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         if (pgprot_val(flags))
  99                 /* <pfn,flags> stored as-is, to permit clearing entries */
 100                 set_pte(pte, pfn_pte(pfn, flags));
 101         else
 102                 pte_clear(&init_mm, vaddr, pte);
 103
 104         /*
 105          * It's enough to flush this one mapping.
 106          * (PGE mappings get flushed as well)
 107          */
 108         __flush_tlb_one(vaddr);
 109 }
 110
 111 /*
 112  * Associate a large virtual page frame with a given physical page frame
 113  * and protection flags for that frame. pfn is for the base of the page,
 114  * vaddr is what the page gets mapped to - both must be properly aligned.
 115  * The pmd must already be instantiated. Assumes PAE mode.
 116  */
 117 void set_pmd_pfn(unsigned long vaddr, unsigned long pfn, pgprot_t flags)
 118 {
 119         pgd_t *pgd;
 120         pud_t *pud;
 121         pmd_t *pmd;
 122
 123         if (vaddr & (PMD_SIZE-1)) {             /* vaddr is misaligned */
 124                 printk(KERN_WARNING "set_pmd_pfn: vaddr misaligned\n");
 125                 return; /* BUG(); */
 126         }
 127         if (pfn & (PTRS_PER_PTE-1)) {           /* pfn is misaligned */
 128                 printk(KERN_WARNING "set_pmd_pfn: pfn misaligned\n");
 129                 return; /* BUG(); */
 130         }
 131         pgd = swapper_pg_dir + pgd_index(vaddr);
 132         if (pgd_none(*pgd)) {
 133                 printk(KERN_WARNING "set_pmd_pfn: pgd_none\n");
 134                 return; /* BUG(); */
 135         }
 136         pud = pud_offset(pgd, vaddr);
 137         pmd = pmd_offset(pud, vaddr);
 138         set_pmd(pmd, pfn_pmd(pfn, flags));
 139         /*
 140          * It's enough to flush this one mapping.
 141          * (PGE mappings get flushed as well)
 142          */
 143         __flush_tlb_one(vaddr);
 144 }
 145
 146 static int fixmaps;
 147 #ifndef CONFIG_COMPAT_VDSO
 148 unsigned long __FIXADDR_TOP = 0xfffff000;
 149 EXPORT_SYMBOL(__FIXADDR_TOP);
 150 #endif
 151
 152 void __set_fixmap (enum fixed_addresses idx, unsigned long phys, pgprot_t flags)
 153 {
 154         unsigned long address = __fix_to_virt(idx);
 155
 156         if (idx >= __end_of_fixed_addresses) {
 157                 BUG();
 158                 return;
 159         }
 160         set_pte_pfn(address, phys >> PAGE_SHIFT, flags);
 161         fixmaps++;
 162 }
 163
 164 /**
 165  * reserve_top_address - reserves a hole in the top of kernel address space
 166  * @reserve - size of hole to reserve
 167  *
 168  * Can be used to relocate the fixmap area and poke a hole in the top
 169  * of kernel address space to make room for a hypervisor.
 170  */
 171 void reserve_top_address(unsigned long reserve)
 172 {
 173         BUG_ON(fixmaps > 0);
 174 #ifdef CONFIG_COMPAT_VDSO
 175         BUG_ON(reserve != 0);
 176 #else
 177         __FIXADDR_TOP = -reserve - PAGE_SIZE;
 178         __VMALLOC_RESERVE += reserve;
 179 #endif
 180 }
 181
 182 pte_t *pte_alloc_one_kernel(struct mm_struct *mm, unsigned long address)
 183 {
 184         return (pte_t *)__get_free_page(GFP_KERNEL|__GFP_REPEAT|__GFP_ZERO);
 185 }
 186
 187 struct page *pte_alloc_one(struct mm_struct *mm, unsigned long address)
 188 {
 189         struct page *pte;
 190
 191 #ifdef CONFIG_HIGHPTE
 192         pte = alloc_pages(GFP_KERNEL|__GFP_HIGHMEM|__GFP_REPEAT|__GFP_ZERO, 0);
 193 #else
 194         pte = alloc_pages(GFP_KERNEL|__GFP_REPEAT|__GFP_ZERO, 0);
 195 #endif
 196         return pte;
 197 }
 198
 199 void pmd_ctor(void *pmd, kmem_cache_t *cache, unsigned long flags)
 200 {
 201         memset(pmd, 0, PTRS_PER_PMD*sizeof(pmd_t));
 202 }
 203
 204 /*
 205  * List of all pgd's needed for non-PAE so it can invalidate entries
 206  * in both cached and uncached pgd's; not needed for PAE since the
 207  * kernel pmd is shared. If PAE were not to share the pmd a similar
 208  * tactic would be needed. This is essentially codepath-based locking
 209  * against pageattr.c; it is the unique case in which a valid change
 210  * of kernel pagetables can't be lazily synchronized by vmalloc faults.
