1 /* 2 * Copyright IBM Corp. 2011 3 * Author(s): Jan Glauber <jang@linux.vnet.ibm.com> 4 */ 5 #include <linux/hugetlb.h> 6 #include <linux/mm.h> 7 #include <asm/cacheflush.h> 8 #include <asm/facility.h> 9 #include <asm/pgtable.h> 10 #include <asm/pgalloc.h> 11 #include <asm/page.h> 12 #include <asm/set_memory.h> 13 14 static inline unsigned long sske_frame(unsigned long addr, unsigned char skey) 15 { 16 asm volatile(".insn rrf,0xb22b0000,%[skey],%[addr],9,0" 17 : [addr] "+a" (addr) : [skey] "d" (skey)); 18 return addr; 19 } 20 21 void __storage_key_init_range(unsigned long start, unsigned long end) 22 { 23 unsigned long boundary, size; 24 25 if (!PAGE_DEFAULT_KEY) 26 return; 27 while (start < end) { 28 if (MACHINE_HAS_EDAT1) { 29 /* set storage keys for a 1MB frame */ 30 size = 1UL << 20; 31 boundary = (start + size) & ~(size - 1); 32 if (boundary <= end) { 33 do { 34 start = sske_frame(start, PAGE_DEFAULT_KEY); 35 } while (start < boundary); 36 continue; 37 } 38 } 39 page_set_storage_key(start, PAGE_DEFAULT_KEY, 0); 40 start += PAGE_SIZE; 41 } 42 } 43 44 #ifdef CONFIG_PROC_FS 45 atomic_long_t direct_pages_count[PG_DIRECT_MAP_MAX]; 46 47 void arch_report_meminfo(struct seq_file *m) 48 { 49 seq_printf(m, "DirectMap4k: %8lu kB\n", 50 atomic_long_read(&direct_pages_count[PG_DIRECT_MAP_4K]) << 2); 51 seq_printf(m, "DirectMap1M: %8lu kB\n", 52 atomic_long_read(&direct_pages_count[PG_DIRECT_MAP_1M]) << 10); 53 seq_printf(m, "DirectMap2G: %8lu kB\n", 54 atomic_long_read(&direct_pages_count[PG_DIRECT_MAP_2G]) << 21); 55 } 56 #endif /* CONFIG_PROC_FS */ 57 58 static void pgt_set(unsigned long *old, unsigned long new, unsigned long addr, 59 unsigned long dtt) 60 { 61 unsigned long table, mask; 62 63 mask = 0; 64 if (MACHINE_HAS_EDAT2) { 65 switch (dtt) { 66 case CRDTE_DTT_REGION3: 67 mask = ~(PTRS_PER_PUD * sizeof(pud_t) - 1); 68 break; 69 case CRDTE_DTT_SEGMENT: 70 mask = ~(PTRS_PER_PMD * sizeof(pmd_t) - 1); 71 break; 72 case CRDTE_DTT_PAGE: 73 mask = ~(PTRS_PER_PTE * sizeof(pte_t) - 1); 74 break; 75 } 76 table = (unsigned long)old & mask; 77 crdte(*old, new, table, dtt, addr, S390_lowcore.kernel_asce); 78 } else if (MACHINE_HAS_IDTE) { 79 cspg(old, *old, new); 80 } else { 81 csp((unsigned int *)old + 1, *old, new); 82 } 83 } 84 85 static int walk_pte_level(pmd_t *pmdp, unsigned long addr, unsigned long end, 86 unsigned long flags) 87 { 88 pte_t *ptep, new; 89 90 ptep = pte_offset(pmdp, addr); 91 do { 92 new = *ptep; 93 if (pte_none(new)) 94 return -EINVAL; 95 if (flags & SET_MEMORY_RO) 96 new = pte_wrprotect(new); 97 else if (flags & SET_MEMORY_RW) 98 new = pte_mkwrite(pte_mkdirty(new)); 99 if (flags & SET_MEMORY_NX) 100 pte_val(new) |= _PAGE_NOEXEC; 101 else if (flags & SET_MEMORY_X) 102 pte_val(new) &= ~_PAGE_NOEXEC; 103 pgt_set((unsigned long *)ptep, pte_val(new), addr, CRDTE_DTT_PAGE); 104 ptep++; 105 addr += PAGE_SIZE; 106 cond_resched(); 107 } while (addr < end); 108 return 0; 109 } 110 111 static int split_pmd_page(pmd_t *pmdp, unsigned long addr) 112 { 113 unsigned long pte_addr, prot; 114 pte_t *pt_dir, *ptep; 115 pmd_t new; 116 int i, ro, nx; 117 118 pt_dir = vmem_pte_alloc(); 119 if (!pt_dir) 120 return -ENOMEM; 121 pte_addr = pmd_pfn(*pmdp) << PAGE_SHIFT; 122 ro = !!(pmd_val(*pmdp) & _SEGMENT_ENTRY_PROTECT); 123 nx = !!