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 <linux/slab.h>
  15 #include <asm/pgalloc.h>
  16 #include <asm/pgtable.h>
  17 #include <asm/setup.h>
  18 #include <asm/tlbflush.h>
  19 #include <asm/sections.h>
  20
  21 static DEFINE_MUTEX(vmem_mutex);
  22
  23 struct memory_segment {
  24         struct list_head list;
  25         unsigned long start;
  26         unsigned long size;
  27 };
  28
  29 static LIST_HEAD(mem_segs);
  30
  31 static void __ref *vmem_alloc_pages(unsigned int order)
  32 {
  33         if (slab_is_available())
  34                 return (void *)__get_free_pages(GFP_KERNEL, order);
  35         return alloc_bootmem_pages((1 << order) * PAGE_SIZE);
  36 }
  37
  38 static inline pud_t *vmem_pud_alloc(void)
  39 {
  40         pud_t *pud = NULL;
  41
  42 #ifdef CONFIG_64BIT
  43         pud = vmem_alloc_pages(2);
  44         if (!pud)
  45                 return NULL;
  46         clear_table((unsigned long *) pud, _REGION3_ENTRY_EMPTY, PAGE_SIZE * 4);
  47 #endif
  48         return pud;
  49 }
  50
  51 static inline pmd_t *vmem_pmd_alloc(void)
  52 {
  53         pmd_t *pmd = NULL;
  54
  55 #ifdef CONFIG_64BIT
  56         pmd = vmem_alloc_pages(2);
  57         if (!pmd)
  58                 return NULL;
  59         clear_table((unsigned long *) pmd, _SEGMENT_ENTRY_EMPTY, PAGE_SIZE * 4);
  60 #endif
  61         return pmd;
  62 }
  63
  64 static pte_t __ref *vmem_pte_alloc(unsigned long address)
  65 {
  66         pte_t *pte;
  67
  68         if (slab_is_available())
  69                 pte = (pte_t *) page_table_alloc(&init_mm, address);
  70         else
  71                 pte = alloc_bootmem(PTRS_PER_PTE * sizeof(pte_t));
  72         if (!pte)
  73                 return NULL;
  74         clear_table((unsigned long *) pte, _PAGE_TYPE_EMPTY,
  75                     PTRS_PER_PTE * sizeof(pte_t));
  76         return pte;
  77 }
  78
  79 /*
  80  * Add a physical memory range to the 1:1 mapping.
  81  */
  82 static int vmem_add_mem(unsigned long start, unsigned long size, int ro)
  83 {
  84         unsigned long address;
  85         pgd_t *pg_dir;
  86         pud_t *pu_dir;
  87         pmd_t *pm_dir;
  88         pte_t *pt_dir;
  89         pte_t  pte;
  90         int ret = -ENOMEM;
  91
  92         for (address = start; address < start + size; address += PAGE_SIZE) {
  93                 pg_dir = pgd_offset_k(address);
  94                 if (pgd_none(*pg_dir)) {
  95                         pu_dir = vmem_pud_alloc();
  96                         if (!pu_dir)
  97                                 goto out;
  98                         pgd_populate(&init_mm, pg_dir, pu_dir);
  99                 }
 100
 101                 pu_dir = pud_offset(pg_dir, address);
 102                 if (pud_none(*pu_dir)) {
 103                         pm_dir = vmem_pmd_alloc();
 104                         if (!pm_dir)
 105                                 goto out;
 106                         pud_populate(&init_mm, pu_dir, pm_dir);
 107                 }
 108
 109                 pte = mk_pte_phys(address, __pgprot(ro ? _PAGE_RO : 0));
 110                 pm_dir = pmd_offset(pu_dir, address);
 111
 112 #ifdef __s390x__
 113                 if (MACHINE_HAS_HPAGE && !(address & ~HPAGE_MASK) &&
 114                     (address + HPAGE_SIZE <= start + size) &&
 115                     (address >= HPAGE_SIZE)) {
 116                         pte_val(pte) |= _SEGMENT_ENTRY_LARGE;
 117                         pmd_val(*pm_dir) = pte_val(pte);
 118                         address += HPAGE_SIZE - PAGE_SIZE;
 119                         continue;
 120                 }
 121 #endif
 122                 if (pmd_none(*pm_dir)) {
 123                         pt_dir = vmem_pte_alloc(address);
 124                         if (!pt_dir)
 125                                 goto out;
 126                         pmd_populate(&init_mm, pm_dir, pt_dir);
 127                 }
 128
 129                 pt_dir = pte_offset_kernel(pm_dir, address);
 130                 *pt_dir = pte;
 131         }
 132         ret = 0;
 133 out:
 134         flush_tlb_kernel_range(start, start + size);
 135         return ret;
 136 }
 137
 138 /*
 139  * Remove a physical memory range from the 1:1 mapping.
