arch/x86/mm/pgtable_32.c

   1 #include <linux/sched.h>
   2 #include <linux/kernel.h>
   3 #include <linux/errno.h>
   4 #include <linux/mm.h>
   5 #include <linux/nmi.h>
   6 #include <linux/swap.h>
   7 #include <linux/smp.h>
   8 #include <linux/highmem.h>
   9 #include <linux/slab.h>
  10 #include <linux/pagemap.h>
  11 #include <linux/spinlock.h>
  12 #include <linux/module.h>
  13 #include <linux/quicklist.h>
  14
  15 #include <asm/system.h>
  16 #include <asm/pgtable.h>
  17 #include <asm/pgalloc.h>
  18 #include <asm/fixmap.h>
  19 #include <asm/e820.h>
  20 #include <asm/tlb.h>
  21 #include <asm/tlbflush.h>
  22
  23 void show_mem(void)
  24 {
  25         int total = 0, reserved = 0;
  26         int shared = 0, cached = 0;
  27         int highmem = 0;
  28         struct page *page;
  29         pg_data_t *pgdat;
  30         unsigned long i;
  31         unsigned long flags;
  32
  33         printk(KERN_INFO "Mem-info:\n");
  34         show_free_areas();
  35         for_each_online_pgdat(pgdat) {
  36                 pgdat_resize_lock(pgdat, &flags);
  37                 for (i = 0; i < pgdat->node_spanned_pages; ++i) {
  38                         if (unlikely(i % MAX_ORDER_NR_PAGES == 0))
  39                                 touch_nmi_watchdog();
  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         if (pgprot_val(flags))
  98                 set_pte_present(&init_mm, vaddr, pte, pfn_pte(pfn, flags));
  99         else
 100                 pte_clear(&init_mm, vaddr, pte);
 101
 102         /*
 103          * It's enough to flush this one mapping.
 104          * (PGE mappings get flushed as well)
 105          */
 106         __flush_tlb_one(vaddr);
 107 }
 108
 109 /*
 110  * Associate a large virtual page frame with a given physical page frame
 111  * and protection flags for that frame. pfn is for the base of the page,
 112  * vaddr is what the page gets mapped to - both must be properly aligned.
 113  * The pmd must already be instantiated. Assumes PAE mode.
 114  */
 115 void set_pmd_pfn(unsigned long vaddr, unsigned long pfn, pgprot_t flags)
 116 {
 117         pgd_t *pgd;
 118         pud_t *pud;
 119         pmd_t *pmd;
 120
 121         if (vaddr & (PMD_SIZE-1)) {             /* vaddr is misaligned */
 122                 printk(KERN_WARNING "set_pmd_pfn: vaddr misaligned\n");
 123                 return; /* BUG(); */
 124         }
 125         if (pfn & (PTRS_PER_PTE-1)) {           /* pfn is misaligned */
 126                 printk(KERN_WARNING "set_pmd_pfn: pfn misaligned\n");
 127                 return; /* BUG(); */
 128         }
 129         pgd = swapper_pg_dir + pgd_index(vaddr);
 130         if (pgd_none(*pgd)) {
 131                 printk(KERN_WARNING "set_pmd_pfn: pgd_none\n");
 132                 return; /* BUG(); */
 133         }
 134         pud = pud_offset(pgd, vaddr);
 135         pmd = pmd_offset(pud, vaddr);
 136         set_pmd(pmd, pfn_pmd(pfn, flags));
 137         /*
 138          * It's enough to flush this one mapping.
 139          * (PGE mappings get flushed as well)
 140          */
 141         __flush_tlb_one(vaddr);
 142 }
 143
 144 static int fixmaps;
 145 unsigned long __FIXADDR_TOP = 0xfffff000;
 146 EXPORT_SYMBOL(__FIXADDR_TOP);
 147
 148 void __set_fixmap (enum fixed_addresses idx, unsigned long phys, pgprot_t flags)
 149 {
 150         unsigned long address = __fix_to_virt(idx);
 151
 152         if (idx >= __end_of_fixed_addresses) {
 153                 BUG();
 154                 return;
 155         }
 156         set_pte_pfn(address, phys >> PAGE_SHIFT, flags);
 157         fixmaps++;
 158 }
 159
 160 /**
 161  * reserve_top_address - reserves a hole in the top of kernel address space
 162  * @reserve - size of hole to reserve
 163  *
 164  * Can be used to relocate the fixmap area and poke a hole in the top
 165  * of kernel address space to make room for a hypervisor.
