2 * linux/arch/i386/mm/pgtable.c
5 #include <linux/sched.h>
6 #include <linux/kernel.h>
7 #include <linux/errno.h>
10 #include <linux/swap.h>
11 #include <linux/smp.h>
12 #include <linux/highmem.h>
13 #include <linux/slab.h>
14 #include <linux/pagemap.h>
15 #include <linux/spinlock.h>
16 #include <linux/module.h>
17 #include <linux/quicklist.h>
19 #include <asm/system.h>
20 #include <asm/pgtable.h>
21 #include <asm/pgalloc.h>
22 #include <asm/fixmap.h>
25 #include <asm/tlbflush.h>
29 int total = 0, reserved = 0;
30 int shared = 0, cached = 0;
37 printk(KERN_INFO "Mem-info:\n");
39 printk(KERN_INFO "Free swap: %6ldkB\n", nr_swap_pages<<(PAGE_SHIFT-10));
40 for_each_online_pgdat(pgdat) {
41 pgdat_resize_lock(pgdat, &flags);
42 for (i = 0; i < pgdat->node_spanned_pages; ++i) {
43 if (unlikely(i % MAX_ORDER_NR_PAGES == 0))
45 page = pgdat_page_nr(pgdat, i);
47 if (PageHighMem(page))
49 if (PageReserved(page))
51 else if (PageSwapCache(page))
53 else if (page_count(page))
54 shared += page_count(page) - 1;
56 pgdat_resize_unlock(pgdat, &flags);
58 printk(KERN_INFO "%d pages of RAM\n", total);
59 printk(KERN_INFO "%d pages of HIGHMEM\n", highmem);
60 printk(KERN_INFO "%d reserved pages\n", reserved);
61 printk(KERN_INFO "%d pages shared\n", shared);
62 printk(KERN_INFO "%d pages swap cached\n", cached);
64 printk(KERN_INFO "%lu pages dirty\n", global_page_state(NR_FILE_DIRTY));
65 printk(KERN_INFO "%lu pages writeback\n",
66 global_page_state(NR_WRITEBACK));
67 printk(KERN_INFO "%lu pages mapped\n", global_page_state(NR_FILE_MAPPED));
68 printk(KERN_INFO "%lu pages slab\n",
69 global_page_state(NR_SLAB_RECLAIMABLE) +
70 global_page_state(NR_SLAB_UNRECLAIMABLE));
71 printk(KERN_INFO "%lu pages pagetables\n",
72 global_page_state(NR_PAGETABLE));
76 * Associate a virtual page frame with a given physical page frame
77 * and protection flags for that frame.
79 static void set_pte_pfn(unsigned long vaddr, unsigned long pfn, pgprot_t flags)
86 pgd = swapper_pg_dir + pgd_index(vaddr);
91 pud = pud_offset(pgd, vaddr);
96 pmd = pmd_offset(pud, vaddr);
101 pte = pte_offset_kernel(pmd, vaddr);
102 if (pgprot_val(flags))
103 set_pte_present(&init_mm, vaddr, pte, pfn_pte(pfn, flags));
105 pte_clear(&init_mm, vaddr, pte);
108 * It's enough to flush this one mapping.
109 * (PGE mappings get flushed as well)
111 __flush_tlb_one(vaddr);
115 * Associate a large virtual page frame with a given physical page frame
116 * and protection flags for that frame. pfn is for the base of the page,
117 * vaddr is what the page gets mapped to - both must be properly aligned.
118 * The pmd must already be instantiated. Assumes PAE mode.