(pmd_val(*pmdp) & _SEGMENT_ENTRY_NOEXEC); 124 prot = pgprot_val(ro ? PAGE_KERNEL_RO : PAGE_KERNEL); 125 if (!nx) 126 prot &= ~_PAGE_NOEXEC; 127 ptep = pt_dir; 128 for (i = 0; i < PTRS_PER_PTE; i++) { 129 pte_val(*ptep) = pte_addr | prot; 130 pte_addr += PAGE_SIZE; 131 ptep++; 132 } 133 pmd_val(new) = __pa(pt_dir) | _SEGMENT_ENTRY; 134 pgt_set((unsigned long *)pmdp, pmd_val(new), addr, CRDTE_DTT_SEGMENT); 135 update_page_count(PG_DIRECT_MAP_4K, PTRS_PER_PTE); 136 update_page_count(PG_DIRECT_MAP_1M, -1); 137 return 0; 138 } 139 140 static void modify_pmd_page(pmd_t *pmdp, unsigned long addr, 141 unsigned long flags) 142 { 143 pmd_t new = *pmdp; 144 145 if (flags & SET_MEMORY_RO) 146 new = pmd_wrprotect(new); 147 else if (flags & SET_MEMORY_RW) 148 new = pmd_mkwrite(pmd_mkdirty(new)); 149 if (flags & SET_MEMORY_NX) 150 pmd_val(new) |= _SEGMENT_ENTRY_NOEXEC; 151 else if (flags & SET_MEMORY_X) 152 pmd_val(new) &= ~_SEGMENT_ENTRY_NOEXEC; 153 pgt_set((unsigned long *)pmdp, pmd_val(new), addr, CRDTE_DTT_SEGMENT); 154 } 155 156 static int walk_pmd_level(pud_t *pudp, unsigned long addr, unsigned long end, 157 unsigned long flags) 158 { 159 unsigned long next; 160 pmd_t *pmdp; 161 int rc = 0; 162 163 pmdp = pmd_offset(pudp, addr); 164 do { 165 if (pmd_none(*pmdp)) 166 return -EINVAL; 167 next = pmd_addr_end(addr, end); 168 if (pmd_large(*pmdp)) { 169 if (addr & ~PMD_MASK || addr + PMD_SIZE > next) { 170 rc = split_pmd_page(pmdp, addr); 171 if (rc) 172 return rc; 173 continue; 174 } 175 modify_pmd_page(pmdp, addr, flags); 176 } else { 177 rc = walk_pte_level(pmdp, addr, next, flags); 178 if (rc) 179 return rc; 180 } 181 pmdp++; 182 addr = next; 183 cond_resched(); 184 } while (addr < end); 185 return rc; 186 } 187 188 static int split_pud_page(pud_t *pudp, unsigned long addr) 189 { 190 unsigned long pmd_addr, prot; 191 pmd_t *pm_dir, *pmdp; 192 pud_t new; 193 int i, ro, nx; 194 195 pm_dir = vmem_crst_alloc(_SEGMENT_ENTRY_EMPTY); 196 if (!pm_dir) 197 return -ENOMEM; 198 pmd_addr = pud_pfn(*pudp) << PAGE_SHIFT; 199 ro = !!(pud_val(*pudp) & _REGION_ENTRY_PROTECT); 200 nx = !!(pud_val(*pudp) & _REGION_ENTRY_NOEXEC); 201 prot = pgprot_val(ro ? SEGMENT_KERNEL_RO : SEGMENT_KERNEL); 202 if (!nx) 203 prot &= ~_SEGMENT_ENTRY_NOEXEC; 204 pmdp = pm_dir; 205 for (i = 0; i < PTRS_PER_PMD; i++) { 206 pmd_val(*pmdp) = pmd_addr | prot; 207 pmd_addr += PMD_SIZE; 208 pmdp++; 209 } 210 pud_val(new) = __pa(pm_dir) | _REGION3_ENTRY; 211 pgt_set((unsigned long *)pudp, pud_val(new), addr, CRDTE_DTT_REGION3); 212 update_page_count(PG_DIRECT_MAP_1M, PTRS_PER_PMD); 213 update_page_count(PG_DIRECT_MAP_2G, -1); 214 return 0; 215 } 216 217 static void modify_pud_page(pud_t *pudp, unsigned long addr, 218 unsigned long flags) 219 { 220 pud_t new = *pudp; 221 222 if (flags & SET_MEMORY_RO) 223 new = pud_wrprotect(new); 224 else if (flags & SET_MEMORY_RW) 225 new = pud_mkwrite(pud_mkdirty(new)); 226 if (flags & SET_MEMORY_NX) 227 pud_val(new) |= _REGION_ENTRY_NOEXEC; 228 else if (flags & SET_MEMORY_X) 229 pud_val(new) &= ~_REGION_ENTRY_NOEXEC; 230 pgt_set((unsigned long *)pudp, pud_val(new), addr, CRDTE_DTT_REGION3); 231 } 232 233 static int walk_pud_level(p4d_t *p4d, unsigned long addr, unsigned long end, 234 unsigned long flags) 235 { 236 unsigned long next; 237 pud_t *pudp; 238 int rc = 0; 239 240 pudp = pud_offset(p4d, addr); 241 do { 242 if (pud_none(*pudp)) 243 return -EINVAL; 