 166  */
 167 void reserve_top_address(unsigned long reserve)
 168 {
 169         BUG_ON(fixmaps > 0);
 170         printk(KERN_INFO "Reserving virtual address space above 0x%08x\n",
 171                (int)-reserve);
 172         __FIXADDR_TOP = -reserve - PAGE_SIZE;
 173         __VMALLOC_RESERVE += reserve;
 174 }
 175
 176 pte_t *pte_alloc_one_kernel(struct mm_struct *mm, unsigned long address)
 177 {
 178         return (pte_t *)__get_free_page(GFP_KERNEL|__GFP_REPEAT|__GFP_ZERO);
 179 }
 180
 181 pgtable_t pte_alloc_one(struct mm_struct *mm, unsigned long address)
 182 {
 183         struct page *pte;
 184
 185 #ifdef CONFIG_HIGHPTE
 186         pte = alloc_pages(GFP_KERNEL|__GFP_HIGHMEM|__GFP_REPEAT|__GFP_ZERO, 0);
 187 #else
 188         pte = alloc_pages(GFP_KERNEL|__GFP_REPEAT|__GFP_ZERO, 0);
 189 #endif
 190         if (pte)
 191                 pgtable_page_ctor(pte);
 192         return pte;
 193 }
 194
 195 /*
 196  * List of all pgd's needed for non-PAE so it can invalidate entries
 197  * in both cached and uncached pgd's; not needed for PAE since the
 198  * kernel pmd is shared. If PAE were not to share the pmd a similar
 199  * tactic would be needed. This is essentially codepath-based locking
 200  * against pageattr.c; it is the unique case in which a valid change
 201  * of kernel pagetables can't be lazily synchronized by vmalloc faults.
 202  * vmalloc faults work because attached pagetables are never freed.
 203  * -- wli
 204  */
 205 static inline void pgd_list_add(pgd_t *pgd)
 206 {
 207         struct page *page = virt_to_page(pgd);
 208
 209         list_add(&page->lru, &pgd_list);
 210 }
 211
 212 static inline void pgd_list_del(pgd_t *pgd)
 213 {
 214         struct page *page = virt_to_page(pgd);
 215
 216         list_del(&page->lru);
 217 }
 218
 219 #define UNSHARED_PTRS_PER_PGD                           \
 220         (SHARED_KERNEL_PMD ? USER_PTRS_PER_PGD : PTRS_PER_PGD)
 221
 222 static void pgd_ctor(void *p)
 223 {
 224         pgd_t *pgd = p;
 225         unsigned long flags;
 226
 227         /* Clear usermode parts of PGD */
 228         memset(pgd, 0, USER_PTRS_PER_PGD*sizeof(pgd_t));
 229
 230         spin_lock_irqsave(&pgd_lock, flags);
 231
 232         /* If the pgd points to a shared pagetable level (either the
 233            ptes in non-PAE, or shared PMD in PAE), then just copy the
 234            references from swapper_pg_dir. */
 235         if (PAGETABLE_LEVELS == 2 ||
 236             (PAGETABLE_LEVELS == 3 && SHARED_KERNEL_PMD)) {
 237                 clone_pgd_range(pgd + USER_PTRS_PER_PGD,
 238                                 swapper_pg_dir + USER_PTRS_PER_PGD,
 239                                 KERNEL_PGD_PTRS);
 240                 paravirt_alloc_pd_clone(__pa(pgd) >> PAGE_SHIFT,
 241                                         __pa(swapper_pg_dir) >> PAGE_SHIFT,
 242                                         USER_PTRS_PER_PGD,
 243                                         KERNEL_PGD_PTRS);
 244         }
 245
 246         /* list required to sync kernel mapping updates */
 247         if (!SHARED_KERNEL_PMD)
 248                 pgd_list_add(pgd);
 249
 250         spin_unlock_irqrestore(&pgd_lock, flags);
 251 }
 252
 253 static void pgd_dtor(void *pgd)
 254 {
 255         unsigned long flags; /* can be called from interrupt context */
 256
 257         if (SHARED_KERNEL_PMD)
 258                 return;
 259
 260         spin_lock_irqsave(&pgd_lock, flags);
 261         pgd_list_del(pgd);
 262         spin_unlock_irqrestore(&pgd_lock, flags);
 263 }
 264
 265 #ifdef CONFIG_X86_PAE
 266 /*
 267  * Mop up any pmd pages which may still be attached to the pgd.
 268  * Normally they will be freed by munmap/exit_mmap, but any pmd we
 269  * preallocate which never got a corresponding vma will need to be
 270  * freed manually.
