arch/x86_64/mm/pageattr.c

   1 /*
   2  * Copyright 2002 Andi Kleen, SuSE Labs.
   3  * Thanks to Ben LaHaise for precious feedback.
   4  */
   5
   6 #include <linux/config.h>
   7 #include <linux/mm.h>
   8 #include <linux/sched.h>
   9 #include <linux/highmem.h>
  10 #include <linux/module.h>
  11 #include <linux/slab.h>
  12 #include <asm/uaccess.h>
  13 #include <asm/processor.h>
  14 #include <asm/tlbflush.h>
  15 #include <asm/io.h>
  16
  17 static inline pte_t *lookup_address(unsigned long address)
  18 {
  19         pgd_t *pgd = pgd_offset_k(address);
  20         pud_t *pud;
  21         pmd_t *pmd;
  22         pte_t *pte;
  23         if (pgd_none(*pgd))
  24                 return NULL;
  25         pud = pud_offset(pgd, address);
  26         if (!pud_present(*pud))
  27                 return NULL;
  28         pmd = pmd_offset(pud, address);
  29         if (!pmd_present(*pmd))
  30                 return NULL;
  31         if (pmd_large(*pmd))
  32                 return (pte_t *)pmd;
  33         pte = pte_offset_kernel(pmd, address);
  34         if (pte && !pte_present(*pte))
  35                 pte = NULL;
  36         return pte;
  37 }
  38
  39 static struct page *split_large_page(unsigned long address, pgprot_t prot,
  40                                      pgprot_t ref_prot)
  41 {
  42         int i;
  43         unsigned long addr;
  44         struct page *base = alloc_pages(GFP_KERNEL, 0);
  45         pte_t *pbase;
  46         if (!base)
  47                 return NULL;
  48         address = __pa(address);
  49         addr = address & LARGE_PAGE_MASK;
  50         pbase = (pte_t *)page_address(base);
  51         for (i = 0; i < PTRS_PER_PTE; i++, addr += PAGE_SIZE) {
  52                 pbase[i] = pfn_pte(addr >> PAGE_SHIFT,
  53                                    addr == address ? prot : ref_prot);
  54         }
  55         return base;
  56 }
  57
  58
  59 static void flush_kernel_map(void *address)
  60 {
  61         if (0 && address && cpu_has_clflush) {
  62                 /* is this worth it? */
  63                 int i;
  64                 for (i = 0; i < PAGE_SIZE; i += boot_cpu_data.x86_clflush_size)
  65                         asm volatile("clflush (%0)" :: "r" (address + i));
  66         } else
  67                 asm volatile("wbinvd":::"memory");
  68         if (address)
  69                 __flush_tlb_one(address);
  70         else
  71                 __flush_tlb_all();
  72 }
  73
  74
  75 static inline void flush_map(unsigned long address)
  76 {
  77         on_each_cpu(flush_kernel_map, (void *)address, 1, 1);
  78 }
  79
  80 struct deferred_page {
  81         struct deferred_page *next;
  82         struct page *fpage;
  83         unsigned long address;
  84 };
  85 static struct deferred_page *df_list; /* protected by init_mm.mmap_sem */
  86
  87 static inline void save_page(unsigned long address, struct page *fpage)
  88 {
  89         struct deferred_page *df;
  90         df = kmalloc(sizeof(struct deferred_page), GFP_KERNEL);
  91         if (!df) {
  92                 flush_map(address);
  93                 __free_page(fpage);
  94         } else {
  95                 df->next = df_list;
  96                 df->fpage = fpage;
  97                 df->address = address;
  98                 df_list = df;
  99         }
 100 }
 101
 102 /*
 103  * No more special protections in this 2/4MB area - revert to a
 104  * large page again.
 105  */
 106 static void revert_page(unsigned long address, pgprot_t ref_prot)
 107 {
 108         pgd_t *pgd;
 109         pud_t *pud;
 110         pmd_t *pmd;
 111         pte_t large_pte;
 112
 113         pgd = pgd_offset_k(address);
 114         BUG_ON(pgd_none(*pgd));
 115         pud = pud_offset(pgd,address);
 116         BUG_ON(pud_none(*pud));
 117         pmd = pmd_offset(pud, address);
 118         BUG_ON(pmd_val(*pmd) & _PAGE_PSE);
 119         pgprot_val(ref_prot) |= _PAGE_PSE;
 120         large_pte = mk_pte_phys(__pa(address) & LARGE_PAGE_MASK, ref_prot);
 121         set_pte((pte_t *)pmd, large_pte);
 122 }
 123
 124 static int
 125 __change_page_attr(unsigned long address, unsigned long pfn, pgprot_t prot,
 126                                    pgprot_t ref_prot)
 127 {
 128         pte_t *kpte;
 129         struct page *kpte_page;
 130         unsigned kpte_flags;
 131         pgprot_t ref_prot2;
 132         kpte = lookup_address(address);
 133         if (!kpte) return 0;
 134         kpte_page = virt_to_page(((unsigned long)kpte) & PAGE_MASK);
 135         kpte_flags = pte_val(*kpte);
 136         if (pgprot_val(prot) != pgprot_val(ref_prot)) {
 137                 if ((kpte_flags & _PAGE_PSE) == 0) {
 138                         set_pte(kpte, pfn_pte(pfn, prot));
 139                 } else {
 140                         /*
 141                          * split_large_page will take the reference for this change_page_attr
 142                          * on the split page.
