1 /* 2 * Copyright IBM Corp. 2006 3 * Author(s): Heiko Carstens <heiko.carstens@de.ibm.com> 4 */ 5 6 #include <linux/bootmem.h> 7 #include <linux/pfn.h> 8 #include <linux/mm.h> 9 #include <linux/init.h> 10 #include <linux/list.h> 11 #include <linux/hugetlb.h> 12 #include <linux/slab.h> 13 #include <linux/memblock.h> 14 #include <asm/cacheflush.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 #include <asm/set_memory.h> 21 22 static DEFINE_MUTEX(vmem_mutex); 23 24 struct memory_segment { 25 struct list_head list; 26 unsigned long start; 27 unsigned long size; 28 }; 29 30 static LIST_HEAD(mem_segs); 31 32 static void __ref *vmem_alloc_pages(unsigned int order) 33 { 34 unsigned long size = PAGE_SIZE << order; 35 36 if (slab_is_available()) 37 return (void *)__get_free_pages(GFP_KERNEL, order); 38 return (void *) memblock_alloc(size, size); 39 } 40 41 void *vmem_crst_alloc(unsigned long val) 42 { 43 unsigned long *table; 44 45 table = vmem_alloc_pages(CRST_ALLOC_ORDER); 46 if (table) 47 crst_table_init(table, val); 48 return table; 49 } 50 51 pte_t __ref *vmem_pte_alloc(void) 52 { 53 unsigned long size = PTRS_PER_PTE * sizeof(pte_t); 54 pte_t *pte; 55 56 if (slab_is_available()) 57 pte = (pte_t *) page_table_alloc(&init_mm); 58 else 59 pte = (pte_t *) memblock_alloc(size, size); 60 if (!pte) 61 return NULL; 62 clear_table((unsigned long *) pte, _PAGE_INVALID, size); 63 return pte; 64 } 65 66 /* 67 * Add a physical memory range to the 1:1 mapping. 68 */ 69 static int vmem_add_mem(unsigned long start, unsigned long size) 70 { 71 unsigned long pgt_prot, sgt_prot, r3_prot; 72 unsigned long pages4k, pages1m, pages2g; 73 unsigned long end = start + size; 74 unsigned long address = start; 75 pgd_t *pg_dir; 76 p4d_t *p4_dir; 77 pud_t *pu_dir; 78 pmd_t *pm_dir; 79 pte_t *pt_dir; 80 int ret = -ENOMEM; 81 82 pgt_prot = pgprot_val(PAGE_KERNEL); 83 sgt_prot = pgprot_val(SEGMENT_KERNEL); 84 r3_prot = pgprot_val(REGION3_KERNEL); 85 if (!MACHINE_HAS_NX) { 86 pgt_prot &= ~_PAGE_NOEXEC; 87 sgt_prot &= ~_SEGMENT_ENTRY_NOEXEC; 88 r3_prot &= ~_REGION_ENTRY_NOEXEC; 89 } 90 pages4k = pages1m = pages2g = 0; 91 while (address < end) { 92 pg_dir = pgd_offset_k(address); 93 if (pgd_none(*pg_dir)) { 94 p4_dir = vmem_crst_alloc(_REGION2_ENTRY_EMPTY); 95 if (!p4_dir) 96 goto out; 97 pgd_populate(&init_mm, pg_dir, p4_dir); 98 } 99 p4_dir = p4d_offset(pg_dir, address); 100 if (p4d_none(*p4_dir)) { 101 pu_dir = vmem_crst_alloc(_REGION3_ENTRY_EMPTY); 102 if (!pu_dir) 103 goto out; 104 p4d_populate(&init_mm, p4_dir, pu_dir); 105 } 106 pu_dir = pud_offset(p4_dir, address); 107 if (MACHINE_HAS_EDAT2 && pud_none(*pu_dir) && address && 108 !(address & ~PUD_MASK) && (address + PUD_SIZE <= end) && 109 !debug_pagealloc_enabled()) { 110 pud_val(*pu_dir) = address | r3_prot; 111 address += PUD_SIZE; 112 pages2g++; 113 continue; 114 } 115 if (pud_none(*pu_dir)) { 116 pm_dir = vmem_crst_alloc(_SEGMENT_ENTRY_EMPTY); 117 if (!pm_dir) 118 goto out; 119 pud_populate(&init_mm, pu_dir, pm_dir); 120 } 121 pm_dir = pmd_offset(pu_dir, address); 122 if (MACHINE_HAS_EDAT1 && pmd_none(*pm_dir) && address && 123 !(address & ~PMD_MASK) && (address + PMD_SIZE <= end) && 124 !debug_pagealloc_enabled()) { 125 pmd_val(*pm_dir) = address | sgt_prot; 126 address += PMD_SIZE; 127 pages1m++; 128 continue; 129 } 130 if (pmd_none(*pm_dir)) { 131 pt_dir = vmem_pte_alloc(); 132 if (!pt_dir) 133 goto out; 134 pmd_populate(&init_mm, pm_dir, pt_dir); 135 } 136 137 pt_dir = pte_offset_kernel(pm_dir, address); 138 pte_val(*pt_dir) = address | pgt_prot; 139 address += PAGE_SIZE; 140 pages4k++; 141 } 142 ret = 0; 143 out: 144 update_page_count(PG_DIRECT_MAP_4K, pages4k); 145 update_page_count(PG_DIRECT_MAP_1M, pages1m); 146 update_page_count(PG_DIRECT_MAP_2G, pages2g); 147 return ret; 148 } 149 150 /* 151 * Remove a physical memory range from the 1:1 mapping. 