 140  * Currently only invalidates page table entries.
 141  */
 142 static void vmem_remove_range(unsigned long start, unsigned long size)
 143 {
 144         unsigned long address;
 145         pgd_t *pg_dir;
 146         pud_t *pu_dir;
 147         pmd_t *pm_dir;
 148         pte_t *pt_dir;
 149         pte_t  pte;
 150
 151         pte_val(pte) = _PAGE_TYPE_EMPTY;
 152         for (address = start; address < start + size; address += PAGE_SIZE) {
 153                 pg_dir = pgd_offset_k(address);
 154                 pu_dir = pud_offset(pg_dir, address);
 155                 if (pud_none(*pu_dir))
 156                         continue;
 157                 pm_dir = pmd_offset(pu_dir, address);
 158                 if (pmd_none(*pm_dir))
 159                         continue;
 160
 161                 if (pmd_huge(*pm_dir)) {
 162                         pmd_clear(pm_dir);
 163                         address += HPAGE_SIZE - PAGE_SIZE;
 164                         continue;
 165                 }
 166
 167                 pt_dir = pte_offset_kernel(pm_dir, address);
 168                 *pt_dir = pte;
 169         }
 170         flush_tlb_kernel_range(start, start + size);
 171 }
 172
 173 /*
 174  * Add a backed mem_map array to the virtual mem_map array.
 175  */
 176 int __meminit vmemmap_populate(struct page *start, unsigned long nr, int node)
 177 {
 178         unsigned long address, start_addr, end_addr;
 179         pgd_t *pg_dir;
 180         pud_t *pu_dir;
 181         pmd_t *pm_dir;
 182         pte_t *pt_dir;
 183         pte_t  pte;
 184         int ret = -ENOMEM;
 185
 186         start_addr = (unsigned long) start;
 187         end_addr = (unsigned long) (start + nr);
 188
 189         for (address = start_addr; address < end_addr; address += PAGE_SIZE) {
 190                 pg_dir = pgd_offset_k(address);
 191                 if (pgd_none(*pg_dir)) {
 192                         pu_dir = vmem_pud_alloc();
 193                         if (!pu_dir)
 194                                 goto out;
 195                         pgd_populate(&init_mm, pg_dir, pu_dir);
 196                 }
 197
 198                 pu_dir = pud_offset(pg_dir, address);
 199                 if (pud_none(*pu_dir)) {
 200                         pm_dir = vmem_pmd_alloc();
 201                         if (!pm_dir)
 202                                 goto out;
 203                         pud_populate(&init_mm, pu_dir, pm_dir);
 204                 }
 205
 206                 pm_dir = pmd_offset(pu_dir, address);
 207                 if (pmd_none(*pm_dir)) {
 208                         pt_dir = vmem_pte_alloc(address);
 209                         if (!pt_dir)
 210                                 goto out;
 211                         pmd_populate(&init_mm, pm_dir, pt_dir);
 212                 }
 213
 214                 pt_dir = pte_offset_kernel(pm_dir, address);
 215                 if (pte_none(*pt_dir)) {
 216                         unsigned long new_page;
 217
 218                         new_page =__pa(vmem_alloc_pages(0));
 219                         if (!new_page)
 220                                 goto out;
 221                         pte = pfn_pte(new_page >> PAGE_SHIFT, PAGE_KERNEL);
 222                         *pt_dir = pte;
 223                 }
 224         }
 225         memset(start, 0, nr * sizeof(struct page));
 226         ret = 0;
 227 out:
 228         flush_tlb_kernel_range(start_addr, end_addr);
 229         return ret;
 230 }
 231
 232 /*
 233  * Add memory segment to the segment list if it doesn't overlap with
 234  * an already present segment.
