arch/i386/mm/pgtable.c

   1 /*
   2  *  linux/arch/i386/mm/pgtable.c
   3  */
   4
   5 #include <linux/sched.h>
   6 #include <linux/kernel.h>
   7 #include <linux/errno.h>
   8 #include <linux/mm.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
  16 #include <asm/system.h>
  17 #include <asm/pgtable.h>
  18 #include <asm/pgalloc.h>
  19 #include <asm/fixmap.h>
  20 #include <asm/e820.h>
  21 #include <asm/tlb.h>
  22 #include <asm/tlbflush.h>
  23
  24 void show_mem(void)
  25 {
  26         int total = 0, reserved = 0;
  27         int shared = 0, cached = 0;
  28         int highmem = 0;
  29         struct page *page;
  30         pg_data_t *pgdat;
  31         unsigned long i;
  32         unsigned long flags;
  33
  34         printk(KERN_INFO "Mem-info:\n");
  35         show_free_areas();
  36         printk(KERN_INFO "Free swap:       %6ldkB\n", nr_swap_pages<<(PAGE_SHIFT-10));
  37         for_each_online_pgdat(pgdat) {
  38                 pgdat_resize_lock(pgdat, &flags);
  39                 for (i = 0; i < pgdat->node_spanned_pages; ++i) {
  40                         page = pgdat_page_nr(pgdat, i);
  41                         total++;
  42                         if (PageHighMem(page))
  43                                 highmem++;
  44                         if (PageReserved(page))
  45                                 reserved++;
  46                         else if (PageSwapCache(page))
  47                                 cached++;
  48                         else if (page_count(page))
  49                                 shared += page_count(page) - 1;
  50                 }
  51                 pgdat_resize_unlock(pgdat, &flags);
  52         }
  53         printk(KERN_INFO "%d pages of RAM\n", total);
  54         printk(KERN_INFO "%d pages of HIGHMEM\n", highmem);
  55         printk(KERN_INFO "%d reserved pages\n", reserved);
  56         printk(KERN_INFO "%d pages shared\n", shared);
  57         printk(KERN_INFO "%d pages swap cached\n", cached);
  58
  59         printk(KERN_INFO "%lu pages dirty\n", global_page_state(NR_FILE_DIRTY));
  60         printk(KERN_INFO "%lu pages writeback\n",
  61                                         global_page_state(NR_WRITEBACK));
  62         printk(KERN_INFO "%lu pages mapped\n", global_page_state(NR_FILE_MAPPED));
  63         printk(KERN_INFO "%lu pages slab\n", global_page_state(NR_SLAB));
  64         printk(KERN_INFO "%lu pages pagetables\n",
  65                                         global_page_state(NR_PAGETABLE));
  66 }
  67
  68 /*
  69  * Associate a virtual page frame with a given physical page frame
  70  * and protection flags for that frame.
  71  */
  72 static void set_pte_pfn(unsigned long vaddr, unsigned long pfn, pgprot_t flags)
  73 {
  74         pgd_t *pgd;
  75         pud_t *pud;
  76         pmd_t *pmd;
  77         pte_t *pte;
  78
  79         pgd = swapper_pg_dir + pgd_index(vaddr);
  80         if (pgd_none(*pgd)) {
  81                 BUG();
  82                 return;
  83         }
  84         pud = pud_offset(pgd, vaddr);
  85         if (pud_none(*pud)) {
  86                 BUG();
  87                 return;
  88         }
  89         pmd = pmd_offset(pud, vaddr);
  90         if (pmd_none(*pmd)) {
  91                 BUG();
  92                 return;
  93         }
  94         pte = pte_offset_kernel(pmd, vaddr);
  95         /* <pfn,flags> stored as-is, to permit clearing entries */
  96         set_pte(pte, pfn_pte(pfn, flags));
  97
  98         /*
  99          * It's enough to flush this one mapping.
 100          * (PGE mappings get flushed as well)
 101          */
 102         __flush_tlb_one(vaddr);
 103 }
 104
 105 /*
 106  * Associate a large virtual page frame with a given physical page frame
 107  * and protection flags for that frame. pfn is for the base of the page,
 108  * vaddr is what the page gets mapped to - both must be properly aligned.
 109  * The pmd must already be instantiated. Assumes PAE mode.
 110  */
 111 void set_pmd_pfn(unsigned long vaddr, unsigned long pfn, pgprot_t flags)
 112 {
 113         pgd_t *pgd;
 114         pud_t *pud;
 115         pmd_t *pmd;
 116
 117         if (vaddr & (PMD_SIZE-1)) {             /* vaddr is misaligned */
 118                 printk(KERN_WARNING "set_pmd_pfn: vaddr misaligned\n");
 119                 return; /* BUG(); */
 120         }
 121         if (pfn & (PTRS_PER_PTE-1)) {           /* pfn is misaligned */
 122                 printk(KERN_WARNING "set_pmd_pfn: pfn misaligned\n");
 123                 return; /* BUG(); */
 124         }
 125         pgd = swapper_pg_dir + pgd_index(vaddr);
 126         if (pgd_none(*pgd)) {
 127                 printk(KERN_WARNING "set_pmd_pfn: pgd_none\n");
 128                 return; /* BUG(); */
 129         }
 130         pud = pud_offset(pgd, vaddr);
 131         pmd = pmd_offset(pud, vaddr);
 132         set_pmd(pmd, pfn_pmd(pfn, flags));
 133         /*
 134          * It's enough to flush this one mapping.
