1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Copyright IBM Corp. 2011 4 * Author(s): Jan Glauber <jang@linux.vnet.ibm.com> 5 */ 6 #include <linux/hugetlb.h> 7 #include <linux/mm.h> 8 #include <asm/cacheflush.h> 9 #include <asm/facility.h> 10 #include <asm/pgalloc.h> 11 #include <asm/kfence.h> 12 #include <asm/page.h> 13 #include <asm/set_memory.h> 14 15 static inline unsigned long sske_frame(unsigned long addr, unsigned char skey) 16 { 17 asm volatile(".insn rrf,0xb22b0000,%[skey],%[addr],1,0" 18 : [addr] "+a" (addr) : [skey] "d" (skey)); 19 return addr; 20 } 21 22 void __storage_key_init_range(unsigned long start, unsigned long end) 23 { 24 unsigned long boundary, size; 25 26 while (start < end) { 27 if (MACHINE_HAS_EDAT1) { 28 /* set storage keys for a 1MB frame */ 29 size = 1UL << 20; 30 boundary = (start + size) & ~(size - 1); 31 if (boundary <= end) { 32 do { 33 start = sske_frame(start, PAGE_DEFAULT_KEY); 34 } while (start < boundary); 35 continue; 36 } 37 } 38 page_set_storage_key(start, PAGE_DEFAULT_KEY, 1); 39 start += PAGE_SIZE; 40 } 41 } 42 43 #ifdef CONFIG_PROC_FS 44 atomic_long_t direct_pages_count[PG_DIRECT_MAP_MAX]; 45 46 void arch_report_meminfo(struct seq_file *m) 47 { 48 seq_printf(m, "DirectMap4k: %8lu kB\n", 49 atomic_long_read(&direct_pages_count[PG_DIRECT_MAP_4K]) << 2); 50 seq_printf(m, "DirectMap1M: %8lu kB\n", 51 atomic_long_read(&direct_pages_count[PG_DIRECT_MAP_1M]) << 10); 52 seq_printf(m, "DirectMap2G: %8lu kB\n", 53 atomic_long_read(&direct_pages_count[PG_DIRECT_MAP_2G]) << 21); 54 } 55 #endif /* CONFIG_PROC_FS */ 56 57 static void pgt_set(unsigned long *old, unsigned long new, unsigned long addr, 58 unsigned long dtt) 59 { 60 unsigned long *table, mask; 61 62 mask = 0; 63 if (MACHINE_HAS_EDAT2) { 64 switch (dtt) { 65 case CRDTE_DTT_REGION3: 66 mask = ~(PTRS_PER_PUD * sizeof(pud_t) - 1); 67 break; 68 case CRDTE_DTT_SEGMENT: 69 mask = ~(PTRS_PER_PMD * sizeof(pmd_t) - 1); 70 break; 71 case CRDTE_DTT_PAGE: 72 mask = ~(PTRS_PER_PTE * sizeof(pte_t) - 1); 73 break; 74 } 75 table = (unsigned long *)((unsigned long)old & mask); 76 crdte(*old, new, table, dtt, addr, S390_lowcore.kernel_asce); 77 } else if (MACHINE_HAS_IDTE) { 78 cspg(old, *old, new); 79 } else { 80 csp((unsigned int *)old + 1, *old, new); 81 } 82 } 83 84 static int walk_pte_level(pmd_t *pmdp, unsigned long addr, unsigned long end, 85 unsigned long flags) 86 { 87 pte_t *ptep, new; 88 89 if (flags == SET_MEMORY_4K) 90 return 0; 91 ptep = pte_offset_kernel(pmdp, addr); 92 do { 93 new = *ptep; 94 if (pte_none(new)) 95 return -EINVAL; 96 if (flags & SET_MEMORY_RO) 97 new = pte_wrprotect(new); 98 else if (flags & SET_MEMORY_RW) 99 new = pte_mkwrite(pte_mkdirty(new)); 100 if (flags & SET_MEMORY_NX) 101 pte_val(new) |= _PAGE_NOEXEC; 102 else if (flags & SET_MEMORY_X) 103 pte_val(new) &= ~_PAGE_NOEXEC; 104 pgt_set((unsigned long *)ptep, pte_val(new), addr, CRDTE_DTT_PAGE); 105 ptep++; 106 addr += PAGE_SIZE; 107 cond_resched(); 108 } while (addr < end); 109 return 0; 110 } 111 112 static int split_pmd_page(pmd_t *pmdp, unsigned long addr) 113 { 114 unsigned long pte_addr, prot; 115 pte_t *pt_dir, *ptep; 116 pmd_t new; 117 int i, ro, nx; 118 119 pt_dir = vmem_pte_alloc(); 120 if (!pt_dir) 121 return -ENOMEM; 122 pte_addr = pmd_pfn(*pmdp) << PAGE_SHIFT; 123 ro = !!