120 void set_pmd_pfn(unsigned long vaddr, unsigned long pfn, pgprot_t flags)
126 if (vaddr & (PMD_SIZE-1)) { /* vaddr is misaligned */
127 printk(KERN_WARNING "set_pmd_pfn: vaddr misaligned\n");
130 if (pfn & (PTRS_PER_PTE-1)) { /* pfn is misaligned */
131 printk(KERN_WARNING "set_pmd_pfn: pfn misaligned\n");
134 pgd = swapper_pg_dir + pgd_index(vaddr);
135 if (pgd_none(*pgd)) {
136 printk(KERN_WARNING "set_pmd_pfn: pgd_none\n");
139 pud = pud_offset(pgd, vaddr);
140 pmd = pmd_offset(pud, vaddr);
141 set_pmd(pmd, pfn_pmd(pfn, flags));
143 * It's enough to flush this one mapping.
144 * (PGE mappings get flushed as well)
146 __flush_tlb_one(vaddr);
150 unsigned long __FIXADDR_TOP = 0xfffff000;
151 EXPORT_SYMBOL(__FIXADDR_TOP);
153 void __set_fixmap (enum fixed_addresses idx, unsigned long phys, pgprot_t flags)
155 unsigned long address = __fix_to_virt(idx);
157 if (idx >= __end_of_fixed_addresses) {
161 set_pte_pfn(address, phys >> PAGE_SHIFT, flags);
166 * reserve_top_address - reserves a hole in the top of kernel address space
167 * @reserve - size of hole to reserve
169 * Can be used to relocate the fixmap area and poke a hole in the top
170 * of kernel address space to make room for a hypervisor.
172 void reserve_top_address(unsigned long reserve)
175 printk(KERN_INFO "Reserving virtual address space above 0x%08x\n",
177 __FIXADDR_TOP = -reserve - PAGE_SIZE;
178 __VMALLOC_RESERVE += reserve;
181 pte_t *pte_alloc_one_kernel(struct mm_struct *mm, unsigned long address)
183 return (pte_t *)__get_free_page(GFP_KERNEL|__GFP_REPEAT|__GFP_ZERO);
186 struct page *pte_alloc_one(struct mm_struct *mm, unsigned long address)
190 #ifdef CONFIG_HIGHPTE
191 pte = alloc_pages(GFP_KERNEL|__GFP_HIGHMEM|__GFP_REPEAT|__GFP_ZERO, 0);
193 pte = alloc_pages(GFP_KERNEL|__GFP_REPEAT|__GFP_ZERO, 0);
199 * List of all pgd's needed for non-PAE so it can invalidate entries
200 * in both cached and uncached pgd's; not needed for PAE since the
201 * kernel pmd is shared. If PAE were not to share the pmd a similar
202 * tactic would be needed. This is essentially codepath-based locking
203 * against pageattr.c; it is the unique case in which a valid change
204 * of kernel pagetables can't be lazily synchronized by vmalloc faults.
205 * vmalloc faults work because attached pagetables are never freed.
208 static inline void pgd_list_add(pgd_t *pgd)
210 struct page *page = virt_to_page(pgd);
212 list_add(&page->lru, &pgd_list);
215 static inline void pgd_list_del(pgd_t *pgd)
217 struct page *page = virt_to_page(pgd);
219 list_del(&page->lru);
222 #define UNSHARED_PTRS_PER_PGD \
223 (SHARED_KERNEL_PMD ? USER_PTRS_PER_PGD : PTRS_PER_PGD)
225 static void pgd_ctor(void *p)
230 /* Clear usermode parts of PGD */
231 memset(pgd, 0, USER_PTRS_PER_PGD*sizeof(pgd_t));
233 spin_lock_irqsave(&pgd_lock, flags);
235 /* If the pgd points to a shared pagetable level (either the
236 ptes in non-PAE, or shared PMD in PAE), then just copy the
237 references from swapper_pg_dir. */
238 if (PAGETABLE_LEVELS == 2 ||
239 (PAGETABLE_LEVELS == 3 && SHARED_KERNEL_PMD)) {
240 clone_pgd_range(pgd + USER_PTRS_PER_PGD,
241 swapper_pg_dir + USER_PTRS_PER_PGD,
243 paravirt_alloc_pd_clone(__pa(pgd) >> PAGE_SHIFT,
244 __pa(swapper_pg_dir) >> PAGE_SHIFT,
249 /* list required to sync kernel mapping updates */
250 if (!SHARED_KERNEL_PMD)
253 spin_unlock_irqrestore(&pgd_lock, flags);
256 static void pgd_dtor(void *pgd)
258 unsigned long flags; /* can be called from interrupt context */
260 if (SHARED_KERNEL_PMD)
263 spin_lock_irqsave(&pgd_lock, flags);
265 spin_unlock_irqrestore(&pgd_lock, flags);
268 #ifdef CONFIG_X86_PAE
270 * Mop up any pmd pages which may still be attached to the pgd.
