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/proc_fs.h> 8 #include <linux/vmalloc.h> 9 #include <linux/mm.h> 10 #include <asm/cacheflush.h> 11 #include <asm/facility.h> 12 #include <asm/pgalloc.h> 13 #include <asm/kfence.h> 14 #include <asm/page.h> 15 #include <asm/asm.h> 16 #include <asm/set_memory.h> 17 18 static inline unsigned long sske_frame(unsigned long addr, unsigned char skey) 19 { 20 asm volatile(".insn rrf,0xb22b0000,%[skey],%[addr],1,0" 21 : [addr] "+a" (addr) : [skey] "d" (skey)); 22 return addr; 23 } 24 25 void __storage_key_init_range(unsigned long start, unsigned long end) 26 { 27 unsigned long boundary, size; 28 29 while (start < end) { 30 if (MACHINE_HAS_EDAT1) { 31 /* set storage keys for a 1MB frame */ 32 size = 1UL << 20; 33 boundary = (start + size) & ~(size - 1); 34 if (boundary <= end) { 35 do { 36 start = sske_frame(start, PAGE_DEFAULT_KEY); 37 } while (start < boundary); 38 continue; 39 } 40 } 41 page_set_storage_key(start, PAGE_DEFAULT_KEY, 1); 42 start += PAGE_SIZE; 43 } 44 } 45 46 #ifdef CONFIG_PROC_FS 47 atomic_long_t __bootdata_preserved(direct_pages_count[PG_DIRECT_MAP_MAX]); 48 49 void arch_report_meminfo(struct seq_file *m) 50 { 51 seq_printf(m, "DirectMap4k: %8lu kB\n", 52 atomic_long_read(&direct_pages_count[PG_DIRECT_MAP_4K]) << 2); 53 seq_printf(m, "DirectMap1M: %8lu kB\n", 54 atomic_long_read(&direct_pages_count[PG_DIRECT_MAP_1M]) << 10); 55 seq_printf(m, "DirectMap2G: %8lu kB\n", 56 atomic_long_read(&direct_pages_count[PG_DIRECT_MAP_2G]) << 21); 57 } 58 #endif /* CONFIG_PROC_FS */ 59 60 static void pgt_set(unsigned long *old, unsigned long new, unsigned long addr, 61 unsigned long dtt) 62 { 63 unsigned long *table, mask; 64 65 mask = 0; 66 if (MACHINE_HAS_EDAT2) { 67 switch (dtt) { 68 case CRDTE_DTT_REGION3: 69 mask = ~(PTRS_PER_PUD * sizeof(pud_t) - 1); 70 break; 71 case CRDTE_DTT_SEGMENT: 72 mask = ~(PTRS_PER_PMD * sizeof(pmd_t) - 1); 73 break; 74 case CRDTE_DTT_PAGE: 75 mask = ~(PTRS_PER_PTE * sizeof(pte_t) - 1); 76 break; 77 } 78 table = (unsigned long *)((unsigned long)old & mask); 79 crdte(*old, new, table, dtt, addr, get_lowcore()->kernel_asce.val); 80 } else if (MACHINE_HAS_IDTE) { 81 cspg(old, *old, new); 82 } else { 83 csp((unsigned int *)old + 1, *old, new); 84 } 85 } 86 87 static int walk_pte_level(pmd_t *pmdp, unsigned long addr, unsigned long end, 88 unsigned long flags) 89 { 90 pte_t *ptep, new; 91 92 if (flags == SET_MEMORY_4K) 93 return 0; 94 ptep = pte_offset_kernel(pmdp, addr); 95 do { 96 new = *ptep; 97 if (pte_none(new)) 98 return -EINVAL; 99 if (flags & SET_MEMORY_RO) 100 new = pte_wrprotect(new); 101 else if (flags & SET_MEMORY_RW) 102 new = pte_mkwrite_novma(pte_mkdirty(new)); 103 if (flags & SET_MEMORY_NX) 104 new = set_pte_bit(new, __pgprot(_PAGE_NOEXEC)); 105 else if (flags & SET_MEMORY_X) 106 new = clear_pte_bit(new, __pgprot(_PAGE_NOEXEC)); 107 if (flags & SET_MEMORY_INV) { 108 new = set_pte_bit(new, __pgprot(_PAGE_INVALID)); 109 } else if (flags & SET_MEMORY_DEF) { 110 new = __pte(pte_val(new) & PAGE_MASK); 111 new = set_pte_bit(new, PAGE_KERNEL); 112 } 113 pgt_set((unsigned long *)ptep, pte_val(new), addr, CRDTE_DTT_PAGE); 114 ptep++; 115 addr += PAGE_SIZE; 116 cond_resched(); 117 } while (addr < end); 118 return 0; 119 } 120 121 static int split_pmd_page(pmd_t *pmdp, unsigned long addr) 122 { 123 unsigned long pte_addr, prot; 124 pte_t *pt_dir, *ptep; 125 pmd_t new; 126 int i, ro, nx; 127 128 pt_dir = vmem_pte_alloc(); 129 if (!pt_dir) 130 return -ENOMEM; 131 pte_addr = pmd_pfn(*pmdp) << PAGE_SHIFT; 132 ro = !!