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 #include <linux/module.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         unsigned long flags;
  34
  35         printk(KERN_INFO "Mem-info:\n");
  36         show_free_areas();
  37         printk(KERN_INFO "Free swap:       %6ldkB\n", nr_swap_pages<<(PAGE_SHIFT-10));
  38         for_each_online_pgdat(pgdat) {
  39                 pgdat_resize_lock(pgdat, &flags);
  40                 for (i = 0; i < pgdat->node_spanned_pages; ++i) {
  41                         page = pgdat_page_nr(pgdat, i);
  42                         total++;
  43                         if (PageHighMem(page))
  44                                 highmem++;
  45                         if (PageReserved(page))
  46                                 reserved++;
  47                         else if (PageSwapCache(page))
  48                                 cached++;
  49                         else if (page_count(page))
  50                                 shared += page_count(page) - 1;
  51                 }
  52                 pgdat_resize_unlock(pgdat, &flags);
  53         }
  54         printk(KERN_INFO "%d pages of RAM\n", total);
  55         printk(KERN_INFO "%d pages of HIGHMEM\n", highmem);
  56         printk(KERN_INFO "%d reserved pages\n", reserved);
  57         printk(KERN_INFO "%d pages shared\n", shared);
  58         printk(KERN_INFO "%d pages swap cached\n", cached);
  59
  60         printk(KERN_INFO "%lu pages dirty\n", global_page_state(NR_FILE_DIRTY));
  61         printk(KERN_INFO "%lu pages writeback\n",
  62                                         global_page_state(NR_WRITEBACK));
  63         printk(KERN_INFO "%lu pages mapped\n", global_page_state(NR_FILE_MAPPED));
  64         printk(KERN_INFO "%lu pages slab\n",
  65                 global_page_state(NR_SLAB_RECLAIMABLE) +
  66                 global_page_state(NR_SLAB_UNRECLAIMABLE));
  67         printk(KERN_INFO "%lu pages pagetables\n",
  68                                         global_page_state(NR_PAGETABLE));
  69 }
  70
  71 /*
  72  * Associate a virtual page frame with a given physical page frame
  73  * and protection flags for that frame.
  74  */
  75 static void set_pte_pfn(unsigned long vaddr, unsigned long pfn, pgprot_t flags)
  76 {
  77         pgd_t *pgd;
  78         pud_t *pud;
  79         pmd_t *pmd;
  80         pte_t *pte;
  81
  82         pgd = swapper_pg_dir + pgd_index(vaddr);
  83         if (pgd_none(*pgd)) {
  84                 BUG();
  85                 return;
  86         }
  87         pud = pud_offset(pgd, vaddr);
  88         if (pud_none(*pud)) {
  89                 BUG();
  90                 return;
  91         }
  92         pmd = pmd_offset(pud, vaddr);
  93         if (pmd_none(*pmd)) {
  94                 BUG();
  95                 return;
  96         }
  97         pte = pte_offset_kernel(pmd, vaddr);
  98         /* <pfn,flags> stored as-is, to permit clearing entries */
  99         set_pte(pte, pfn_pte(pfn, flags));
 100
 101         /*
 102          * It's enough to flush this one mapping.
 103          * (PGE mappings get flushed as well)
 104          */
 105         __flush_tlb_one(vaddr);
 106 }
 107
 108 /*
 109  * Associate a large virtual page frame with a given physical page frame
 110  * and protection flags for that frame. pfn is for the base of the page,
 111  * vaddr is what the page gets mapped to - both must be properly aligned.
 112  * The pmd must already be instantiated. Assumes PAE mode.
