_pg_dir + USER_PTRS_PER_PGD,
250 paravirt_alloc_pd_clone(__pa(pgd) >> PAGE_SHIFT,
251 __pa(swapper_pg_dir) >> PAGE_SHIFT,
255 spin_unlock_irqrestore(&pgd_lock, flags);
257 #else /* PTRS_PER_PMD > 1 */
258 /* PAE pgd constructor */
259 static void pgd_ctor(void *pgd)
261 /* PAE, kernel PMD may be shared */
263 if (SHARED_KERNEL_PMD) {
264 clone_pgd_range((pgd_t *)pgd + USER_PTRS_PER_PGD,
265 swapper_pg_dir + USER_PTRS_PER_PGD,
270 memset(pgd, 0, USER_PTRS_PER_PGD*sizeof(pgd_t));
271 spin_lock_irqsave(&pgd_lock, flags);
273 spin_unlock_irqrestore(&pgd_lock, flags);
276 #endif /* PTRS_PER_PMD */
278 static void pgd_dtor(void *pgd)
280 unsigned long flags; /* can be called from interrupt context */
282 if (SHARED_KERNEL_PMD)
285 paravirt_release_pd(__pa(pgd) >> PAGE_SHIFT);
286 spin_lock_irqsave(&pgd_lock, flags);
288 spin_unlock_irqrestore(&pgd_lock, flags);
291 #define UNSHARED_PTRS_PER_PGD \
292 (SHARED_KERNEL_PMD ? USER_PTRS_PER_PGD : PTRS_PER_PGD)
294 /* If we allocate a pmd for part of the kernel address space, then
295 make sure its initialized with the appropriate kernel mappings.
296 Otherwise use a cached zeroed pmd. */
297 static pmd_t *pmd_cache_alloc(int idx)
301 if (idx >= USER_PTRS_PER_PGD) {
302 pmd = (pmd_t *)__get_free_page(GFP_KERNEL);
306 (void *)pgd_page_vaddr(swapper_pg_dir[idx]),
307 sizeof(pmd_t) * PTRS_PER_PMD);
309 pmd = kmem_cache_alloc(pmd_cache, GFP_KERNEL);
314 static void pmd_cache_free(pmd_t *pmd, int idx)
316 if (idx >= USER_PTRS_PER_PGD)
317 free_page((unsigned long)pmd);
319 kmem_cache_free(pmd_cache, pmd);
322 pgd_t *pgd_alloc(struct mm_struct *mm)
325 pgd_t *pgd = quicklist_alloc(0, GFP_KERNEL, pgd_ctor);
327 if (PTRS_PER_PMD == 1 || !pgd)
330 for (i = 0; i < UNSHARED_PTRS_PER_PGD; ++i) {
331 pmd_t *pmd = pmd_cache_alloc(i);
336 paravirt_alloc_pd(__pa(pmd) >> PAGE_SHIFT);
337 set_pgd(&pgd[i], __pgd(1 + __pa(pmd)));
342 for (i--; i >= 0; i--) {
343 pgd_t pgdent = pgd[i];
344 void* pmd = (void *)__va(pgd_val(pgdent)-1);
345 paravirt_release_pd(__pa(pmd) >> PAGE_SHIFT);
346 pmd_cache_free(pmd, i);
348 quicklist_free(0, pgd_dtor, pgd);
352 void pgd_free(pgd_t *pgd)
356 /* in the PAE case user pgd entries are overwritten before usage */
357 if (PTRS_PER_PMD > 1)
358 for (i = 0; i < UNSHARED_PTRS_PER_PGD; ++i) {
359 pgd_t pgdent = pgd[i];
360 void* pmd = (void *)__va(pgd_val(pgdent)-1);
361 paravirt_release_pd(__pa(pmd) >> PAGE_SHIFT);
362 pmd_cache_free(pmd, i);
364 /* in the non-PAE case, free_pgtables() clears user pgd entries */
365 quicklist_free(0, pgd_dtor, pgd);
368 void check_pgt_cache(void)
370 quicklist_trim(0, pgd_dtor, 25, 16);