152 * Currently only invalidates page table entries. 153 */ 154 static void vmem_remove_range(unsigned long start, unsigned long size) 155 { 156 unsigned long pages4k, pages1m, pages2g; 157 unsigned long end = start + size; 158 unsigned long address = start; 159 pgd_t *pg_dir; 160 p4d_t *p4_dir; 161 pud_t *pu_dir; 162 pmd_t *pm_dir; 163 pte_t *pt_dir; 164 165 pages4k = pages1m = pages2g = 0; 166 while (address < end) { 167 pg_dir = pgd_offset_k(address); 168 if (pgd_none(*pg_dir)) { 169 address += PGDIR_SIZE; 170 continue; 171 } 172 p4_dir = p4d_offset(pg_dir, address); 173 if (p4d_none(*p4_dir)) { 174 address += P4D_SIZE; 175 continue; 176 } 177 pu_dir = pud_offset(p4_dir, address); 178 if (pud_none(*pu_dir)) { 179 address += PUD_SIZE; 180 continue; 181 } 182 if (pud_large(*pu_dir)) { 183 pud_clear(pu_dir); 184 address += PUD_SIZE; 185 pages2g++; 186 continue; 187 } 188 pm_dir = pmd_offset(pu_dir, address); 189 if (pmd_none(*pm_dir)) { 190 address += PMD_SIZE; 191 continue; 192 } 193 if (pmd_large(*pm_dir)) { 194 pmd_clear(pm_dir); 195 address += PMD_SIZE; 196 pages1m++; 197 continue; 198 } 199 pt_dir = pte_offset_kernel(pm_dir, address); 200 pte_clear(&init_mm, address, pt_dir); 201 address += PAGE_SIZE; 202 pages4k++; 203 } 204 flush_tlb_kernel_range(start, end); 205 update_page_count(PG_DIRECT_MAP_4K, -pages4k); 206 update_page_count(PG_DIRECT_MAP_1M, -pages1m); 207 update_page_count(PG_DIRECT_MAP_2G, -pages2g); 208 } 209 210 /* 211 * Add a backed mem_map array to the virtual mem_map array. 212 */ 213 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node) 214 { 215 unsigned long pgt_prot, sgt_prot; 216 unsigned long address = start; 217 pgd_t *pg_dir; 218 p4d_t *p4_dir; 219 pud_t *pu_dir; 220 pmd_t *pm_dir; 221 pte_t *pt_dir; 222 int ret = -ENOMEM; 223 224 pgt_prot = pgprot_val(PAGE_KERNEL); 225 sgt_prot = pgprot_val(SEGMENT_KERNEL); 226 if (!MACHINE_HAS_NX) { 227 pgt_prot &= ~_PAGE_NOEXEC; 228 sgt_prot &= ~_SEGMENT_ENTRY_NOEXEC; 229 } 230 for (address = start; address < end;) { 231 pg_dir = pgd_offset_k(address); 232 if (pgd_none(*pg_dir)) { 233 p4_dir = vmem_crst_alloc(_REGION2_ENTRY_EMPTY); 234 if (!p4_dir) 235 goto out; 236 pgd_populate(&init_mm, pg_dir, p4_dir); 237 } 238 239 p4_dir = p4d_offset(pg_dir, address); 240 if (p4d_none(*p4_dir)) { 241 pu_dir = vmem_crst_alloc(_REGION3_ENTRY_EMPTY); 242 if (!pu_dir) 243 goto out; 244 p4d_populate(&init_mm, p4_dir, pu_dir); 245 } 246 247 pu_dir = pud_offset(p4_dir, address); 248 if (pud_none(*pu_dir)) { 249 pm_dir = vmem_crst_alloc(_SEGMENT_ENTRY_EMPTY); 250 if (!pm_dir) 251 goto out; 252 pud_populate(&init_mm, pu_dir, pm_dir); 253 } 254 255 pm_dir = pmd_offset(pu_dir, address); 256 if (pmd_none(*pm_dir)) { 257 /* Use 1MB frames for vmemmap if available. We always 258 * use large frames even if they are only partially 259 * used. 260 * Otherwise we would have also page tables since 261 * vmemmap_populate gets called for each section 262 * separately. */ 263 if (MACHINE_HAS_EDAT1) { 264 void *new_page; 265 266 new_page = vmemmap_alloc_block(PMD_SIZE, node); 267 if (!new_page) 268 goto out; 269 pmd_val(*pm_dir) = __pa(new_page) | sgt_prot; 270 address = (address + PMD_SIZE) & PMD_MASK; 271 continue; 272 } 273 pt_dir = vmem_pte_alloc(); 274 if (!pt_dir) 275 goto out; 276 pmd_populate(&init_mm, pm_dir, pt_dir); 277 } else if (pmd_large(*pm_dir)) { 278 address = (address + PMD_SIZE) & PMD_MASK; 279 continue; 280 } 281 282 pt_dir = pte_offset_kernel(pm_dir, address); 283 if (pte_none(*pt_dir)) { 284 void *new_page; 285 286 new_page = vmemmap_alloc_block(PAGE_SIZE, node); 287 if (!new_page) 288 goto out; 289 pte_val(*pt_dir) = __pa(new_page) | pgt_prot; 290 } 291 address += PAGE_SIZE; 292 } 293 ret = 0; 294 out: 295 return ret; 296 } 297 298 void vmemmap_free(unsigned long start, unsigned long end) 299 { 300 } 301 302 /* 303 * Add memory segment to the segment list if it doesn't overlap with 304 * an already present segment. 