 113  */
 114 void set_pmd_pfn(unsigned long vaddr, unsigned long pfn, pgprot_t flags)
 115 {
 116         pgd_t *pgd;
 117         pud_t *pud;
 118         pmd_t *pmd;
 119
 120         if (vaddr & (PMD_SIZE-1)) {             /* vaddr is misaligned */
 121                 printk(KERN_WARNING "set_pmd_pfn: vaddr misaligned\n");
 122                 return; /* BUG(); */
 123         }
 124         if (pfn & (PTRS_PER_PTE-1)) {           /* pfn is misaligned */
 125                 printk(KERN_WARNING "set_pmd_pfn: pfn misaligned\n");
 126                 return; /* BUG(); */
 127         }
 128         pgd = swapper_pg_dir + pgd_index(vaddr);
 129         if (pgd_none(*pgd)) {
 130                 printk(KERN_WARNING "set_pmd_pfn: pgd_none\n");
 131                 return; /* BUG(); */
 132         }
 133         pud = pud_offset(pgd, vaddr);
 134         pmd = pmd_offset(pud, vaddr);
 135         set_pmd(pmd, pfn_pmd(pfn, flags));
 136         /*
 137          * It's enough to flush this one mapping.
 138          * (PGE mappings get flushed as well)
 139          */
 140         __flush_tlb_one(vaddr);
 141 }
 142
 143 static int fixmaps;
 144 #ifndef CONFIG_COMPAT_VDSO
 145 unsigned long __FIXADDR_TOP = 0xfffff000;
 146 EXPORT_SYMBOL(__FIXADDR_TOP);
 147 #endif
 148
 149 void __set_fixmap (enum fixed_addresses idx, unsigned long phys, pgprot_t flags)
 150 {
 151         unsigned long address = __fix_to_virt(idx);
 152
 153         if (idx >= __end_of_fixed_addresses) {
 154                 BUG();
 155                 return;
 156         }
 157         set_pte_pfn(address, phys >> PAGE_SHIFT, flags);
 158         fixmaps++;
 159 }
 160
 161 /**
 162  * reserve_top_address - reserves a hole in the top of kernel address space
 163  * @reserve - size of hole to reserve
 164  *
 165  * Can be used to relocate the fixmap area and poke a hole in the top
 166  * of kernel address space to make room for a hypervisor.
 167  */
 168 void reserve_top_address(unsigned long reserve)
 169 {
 170         BUG_ON(fixmaps > 0);
 171 #ifdef CONFIG_COMPAT_VDSO
 172         BUG_ON(reserve != 0);
 173 #else
 174         __FIXADDR_TOP = -reserve - PAGE_SIZE;
 175         __VMALLOC_RESERVE += reserve;
 176 #endif
 177 }
 178
 179 pte_t *pte_alloc_one_kernel(struct mm_struct *mm, unsigned long address)
 180 {
 181         return (pte_t *)__get_free_page(GFP_KERNEL|__GFP_REPEAT|__GFP_ZERO);
 182 }
 183
 184 struct page *pte_alloc_one(struct mm_struct *mm, unsigned long address)
 185 {
 186         struct page *pte;
 187
 188 #ifdef CONFIG_HIGHPTE
 189         pte = alloc_pages(GFP_KERNEL|__GFP_HIGHMEM|__GFP_REPEAT|__GFP_ZERO, 0);
 190 #else
 191         pte = alloc_pages(GFP_KERNEL|__GFP_REPEAT|__GFP_ZERO, 0);
 192 #endif
 193         return pte;
 194 }
 195
 196 void pmd_ctor(void *pmd, struct kmem_cache *cache, unsigned long flags)
 197 {
 198         memset(pmd, 0, PTRS_PER_PMD*sizeof(pmd_t));
 199 }
 200
 201 /*
 202  * List of all pgd's needed for non-PAE so it can invalidate entries
 203  * in both cached and uncached pgd's; not needed for PAE since the
 204  * kernel pmd is shared. If PAE were not to share the pmd a similar
 205  * tactic would be needed. This is essentially codepath-based locking
 206  * against pageattr.c; it is the unique case in which a valid change
 207  * of kernel pagetables can't be lazily synchronized by vmalloc faults.