 135          * (PGE mappings get flushed as well)
 136          */
 137         __flush_tlb_one(vaddr);
 138 }
 139
 140 void __set_fixmap (enum fixed_addresses idx, unsigned long phys, pgprot_t flags)
 141 {
 142         unsigned long address = __fix_to_virt(idx);
 143
 144         if (idx >= __end_of_fixed_addresses) {
 145                 BUG();
 146                 return;
 147         }
 148         set_pte_pfn(address, phys >> PAGE_SHIFT, flags);
 149 }
 150
 151 pte_t *pte_alloc_one_kernel(struct mm_struct *mm, unsigned long address)
 152 {
 153         return (pte_t *)__get_free_page(GFP_KERNEL|__GFP_REPEAT|__GFP_ZERO);
 154 }
 155
 156 struct page *pte_alloc_one(struct mm_struct *mm, unsigned long address)
 157 {
 158         struct page *pte;
 159
 160 #ifdef CONFIG_HIGHPTE
 161         pte = alloc_pages(GFP_KERNEL|__GFP_HIGHMEM|__GFP_REPEAT|__GFP_ZERO, 0);
 162 #else
 163         pte = alloc_pages(GFP_KERNEL|__GFP_REPEAT|__GFP_ZERO, 0);
 164 #endif
 165         return pte;
 166 }
 167
 168 void pmd_ctor(void *pmd, kmem_cache_t *cache, unsigned long flags)
 169 {
 170         memset(pmd, 0, PTRS_PER_PMD*sizeof(pmd_t));
 171 }
 172
 173 /*
 174  * List of all pgd's needed for non-PAE so it can invalidate entries
 175  * in both cached and uncached pgd's; not needed for PAE since the
 176  * kernel pmd is shared. If PAE were not to share the pmd a similar
 177  * tactic would be needed. This is essentially codepath-based locking
 178  * against pageattr.c; it is the unique case in which a valid change
 179  * of kernel pagetables can't be lazily synchronized by vmalloc faults.
 180  * vmalloc faults work because attached pagetables are never freed.
 181  * The locking scheme was chosen on the basis of manfred's
 182  * recommendations and having no core impact whatsoever.
 183  * -- wli
 184  */
 185 DEFINE_SPINLOCK(pgd_lock);
 186 struct page *pgd_list;
 187
 188 static inline void pgd_list_add(pgd_t *pgd)
 189 {
 190         struct page *page = virt_to_page(pgd);
 191         page->index = (unsigned long)pgd_list;
 192         if (pgd_list)
 193                 set_page_private(pgd_list, (unsigned long)&page->index);
 194         pgd_list = page;
 195         set_page_private(page, (unsigned long)&pgd_list);
 196 }
 197
 198 static inline void pgd_list_del(pgd_t *pgd)
 199 {
 200         struct page *next, **pprev, *page = virt_to_page(pgd);
 201         next = (struct page *)page->index;
 202         pprev = (struct page **)page_private(page);
 203         *pprev = next;
 204         if (next)
 205                 set_page_private(next, (unsigned long)pprev);
 206 }
 207
 208 void pgd_ctor(void *pgd, kmem_cache_t *cache, unsigned long unused)
 209 {
 210         unsigned long flags;
 211
 212         if (PTRS_PER_PMD == 1) {
 213                 memset(pgd, 0, USER_PTRS_PER_PGD*sizeof(pgd_t));
 214                 spin_lock_irqsave(&pgd_lock, flags);
 215         }
 216
 217         clone_pgd_range((pgd_t *)pgd + USER_PTRS_PER_PGD,
 218                         swapper_pg_dir + USER_PTRS_PER_PGD,
 219                         KERNEL_PGD_PTRS);
 220         if (PTRS_PER_PMD > 1)
 221                 return;
 222
 223         pgd_list_add(pgd);
 224         spin_unlock_irqrestore(&pgd_lock, flags);
 225 }
 226
 227 /* never called when PTRS_PER_PMD > 1 */
 228 void pgd_dtor(void *pgd, kmem_cache_t *cache, unsigned long unused)
 229 {
 230         unsigned long flags; /* can be called from interrupt context */
 231
 232         spin_lock_irqsave(&pgd_lock, flags);
 233         pgd_list_del(pgd);
 234         spin_unlock_irqrestore(&pgd_lock, flags);
 235 }
 236
 237 pgd_t *pgd_alloc(struct mm_struct *mm)
 238 {
 239         int i;
 240         pgd_t *pgd = kmem_cache_alloc(pgd_cache, GFP_KERNEL);
 241
 242         if (PTRS_PER_PMD == 1 || !pgd)
 243                 return pgd;
 244
 245         for (i = 0; i < USER_PTRS_PER_PGD; ++i) {
 246                 pmd_t *pmd = kmem_cache_alloc(pmd_cache, GFP_KERNEL);
 247                 if (!pmd)
 248                         goto out_oom;
 249                 set_pgd(&pgd[i], __pgd(1 + __pa(pmd)));
 250         }
 251         return pgd;
 252
 253 out_oom:
 254         for (i--; i >= 0; i--)
 255                 kmem_cache_free(pmd_cache, (void *)__va(pgd_val(pgd[i])-1));
 256         kmem_cache_free(pgd_cache, pgd);
 257         return NULL;
 258 }
 259
 260 void pgd_free(pgd_t *pgd)
 261 {
 262         int i;
 263
 264         /* in the PAE case user pgd entries are overwritten before usage */
 265         if (PTRS_PER_PMD > 1)
 266                 for (i = 0; i < USER_PTRS_PER_PGD; ++i)
 267                         kmem_cache_free(pmd_cache, (void *)__va(pgd_val(pgd[i])-1));
 268         /* in the non-PAE case, free_pgtables() clears user pgd entries */
 269         kmem_cache_free(pgd_cache, pgd);
 270 }