244 next = pud_addr_end(addr, end); 245 if (pud_large(*pudp)) { 246 if (addr & ~PUD_MASK || addr + PUD_SIZE > next) { 247 rc = split_pud_page(pudp, addr); 248 if (rc) 249 break; 250 continue; 251 } 252 modify_pud_page(pudp, addr, flags); 253 } else { 254 rc = walk_pmd_level(pudp, addr, next, flags); 255 } 256 pudp++; 257 addr = next; 258 cond_resched(); 259 } while (addr < end && !rc); 260 return rc; 261 } 262 263 static int walk_p4d_level(pgd_t *pgd, unsigned long addr, unsigned long end, 264 unsigned long flags) 265 { 266 unsigned long next; 267 p4d_t *p4dp; 268 int rc = 0; 269 270 p4dp = p4d_offset(pgd, addr); 271 do { 272 if (p4d_none(*p4dp)) 273 return -EINVAL; 274 next = p4d_addr_end(addr, end); 275 rc = walk_pud_level(p4dp, addr, next, flags); 276 p4dp++; 277 addr = next; 278 cond_resched(); 279 } while (addr < end && !rc); 280 return rc; 281 } 282 283 static DEFINE_MUTEX(cpa_mutex); 284 285 static int change_page_attr(unsigned long addr, unsigned long end, 286 unsigned long flags) 287 { 288 unsigned long next; 289 int rc = -EINVAL; 290 pgd_t *pgdp; 291 292 if (addr == end) 293 return 0; 294 if (end >= MODULES_END) 295 return -EINVAL; 296 mutex_lock(&cpa_mutex); 297 pgdp = pgd_offset_k(addr); 298 do { 299 if (pgd_none(*pgdp)) 300 break; 301 next = pgd_addr_end(addr, end); 302 rc = walk_p4d_level(pgdp, addr, next, flags); 303 if (rc) 304 break; 305 cond_resched(); 306 } while (pgdp++, addr = next, addr < end && !rc); 307 mutex_unlock(&cpa_mutex); 308 return rc; 309 } 310 311 int __set_memory(unsigned long addr, int numpages, unsigned long flags) 312 { 313 if (!MACHINE_HAS_NX) 314 flags &= ~(SET_MEMORY_NX | SET_MEMORY_X); 315 if (!flags) 316 return 0; 317 addr &= PAGE_MASK; 318 return change_page_attr(addr, addr + numpages * PAGE_SIZE, flags); 319 } 320 321 #ifdef CONFIG_DEBUG_PAGEALLOC 322 323 static void ipte_range(pte_t *pte, unsigned long address, int nr) 324 { 325 int i; 326 327 if (test_facility(13)) { 328 __ptep_ipte_range(address, nr - 1, pte, IPTE_GLOBAL); 329 return; 330 } 331 for (i = 0; i < nr; i++) { 332 __ptep_ipte(address, pte, 0, 0, IPTE_GLOBAL); 333 address += PAGE_SIZE; 334 pte++; 335 } 336 } 337 338 void __kernel_map_pages(struct page *page, int numpages, int enable) 339 { 340 unsigned long address; 341 int nr, i, j; 342 pgd_t *pgd; 343 p4d_t *p4d; 344 pud_t *pud; 345 pmd_t *pmd; 346 pte_t *pte; 347 348 for (i = 0; i < numpages;) { 349 address = page_to_phys(page + i); 350 pgd = pgd_offset_k(address); 351 p4d = p4d_offset(pgd, address); 352 pud = pud_offset(p4d, address); 353 pmd = pmd_offset(pud, address); 354 pte = pte_offset_kernel(pmd, address); 355 nr = (unsigned long)pte >> ilog2(sizeof(long)); 356 nr = PTRS_PER_PTE - (nr & (PTRS_PER_PTE - 1)); 357 nr = min(numpages - i, nr); 358 if (enable) { 359 for (j = 0; j < nr; j++) { 360 pte_val(*pte) &= ~_PAGE_INVALID; 361 address += PAGE_SIZE; 362 pte++; 363 } 364 } else { 365 ipte_range(pte, address, nr); 366 } 367 i += nr; 368 } 369 } 370 371 #ifdef CONFIG_HIBERNATION 372 bool kernel_page_present(struct page *page) 373 { 374 unsigned long addr; 375 int cc; 376 377 addr = page_to_phys(page); 378 asm volatile( 379 " lra %1,0(%1)\n" 380 " ipm %0\n" 381 " srl %0,28" 382 : "=d" (cc), "+a" (addr) : : "cc"); 383 return cc == 0; 384 } 385 #endif /* CONFIG_HIBERNATION */ 386 387 #endif /* CONFIG_DEBUG_PAGEALLOC */ 388