(pmd_val(*pmdp) & _SEGMENT_ENTRY_PROTECT); 124 nx = !!(pmd_val(*pmdp) & _SEGMENT_ENTRY_NOEXEC); 125 prot = pgprot_val(ro ? PAGE_KERNEL_RO : PAGE_KERNEL); 126 if (!nx) 127 prot &= ~_PAGE_NOEXEC; 128 ptep = pt_dir; 129 for (i = 0; i < PTRS_PER_PTE; i++) { 130 pte_val(*ptep) = pte_addr | prot; 131 pte_addr += PAGE_SIZE; 132 ptep++; 133 } 134 pmd_val(new) = __pa(pt_dir) | _SEGMENT_ENTRY; 135 pgt_set((unsigned long *)pmdp, pmd_val(new), addr, CRDTE_DTT_SEGMENT); 136 update_page_count(PG_DIRECT_MAP_4K, PTRS_PER_PTE); 137 update_page_count(PG_DIRECT_MAP_1M, -1); 138 return 0; 139 } 140 141 static void modify_pmd_page(pmd_t *pmdp, unsigned long addr, 142 unsigned long flags) 143 { 144 pmd_t new = *pmdp; 145 146 if (flags & SET_MEMORY_RO) 147 new = pmd_wrprotect(new); 148 else if (flags & SET_MEMORY_RW) 149 new = pmd_mkwrite(pmd_mkdirty(new)); 150 if (flags & SET_MEMORY_NX) 151 pmd_val(new) |= _SEGMENT_ENTRY_NOEXEC; 152 else if (flags & SET_MEMORY_X) 153 pmd_val(new) &= ~_SEGMENT_ENTRY_NOEXEC; 154 pgt_set((unsigned long *)pmdp, pmd_val(new), addr, CRDTE_DTT_SEGMENT); 155 } 156 157 static int walk_pmd_level(pud_t *pudp, unsigned long addr, unsigned long end, 158 unsigned long flags) 159 { 160 unsigned long next; 161 int need_split; 162 pmd_t *pmdp; 163 int rc = 0; 164 165 pmdp = pmd_offset(pudp, addr); 166 do { 167 if (pmd_none(*pmdp)) 168 return -EINVAL; 169 next = pmd_addr_end(addr, end); 170 if (pmd_large(*pmdp)) { 171 need_split = !!(flags & SET_MEMORY_4K); 172 need_split |= !!(addr & ~PMD_MASK); 173 need_split |= !!(addr + PMD_SIZE > next); 174 if (need_split) { 175 rc = split_pmd_page(pmdp, addr); 176 if (rc) 177 return rc; 178 continue; 179 } 180 modify_pmd_page(pmdp, addr, flags); 181 } else { 182 rc = walk_pte_level(pmdp, addr, next, flags); 183 if (rc) 184 return rc; 185 } 186 pmdp++; 187 addr = next; 188 cond_resched(); 189 } while (addr < end); 190 return rc; 191 } 192 193 static int split_pud_page(pud_t *pudp, unsigned long addr) 194 { 195 unsigned long pmd_addr, prot; 196 pmd_t *pm_dir, *pmdp; 197 pud_t new; 198 int i, ro, nx; 199 200 pm_dir = vmem_crst_alloc(_SEGMENT_ENTRY_EMPTY); 201 if (!pm_dir) 202 return -ENOMEM; 203 pmd_addr = pud_pfn(*pudp) << PAGE_SHIFT; 204 ro = !!(pud_val(*pudp) & _REGION_ENTRY_PROTECT); 205 nx = !!(pud_val(*pudp) & _REGION_ENTRY_NOEXEC); 206 prot = pgprot_val(ro ? SEGMENT_KERNEL_RO : SEGMENT_KERNEL); 207 if (!nx) 208 prot &= ~_SEGMENT_ENTRY_NOEXEC; 209 pmdp = pm_dir; 210 for (i = 0; i < PTRS_PER_PMD; i++) { 211 pmd_val(*pmdp) = pmd_addr | prot; 212 pmd_addr += PMD_SIZE; 213 pmdp++; 214 } 215 pud_val(new) = __pa(pm_dir) | _REGION3_ENTRY; 216 pgt_set((unsigned long *)pudp, pud_val(new), addr, CRDTE_DTT_REGION3); 217 update_page_count(PG_DIRECT_MAP_1M, PTRS_PER_PMD); 218 update_page_count(PG_DIRECT_MAP_2G, -1); 219 return 0; 220 } 221 222 static void modify_pud_page(pud_t *pudp, unsigned long addr, 223 unsigned long flags) 224 { 225 pud_t new = *pudp; 226 227 if (flags & SET_MEMORY_RO) 228 new = pud_wrprotect(new); 229 else if (flags & SET_MEMORY_RW) 230 new = pud_mkwrite(pud_mkdirty(new)); 231 if (flags & SET_MEMORY_NX) 232 pud_val(new) |= _REGION_ENTRY_NOEXEC; 233 else if (flags & SET_MEMORY_X) 234 pud_val(new) &= ~_REGION_ENTRY_NOEXEC; 235 pgt_set((unsigned long *)pudp, pud_val(new), addr, CRDTE_DTT_REGION3); 236 } 237 238 static int walk_pud_level(p4d_t *p4d, unsigned long addr, unsigned long end, 239 unsigned long flags) 240 { 241 unsigned long next; 242 int need_split; 243 pud_t *pudp; 244 int rc = 0; 245 246 pudp = pud_offset(p4d, addr); 247 do { 248 if (pud_none(*pudp)) 249 return -EINVAL; 250 next = pud_addr_end(addr, end); 251 if (pud_large(*pudp)) { 252 need_split = !!