include/asm-x86_64/mach-xen/asm/pgalloc.h

   1 #ifndef _X86_64_PGALLOC_H
   2 #define _X86_64_PGALLOC_H
   3
   4 #include <asm/fixmap.h>
   5 #include <asm/pda.h>
   6 #include <linux/threads.h>
   7 #include <linux/mm.h>
   8 #include <asm/io.h>             /* for phys_to_virt and page_to_pseudophys */
   9
  10 #include <xen/features.h>
  11
  12 #define arch_add_exec_range(mm, limit) \
  13                 do { (void)(mm), (void)(limit); } while (0)
  14 #define arch_flush_exec_range(mm) \
  15                 do { (void)(mm); } while (0)
  16 #define arch_remove_exec_range(mm, limit) \
  17                 do { (void)(mm), (void)(limit); } while (0)
  18
  19 void make_page_readonly(void *va, unsigned int feature);
  20 void make_page_writable(void *va, unsigned int feature);
  21 void make_pages_readonly(void *va, unsigned int nr, unsigned int feature);
  22 void make_pages_writable(void *va, unsigned int nr, unsigned int feature);
  23
  24 #define __user_pgd(pgd) ((pgd) + PTRS_PER_PGD)
  25
  26 static inline void pmd_populate_kernel(struct mm_struct *mm, pmd_t *pmd, pte_t *pte)
  27 {
  28         set_pmd(pmd, __pmd(_PAGE_TABLE | __pa(pte)));
  29 }
  30
  31 static inline void pud_populate(struct mm_struct *mm, pud_t *pud, pmd_t *pmd)
  32 {
  33         if (unlikely((mm)->context.pinned)) {
  34                 BUG_ON(HYPERVISOR_update_va_mapping(
  35                                (unsigned long)pmd,
  36                                pfn_pte(virt_to_phys(pmd)>>PAGE_SHIFT,
  37                                        PAGE_KERNEL_RO), 0));
  38                 set_pud(pud, __pud(_PAGE_TABLE | __pa(pmd)));
  39         } else {
  40                 *(pud) =  __pud(_PAGE_TABLE | __pa(pmd));
  41         }
  42 }
  43
  44 /*
  45  * We need to use the batch mode here, but pgd_pupulate() won't be
  46  * be called frequently.
  47  */
  48 static inline void pgd_populate(struct mm_struct *mm, pgd_t *pgd, pud_t *pud)
  49 {
  50         if (unlikely((mm)->context.pinned)) {
  51                 BUG_ON(HYPERVISOR_update_va_mapping(
  52                                (unsigned long)pud,
  53                                pfn_pte(virt_to_phys(pud)>>PAGE_SHIFT,
  54                                        PAGE_KERNEL_RO), 0));
  55                 set_pgd(pgd, __pgd(_PAGE_TABLE | __pa(pud)));
  56                 set_pgd(__user_pgd(pgd), __pgd(_PAGE_TABLE | __pa(pud)));
  57         } else {
  58                 *(pgd) =  __pgd(_PAGE_TABLE | __pa(pud));
  59                 *(__user_pgd(pgd)) = *(pgd);
  60         }
  61 }
  62
  63 static inline void pmd_populate(struct mm_struct *mm, pmd_t *pmd, struct page *pte)
  64 {
  65         if (unlikely((mm)->context.pinned)) {
  66                 BUG_ON(HYPERVISOR_update_va_mapping(
  67                                (unsigned long)__va(page_to_pfn(pte) << PAGE_SHIFT),
  68                                pfn_pte(page_to_pfn(pte), PAGE_KERNEL_RO), 0));
  69                 set_pmd(pmd, __pmd(_PAGE_TABLE | (page_to_pfn(pte) << PAGE_SHIFT)));
  70         } else {
  71                 *(pmd) = __pmd(_PAGE_TABLE | (page_to_pfn(pte) << PAGE_SHIFT));
  72         }
  73 }
  74
  75 static inline void pmd_free(pmd_t *pmd)
  76 {
  77         pte_t *ptep = virt_to_ptep(pmd);
  78
  79         if (!pte_write(*ptep)) {
  80                 BUG_ON(HYPERVISOR_update_va_mapping(
  81                         (unsigned long)pmd,
  82                         pfn_pte(virt_to_phys(pmd)>>PAGE_SHIFT, PAGE_KERNEL),
  83                         0));
  84         }
  85         free_page((unsigned long)pmd);
  86 }
  87
  88 static inline pmd_t *pmd_alloc_one (struct mm_struct *mm, unsigned long addr)
  89 {
  90         return (pmd_t *)get_zeroed_page(GFP_KERNEL|__GFP_REPEAT);
  91 }
  92
  93 static inline pud_t *pud_alloc_one(struct mm_struct *mm, unsigned long addr)
  94 {
  95         return (pud_t *)get_zeroed_page(GFP_KERNEL|__GFP_REPEAT);
  96 }
  97
  98 static inline void pud_free (pud_t *pud)
  99 {
 100         pte_t *ptep = virt_to_ptep(pud);
 101
 102         if (!pte_write(*ptep)) {
 103                 BUG_ON(HYPERVISOR_update_va_mapping(
 104                         (unsigned long)pud,
 105                         pfn_pte(virt_to_phys(pud)>>PAGE_SHIFT, PAGE_KERNEL),
 106                         0));
 107         }
 108         free_page((unsigned long)pud);
 109 }
 110
 111 static inline void pgd_list_add(pgd_t *pgd)
 112 {
 113         struct page *page = virt_to_page(pgd);
 114
 115         spin_lock(&pgd_lock);
 116         page->index = (pgoff_t)pgd_list;
 117         if (pgd_list)
 118                 pgd_list->private = (unsigned long)&page->index;
 119         pgd_list = page;
 120         page->private = (unsigned long)&pgd_list;
 121         spin_unlock(&pgd_lock);
 122 }
 123
 124 static inline void pgd_list_del(pgd_t *pgd)
 125 {
 126         struct page *next, **pprev, *page = virt_to_page(pgd);
 127
 128         spin_lock(&pgd_lock);
 129         next = (struct page *)page->index;
 130         pprev = (struct page **)page->private;
 131         *pprev = next;
 132         if (next)
 133                 next->private = (unsigned long)pprev;
 134         spin_unlock(&pgd_lock);
 135 }
 136
 137 static inline pgd_t *pgd_alloc(struct mm_struct *mm)
 138 {
 139         /*
 140          * We allocate two contiguous pages for kernel and user.