(pmd_val(*pmdp) & _SEGMENT_ENTRY_PROTECT); 133 nx = !!(pmd_val(*pmdp) & _SEGMENT_ENTRY_NOEXEC); 134 prot = pgprot_val(ro ? PAGE_KERNEL_RO : PAGE_KERNEL); 135 if (!nx) 136 prot &= ~_PAGE_NOEXEC; 137 ptep = pt_dir; 138 for (i = 0; i < PTRS_PER_PTE; i++) { 139 set_pte(ptep, __pte(pte_addr | prot)); 140 pte_addr += PAGE_SIZE; 141 ptep++; 142 } 143 new = __pmd(__pa(pt_dir) | _SEGMENT_ENTRY); 144 pgt_set((unsigned long *)pmdp, pmd_val(new), addr, CRDTE_DTT_SEGMENT); 145 update_page_count(PG_DIRECT_MAP_4K, PTRS_PER_PTE); 146 update_page_count(PG_DIRECT_MAP_1M, -1); 147 return 0; 148 } 149 150 static void modify_pmd_page(pmd_t *pmdp, unsigned long addr, 151 unsigned long flags) 152 { 153 pmd_t new = *pmdp; 154 155 if (flags & SET_MEMORY_RO) 156 new = pmd_wrprotect(new); 157 else if (flags & SET_MEMORY_RW) 158 new = pmd_mkwrite_novma(pmd_mkdirty(new)); 159 if (flags & SET_MEMORY_NX) 160 new = set_pmd_bit(new, __pgprot(_SEGMENT_ENTRY_NOEXEC)); 161 else if (flags & SET_MEMORY_X) 162 new = clear_pmd_bit(new, __pgprot(_SEGMENT_ENTRY_NOEXEC)); 163 if (flags & SET_MEMORY_INV) { 164 new = set_pmd_bit(new, __pgprot(_SEGMENT_ENTRY_INVALID)); 165 } else if (flags & SET_MEMORY_DEF) { 166 new = __pmd(pmd_val(new) & PMD_MASK); 167 new = set_pmd_bit(new, SEGMENT_KERNEL); 168 } 169 pgt_set((unsigned long *)pmdp, pmd_val(new), addr, CRDTE_DTT_SEGMENT); 170 } 171 172 static int walk_pmd_level(pud_t *pudp, unsigned long addr, unsigned long end, 173 unsigned long flags) 174 { 175 unsigned long next; 176 int need_split; 177 pmd_t *pmdp; 178 int rc = 0; 179 180 pmdp = pmd_offset(pudp, addr); 181 do { 182 if (pmd_none(*pmdp)) 183 return -EINVAL; 184 next = pmd_addr_end(addr, end); 185 if (pmd_leaf(*pmdp)) { 186 need_split = !!(flags & SET_MEMORY_4K); 187 need_split |= !!(addr & ~PMD_MASK); 188 need_split |= !!(addr + PMD_SIZE > next); 189 if (need_split) { 190 rc = split_pmd_page(pmdp, addr); 191 if (rc) 192 return rc; 193 continue; 194 } 195 modify_pmd_page(pmdp, addr, flags); 196 } else { 197 rc = walk_pte_level(pmdp, addr, next, flags); 198 if (rc) 199 return rc; 200 } 201 pmdp++; 202 addr = next; 203 cond_resched(); 204 } while (addr < end); 205 return rc; 206 } 207 208 static int split_pud_page(pud_t *pudp, unsigned long addr) 209 { 210 unsigned long pmd_addr, prot; 211 pmd_t *pm_dir, *pmdp; 212 pud_t new; 213 int i, ro, nx; 214 215 pm_dir = vmem_crst_alloc(_SEGMENT_ENTRY_EMPTY); 216 if (!pm_dir) 217 return -ENOMEM; 218 pmd_addr = pud_pfn(*pudp) << PAGE_SHIFT; 219 ro = !!(pud_val(*pudp) & _REGION_ENTRY_PROTECT); 220 nx = !!