 271  */
 272 static void pgd_mop_up_pmds(struct mm_struct *mm, pgd_t *pgdp)
 273 {
 274         int i;
 275
 276         for(i = 0; i < UNSHARED_PTRS_PER_PGD; i++) {
 277                 pgd_t pgd = pgdp[i];
 278
 279                 if (pgd_val(pgd) != 0) {
 280                         pmd_t *pmd = (pmd_t *)pgd_page_vaddr(pgd);
 281
 282                         pgdp[i] = native_make_pgd(0);
 283
 284                         paravirt_release_pd(pgd_val(pgd) >> PAGE_SHIFT);
 285                         pmd_free(mm, pmd);
 286                 }
 287         }
 288 }
 289
 290 /*
 291  * In PAE mode, we need to do a cr3 reload (=tlb flush) when
 292  * updating the top-level pagetable entries to guarantee the
 293  * processor notices the update.  Since this is expensive, and
 294  * all 4 top-level entries are used almost immediately in a
 295  * new process's life, we just pre-populate them here.
 296  *
 297  * Also, if we're in a paravirt environment where the kernel pmd is
 298  * not shared between pagetables (!SHARED_KERNEL_PMDS), we allocate
 299  * and initialize the kernel pmds here.
 300  */
 301 static int pgd_prepopulate_pmd(struct mm_struct *mm, pgd_t *pgd)
 302 {
 303         pud_t *pud;
 304         unsigned long addr;
 305         int i;
 306
 307         pud = pud_offset(pgd, 0);
 308         for (addr = i = 0; i < UNSHARED_PTRS_PER_PGD;
 309              i++, pud++, addr += PUD_SIZE) {
 310                 pmd_t *pmd = pmd_alloc_one(mm, addr);
 311
 312                 if (!pmd) {
 313                         pgd_mop_up_pmds(mm, pgd);
 314                         return 0;
 315                 }
 316
 317                 if (i >= USER_PTRS_PER_PGD)
 318                         memcpy(pmd, (pmd_t *)pgd_page_vaddr(swapper_pg_dir[i]),
 319                                sizeof(pmd_t) * PTRS_PER_PMD);
 320
 321                 pud_populate(mm, pud, pmd);
 322         }
 323
 324         return 1;
 325 }
 326 #else  /* !CONFIG_X86_PAE */
 327 /* No need to prepopulate any pagetable entries in non-PAE modes. */
 328 static int pgd_prepopulate_pmd(struct mm_struct *mm, pgd_t *pgd)
 329 {
 330         return 1;
 331 }
 332
 333 static void pgd_mop_up_pmds(struct mm_struct *mm, pgd_t *pgdp)
 334 {
 335 }
 336 #endif  /* CONFIG_X86_PAE */
 337
 338 pgd_t *pgd_alloc(struct mm_struct *mm)
 339 {
 340         pgd_t *pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
 341
 342         /* so that alloc_pd can use it */
 343         mm->pgd = pgd;
 344         if (pgd)
 345                 pgd_ctor(pgd);
 346
 347         if (pgd && !pgd_prepopulate_pmd(mm, pgd)) {
 348                 pgd_dtor(pgd);
 349                 free_page((unsigned long)pgd);
 350                 pgd = NULL;
 351         }
 352
 353         return pgd;
 354 }
 355
 356 void pgd_free(struct mm_struct *mm, pgd_t *pgd)
 357 {
 358         pgd_mop_up_pmds(mm, pgd);
 359         pgd_dtor(pgd);
 360         free_page((unsigned long)pgd);
 361 }
 362
 363 void __pte_free_tlb(struct mmu_gather *tlb, struct page *pte)
 364 {
 365         pgtable_page_dtor(pte);
 366         paravirt_release_pt(page_to_pfn(pte));
 367         tlb_remove_page(tlb, pte);
 368 }
 369
 370 #ifdef CONFIG_X86_PAE
 371
 372 void __pmd_free_tlb(struct mmu_gather *tlb, pmd_t *pmd)
 373 {
 374         paravirt_release_pd(__pa(pmd) >> PAGE_SHIFT);
 375         tlb_remove_page(tlb, virt_to_page(pmd));
 376 }
 377
 378 #endif
 379
 380 int pmd_bad(pmd_t pmd)
 381 {
 382         WARN_ON_ONCE(pmd_bad_v1(pmd) != pmd_bad_v2(pmd));
 383
 384         return pmd_bad_v1(pmd);
 385 }