arch/s390/mm/pgtable.c

   1 /*
   2  *    Copyright IBM Corp. 2007,2009
   3  *    Author(s): Martin Schwidefsky <schwidefsky@de.ibm.com>
   4  */
   5
   6 #include <linux/sched.h>
   7 #include <linux/kernel.h>
   8 #include <linux/errno.h>
   9 #include <linux/mm.h>
  10 #include <linux/swap.h>
  11 #include <linux/smp.h>
  12 #include <linux/highmem.h>
  13 #include <linux/slab.h>
  14 #include <linux/pagemap.h>
  15 #include <linux/spinlock.h>
  16 #include <linux/module.h>
  17 #include <linux/quicklist.h>
  18
  19 #include <asm/system.h>
  20 #include <asm/pgtable.h>
  21 #include <asm/pgalloc.h>
  22 #include <asm/tlb.h>
  23 #include <asm/tlbflush.h>
  24 #include <asm/mmu_context.h>
  25
  26 #ifndef CONFIG_64BIT
  27 #define ALLOC_ORDER     1
  28 #define TABLES_PER_PAGE 4
  29 #define FRAG_MASK       15UL
  30 #define SECOND_HALVES   10UL
  31
  32 void clear_table_pgstes(unsigned long *table)
  33 {
  34         clear_table(table, _PAGE_TYPE_EMPTY, PAGE_SIZE/4);
  35         memset(table + 256, 0, PAGE_SIZE/4);
  36         clear_table(table + 512, _PAGE_TYPE_EMPTY, PAGE_SIZE/4);
  37         memset(table + 768, 0, PAGE_SIZE/4);
  38 }
  39
  40 #else
  41 #define ALLOC_ORDER     2
  42 #define TABLES_PER_PAGE 2
  43 #define FRAG_MASK       3UL
  44 #define SECOND_HALVES   2UL
  45
  46 void clear_table_pgstes(unsigned long *table)
  47 {
  48         clear_table(table, _PAGE_TYPE_EMPTY, PAGE_SIZE/2);
  49         memset(table + 256, 0, PAGE_SIZE/2);
  50 }
  51
  52 #endif
  53
  54 unsigned long VMALLOC_START = VMALLOC_END - VMALLOC_SIZE;
  55 EXPORT_SYMBOL(VMALLOC_START);
  56
  57 static int __init parse_vmalloc(char *arg)
  58 {
  59         if (!arg)
  60                 return -EINVAL;
  61         VMALLOC_START = (VMALLOC_END - memparse(arg, &arg)) & PAGE_MASK;
  62         return 0;
  63 }
  64 early_param("vmalloc", parse_vmalloc);
  65
  66 unsigned long *crst_table_alloc(struct mm_struct *mm, int noexec)
  67 {
  68         struct page *page = alloc_pages(GFP_KERNEL, ALLOC_ORDER);
  69
  70         if (!page)
  71                 return NULL;
  72         page->index = 0;
  73         if (noexec) {
  74                 struct page *shadow = alloc_pages(GFP_KERNEL, ALLOC_ORDER);
  75                 if (!shadow) {
  76                         __free_pages(page, ALLOC_ORDER);
  77                         return NULL;
  78                 }
  79                 page->index = page_to_phys(shadow);
  80         }
  81         spin_lock(&mm->context.list_lock);
  82         list_add(&page->lru, &mm->context.crst_list);
  83         spin_unlock(&mm->context.list_lock);
  84         return (unsigned long *) page_to_phys(page);
  85 }
  86
  87 void crst_table_free(struct mm_struct *mm, unsigned long *table)
  88 {
  89         unsigned long *shadow = get_shadow_table(table);
  90         struct page *page = virt_to_page(table);
  91
  92         spin_lock(&mm->context.list_lock);
  93         list_del(&page->lru);
  94         spin_unlock(&mm->context.list_lock);
  95         if (shadow)
  96                 free_pages((unsigned long) shadow, ALLOC_ORDER);
  97         free_pages((unsigned long) table, ALLOC_ORDER);
  98 }
  99
 100 #ifdef CONFIG_64BIT
 101 int crst_table_upgrade(struct mm_struct *mm, unsigned long limit)
 102 {
 103         unsigned long *table, *pgd;
 104         unsigned long entry;
 105
 106         BUG_ON(limit > (1UL << 53));
 107 repeat:
 108         table = crst_table_alloc(mm, mm->context.noexec);
 109         if (!table)
 110                 return -ENOMEM;
 111         spin_lock(&mm->page_table_lock);
 112         if (mm->context.asce_limit < limit) {
 113                 pgd = (unsigned long *) mm->pgd;
 114                 if (mm->context.asce_limit <= (1UL << 31)) {
 115                         entry = _REGION3_ENTRY_EMPTY;
 116                         mm->context.asce_limit = 1UL << 42;
 117                         mm->context.asce_bits = _ASCE_TABLE_LENGTH |
 118                                                 _ASCE_USER_BITS |
 119                                                 _ASCE_TYPE_REGION3;
 120                 } else {
 121                         entry = _REGION2_ENTRY_EMPTY;
 122                         mm->context.asce_limit = 1UL << 53;
