arch/s390/mm/vmem.c

   1 /*
   2  *  arch/s390/mm/vmem.c
   3  *
   4  *    Copyright IBM Corp. 2006
   5  *    Author(s): Heiko Carstens <heiko.carstens@de.ibm.com>
   6  */
   7
   8 #include <linux/bootmem.h>
   9 #include <linux/pfn.h>
  10 #include <linux/mm.h>
  11 #include <linux/module.h>
  12 #include <linux/list.h>
  13 #include <linux/hugetlb.h>
  14 #include <asm/pgalloc.h>
  15 #include <asm/pgtable.h>
  16 #include <asm/setup.h>
  17 #include <asm/tlbflush.h>
  18 #include <asm/sections.h>
  19
  20 static DEFINE_MUTEX(vmem_mutex);
  21
  22 struct memory_segment {
  23         struct list_head list;
  24         unsigned long start;
  25         unsigned long size;
  26 };
  27
  28 static LIST_HEAD(mem_segs);
  29
  30 static void __ref *vmem_alloc_pages(unsigned int order)
  31 {
  32         if (slab_is_available())
  33                 return (void *)__get_free_pages(GFP_KERNEL, order);
  34         return alloc_bootmem_pages((1 << order) * PAGE_SIZE);
  35 }
  36
  37 static inline pud_t *vmem_pud_alloc(void)
  38 {
  39         pud_t *pud = NULL;
  40
  41 #ifdef CONFIG_64BIT
  42         pud = vmem_alloc_pages(2);
  43         if (!pud)
  44                 return NULL;
  45         clear_table((unsigned long *) pud, _REGION3_ENTRY_EMPTY, PAGE_SIZE * 4);
  46 #endif
  47         return pud;
  48 }
  49
  50 static inline pmd_t *vmem_pmd_alloc(void)
  51 {
  52         pmd_t *pmd = NULL;
  53
  54 #ifdef CONFIG_64BIT
  55         pmd = vmem_alloc_pages(2);
  56         if (!pmd)
  57                 return NULL;
  58         clear_table((unsigned long *) pmd, _SEGMENT_ENTRY_EMPTY, PAGE_SIZE * 4);
  59 #endif
  60         return pmd;
  61 }
  62
  63 static pte_t __ref *vmem_pte_alloc(void)
  64 {
  65         pte_t *pte;
  66
  67         if (slab_is_available())
  68                 pte = (pte_t *) page_table_alloc(&init_mm);
  69         else
  70                 pte = alloc_bootmem(PTRS_PER_PTE * sizeof(pte_t));
  71         if (!pte)
  72                 return NULL;
  73         clear_table((unsigned long *) pte, _PAGE_TYPE_EMPTY,
  74                     PTRS_PER_PTE * sizeof(pte_t));
  75         return pte;
  76 }
  77
  78 /*
  79  * Add a physical memory range to the 1:1 mapping.
  80  */
  81 static int vmem_add_mem(unsigned long start, unsigned long size, int ro)
  82 {
  83         unsigned long address;
  84         pgd_t *pg_dir;
  85         pud_t *pu_dir;
  86         pmd_t *pm_dir;
  87         pte_t *pt_dir;
  88         pte_t  pte;
  89         int ret = -ENOMEM;
  90
  91         for (address = start; address < start + size; address += PAGE_SIZE) {
  92                 pg_dir = pgd_offset_k(address);
  93                 if (pgd_none(*pg_dir)) {
  94                         pu_dir = vmem_pud_alloc();
  95                         if (!pu_dir)
  96                                 goto out;
  97                         pgd_populate_kernel(&init_mm, pg_dir, pu_dir);
  98                 }
  99
 100                 pu_dir = pud_offset(pg_dir, address);
 101                 if (pud_none(*pu_dir)) {
 102                         pm_dir = vmem_pmd_alloc();
 103                         if (!pm_dir)
 104                                 goto out;
 105                         pud_populate_kernel(&init_mm, pu_dir, pm_dir);
 106                 }
 107
 108                 pte = mk_pte_phys(address, __pgprot(ro ? _PAGE_RO : 0));
 109                 pm_dir = pmd_offset(pu_dir, address);
 110
 111 #ifdef __s390x__
 112                 if (MACHINE_HAS_HPAGE && !(address & ~HPAGE_MASK) &&
 113                     (address + HPAGE_SIZE <= start + size) &&
 114                     (address >= HPAGE_SIZE)) {
 115                         pte_val(pte) |= _SEGMENT_ENTRY_LARGE;
 116                         pmd_val(*pm_dir) = pte_val(pte);
 117                         address += HPAGE_SIZE - PAGE_SIZE;
 118                         continue;
 119                 }
 120 #endif
 121                 if (pmd_none(*pm_dir)) {
 122                         pt_dir = vmem_pte_alloc();
 123                         if (!pt_dir)
 124                                 goto out;
 125                         pmd_populate_kernel(&init_mm, pm_dir, pt_dir);
 126                 }
 127
 128                 pt_dir = pte_offset_kernel(pm_dir, address);
 129                 *pt_dir = pte;
 130         }
 131         ret = 0;
 132 out:
 133         flush_tlb_kernel_range(start, start + size);
 134         return ret;
 135 }
 136
 137 /*
 138  * Remove a physical memory range from the 1:1 mapping.
