arch/s390/mm/pgtable.c

   1 /*
   2  *  arch/s390/mm/pgtable.c
   3  *
   4  *    Copyright IBM Corp. 2007
   5  *    Author(s): Martin Schwidefsky <schwidefsky@de.ibm.com>
   6  */
   7
   8 #include <linux/sched.h>
   9 #include <linux/kernel.h>
  10 #include <linux/errno.h>
  11 #include <linux/mm.h>
  12 #include <linux/swap.h>
  13 #include <linux/smp.h>
  14 #include <linux/highmem.h>
  15 #include <linux/slab.h>
  16 #include <linux/pagemap.h>
  17 #include <linux/spinlock.h>
  18 #include <linux/module.h>
  19 #include <linux/quicklist.h>
  20
  21 #include <asm/system.h>
  22 #include <asm/pgtable.h>
  23 #include <asm/pgalloc.h>
  24 #include <asm/tlb.h>
  25 #include <asm/tlbflush.h>
  26 #include <asm/mmu_context.h>
  27
  28 #ifndef CONFIG_64BIT
  29 #define ALLOC_ORDER     1
  30 #define TABLES_PER_PAGE 4
  31 #define FRAG_MASK       15UL
  32 #define SECOND_HALVES   10UL
  33
  34 void clear_table_pgstes(unsigned long *table)
  35 {
  36         clear_table(table, _PAGE_TYPE_EMPTY, PAGE_SIZE/4);
  37         memset(table + 256, 0, PAGE_SIZE/4);
  38         clear_table(table + 512, _PAGE_TYPE_EMPTY, PAGE_SIZE/4);
  39         memset(table + 768, 0, PAGE_SIZE/4);
  40 }
  41
  42 #else
  43 #define ALLOC_ORDER     2
  44 #define TABLES_PER_PAGE 2
  45 #define FRAG_MASK       3UL
  46 #define SECOND_HALVES   2UL
  47
  48 void clear_table_pgstes(unsigned long *table)
  49 {
  50         clear_table(table, _PAGE_TYPE_EMPTY, PAGE_SIZE/2);
  51         memset(table + 256, 0, PAGE_SIZE/2);
  52 }
  53
  54 #endif
  55
  56 unsigned long *crst_table_alloc(struct mm_struct *mm, int noexec)
  57 {
  58         struct page *page = alloc_pages(GFP_KERNEL, ALLOC_ORDER);
  59
  60         if (!page)
  61                 return NULL;
  62         page->index = 0;
  63         if (noexec) {
  64                 struct page *shadow = alloc_pages(GFP_KERNEL, ALLOC_ORDER);
  65                 if (!shadow) {
  66                         __free_pages(page, ALLOC_ORDER);
  67                         return NULL;
  68                 }
  69                 page->index = page_to_phys(shadow);
  70         }
  71         spin_lock(&mm->page_table_lock);
  72         list_add(&page->lru, &mm->context.crst_list);
  73         spin_unlock(&mm->page_table_lock);
  74         return (unsigned long *) page_to_phys(page);
  75 }
  76
  77 void crst_table_free(struct mm_struct *mm, unsigned long *table)
  78 {
  79         unsigned long *shadow = get_shadow_table(table);
  80         struct page *page = virt_to_page(table);
  81
  82         spin_lock(&mm->page_table_lock);
  83         list_del(&page->lru);
  84         spin_unlock(&mm->page_table_lock);
  85         if (shadow)
  86                 free_pages((unsigned long) shadow, ALLOC_ORDER);
  87         free_pages((unsigned long) table, ALLOC_ORDER);
  88 }
  89
  90 #ifdef CONFIG_64BIT
  91 int crst_table_upgrade(struct mm_struct *mm, unsigned long limit)
  92 {
  93         unsigned long *table, *pgd;
  94         unsigned long entry;
  95
  96         BUG_ON(limit > (1UL << 53));
  97 repeat:
  98         table = crst_table_alloc(mm, mm->context.noexec);
  99         if (!table)
 100                 return -ENOMEM;
 101         spin_lock(&mm->page_table_lock);
 102         if (mm->context.asce_limit < limit) {
 103                 pgd = (unsigned long *) mm->pgd;
 104                 if (mm->context.asce_limit <= (1UL << 31)) {
 105                         entry = _REGION3_ENTRY_EMPTY;
 106                         mm->context.asce_limit = 1UL << 42;
 107                         mm->context.asce_bits = _ASCE_TABLE_LENGTH |
 108                                                 _ASCE_USER_BITS |
 109                                                 _ASCE_TYPE_REGION3;
 110                 } else {
 111                         entry = _REGION2_ENTRY_EMPTY;