 211  * vmalloc faults work because attached pagetables are never freed.
 212  * The locking scheme was chosen on the basis of manfred's
 213  * recommendations and having no core impact whatsoever.
 214  * -- wli
 215  */
 216 DEFINE_SPINLOCK(pgd_lock);
 217 struct page *pgd_list;
 218
 219 static inline void pgd_list_add(pgd_t *pgd)
 220 {
 221         struct page *page = virt_to_page(pgd);
 222         page->index = (unsigned long)pgd_list;
 223         if (pgd_list)
 224                 set_page_private(pgd_list, (unsigned long)&page->index);
 225         pgd_list = page;
 226         set_page_private(page, (unsigned long)&pgd_list);
 227 }
 228
 229 static inline void pgd_list_del(pgd_t *pgd)
 230 {
 231         struct page *next, **pprev, *page = virt_to_page(pgd);
 232         next = (struct page *)page->index;
 233         pprev = (struct page **)page_private(page);
 234         *pprev = next;
 235         if (next)
 236                 set_page_private(next, (unsigned long)pprev);
 237 }
 238
 239 void pgd_ctor(void *pgd, kmem_cache_t *cache, unsigned long unused)
 240 {
 241         unsigned long flags;
 242
 243         if (PTRS_PER_PMD == 1) {
 244                 memset(pgd, 0, USER_PTRS_PER_PGD*sizeof(pgd_t));
 245                 spin_lock_irqsave(&pgd_lock, flags);
 246         }
 247
 248         clone_pgd_range((pgd_t *)pgd + USER_PTRS_PER_PGD,
 249                         swapper_pg_dir + USER_PTRS_PER_PGD,
 250                         KERNEL_PGD_PTRS);
 251         if (PTRS_PER_PMD > 1)
 252                 return;
 253
 254         pgd_list_add(pgd);
 255         spin_unlock_irqrestore(&pgd_lock, flags);
 256 }
 257
 258 /* never called when PTRS_PER_PMD > 1 */
 259 void pgd_dtor(void *pgd, kmem_cache_t *cache, unsigned long unused)
 260 {
 261         unsigned long flags; /* can be called from interrupt context */
 262
 263         spin_lock_irqsave(&pgd_lock, flags);
 264         pgd_list_del(pgd);
 265         spin_unlock_irqrestore(&pgd_lock, flags);
 266 }
 267
 268 pgd_t *pgd_alloc(struct mm_struct *mm)
 269 {
 270         int i;
 271         pgd_t *pgd = kmem_cache_alloc(pgd_cache, GFP_KERNEL);
 272
 273         if (PTRS_PER_PMD == 1 || !pgd)
 274                 return pgd;
 275
 276         for (i = 0; i < USER_PTRS_PER_PGD; ++i) {
 277                 pmd_t *pmd = kmem_cache_alloc(pmd_cache, GFP_KERNEL);
 278                 if (!pmd)
 279                         goto out_oom;
 280                 set_pgd(&pgd[i], __pgd(1 + __pa(pmd)));
 281         }
 282         return pgd;
 283
 284 out_oom:
 285         for (i--; i >= 0; i--)
 286                 kmem_cache_free(pmd_cache, (void *)__va(pgd_val(pgd[i])-1));
 287         kmem_cache_free(pgd_cache, pgd);
 288         return NULL;
 289 }
 290
 291 void pgd_free(pgd_t *pgd)
 292 {
 293         int i;
 294
 295         /* in the PAE case user pgd entries are overwritten before usage */
 296         if (PTRS_PER_PMD > 1)
 297                 for (i = 0; i < USER_PTRS_PER_PGD; ++i)
 298                         kmem_cache_free(pmd_cache, (void *)__va(pgd_val(pgd[i])-1));
 299         /* in the non-PAE case, free_pgtables() clears user pgd entries */
 300         kmem_cache_free(pgd_cache, pgd);
 301 }