 123                         mm->context.asce_bits = _ASCE_TABLE_LENGTH |
 124                                                 _ASCE_USER_BITS |
 125                                                 _ASCE_TYPE_REGION2;
 126                 }
 127                 crst_table_init(table, entry);
 128                 pgd_populate(mm, (pgd_t *) table, (pud_t *) pgd);
 129                 mm->pgd = (pgd_t *) table;
 130                 mm->task_size = mm->context.asce_limit;
 131                 table = NULL;
 132         }
 133         spin_unlock(&mm->page_table_lock);
 134         if (table)
 135                 crst_table_free(mm, table);
 136         if (mm->context.asce_limit < limit)
 137                 goto repeat;
 138         update_mm(mm, current);
 139         return 0;
 140 }
 141
 142 void crst_table_downgrade(struct mm_struct *mm, unsigned long limit)
 143 {
 144         pgd_t *pgd;
 145
 146         if (mm->context.asce_limit <= limit)
 147                 return;
 148         __tlb_flush_mm(mm);
 149         while (mm->context.asce_limit > limit) {
 150                 pgd = mm->pgd;
 151                 switch (pgd_val(*pgd) & _REGION_ENTRY_TYPE_MASK) {
 152                 case _REGION_ENTRY_TYPE_R2:
 153                         mm->context.asce_limit = 1UL << 42;
 154                         mm->context.asce_bits = _ASCE_TABLE_LENGTH |
 155                                                 _ASCE_USER_BITS |
 156                                                 _ASCE_TYPE_REGION3;
 157                         break;
 158                 case _REGION_ENTRY_TYPE_R3:
 159                         mm->context.asce_limit = 1UL << 31;
 160                         mm->context.asce_bits = _ASCE_TABLE_LENGTH |
 161                                                 _ASCE_USER_BITS |
 162                                                 _ASCE_TYPE_SEGMENT;
 163                         break;
 164                 default:
 165                         BUG();
 166                 }
 167                 mm->pgd = (pgd_t *) (pgd_val(*pgd) & _REGION_ENTRY_ORIGIN);
 168                 mm->task_size = mm->context.asce_limit;
 169                 crst_table_free(mm, (unsigned long *) pgd);
 170         }
 171         update_mm(mm, current);
 172 }
 173 #endif
 174
 175 /*
 176  * page table entry allocation/free routines.
 177  */
 178 unsigned long *page_table_alloc(struct mm_struct *mm)
 179 {
 180         struct page *page;
 181         unsigned long *table;
 182         unsigned long bits;
 183
 184         bits = (mm->context.noexec || mm->context.has_pgste) ? 3UL : 1UL;
 185         spin_lock(&mm->context.list_lock);
 186         page = NULL;
 187         if (!list_empty(&mm->context.pgtable_list)) {
 188                 page = list_first_entry(&mm->context.pgtable_list,
 189                                         struct page, lru);
 190                 if ((page->flags & FRAG_MASK) == ((1UL << TABLES_PER_PAGE) - 1))
 191                         page = NULL;
 192         }
 193         if (!page) {
 194                 spin_unlock(&mm->context.list_lock);
 195                 page = alloc_page(GFP_KERNEL|__GFP_REPEAT);
 196                 if (!page)
 197                         return NULL;
 198                 pgtable_page_ctor(page);
 199                 page->flags &= ~FRAG_MASK;
 200                 table = (unsigned long *) page_to_phys(page);
 201                 if (mm->context.has_pgste)
 202                         clear_table_pgstes(table);
 203                 else
 204                         clear_table(table, _PAGE_TYPE_EMPTY, PAGE_SIZE);
 205                 spin_lock(&mm->context.list_lock);
 206                 list_add(&page->lru, &mm->context.pgtable_list);
 207         }
 208         table = (unsigned long *) page_to_phys(page);
 209         while (page->flags & bits) {
 210                 table += 256;
 211                 bits <<= 1;
 212         }
 213         page->flags |= bits;
 214         if ((page->flags & FRAG_MASK) == ((1UL << TABLES_PER_PAGE) - 1))
 215                 list_move_tail(&page->lru, &mm->context.pgtable_list);
 216         spin_unlock(&mm->context.list_lock);
 217         return table;
 218 }
 219
 220 void page_table_free(struct mm_struct *mm, unsigned long *table)
 221 {