 208  * vmalloc faults work because attached pagetables are never freed.
 209  * The locking scheme was chosen on the basis of manfred's
 210  * recommendations and having no core impact whatsoever.
 211  * -- wli
 212  */
 213 DEFINE_SPINLOCK(pgd_lock);
 214 struct page *pgd_list;
 215
 216 static inline void pgd_list_add(pgd_t *pgd)
 217 {
 218         struct page *page = virt_to_page(pgd);
 219         page->index = (unsigned long)pgd_list;
 220         if (pgd_list)
 221                 set_page_private(pgd_list, (unsigned long)&page->index);
 222         pgd_list = page;
 223         set_page_private(page, (unsigned long)&pgd_list);
 224 }
 225
 226 static inline void pgd_list_del(pgd_t *pgd)
 227 {
 228         struct page *next, **pprev, *page = virt_to_page(pgd);
 229         next = (struct page *)page->index;
 230         pprev = (struct page **)page_private(page);
 231         *pprev = next;
 232         if (next)
 233                 set_page_private(next, (unsigned long)pprev);
 234 }
 235
 236 void pgd_ctor(void *pgd, struct kmem_cache *cache, unsigned long unused)
 237 {
 238         unsigned long flags;
 239
 240         if (PTRS_PER_PMD == 1) {
 241                 memset(pgd, 0, USER_PTRS_PER_PGD*sizeof(pgd_t));
 242                 spin_lock_irqsave(&pgd_lock, flags);
 243         }
 244
 245         clone_pgd_range((pgd_t *)pgd + USER_PTRS_PER_PGD,
 246                         swapper_pg_dir + USER_PTRS_PER_PGD,
 247                         KERNEL_PGD_PTRS);
 248         if (PTRS_PER_PMD > 1)
 249                 return;
 250
 251         pgd_list_add(pgd);
 252         spin_unlock_irqrestore(&pgd_lock, flags);
 253 }
 254
 255 /* never called when PTRS_PER_PMD > 1 */
 256 void pgd_dtor(void *pgd, struct kmem_cache *cache, unsigned long unused)
 257 {
 258         unsigned long flags; /* can be called from interrupt context */
 259
 260         spin_lock_irqsave(&pgd_lock, flags);
 261         pgd_list_del(pgd);
 262         spin_unlock_irqrestore(&pgd_lock, flags);
 263 }
 264
 265 pgd_t *pgd_alloc(struct mm_struct *mm)
 266 {
 267         int i;
 268         pgd_t *pgd = kmem_cache_alloc(pgd_cache, GFP_KERNEL);
 269
 270         if (PTRS_PER_PMD == 1 || !pgd)
 271                 return pgd;
 272
 273         for (i = 0; i < USER_PTRS_PER_PGD; ++i) {
 274                 pmd_t *pmd = kmem_cache_alloc(pmd_cache, GFP_KERNEL);
 275                 if (!pmd)
 276                         goto out_oom;
 277                 set_pgd(&pgd[i], __pgd(1 + __pa(pmd)));
 278         }
 279         return pgd;
 280
 281 out_oom:
 282         for (i--; i >= 0; i--)
 283                 kmem_cache_free(pmd_cache, (void *)__va(pgd_val(pgd[i])-1));
 284         kmem_cache_free(pgd_cache, pgd);
 285         return NULL;
 286 }
 287
 288 void pgd_free(pgd_t *pgd)
 289 {
 290         int i;
 291
 292         /* in the PAE case user pgd entries are overwritten before usage */
 293         if (PTRS_PER_PMD > 1)
 294                 for (i = 0; i < USER_PTRS_PER_PGD; ++i)
 295                         kmem_cache_free(pmd_cache, (void *)__va(pgd_val(pgd[i])-1));
 296         /* in the non-PAE case, free_pgtables() clears user pgd entries */
 297         kmem_cache_free(pgd_cache, pgd);
 298 }