arch/i386/mm/pgtable.c

   1 /*
   2  *  linux/arch/i386/mm/pgtable.c
   3  */
   4
   5 #include <linux/config.h>
   6 #include <linux/sched.h>
   7 #include <linux/kernel.h>
   8 #include <linux/errno.h>
   9 #include <linux/mm.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
  17 #include <asm/system.h>
  18 #include <asm/pgtable.h>
  19 #include <asm/pgalloc.h>
  20 #include <asm/fixmap.h>
  21 #include <asm/e820.h>
  22 #include <asm/tlb.h>
  23 #include <asm/tlbflush.h>
  24
  25 void show_mem(void)
  26 {
  27         int total = 0, reserved = 0;
  28         int shared = 0, cached = 0;
  29         int highmem = 0;
  30         struct page *page;
  31         pg_data_t *pgdat;
  32         unsigned long i;
  33
  34         printk("Mem-info:\n");
  35         show_free_areas();
  36         printk("Free swap:       %6ldkB\n", nr_swap_pages<<(PAGE_SHIFT-10));
  37         for_each_pgdat(pgdat) {
  38                 for (i = 0; i < pgdat->node_spanned_pages; ++i) {
  39                         page = pgdat->node_mem_map + i;
  40                         total++;
  41                         if (PageHighMem(page))
  42                                 highmem++;
  43                         if (PageReserved(page))
  44                                 reserved++;
  45                         else if (PageSwapCache(page))
  46                                 cached++;
  47                         else if (page_count(page))
  48                                 shared += page_count(page) - 1;
  49                 }
  50         }
  51         printk("%d pages of RAM\n", total);
  52         printk("%d pages of HIGHMEM\n",highmem);
  53         printk("%d reserved pages\n",reserved);
  54         printk("%d pages shared\n",shared);
  55         printk("%d pages swap cached\n",cached);
  56 }
  57
  58 /*
  59  * Associate a virtual page frame with a given physical page frame
  60  * and protection flags for that frame.
  61  */
  62 static void set_pte_pfn(unsigned long vaddr, unsigned long pfn, pgprot_t flags)
  63 {
  64         pgd_t *pgd;
  65         pmd_t *pmd;
  66         pte_t *pte;
  67
  68         pgd = swapper_pg_dir + pgd_index(vaddr);
  69         if (pgd_none(*pgd)) {
  70                 BUG();
  71                 return;
  72         }
  73         pmd = pmd_offset(pgd, vaddr);
  74         if (pmd_none(*pmd)) {
  75                 BUG();
  76                 return;
  77         }
  78         pte = pte_offset_kernel(pmd, vaddr);
  79         /* <pfn,flags> stored as-is, to permit clearing entries */
  80         set_pte(pte, pfn_pte(pfn, flags));
  81
  82         /*
  83          * It's enough to flush this one mapping.
  84          * (PGE mappings get flushed as well)
  85          */
  86         __flush_tlb_one(vaddr);
  87 }
  88
  89 /*
  90  * Associate a large virtual page frame with a given physical page frame
  91  * and protection flags for that frame. pfn is for the base of the page,
  92  * vaddr is what the page gets mapped to - both must be properly aligned.
  93  * The pmd must already be instantiated. Assumes PAE mode.
  94  */
  95 void set_pmd_pfn(unsigned long vaddr, unsigned long pfn, pgprot_t flags)
  96 {
  97         pgd_t *pgd;
  98         pmd_t *pmd;
  99
 100         if (vaddr & (PMD_SIZE-1)) {             /* vaddr is misaligned */
 101                 printk ("set_pmd_pfn: vaddr misaligned\n");
 102                 return; /* BUG(); */
 103         }
 104         if (pfn & (PTRS_PER_PTE-1)) {           /* pfn is misaligned */
 105                 printk ("set_pmd_pfn: pfn misaligned\n");
 106                 return; /* BUG(); */
 107         }
 108         pgd = swapper_pg_dir + pgd_index(vaddr);
 109         if (pgd_none(*pgd)) {
 110                 printk ("set_pmd_pfn: pgd_none\n");
 111                 return; /* BUG(); */
 112         }
 113         pmd = pmd_offset(pgd, vaddr);
 114         set_pmd(pmd, pfn_pmd(pfn, flags));
 115         /*
 116          * It's enough to flush this one mapping.