 139  * Currently only invalidates page table entries.
 140  */
 141 static void vmem_remove_range(unsigned long start, unsigned long size)
 142 {
 143         unsigned long address;
 144         pgd_t *pg_dir;
 145         pud_t *pu_dir;
 146         pmd_t *pm_dir;
 147         pte_t *pt_dir;
 148         pte_t  pte;
 149
 150         pte_val(pte) = _PAGE_TYPE_EMPTY;
 151         for (address = start; address < start + size; address += PAGE_SIZE) {
 152                 pg_dir = pgd_offset_k(address);
 153                 pu_dir = pud_offset(pg_dir, address);
 154                 if (pud_none(*pu_dir))
 155                         continue;
 156                 pm_dir = pmd_offset(pu_dir, address);
 157                 if (pmd_none(*pm_dir))
 158                         continue;
 159
 160                 if (pmd_huge(*pm_dir)) {
 161                         pmd_clear_kernel(pm_dir);
 162                         address += HPAGE_SIZE - PAGE_SIZE;
 163                         continue;
 164                 }
 165
 166                 pt_dir = pte_offset_kernel(pm_dir, address);
 167                 *pt_dir = pte;
 168         }
 169         flush_tlb_kernel_range(start, start + size);
 170 }
 171
 172 /*
 173  * Add a backed mem_map array to the virtual mem_map array.
 174  */
 175 int __meminit vmemmap_populate(struct page *start, unsigned long nr, int node)
 176 {
 177         unsigned long address, start_addr, end_addr;
 178         pgd_t *pg_dir;
 179         pud_t *pu_dir;
 180         pmd_t *pm_dir;
 181         pte_t *pt_dir;
 182         pte_t  pte;
 183         int ret = -ENOMEM;
 184
 185         start_addr = (unsigned long) start;
 186         end_addr = (unsigned long) (start + nr);
 187
 188         for (address = start_addr; address < end_addr; address += PAGE_SIZE) {
 189                 pg_dir = pgd_offset_k(address);
 190                 if (pgd_none(*pg_dir)) {
 191                         pu_dir = vmem_pud_alloc();
 192                         if (!pu_dir)
 193                                 goto out;
 194                         pgd_populate_kernel(&init_mm, pg_dir, pu_dir);
 195                 }
 196
 197                 pu_dir = pud_offset(pg_dir, address);
 198                 if (pud_none(*pu_dir)) {
 199                         pm_dir = vmem_pmd_alloc();
 200                         if (!pm_dir)
 201                                 goto out;
 202                         pud_populate_kernel(&init_mm, pu_dir, pm_dir);
 203                 }
 204
 205                 pm_dir = pmd_offset(pu_dir, address);
 206                 if (pmd_none(*pm_dir)) {
 207                         pt_dir = vmem_pte_alloc();
 208                         if (!pt_dir)
 209                                 goto out;
 210                         pmd_populate_kernel(&init_mm, pm_dir, pt_dir);
 211                 }
 212
 213                 pt_dir = pte_offset_kernel(pm_dir, address);
 214                 if (pte_none(*pt_dir)) {
 215                         unsigned long new_page;
 216
 217                         new_page =__pa(vmem_alloc_pages(0));
 218                         if (!new_page)
 219                                 goto out;
 220                         pte = pfn_pte(new_page >> PAGE_SHIFT, PAGE_KERNEL);
 221                         *pt_dir = pte;
 222                 }
 223         }
 224         memset(start, 0, nr * sizeof(struct page));
 225         ret = 0;
 226 out:
 227         flush_tlb_kernel_range(start_addr, end_addr);
 228         return ret;
 229 }
 230
 231 /*
 232  * Add memory segment to the segment list if it doesn't overlap with
 233  * an already present segment.
 234  */
 235 static int insert_memory_segment(struct memory_segment *seg)
 236 {
 237         struct memory_segment *tmp;
 238
 239         if (seg->start + seg->size > VMEM_MAX_PHYS ||
 240             seg->start + seg->size < seg->start)
 241                 return -ERANGE;
 242
 243         list_for_each_entry(tmp, &mem_segs, list) {
 244                 if (seg->start >= tmp->start + tmp->size)
 245                         continue;
 246                 if (seg->start + seg->size <= tmp->start)
 247                         continue;
 248                 return -ENOSPC;
 249         }
 250         list_add(&seg->list, &mem_segs);
 251         return 0;
 252 }
 253
 254 /*
 255  * Remove memory segment from the segment list.