271 * Normally they will be freed by munmap/exit_mmap, but any pmd we
272 * preallocate which never got a corresponding vma will need to be
275 static void pgd_mop_up_pmds(struct mm_struct *mm, pgd_t *pgdp)
279 for(i = 0; i < UNSHARED_PTRS_PER_PGD; i++) {
282 if (pgd_val(pgd) != 0) {
283 pmd_t *pmd = (pmd_t *)pgd_page_vaddr(pgd);
285 pgdp[i] = native_make_pgd(0);
287 paravirt_release_pd(pgd_val(pgd) >> PAGE_SHIFT);
294 * In PAE mode, we need to do a cr3 reload (=tlb flush) when
295 * updating the top-level pagetable entries to guarantee the
296 * processor notices the update. Since this is expensive, and
297 * all 4 top-level entries are used almost immediately in a
298 * new process's life, we just pre-populate them here.
300 * Also, if we're in a paravirt environment where the kernel pmd is
301 * not shared between pagetables (!SHARED_KERNEL_PMDS), we allocate
302 * and initialize the kernel pmds here.
304 static int pgd_prepopulate_pmd(struct mm_struct *mm, pgd_t *pgd)
310 pud = pud_offset(pgd, 0);
311 for (addr = i = 0; i < UNSHARED_PTRS_PER_PGD;
312 i++, pud++, addr += PUD_SIZE) {
313 pmd_t *pmd = pmd_alloc_one(mm, addr);
316 pgd_mop_up_pmds(mm, pgd);
320 if (i >= USER_PTRS_PER_PGD)
321 memcpy(pmd, (pmd_t *)pgd_page_vaddr(swapper_pg_dir[i]),
322 sizeof(pmd_t) * PTRS_PER_PMD);
324 pud_populate(mm, pud, pmd);
329 #else /* !CONFIG_X86_PAE */
330 /* No need to prepopulate any pagetable entries in non-PAE modes. */
331 static int pgd_prepopulate_pmd(struct mm_struct *mm, pgd_t *pgd)
336 static void pgd_mop_up_pmds(struct mm_struct *mm, pgd_t *pgdp)
339 #endif /* CONFIG_X86_PAE */
341 pgd_t *pgd_alloc(struct mm_struct *mm)
343 pgd_t *pgd = quicklist_alloc(0, GFP_KERNEL, pgd_ctor);
345 mm->pgd = pgd; /* so that alloc_pd can use it */
347 if (pgd && !pgd_prepopulate_pmd(mm, pgd)) {
348 quicklist_free(0, pgd_dtor, pgd);
355 void pgd_free(struct mm_struct *mm, pgd_t *pgd)
357 pgd_mop_up_pmds(mm, pgd);
358 quicklist_free(0, pgd_dtor, pgd);
361 void check_pgt_cache(void)
363 quicklist_trim(0, pgd_dtor, 25, 16);
366 void __pte_free_tlb(struct mmu_gather *tlb, struct page *pte)
368 paravirt_release_pt(page_to_pfn(pte));
369 tlb_remove_page(tlb, pte);
372 #ifdef CONFIG_X86_PAE
374 void __pmd_free_tlb(struct mmu_gather *tlb, pmd_t *pmd)
376 paravirt_release_pd(__pa(pmd) >> PAGE_SHIFT);
377 tlb_remove_page(tlb, virt_to_page(pmd));