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",
  64                 global_page_state(NR_SLAB_RECLAIMABLE) +
  65                 global_page_state(NR_SLAB_UNRECLAIMABLE));
  66         printk(KERN_INFO "%lu pages pagetables\n",
  67                                         global_page_state(NR_PAGETABLE));
  68 }
  69
  70 /*
  71  * Associate a virtual page frame with a given physical page frame
  72  * and protection flags for that frame.
  73  */
  74 static void set_pte_pfn(unsigned long vaddr, unsigned long pfn, pgprot_t flags)
  75 {
  76         pgd_t *pgd;
  77         pud_t *pud;
  78         pmd_t *pmd;
  79         pte_t *pte;
  80
  81         pgd = swapper_pg_dir + pgd_index(vaddr);
  82         if (pgd_none(*pgd)) {
  83                 BUG();
  84                 return;
  85         }
  86         pud = pud_offset(pgd, vaddr);
  87         if (pud_none(*pud)) {
  88                 BUG();
  89                 return;
  90         }
  91         pmd = pmd_offset(pud, vaddr);
  92         if (pmd_none(*pmd)) {
  93                 BUG();
  94                 return;
  95         }
  96         pte = pte_offset_kernel(pmd, vaddr);
  97         /* <pfn,flags> stored as-is, to permit clearing entries */
  98         set_pte(pte, pfn_pte(pfn, flags));
  99
 100         /*
 101          * It's enough to flush this one mapping.
 102          * (PGE mappings get flushed as well)
 103          */
 104         __flush_tlb_one(vaddr);
 105 }
 106
 107 /*
 108  * Associate a large virtual page frame with a given physical page frame
 109  * and protection flags for that frame. pfn is for the base of the page,
 110  * vaddr is what the page gets mapped to - both must be properly aligned.
 111  * The pmd must already be instantiated. Assumes PAE mode.
 112  */
 113 void set_pmd_pfn(unsigned long vaddr, unsigned long pfn, pgprot_t flags)
 114 {
 115         pgd_t *pgd;
 116         pud_t *pud;
 117         pmd_t *pmd;
 118
 119         if (vaddr & (PMD_SIZE-1)) {             /* vaddr is misaligned */
 120                 printk(KERN_WARNING "set_pmd_pfn: vaddr misaligned\n");
 121                 return; /* BUG(); */
 122         }
 123         if (pfn & (PTRS_PER_PTE-1)) {           /* pfn is misaligned */
 124                 printk(KERN_WARNING "set_pmd_pfn: pfn misaligned\n");
 125                 return; /* BUG(); */
 126         }
 127         pgd = swapper_pg_dir + pgd_index(vaddr);
 128         if (pgd_none(*pgd)) {
 129                 printk(KERN_WARNING "set_pmd_pfn: pgd_none\n");
 130                 return; /* BUG(); */
 131         }
 132         pud = pud_offset(pgd, vaddr);
 133         pmd = pmd_offset(pud, vaddr);
 134         set_pmd(pmd, pfn_pmd(pfn, flags));
 135         /*
 136          * It's enough to flush this one mapping.
 137          * (PGE mappings get flushed as well)
 138          */
 139         __flush_tlb_one(vaddr);
 140 }
 141
 142 void __set_fixmap (enum fixed_addresses idx, unsigned long phys, pgprot_t flags)
 143 {
 144         unsigned long address = __fix_to_virt(idx);
 145
 146         if (idx >= __end_of_fixed_addresses) {
 147                 BUG();
 148                 return;
 149         }
 150         set_pte_pfn(address, phys >> PAGE_SHIFT, flags);
 151 }
 152
 153 pte_t *pte_alloc_one_kernel(struct mm_struct *mm, unsigned long address)
 154 {
 155         return (pte_t *)__get_free_page(GFP_KERNEL|__GFP_REPEAT|__GFP_ZERO);
 156 }
 157
 158 struct page *pte_alloc_one(struct mm_struct *mm, unsigned long address)
 159 {
 160         struct page *pte;
 161
 162 #ifdef CONFIG_HIGHPTE
 163         pte = alloc_pages(GFP_KERNEL|__GFP_HIGHMEM|__GFP_REPEAT|__GFP_ZERO, 0);
 164 #else
 165         pte = alloc_pages(GFP_KERNEL|__GFP_REPEAT|__GFP_ZERO, 0);
 166 #endif
 167         return pte;
 168 }
 169
 170 void pmd_ctor(void *pmd, kmem_cache_t *cache, unsigned long flags)
 171 {
 172         memset(pmd, 0, PTRS_PER_PMD*sizeof(pmd_t));
 173 }
 174
 175 /*
 176  * List of all pgd's needed for non-PAE so it can invalidate entries
 177  * in both cached and uncached pgd's; not needed for PAE since the
 178  * kernel pmd is shared. If PAE were not to share the pmd a similar
 179  * tactic would be needed. This is essentially codepath-based locking
 180  * against pageattr.c; it is the unique case in which a valid change
 181  * of kernel pagetables can't be lazily synchronized by vmalloc faults.