 256  */
 257 static void remove_memory_segment(struct memory_segment *seg)
 258 {
 259         list_del(&seg->list);
 260 }
 261
 262 static void __remove_shared_memory(struct memory_segment *seg)
 263 {
 264         remove_memory_segment(seg);
 265         vmem_remove_range(seg->start, seg->size);
 266 }
 267
 268 int vmem_remove_mapping(unsigned long start, unsigned long size)
 269 {
 270         struct memory_segment *seg;
 271         int ret;
 272
 273         mutex_lock(&vmem_mutex);
 274
 275         ret = -ENOENT;
 276         list_for_each_entry(seg, &mem_segs, list) {
 277                 if (seg->start == start && seg->size == size)
 278                         break;
 279         }
 280
 281         if (seg->start != start || seg->size != size)
 282                 goto out;
 283
 284         ret = 0;
 285         __remove_shared_memory(seg);
 286         kfree(seg);
 287 out:
 288         mutex_unlock(&vmem_mutex);
 289         return ret;
 290 }
 291
 292 int vmem_add_mapping(unsigned long start, unsigned long size)
 293 {
 294         struct memory_segment *seg;
 295         int ret;
 296
 297         mutex_lock(&vmem_mutex);
 298         ret = -ENOMEM;
 299         seg = kzalloc(sizeof(*seg), GFP_KERNEL);
 300         if (!seg)
 301                 goto out;
 302         seg->start = start;
 303         seg->size = size;
 304
 305         ret = insert_memory_segment(seg);
 306         if (ret)
 307                 goto out_free;
 308
 309         ret = vmem_add_mem(start, size, 0);
 310         if (ret)
 311                 goto out_remove;
 312         goto out;
 313
 314 out_remove:
 315         __remove_shared_memory(seg);
 316 out_free:
 317         kfree(seg);
 318 out:
 319         mutex_unlock(&vmem_mutex);
 320         return ret;
 321 }
 322
 323 /*
 324  * map whole physical memory to virtual memory (identity mapping)
 325  * we reserve enough space in the vmalloc area for vmemmap to hotplug
 326  * additional memory segments.
 327  */
 328 void __init vmem_map_init(void)
 329 {
 330         unsigned long ro_start, ro_end;
 331         unsigned long start, end;
 332         int i;
 333
 334         spin_lock_init(&init_mm.context.list_lock);
 335         INIT_LIST_HEAD(&init_mm.context.crst_list);
 336         INIT_LIST_HEAD(&init_mm.context.pgtable_list);
 337         init_mm.context.noexec = 0;
 338         ro_start = ((unsigned long)&_stext) & PAGE_MASK;
 339         ro_end = PFN_ALIGN((unsigned long)&_eshared);
 340         for (i = 0; i < MEMORY_CHUNKS && memory_chunk[i].size > 0; i++) {
 341                 start = memory_chunk[i].addr;
 342                 end = memory_chunk[i].addr + memory_chunk[i].size;
 343                 if (start >= ro_end || end <= ro_start)
 344                         vmem_add_mem(start, end - start, 0);
 345                 else if (start >= ro_start && end <= ro_end)
 346                         vmem_add_mem(start, end - start, 1);
 347                 else if (start >= ro_start) {
 348                         vmem_add_mem(start, ro_end - start, 1);
 349                         vmem_add_mem(ro_end, end - ro_end, 0);
 350                 } else if (end < ro_end) {
 351                         vmem_add_mem(start, ro_start - start, 0);
 352                         vmem_add_mem(ro_start, end - ro_start, 1);
 353                 } else {
 354                         vmem_add_mem(start, ro_start - start, 0);
 355                         vmem_add_mem(ro_start, ro_end - ro_start, 1);
 356                         vmem_add_mem(ro_end, end - ro_end, 0);
 357                 }
 358         }
 359 }
 360
 361 /*
 362  * Convert memory chunk array to a memory segment list so there is a single
 363  * list that contains both r/w memory and shared memory segments.
 364  */
 365 static int __init vmem_convert_memory_chunk(void)
 366 {
 367         struct memory_segment *seg;
 368         int i;
 369
 370         mutex_lock(&vmem_mutex);
 371         for (i = 0; i < MEMORY_CHUNKS; i++) {
 372                 if (!memory_chunk[i].size)
 373                         continue;
 374                 seg = kzalloc(sizeof(*seg), GFP_KERNEL);
 375                 if (!seg)
 376                         panic("Out of memory...\n");
 377                 seg->start = memory_chunk[i].addr;
 378                 seg->size = memory_chunk[i].size;
 379                 insert_memory_segment(seg);
 380         }
 381         mutex_unlock(&vmem_mutex);
 382         return 0;
 383 }
 384
 385 core_initcall(vmem_convert_memory_chunk);