 143                          */
 144
 145                         struct page *split;
 146                         ref_prot2 = __pgprot(pgprot_val(pte_pgprot(*lookup_address(address))) & ~(1<<_PAGE_BIT_PSE));
 147
 148                         split = split_large_page(address, prot, ref_prot2);
 149                         if (!split)
 150                                 return -ENOMEM;
 151                         set_pte(kpte,mk_pte(split, ref_prot2));
 152                         kpte_page = split;
 153                 }
 154                 get_page(kpte_page);
 155         } else if ((kpte_flags & _PAGE_PSE) == 0) {
 156                 set_pte(kpte, pfn_pte(pfn, ref_prot));
 157                 __put_page(kpte_page);
 158         } else
 159                 BUG();
 160
 161         /* on x86-64 the direct mapping set at boot is not using 4k pages */
 162         BUG_ON(PageReserved(kpte_page));
 163
 164         switch (page_count(kpte_page)) {
 165         case 1:
 166                 save_page(address, kpte_page);
 167                 revert_page(address, ref_prot);
 168                 break;
 169         case 0:
 170                 BUG(); /* memleak and failed 2M page regeneration */
 171         }
 172         return 0;
 173 }
 174
 175 /*
 176  * Change the page attributes of an page in the linear mapping.
 177  *
 178  * This should be used when a page is mapped with a different caching policy
 179  * than write-back somewhere - some CPUs do not like it when mappings with
 180  * different caching policies exist. This changes the page attributes of the
 181  * in kernel linear mapping too.
 182  *
 183  * The caller needs to ensure that there are no conflicting mappings elsewhere.
 184  * This function only deals with the kernel linear map.
 185  *
 186  * Caller must call global_flush_tlb() after this.
 187  */
 188 int change_page_attr_addr(unsigned long address, int numpages, pgprot_t prot)
 189 {
 190         int err = 0;
 191         int i;
 192
 193         down_write(&init_mm.mmap_sem);
 194         for (i = 0; i < numpages; i++, address += PAGE_SIZE) {
 195                 unsigned long pfn = __pa(address) >> PAGE_SHIFT;
 196
 197                 err = __change_page_attr(address, pfn, prot, PAGE_KERNEL);
 198                 if (err)
 199                         break;
 200                 /* Handle kernel mapping too which aliases part of the
 201                  * lowmem */
 202                 if (__pa(address) < KERNEL_TEXT_SIZE) {
 203                         unsigned long addr2;
 204                         pgprot_t prot2 = prot;
 205                         addr2 = __START_KERNEL_map + __pa(address);
 206                         pgprot_val(prot2) &= ~_PAGE_NX;
 207                         err = __change_page_attr(addr2, pfn, prot2, PAGE_KERNEL_EXEC);
 208                 }
 209         }
 210         up_write(&init_mm.mmap_sem);
 211         return err;
 212 }
 213
 214 /* Don't call this for MMIO areas that may not have a mem_map entry */
 215 int change_page_attr(struct page *page, int numpages, pgprot_t prot)
 216 {
 217         unsigned long addr = (unsigned long)page_address(page);
 218         return change_page_attr_addr(addr, numpages, prot);
 219 }
 220
 221 void global_flush_tlb(void)
 222 {
 223         struct deferred_page *df, *next_df;
 224
 225         down_read(&init_mm.mmap_sem);
 226         df = xchg(&df_list, NULL);
 227         up_read(&init_mm.mmap_sem);
 228         flush_map((df && !df->next) ? df->address : 0);
 229         for (; df; df = next_df) {
 230                 next_df = df->next;
 231                 if (df->fpage)
 232                         __free_page(df->fpage);
 233                 kfree(df);
 234         }
 235 }
 236
 237 EXPORT_SYMBOL(change_page_attr);
 238 EXPORT_SYMBOL(global_flush_tlb);