 222         struct page *page;
 223         unsigned long bits;
 224
 225         bits = (mm->context.noexec || mm->context.has_pgste) ? 3UL : 1UL;
 226         bits <<= (__pa(table) & (PAGE_SIZE - 1)) / 256 / sizeof(unsigned long);
 227         page = pfn_to_page(__pa(table) >> PAGE_SHIFT);
 228         spin_lock(&mm->context.list_lock);
 229         page->flags ^= bits;
 230         if (page->flags & FRAG_MASK) {
 231                 /* Page now has some free pgtable fragments. */
 232                 list_move(&page->lru, &mm->context.pgtable_list);
 233                 page = NULL;
 234         } else
 235                 /* All fragments of the 4K page have been freed. */
 236                 list_del(&page->lru);
 237         spin_unlock(&mm->context.list_lock);
 238         if (page) {
 239                 pgtable_page_dtor(page);
 240                 __free_page(page);
 241         }
 242 }
 243
 244 void disable_noexec(struct mm_struct *mm, struct task_struct *tsk)
 245 {
 246         struct page *page;
 247
 248         spin_lock(&mm->context.list_lock);
 249         /* Free shadow region and segment tables. */
 250         list_for_each_entry(page, &mm->context.crst_list, lru)
 251                 if (page->index) {
 252                         free_pages((unsigned long) page->index, ALLOC_ORDER);
 253                         page->index = 0;
 254                 }
 255         /* "Free" second halves of page tables. */
 256         list_for_each_entry(page, &mm->context.pgtable_list, lru)
 257                 page->flags &= ~SECOND_HALVES;
 258         spin_unlock(&mm->context.list_lock);
 259         mm->context.noexec = 0;
 260         update_mm(mm, tsk);
 261 }
 262
 263 /*
 264  * switch on pgstes for its userspace process (for kvm)
 265  */
 266 int s390_enable_sie(void)
 267 {
 268         struct task_struct *tsk = current;
 269         struct mm_struct *mm, *old_mm;
 270
 271         /* Do we have switched amode? If no, we cannot do sie */
 272         if (!switch_amode)
 273                 return -EINVAL;
 274
 275         /* Do we have pgstes? if yes, we are done */
 276         if (tsk->mm->context.has_pgste)
 277                 return 0;
 278
 279         /* lets check if we are allowed to replace the mm */
 280         task_lock(tsk);
 281         if (!tsk->mm || atomic_read(&tsk->mm->mm_users) > 1 ||
 282 #ifdef CONFIG_AIO
 283             !hlist_empty(&tsk->mm->ioctx_list) ||
 284 #endif
 285             tsk->mm != tsk->active_mm) {
 286                 task_unlock(tsk);
 287                 return -EINVAL;
 288         }
 289         task_unlock(tsk);
 290
 291         /* we copy the mm and let dup_mm create the page tables with_pgstes */
 292         tsk->mm->context.alloc_pgste = 1;
 293         mm = dup_mm(tsk);
 294         tsk->mm->context.alloc_pgste = 0;
 295         if (!mm)
 296                 return -ENOMEM;
 297
 298         /* Now lets check again if something happened */
 299         task_lock(tsk);
 300         if (!tsk->mm || atomic_read(&tsk->mm->mm_users) > 1 ||
 301 #ifdef CONFIG_AIO
 302             !hlist_empty(&tsk->mm->ioctx_list) ||
 303 #endif
 304             tsk->mm != tsk->active_mm) {
 305                 mmput(mm);
 306                 task_unlock(tsk);
 307                 return -EINVAL;
 308         }
 309
 310         /* ok, we are alone. No ptrace, no threads, etc. */
 311         old_mm = tsk->mm;
 312         tsk->mm = tsk->active_mm = mm;
 313         preempt_disable();
 314         update_mm(mm, tsk);
 315         cpumask_set_cpu(smp_processor_id(), mm_cpumask(mm));
 316         preempt_enable();
 317         task_unlock(tsk);
 318         mmput(old_mm);
 319         return 0;
 320 }
 321 EXPORT_SYMBOL_GPL(s390_enable_sie);
 322
 323 #if defined(CONFIG_DEBUG_PAGEALLOC) && defined(CONFIG_HIBERNATION)
 324 bool kernel_page_present(struct page *page)
 325 {
 326         unsigned long addr;
 327         int cc;
 328
 329         addr = page_to_phys(page);
 330         asm volatile(
 331                 "       lra     %1,0(%1)\n"
 332                 "       ipm     %0\n"
 333                 "       srl     %0,28"
 334                 : "=d" (cc), "+a" (addr) : : "cc");
 335         return cc == 0;
 336 }
 337 #endif /* CONFIG_HIBERNATION && CONFIG_DEBUG_PAGEALLOC */