 182  * vmalloc faults work because attached pagetables are never freed.
 183  * The locking scheme was chosen on the basis of manfred's
 184  * recommendations and having no core impact whatsoever.
 185  * -- wli
 186  */
 187 DEFINE_SPINLOCK(pgd_lock);
 188 struct page *pgd_list;
 189
 190 static inline void pgd_list_add(pgd_t *pgd)
 191 {
 192         struct page *page = virt_to_page(pgd);
 193         page->index = (unsigned long)pgd_list;
 194         if (pgd_list)
 195                 set_page_private(pgd_list, (unsigned long)&page->index);
 196         pgd_list = page;
 197         set_page_private(page, (unsigned long)&pgd_list);
 198 }
 199
 200 static inline void pgd_list_del(pgd_t *pgd)
 201 {
 202         struct page *next, **pprev, *page = virt_to_page(pgd);
 203         next = (struct page *)page->index;
 204         pprev = (struct page **)page_private(page);
 205         *pprev = next;
 206         if (next)
 207                 set_page_private(next, (unsigned long)pprev);
 208 }
 209
 210 void pgd_ctor(void *pgd, kmem_cache_t *cache, unsigned long unused)
 211 {
 212         unsigned long flags;
 213
 214         if (PTRS_PER_PMD == 1) {
 215                 memset(pgd, 0, USER_PTRS_PER_PGD*sizeof(pgd_t));
 216                 spin_lock_irqsave(&pgd_lock, flags);
 217         }
 218
 219         clone_pgd_range((pgd_t *)pgd + USER_PTRS_PER_PGD,
 220                         swapper_pg_dir + USER_PTRS_PER_PGD,
 221                         KERNEL_PGD_PTRS);
 222         if (PTRS_PER_PMD > 1)
 223                 return;
 224
 225         pgd_list_add(pgd);
 226         spin_unlock_irqrestore(&pgd_lock, flags);
 227 }
 228
 229 /* never called when PTRS_PER_PMD > 1 */
 230 void pgd_dtor(void *pgd, kmem_cache_t *cache, unsigned long unused)
 231 {
 232         unsigned long flags; /* can be called from interrupt context */
 233
 234         spin_lock_irqsave(&pgd_lock, flags);
 235         pgd_list_del(pgd);
 236         spin_unlock_irqrestore(&pgd_lock, flags);
 237 }
 238
 239 pgd_t *pgd_alloc(struct mm_struct *mm)
 240 {
 241         int i;
 242         pgd_t *pgd = kmem_cache_alloc(pgd_cache, GFP_KERNEL);
 243
 244         if (PTRS_PER_PMD == 1 || !pgd)
 245                 return pgd;
 246
 247         for (i = 0; i < USER_PTRS_PER_PGD; ++i) {
 248                 pmd_t *pmd = kmem_cache_alloc(pmd_cache, GFP_KERNEL);
 249                 if (!pmd)
 250                         goto out_oom;
 251                 set_pgd(&pgd[i], __pgd(1 + __pa(pmd)));
 252         }
 253         return pgd;
 254
 255 out_oom:
 256         for (i--; i >= 0; i--)
 257                 kmem_cache_free(pmd_cache, (void *)__va(pgd_val(pgd[i])-1));
 258         kmem_cache_free(pgd_cache, pgd);
 259         return NULL;
 260 }
 261
 262 void pgd_free(pgd_t *pgd)
 263 {
 264         int i;
 265
 266         /* in the PAE case user pgd entries are overwritten before usage */
 267         if (PTRS_PER_PMD > 1)
 268                 for (i = 0; i < USER_PTRS_PER_PGD; ++i)
 269                         kmem_cache_free(pmd_cache, (void *)__va(pgd_val(pgd[i])-1));
 270         /* in the non-PAE case, free_pgtables() clears user pgd entries */
 271         kmem_cache_free(pgd_cache, pgd);
 272 }