 117          * (PGE mappings get flushed as well)
 118          */
 119         __flush_tlb_one(vaddr);
 120 }
 121
 122 void __set_fixmap (enum fixed_addresses idx, unsigned long phys, pgprot_t flags)
 123 {
 124         unsigned long address = __fix_to_virt(idx);
 125
 126         if (idx >= __end_of_fixed_addresses) {
 127                 BUG();
 128                 return;
 129         }
 130         set_pte_pfn(address, phys >> PAGE_SHIFT, flags);
 131 }
 132
 133 pte_t *pte_alloc_one_kernel(struct mm_struct *mm, unsigned long address)
 134 {
 135         pte_t *pte = (pte_t *)__get_free_page(GFP_KERNEL|__GFP_REPEAT);
 136         if (pte)
 137                 clear_page(pte);
 138         return pte;
 139 }
 140
 141 struct page *pte_alloc_one(struct mm_struct *mm, unsigned long address)
 142 {
 143         struct page *pte;
 144
 145 #ifdef CONFIG_HIGHPTE
 146         pte = alloc_pages(GFP_KERNEL|__GFP_HIGHMEM|__GFP_REPEAT, 0);
 147 #else
 148         pte = alloc_pages(GFP_KERNEL|__GFP_REPEAT, 0);
 149 #endif
 150         if (pte)
 151                 clear_highpage(pte);
 152         return pte;
 153 }
 154
 155 void pmd_ctor(void *pmd, kmem_cache_t *cache, unsigned long flags)
 156 {
 157         memset(pmd, 0, PTRS_PER_PMD*sizeof(pmd_t));
 158 }
 159
 160 /*
 161  * List of all pgd's needed for non-PAE so it can invalidate entries
 162  * in both cached and uncached pgd's; not needed for PAE since the
 163  * kernel pmd is shared. If PAE were not to share the pmd a similar
 164  * tactic would be needed. This is essentially codepath-based locking
 165  * against pageattr.c; it is the unique case in which a valid change
 166  * of kernel pagetables can't be lazily synchronized by vmalloc faults.
 167  * vmalloc faults work because attached pagetables are never freed.
 168  * If the locking proves to be non-performant, a ticketing scheme with
 169  * checks at dup_mmap(), exec(), and other mmlist addition points
 170  * could be used. The locking scheme was chosen on the basis of
 171  * manfred's recommendations and having no core impact whatsoever.
 172  * -- wli
 173  */
 174 spinlock_t pgd_lock = SPIN_LOCK_UNLOCKED;
 175 struct page *pgd_list;
 176
 177 static inline void pgd_list_add(pgd_t *pgd)
 178 {
 179         struct page *page = virt_to_page(pgd);
 180         page->index = (unsigned long)pgd_list;
 181         if (pgd_list)
 182                 pgd_list->private = (unsigned long)&page->index;
 183         pgd_list = page;
 184         page->private = (unsigned long)&pgd_list;
 185 }
 186
 187 static inline void pgd_list_del(pgd_t *pgd)
 188 {
 189         struct page *next, **pprev, *page = virt_to_page(pgd);
 190         next = (struct page *)page->index;
 191         pprev = (struct page **)page->private;
 192         *pprev = next;
 193         if (next)
 194                 next->private = (unsigned long)pprev;
 195 }
 196
 197 void pgd_ctor(void *pgd, kmem_cache_t *cache, unsigned long unused)
 198 {
 199         unsigned long flags;
 200
 201         if (PTRS_PER_PMD == 1)
 202                 spin_lock_irqsave(&pgd_lock, flags);
 203
 204         memcpy((pgd_t *)pgd + USER_PTRS_PER_PGD,
 205                         swapper_pg_dir + USER_PTRS_PER_PGD,
 206                         (PTRS_PER_PGD - USER_PTRS_PER_PGD) * sizeof(pgd_t));
 207
 208         if (PTRS_PER_PMD > 1)
 209                 return;
 210
 211         pgd_list_add(pgd);
 212         spin_unlock_irqrestore(&pgd_lock, flags);
 213         memset(pgd, 0, USER_PTRS_PER_PGD*sizeof(pgd_t));
 214 }
 215
 216 /* never called when PTRS_PER_PMD > 1 */
 217 void pgd_dtor(void *pgd, kmem_cache_t *cache, unsigned long unused)
 218 {
 219         unsigned long flags; /* can be called from interrupt context */
 220
 221         spin_lock_irqsave(&pgd_lock, flags);
 222         pgd_list_del(pgd);
 223         spin_unlock_irqrestore(&pgd_lock, flags);
 224 }
 225
 226 pgd_t *pgd_alloc(struct mm_struct *mm)
 227 {
 228         int i;
 229         pgd_t *pgd = kmem_cache_alloc(pgd_cache, GFP_KERNEL);
 230
 231         if (PTRS_PER_PMD == 1 || !pgd)
 232                 return pgd;
 233
 234         for (i = 0; i < USER_PTRS_PER_PGD; ++i) {
 235                 pmd_t *pmd = kmem_cache_alloc(pmd_cache, GFP_KERNEL);
 236                 if (!pmd)
 237                         goto out_oom;
 238                 set_pgd(&pgd[i], __pgd(1 + __pa((u64)((u32)pmd))));
 239         }
 240         return pgd;
 241
 242 out_oom:
 243         for (i--; i >= 0; i--)
 244                 kmem_cache_free(pmd_cache, (void *)__va(pgd_val(pgd[i])-1));
 245         kmem_cache_free(pgd_cache, pgd);
 246         return NULL;
 247 }
 248
 249 void pgd_free(pgd_t *pgd)
 250 {
 251         int i;
 252
 253         /* in the PAE case user pgd entries are overwritten before usage */
 254         if (PTRS_PER_PMD > 1)
 255                 for (i = 0; i < USER_PTRS_PER_PGD; ++i)
 256                         kmem_cache_free(pmd_cache, (void *)__va(pgd_val(pgd[i])-1));
 257         /* in the non-PAE case, clear_page_tables() clears user pgd entries */
 258         kmem_cache_free(pgd_cache, pgd);
 259 }