 112                         mm->context.asce_limit = 1UL << 53;
 113                         mm->context.asce_bits = _ASCE_TABLE_LENGTH |
 114                                                 _ASCE_USER_BITS |
 115                                                 _ASCE_TYPE_REGION2;
 116                 }
 117                 crst_table_init(table, entry);
 118                 pgd_populate(mm, (pgd_t *) table, (pud_t *) pgd);
 119                 mm->pgd = (pgd_t *) table;
 120                 table = NULL;
 121         }
 122         spin_unlock(&mm->page_table_lock);
 123         if (table)
 124                 crst_table_free(mm, table);
 125         if (mm->context.asce_limit < limit)
 126                 goto repeat;
 127         update_mm(mm, current);
 128         return 0;
 129 }
 130
 131 void crst_table_downgrade(struct mm_struct *mm, unsigned long limit)
 132 {
 133         pgd_t *pgd;
 134
 135         if (mm->context.asce_limit <= limit)
 136                 return;
 137         __tlb_flush_mm(mm);
 138         while (mm->context.asce_limit > limit) {
 139                 pgd = mm->pgd;
 140                 switch (pgd_val(*pgd) & _REGION_ENTRY_TYPE_MASK) {
 141                 case _REGION_ENTRY_TYPE_R2:
 142                         mm->context.asce_limit = 1UL << 42;
 143                         mm->context.asce_bits = _ASCE_TABLE_LENGTH |
 144                                                 _ASCE_USER_BITS |
 145                                                 _ASCE_TYPE_REGION3;
 146                         break;
 147                 case _REGION_ENTRY_TYPE_R3:
 148                         mm->context.asce_limit = 1UL << 31;
 149                         mm->context.asce_bits = _ASCE_TABLE_LENGTH |
 150                                                 _ASCE_USER_BITS |
 151                                                 _ASCE_TYPE_SEGMENT;
 152                         break;
 153                 default:
 154                         BUG();
 155                 }
 156                 mm->pgd = (pgd_t *) (pgd_val(*pgd) & _REGION_ENTRY_ORIGIN);
 157                 crst_table_free(mm, (unsigned long *) pgd);
 158         }
 159         update_mm(mm, current);
 160 }
 161 #endif
 162
 163 /*
 164  * page table entry allocation/free routines.
 165  */
 166 unsigned long *page_table_alloc(struct mm_struct *mm)
 167 {
 168         struct page *page;
 169         unsigned long *table;
 170         unsigned long bits;
 171
 172         bits = (mm->context.noexec || mm->context.has_pgste) ? 3UL : 1UL;
 173         spin_lock(&mm->page_table_lock);
 174         page = NULL;
 175         if (!list_empty(&mm->context.pgtable_list)) {
 176                 page = list_first_entry(&mm->context.pgtable_list,
 177                                         struct page, lru);
 178                 if ((page->flags & FRAG_MASK) == ((1UL << TABLES_PER_PAGE) - 1))
 179                         page = NULL;
 180         }
 181         if (!page) {
 182                 spin_unlock(&mm->page_table_lock);
 183                 page = alloc_page(GFP_KERNEL|__GFP_REPEAT);
 184                 if (!page)
 185                         return NULL;
 186                 pgtable_page_ctor(page);
 187                 page->flags &= ~FRAG_MASK;
 188                 table = (unsigned long *) page_to_phys(page);
 189                 if (mm->context.has_pgste)
 190                         clear_table_pgstes(table);
 191                 else
 192                         clear_table(table, _PAGE_TYPE_EMPTY, PAGE_SIZE);
 193                 spin_lock(&mm->page_table_lock);
 194                 list_add(&page->lru, &mm->context.pgtable_list);
 195         }
 196         table = (unsigned long *) page_to_phys(page);
 197         while (page->flags & bits) {
 198                 table += 256;