(flags & SET_MEMORY_4K); 253 need_split |= !!(addr & ~PUD_MASK); 254 need_split |= !!(addr + PUD_SIZE > next); 255 if (need_split) { 256 rc = split_pud_page(pudp, addr); 257 if (rc) 258 break; 259 continue; 260 } 261 modify_pud_page(pudp, addr, flags); 262 } else { 263 rc = walk_pmd_level(pudp, addr, next, flags); 264 } 265 pudp++; 266 addr = next; 267 cond_resched(); 268 } while (addr < end && !rc); 269 return rc; 270 } 271 272 static int walk_p4d_level(pgd_t *pgd, unsigned long addr, unsigned long end, 273 unsigned long flags) 274 { 275 unsigned long next; 276 p4d_t *p4dp; 277 int rc = 0; 278 279 p4dp = p4d_offset(pgd, addr); 280 do { 281 if (p4d_none(*p4dp)) 282 return -EINVAL; 283 next = p4d_addr_end(addr, end); 284 rc = walk_pud_level(p4dp, addr, next, flags); 285 p4dp++; 286 addr = next; 287 cond_resched(); 288 } while (addr < end && !rc); 289 return rc; 290 } 291 292 DEFINE_MUTEX(cpa_mutex); 293 294 static int change_page_attr(unsigned long addr, unsigned long end, 295 unsigned long flags) 296 { 297 unsigned long next; 298 int rc = -EINVAL; 299 pgd_t *pgdp; 300 301 if (addr == end) 302 return 0; 303 if (end >= MODULES_END) 304 return -EINVAL; 305 mutex_lock(&cpa_mutex); 306 pgdp = pgd_offset_k(addr); 307 do { 308 if (pgd_none(*pgdp)) 309 break; 310 next = pgd_addr_end(addr, end); 311 rc = walk_p4d_level(pgdp, addr, next, flags); 312 if (rc) 313 break; 314 cond_resched(); 315 } while (pgdp++, addr = next, addr < end && !rc); 316 mutex_unlock(&cpa_mutex); 317 return rc; 318 } 319 320 int __set_memory(unsigned long addr, int numpages, unsigned long flags) 321 { 322 if (!MACHINE_HAS_NX) 323 flags &= ~(SET_MEMORY_NX | SET_MEMORY_X); 324 if (!flags) 325 return 0; 326 addr &= PAGE_MASK; 327 return change_page_attr(addr, addr + numpages * PAGE_SIZE, flags); 328 } 329 330 #if defined(CONFIG_DEBUG_PAGEALLOC) || defined(CONFIG_KFENCE) 331 332 static void ipte_range(pte_t *pte, unsigned long address, int nr) 333 { 334 int i; 335 336 if (test_facility(13)) { 337 __ptep_ipte_range(address, nr - 1, pte, IPTE_GLOBAL); 338 return; 339 } 340 for (i = 0; i < nr; i++) { 341 __ptep_ipte(address, pte, 0, 0, IPTE_GLOBAL); 342 address += PAGE_SIZE; 343 pte++; 344 } 345 } 346 347 void __kernel_map_pages(struct page *page, int numpages, int enable) 348 { 349 unsigned long address; 350 int nr, i, j; 351 pte_t *pte; 352 353 for (i = 0; i < numpages;) { 354 address = (unsigned long)page_to_virt(page + i); 355 pte = virt_to_kpte(address); 356 nr = (unsigned long)pte >> ilog2(sizeof(long)); 357 nr = PTRS_PER_PTE - (nr & (PTRS_PER_PTE - 1)); 358 nr = min(numpages - i, nr); 359 if (enable) { 360 for (j = 0; j < nr; j++) { 361 pte_val(*pte) &= ~_PAGE_INVALID; 362 address += PAGE_SIZE; 363 pte++; 364 } 365 } else { 366 ipte_range(pte, address, nr); 367 } 368 i += nr; 369 } 370 } 371 372 #endif /* CONFIG_DEBUG_PAGEALLOC */ 373