 235  */
 236 static int insert_memory_segment(struct memory_segment *seg)
 237 {
 238         struct memory_segment *tmp;
 239
 240         if (seg->start + seg->size > VMEM_MAX_PHYS ||
 241             seg->start + seg->size < seg->start)
 242                 return -ERANGE;
 243
 244         list_for_each_entry(tmp, &mem_segs, list) {
 245                 if (seg->start >= tmp->start + tmp->size)
 246                         continue;
 247                 if (seg->start + seg->size <= tmp->start)
 248                         continue;
 249                 return -ENOSPC;
 250         }
 251         list_add(&seg->list, &mem_segs);
 252         return 0;
 253 }
 254
 255 /*
 256  * Remove memory segment from the segment list.
 257  */
 258 static void remove_memory_segment(struct memory_segment *seg)
 259 {
 260         list_del(&seg->list);
 261 }
 262
 263 static void __remove_shared_memory(struct memory_segment *seg)
 264 {
 265         remove_memory_segment(seg);
 266         vmem_remove_range(seg->start, seg->size);
 267 }
 268
 269 int vmem_remove_mapping(unsigned long start, unsigned long size)
 270 {
 271         struct memory_segment *seg;
 272         int ret;
 273
 274         mutex_lock(&vmem_mutex);
 275
 276         ret = -ENOENT;
 277         list_for_each_entry(seg, &mem_segs, list) {
 278                 if (seg->start == start && seg->size == size)
 279                         break;
 280         }
 281
 282         if (seg->start != start || seg->size != size)
 283                 goto out;
 284
 285         ret = 0;
 286         __remove_shared_memory(seg);
 287         kfree(seg);
 288 out:
 289         mutex_unlock(&vmem_mutex);
 290         return ret;
 291 }
 292
 293 int vmem_add_mapping(unsigned long start, unsigned long size)
 294 {
 295         struct memory_segment *seg;
 296         int ret;
 297
 298         mutex_lock(&vmem_mutex);
 299         ret = -ENOMEM;
 300         seg = kzalloc(sizeof(*seg), GFP_KERNEL);
 301         if (!seg)
 302                 goto out;
 303         seg->start = start;
 304         seg->size = size;
 305
 306         ret = insert_memory_segment(seg);
 307         if (ret)
 308                 goto out_free;
 309
 310         ret = vmem_add_mem(start, size, 0);
 311         if (ret)
 312                 goto out_remove;
 313         goto out;
 314
 315 out_remove:
 316         __remove_shared_memory(seg);
 317 out_free:
 318         kfree(seg);
 319 out:
 320         mutex_unlock(&vmem_mutex);
 321         return ret;
 322 }
 323
 324 /*
 325  * map whole physical memory to virtual memory (identity mapping)
 326  * we reserve enough space in the vmalloc area for vmemmap to hotplug
 327  * additional memory segments.
 328  */
 329 void __init vmem_map_init(void)
 330 {
 331         unsigned long ro_start, ro_end;
 332         unsigned long start, end;
 333         int i;
 334
 335         ro_start = ((unsigned long)&_stext) & PAGE_MASK;
 336         ro_end = PFN_ALIGN((unsigned long)&_eshared);
 337         for (i = 0; i < MEMORY_CHUNKS && memory_chunk[i].size > 0; i++) {
 338                 if (memory_chunk[i].type == CHUNK_CRASHK ||
 339                     memory_chunk[i].type == CHUNK_OLDMEM)
 340                         continue;
 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                 if (memory_chunk[i].type == CHUNK_CRASHK ||
 375                     memory_chunk[i].type == CHUNK_OLDMEM)
 376                         continue;
 377                 seg = kzalloc(sizeof(*seg), GFP_KERNEL);
 378                 if (!seg)
 379                         panic("Out of memory...\n");
 380                 seg->start = memory_chunk[i].addr;
 381                 seg->size = memory_chunk[i].size;
 382                 insert_memory_segment(seg);
 383         }
 384         mutex_unlock(&vmem_mutex);
 385         return 0;
 386 }
 387
 388 core_initcall(vmem_convert_memory_chunk);