 141          */
 142         unsigned boundary;
 143         pgd_t *pgd = (pgd_t *)__get_free_pages(GFP_KERNEL|__GFP_REPEAT, 1);
 144
 145         if (!pgd)
 146                 return NULL;
 147         pgd_list_add(pgd);
 148         /*
 149          * Copy kernel pointers in from init.
 150          * Could keep a freelist or slab cache of those because the kernel
 151          * part never changes.
 152          */
 153         boundary = pgd_index(__PAGE_OFFSET);
 154         memset(pgd, 0, boundary * sizeof(pgd_t));
 155         memcpy(pgd + boundary,
 156                init_level4_pgt + boundary,
 157                (PTRS_PER_PGD - boundary) * sizeof(pgd_t));
 158
 159         memset(__user_pgd(pgd), 0, PAGE_SIZE); /* clean up user pgd */
 160         /*
 161          * Set level3_user_pgt for vsyscall area
 162          */
 163         set_pgd(__user_pgd(pgd) + pgd_index(VSYSCALL_START),
 164                 mk_kernel_pgd(__pa_symbol(level3_user_pgt)));
 165         return pgd;
 166 }
 167
 168 static inline void pgd_free(pgd_t *pgd)
 169 {
 170         pte_t *ptep = virt_to_ptep(pgd);
 171
 172         if (!pte_write(*ptep)) {
 173                 xen_pgd_unpin(__pa(pgd));
 174                 BUG_ON(HYPERVISOR_update_va_mapping(
 175                                (unsigned long)pgd,
 176                                pfn_pte(virt_to_phys(pgd)>>PAGE_SHIFT, PAGE_KERNEL),
 177                                0));
 178         }
 179
 180         ptep = virt_to_ptep(__user_pgd(pgd));
 181
 182         if (!pte_write(*ptep)) {
 183                 xen_pgd_unpin(__pa(__user_pgd(pgd)));
 184                 BUG_ON(HYPERVISOR_update_va_mapping(
 185                                (unsigned long)__user_pgd(pgd),
 186                                pfn_pte(virt_to_phys(__user_pgd(pgd))>>PAGE_SHIFT,
 187                                        PAGE_KERNEL),
 188                                0));
 189         }
 190
 191         pgd_list_del(pgd);
 192         free_pages((unsigned long)pgd, 1);
 193 }
 194
 195 static inline pte_t *pte_alloc_one_kernel(struct mm_struct *mm, unsigned long address)
 196 {
 197         pte_t *pte = (pte_t *)get_zeroed_page(GFP_KERNEL|__GFP_REPEAT);
 198         if (pte)
 199                 make_page_readonly(pte, XENFEAT_writable_page_tables);
 200
 201         return pte;
 202 }
 203
 204 static inline struct page *pte_alloc_one(struct mm_struct *mm, unsigned long address)
 205 {
 206         struct page *pte;
 207
 208         pte = alloc_pages(GFP_KERNEL|__GFP_REPEAT|__GFP_ZERO, 0);
 209         return pte;
 210 }
 211
 212 /* Should really implement gc for free page table pages. This could be
 213    done with a reference count in struct page. */
 214
 215 static inline void pte_free_kernel(pte_t *pte)
 216 {
 217         BUG_ON((unsigned long)pte & (PAGE_SIZE-1));
 218         make_page_writable(pte, XENFEAT_writable_page_tables);
 219         free_page((unsigned long)pte);
 220 }
 221
 222 extern void pte_free(struct page *pte);
 223
 224 //#define __pte_free_tlb(tlb,pte) tlb_remove_page((tlb),(pte))
 225 //#define __pmd_free_tlb(tlb,x)   tlb_remove_page((tlb),virt_to_page(x))
 226 //#define __pud_free_tlb(tlb,x)   tlb_remove_page((tlb),virt_to_page(x))
 227
 228 #define __pte_free_tlb(tlb,x)   pte_free((x))
 229 #define __pmd_free_tlb(tlb,x)   pmd_free((x))
 230 #define __pud_free_tlb(tlb,x)   pud_free((x))
 231
 232 #endif /* _X86_64_PGALLOC_H */