305 */ 306 static int insert_memory_segment(struct memory_segment *seg) 307 { 308 struct memory_segment *tmp; 309 310 if (seg->start + seg->size > VMEM_MAX_PHYS || 311 seg->start + seg->size < seg->start) 312 return -ERANGE; 313 314 list_for_each_entry(tmp, &mem_segs, list) { 315 if (seg->start >= tmp->start + tmp->size) 316 continue; 317 if (seg->start + seg->size <= tmp->start) 318 continue; 319 return -ENOSPC; 320 } 321 list_add(&seg->list, &mem_segs); 322 return 0; 323 } 324 325 /* 326 * Remove memory segment from the segment list. 327 */ 328 static void remove_memory_segment(struct memory_segment *seg) 329 { 330 list_del(&seg->list); 331 } 332 333 static void __remove_shared_memory(struct memory_segment *seg) 334 { 335 remove_memory_segment(seg); 336 vmem_remove_range(seg->start, seg->size); 337 } 338 339 int vmem_remove_mapping(unsigned long start, unsigned long size) 340 { 341 struct memory_segment *seg; 342 int ret; 343 344 mutex_lock(&vmem_mutex); 345 346 ret = -ENOENT; 347 list_for_each_entry(seg, &mem_segs, list) { 348 if (seg->start == start && seg->size == size) 349 break; 350 } 351 352 if (seg->start != start || seg->size != size) 353 goto out; 354 355 ret = 0; 356 __remove_shared_memory(seg); 357 kfree(seg); 358 out: 359 mutex_unlock(&vmem_mutex); 360 return ret; 361 } 362 363 int vmem_add_mapping(unsigned long start, unsigned long size) 364 { 365 struct memory_segment *seg; 366 int ret; 367 368 mutex_lock(&vmem_mutex); 369 ret = -ENOMEM; 370 seg = kzalloc(sizeof(*seg), GFP_KERNEL); 371 if (!seg) 372 goto out; 373 seg->start = start; 374 seg->size = size; 375 376 ret = insert_memory_segment(seg); 377 if (ret) 378 goto out_free; 379 380 ret = vmem_add_mem(start, size); 381 if (ret) 382 goto out_remove; 383 goto out; 384 385 out_remove: 386 __remove_shared_memory(seg); 387 out_free: 388 kfree(seg); 389 out: 390 mutex_unlock(&vmem_mutex); 391 return ret; 392 } 393 394 /* 395 * map whole physical memory to virtual memory (identity mapping) 396 * we reserve enough space in the vmalloc area for vmemmap to hotplug 397 * additional memory segments. 398 */ 399 void __init vmem_map_init(void) 400 { 401 struct memblock_region *reg; 402 403 for_each_memblock(memory, reg) 404 vmem_add_mem(reg->base, reg->size); 405 __set_memory((unsigned long) _stext, 406 (_etext - _stext) >> PAGE_SHIFT, 407 SET_MEMORY_RO | SET_MEMORY_X); 408 __set_memory((unsigned long) _etext, 409 (_eshared - _etext) >> PAGE_SHIFT, 410 SET_MEMORY_RO); 411 __set_memory((unsigned long) _sinittext, 412 (_einittext - _sinittext) >> PAGE_SHIFT, 413 SET_MEMORY_RO | SET_MEMORY_X); 414 pr_info("Write protected kernel read-only data: %luk\n", 415 (_eshared - _stext) >> 10); 416 } 417 418 /* 419 * Convert memblock.memory to a memory segment list so there is a single 420 * list that contains all memory segments. 421 */ 422 static int __init vmem_convert_memory_chunk(void) 423 { 424 struct memblock_region *reg; 425 struct memory_segment *seg; 426 427 mutex_lock(&vmem_mutex); 428 for_each_memblock(memory, reg) { 429 seg = kzalloc(sizeof(*seg), GFP_KERNEL); 430 if (!seg) 431 panic("Out of memory...\n"); 432 seg->start = reg->base; 433 seg->size = reg->size; 434 insert_memory_segment(seg); 435 } 436 mutex_unlock(&vmem_mutex); 437 return 0; 438 } 439 440 core_initcall(vmem_convert_memory_chunk); 441