(pud_val(*pudp) & _REGION_ENTRY_NOEXEC); 221 prot = pgprot_val(ro ? SEGMENT_KERNEL_RO : SEGMENT_KERNEL); 222 if (!nx) 223 prot &= ~_SEGMENT_ENTRY_NOEXEC; 224 pmdp = pm_dir; 225 for (i = 0; i < PTRS_PER_PMD; i++) { 226 set_pmd(pmdp, __pmd(pmd_addr | prot)); 227 pmd_addr += PMD_SIZE; 228 pmdp++; 229 } 230 new = __pud(__pa(pm_dir) | _REGION3_ENTRY); 231 pgt_set((unsigned long *)pudp, pud_val(new), addr, CRDTE_DTT_REGION3); 232 update_page_count(PG_DIRECT_MAP_1M, PTRS_PER_PMD); 233 update_page_count(PG_DIRECT_MAP_2G, -1); 234 return 0; 235 } 236 237 static void modify_pud_page(pud_t *pudp, unsigned long addr, 238 unsigned long flags) 239 { 240 pud_t new = *pudp; 241 242 if (flags & SET_MEMORY_RO) 243 new = pud_wrprotect(new); 244 else if (flags & SET_MEMORY_RW) 245 new = pud_mkwrite(pud_mkdirty(new)); 246 if (flags & SET_MEMORY_NX) 247 new = set_pud_bit(new, __pgprot(_REGION_ENTRY_NOEXEC)); 248 else if (flags & SET_MEMORY_X) 249 new = clear_pud_bit(new, __pgprot(_REGION_ENTRY_NOEXEC)); 250 if (flags & SET_MEMORY_INV) { 251 new = set_pud_bit(new, __pgprot(_REGION_ENTRY_INVALID)); 252 } else if (flags & SET_MEMORY_DEF) { 253 new = __pud(pud_val(new) & PUD_MASK); 254 new = set_pud_bit(new, REGION3_KERNEL); 255 } 256 pgt_set((unsigned long *)pudp, pud_val(new), addr, CRDTE_DTT_REGION3); 257 } 258 259 static int walk_pud_level(p4d_t *p4d, unsigned long addr, unsigned long end, 260 unsigned long flags) 261 { 262 unsigned long next; 263 int need_split; 264 pud_t *pudp; 265 int rc = 0; 266 267 pudp = pud_offset(p4d, addr); 268 do { 269 if (pud_none(*pudp)) 270 return -EINVAL; 271 next = pud_addr_end(addr, end); 272 if (pud_leaf(*pudp)) { 273 need_split = !!(flags & SET_MEMORY_4K); 274 need_split |= !!(addr & ~PUD_MASK); 275 need_split |= !!(addr + PUD_SIZE > next); 276 if (need_split) { 277 rc = split_pud_page(pudp, addr); 278 if (rc) 279 break; 280 continue; 281 } 282 modify_pud_page(pudp, addr, flags); 283 } else { 284 rc = walk_pmd_level(pudp, addr, next, flags); 285 } 286 pudp++; 287 addr = next; 288 cond_resched(); 289 } while (addr < end && !rc); 290 return rc; 291 } 292 293 static int walk_p4d_level(pgd_t *pgd, unsigned long addr, unsigned long end, 294 unsigned long flags) 295 { 296 unsigned long next; 297 p4d_t *p4dp; 298 int rc = 0; 299 300 p4dp = p4d_offset(pgd, addr); 301 do { 302 if (p4d_none(*p4dp)) 303 return -EINVAL; 304 next = p4d_addr_end(addr, end); 305 rc = walk_pud_level(p4dp, addr, next, flags); 306 p4dp++; 307 addr = next; 308 cond_resched(); 309 } while (addr < end && !rc); 310 return rc; 311 } 312 313 DEFINE_MUTEX(cpa_mutex); 314 315 static int change_page_attr(unsigned long addr, unsigned long end, 316 unsigned long flags) 317 { 318 unsigned long next; 319 int rc = -EINVAL; 320 pgd_t *pgdp; 321 322 pgdp = pgd_offset_k(addr); 323 do { 324 if (pgd_none(*pgdp)) 325 break; 326 next = pgd_addr_end(addr, end); 327 rc = walk_p4d_level(pgdp, addr, next, flags); 328 if (rc) 329 break; 330 cond_resched(); 331 } while (pgdp++, addr = next, addr < end && !rc); 332 return rc; 333 } 334 335 static int change_page_attr_alias(unsigned long addr, unsigned long end, 336 unsigned long flags) 337 { 338 unsigned long alias, offset, va_start, va_end; 339 struct vm_struct *area; 340 int rc = 0; 341 342 /* 343 * Changes to read-only permissions on kernel VA mappings are also 344 * applied to the kernel direct mapping. Execute permissions are 345 * intentionally not transferred to keep all allocated pages within 346 * the direct mapping non-executable. 