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