 199                 bits <<= 1;
 200         }
 201         page->flags |= bits;
 202         if ((page->flags & FRAG_MASK) == ((1UL << TABLES_PER_PAGE) - 1))
 203                 list_move_tail(&page->lru, &mm->context.pgtable_list);
 204         spin_unlock(&mm->page_table_lock);
 205         return table;
 206 }
 207
 208 void page_table_free(struct mm_struct *mm, unsigned long *table)
 209 {
 210         struct page *page;
 211         unsigned long bits;
 212
 213         bits = (mm->context.noexec || mm->context.has_pgste) ? 3UL : 1UL;
 214         bits <<= (__pa(table) & (PAGE_SIZE - 1)) / 256 / sizeof(unsigned long);
 215         page = pfn_to_page(__pa(table) >> PAGE_SHIFT);
 216         spin_lock(&mm->page_table_lock);
 217         page->flags ^= bits;
 218         if (page->flags & FRAG_MASK) {
 219                 /* Page now has some free pgtable fragments. */
 220                 list_move(&page->lru, &mm->context.pgtable_list);
 221                 page = NULL;
 222         } else
 223                 /* All fragments of the 4K page have been freed. */
 224                 list_del(&page->lru);
 225         spin_unlock(&mm->page_table_lock);
 226         if (page) {
 227                 pgtable_page_dtor(page);
 228                 __free_page(page);
 229         }
 230 }
 231
 232 void disable_noexec(struct mm_struct *mm, struct task_struct *tsk)
 233 {
 234         struct page *page;
 235
 236         spin_lock(&mm->page_table_lock);
 237         /* Free shadow region and segment tables. */
 238         list_for_each_entry(page, &mm->context.crst_list, lru)
 239                 if (page->index) {
 240                         free_pages((unsigned long) page->index, ALLOC_ORDER);
 241                         page->index = 0;
 242                 }
 243         /* "Free" second halves of page tables. */
 244         list_for_each_entry(page, &mm->context.pgtable_list, lru)
 245                 page->flags &= ~SECOND_HALVES;
 246         spin_unlock(&mm->page_table_lock);
 247         mm->context.noexec = 0;
 248         update_mm(mm, tsk);
 249 }
 250
 251 /*
 252  * switch on pgstes for its userspace process (for kvm)
 253  */
 254 int s390_enable_sie(void)
 255 {
 256         struct task_struct *tsk = current;
 257         struct mm_struct *mm, *old_mm;
 258
 259         /* Do we have pgstes? if yes, we are done */
 260         if (tsk->mm->context.has_pgste)
 261                 return 0;
 262
 263         /* lets check if we are allowed to replace the mm */
 264         task_lock(tsk);
 265         if (!tsk->mm || atomic_read(&tsk->mm->mm_users) > 1 ||
 266             tsk->mm != tsk->active_mm || tsk->mm->ioctx_list) {
 267                 task_unlock(tsk);
 268                 return -EINVAL;
 269         }
 270         task_unlock(tsk);
 271
 272         /* we copy the mm and let dup_mm create the page tables with_pgstes */
 273         tsk->mm->context.alloc_pgste = 1;
 274         mm = dup_mm(tsk);
 275         tsk->mm->context.alloc_pgste = 0;
 276         if (!mm)
 277                 return -ENOMEM;
 278
 279         /* Now lets check again if something happened */
 280         task_lock(tsk);
 281         if (!tsk->mm || atomic_read(&tsk->mm->mm_users) > 1 ||
 282             tsk->mm != tsk->active_mm || tsk->mm->ioctx_list) {
 283                 mmput(mm);
 284                 task_unlock(tsk);
 285                 return -EINVAL;
 286         }
 287
 288         /* ok, we are alone. No ptrace, no threads, etc. */
 289         old_mm = tsk->mm;
 290         tsk->mm = tsk->active_mm = mm;
 291         preempt_disable();
 292         update_mm(mm, tsk);
 293         cpu_set(smp_processor_id(), mm->cpu_vm_mask);
 294         preempt_enable();
 295         task_unlock(tsk);
 296         mmput(old_mm);
 297         return 0;
 298 }
 299 EXPORT_SYMBOL_GPL(s390_enable_sie);