347 */ 348 flags &= SET_MEMORY_RO | SET_MEMORY_RW; 349 if (!flags) 350 return 0; 351 area = NULL; 352 while (addr < end) { 353 if (!area) 354 area = find_vm_area((void *)addr); 355 if (!area || !(area->flags & VM_ALLOC)) 356 return 0; 357 va_start = (unsigned long)area->addr; 358 va_end = va_start + area->nr_pages * PAGE_SIZE; 359 offset = (addr - va_start) >> PAGE_SHIFT; 360 alias = (unsigned long)page_address(area->pages[offset]); 361 rc = change_page_attr(alias, alias + PAGE_SIZE, flags); 362 if (rc) 363 break; 364 addr += PAGE_SIZE; 365 if (addr >= va_end) 366 area = NULL; 367 } 368 return rc; 369 } 370 371 int __set_memory(unsigned long addr, unsigned long numpages, unsigned long flags) 372 { 373 unsigned long end; 374 int rc; 375 376 if (!MACHINE_HAS_NX) 377 flags &= ~(SET_MEMORY_NX | SET_MEMORY_X); 378 if (!flags) 379 return 0; 380 if (!numpages) 381 return 0; 382 addr &= PAGE_MASK; 383 end = addr + numpages * PAGE_SIZE; 384 mutex_lock(&cpa_mutex); 385 rc = change_page_attr(addr, end, flags); 386 if (rc) 387 goto out; 388 rc = change_page_attr_alias(addr, end, flags); 389 out: 390 mutex_unlock(&cpa_mutex); 391 return rc; 392 } 393 394 int set_direct_map_invalid_noflush(struct page *page) 395 { 396 return __set_memory((unsigned long)page_to_virt(page), 1, SET_MEMORY_INV); 397 } 398 399 int set_direct_map_default_noflush(struct page *page) 400 { 401 return __set_memory((unsigned long)page_to_virt(page), 1, SET_MEMORY_DEF); 402 } 403 404 int set_direct_map_valid_noflush(struct page *page, unsigned nr, bool valid) 405 { 406 unsigned long flags; 407 408 if (valid) 409 flags = SET_MEMORY_DEF; 410 else 411 flags = SET_MEMORY_INV; 412 413 return __set_memory((unsigned long)page_to_virt(page), nr, flags); 414 } 415 416 bool kernel_page_present(struct page *page) 417 { 418 unsigned long addr; 419 unsigned int cc; 420 421 addr = (unsigned long)page_address(page); 422 asm volatile( 423 " lra %[addr],0(%[addr])\n" 424 CC_IPM(cc) 425 : CC_OUT(cc, cc), [addr] "+a" (addr) 426 : 427 : CC_CLOBBER); 428 return CC_TRANSFORM(cc) == 0; 429 } 430 431 #if defined(CONFIG_DEBUG_PAGEALLOC) || defined(CONFIG_KFENCE) 432 433 static void ipte_range(pte_t *pte, unsigned long address, int nr) 434 { 435 int i; 436 437 if (test_facility(13)) { 438 __ptep_ipte_range(address, nr - 1, pte, IPTE_GLOBAL); 439 return; 440 } 441 for (i = 0; i < nr; i++) { 442 __ptep_ipte(address, pte, 0, 0, IPTE_GLOBAL); 443 address += PAGE_SIZE; 444 pte++; 445 } 446 } 447 448 void __kernel_map_pages(struct page *page, int numpages, int enable) 449 { 450 unsigned long address; 451 pte_t *ptep, pte; 452 int nr, i, j; 453 454 for (i = 0; i < numpages;) { 455 address = (unsigned long)page_to_virt(page + i); 456 ptep = virt_to_kpte(address); 457 nr = (unsigned long)ptep >> ilog2(sizeof(long)); 458 nr = PTRS_PER_PTE - (nr & (PTRS_PER_PTE - 1)); 459 nr = min(numpages - i, nr); 460 if (enable) { 461 for (j = 0; j < nr; j++) { 462 pte = clear_pte_bit(*ptep, __pgprot(_PAGE_INVALID)); 463 set_pte(ptep, pte); 464 address += PAGE_SIZE; 465 ptep++; 466 } 467 } else { 468 ipte_range(ptep, address, nr); 469 } 470 i += nr; 471 } 472 } 473 474 #endif /* CONFIG_DEBUG_PAGEALLOC */ 475