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/cpufeature.h> 7 #include <linux/hugetlb.h> 8 #include <linux/proc_fs.h> 9 #include <linux/vmalloc.h> 10 #include <linux/mm.h> 11 #include <asm/cacheflush.h> 12 #include <asm/facility.h> 13 #include <asm/pgalloc.h> 14 #include <asm/kfence.h> 15 #include <asm/page.h> 16 #include <asm/asm.h> 17 #include <asm/set_memory.h> 18 19 static inline unsigned long sske_frame(unsigned long addr, unsigned char skey) 20 { 21 asm volatile(".insn rrf,0xb22b0000,%[skey],%[addr],1,0" 22 : [addr] "+a" (addr) : [skey] "d" (skey)); 23 return addr; 24 } 25 26 void __storage_key_init_range(unsigned long start, unsigned long end) 27 { 28 unsigned long boundary, size; 29 30 while (start < end) { 31 if (cpu_has_edat1()) { 32 /* set storage keys for a 1MB frame */ 33 size = 1UL << 20; 34 boundary = (start + size) & ~(size - 1); 35 if (boundary <= end) { 36 do { 37 start = sske_frame(start, PAGE_DEFAULT_KEY); 38 } while (start < boundary); 39 continue; 40 } 41 } 42 page_set_storage_key(start, PAGE_DEFAULT_KEY, 1); 43 start += PAGE_SIZE; 44 } 45 } 46 47 #ifdef CONFIG_PROC_FS 48 atomic_long_t __bootdata_preserved(direct_pages_count[PG_DIRECT_MAP_MAX]); 49 50 void arch_report_meminfo(struct seq_file *m) 51 { 52 seq_printf(m, "DirectMap4k: %8lu kB\n", 53 atomic_long_read(&direct_pages_count[PG_DIRECT_MAP_4K]) << 2); 54 seq_printf(m, "DirectMap1M: %8lu kB\n", 55 atomic_long_read(&direct_pages_count[PG_DIRECT_MAP_1M]) << 10); 56 seq_printf(m, "DirectMap2G: %8lu kB\n", 57 atomic_long_read(&direct_pages_count[PG_DIRECT_MAP_2G]) << 21); 58 } 59 #endif /* CONFIG_PROC_FS */ 60 61 static void pgt_set(unsigned long *old, unsigned long new, unsigned long addr, 62 unsigned long dtt) 63 { 64 unsigned long *table, mask; 65 66 mask = 0; 67 if (cpu_has_edat2()) { 68 switch (dtt) { 69 case CRDTE_DTT_REGION3: 70 mask = ~(PTRS_PER_PUD * sizeof(pud_t) - 1); 71 break; 72 case CRDTE_DTT_SEGMENT: 73 mask = ~(PTRS_PER_PMD * sizeof(pmd_t) - 1); 74 break; 75 case CRDTE_DTT_PAGE: 76 mask = ~(PTRS_PER_PTE * sizeof(pte_t) - 1); 77 break; 78 } 79 table = (unsigned long *)((unsigned long)old & mask); 80 crdte(*old, new, table, dtt, addr, get_lowcore()->kernel_asce.val); 81 } else if (cpu_has_idte()) { 82 cspg(old, *old, new); 83 } else { 84 csp((unsigned int *)old + 1, *old, new); 85 } 86 } 87 88 static int walk_pte_level(pmd_t *pmdp, unsigned long addr, unsigned long end, 89 unsigned long flags) 90 { 91 pte_t *ptep, new; 92 93 if (flags == SET_MEMORY_4K) 94 return 0; 95 ptep = pte_offset_kernel(pmdp, addr); 96 do { 97 new = *ptep; 98 if (pte_none(new)) 99 return -EINVAL; 100 if (flags & SET_MEMORY_RO) 101 new = pte_wrprotect(new); 102 else if (flags & SET_MEMORY_RW) 103 new = pte_mkwrite_novma(pte_mkdirty(new)); 104 if (flags & SET_MEMORY_NX) 105 new = set_pte_bit(new, __pgprot(_PAGE_NOEXEC)); 106 else if (flags & SET_MEMORY_X) 107 new = clear_pte_bit(new, __pgprot(_PAGE_NOEXEC)); 108 if (flags & SET_MEMORY_INV) { 109 new = set_pte_bit(new, __pgprot(_PAGE_INVALID)); 110 } else if (flags & SET_MEMORY_DEF) { 111 new = __pte(pte_val(new) & PAGE_MASK); 112 new = set_pte_bit(new, PAGE_KERNEL); 113 } 114 pgt_set((unsigned long *)ptep, pte_val(new), addr, CRDTE_DTT_PAGE); 115 ptep++; 116 addr += PAGE_SIZE; 117 cond_resched(); 118 } while (addr < end); 119 return 0; 120 } 121 122 static int split_pmd_page(pmd_t *pmdp, unsigned long addr) 123 { 124 unsigned long pte_addr, prot; 125 pte_t *pt_dir, *ptep; 126 pmd_t new; 127 int i, ro, nx; 128 129 pt_dir = vmem_pte_alloc(); 130 if (!pt_dir) 131 return -ENOMEM; 132 pte_addr = pmd_pfn(*pmdp) << PAGE_SHIFT; 133 ro = !!(pmd_val(*pmdp) & _SEGMENT_ENTRY_PROTECT); 134 nx = !!(pmd_val(*pmdp) & _SEGMENT_ENTRY_NOEXEC); 135 prot = pgprot_val(ro ? PAGE_KERNEL_RO : PAGE_KERNEL); 136 if (!nx) 137 prot &= ~_PAGE_NOEXEC; 138 ptep = pt_dir; 139 for (i = 0; i < PTRS_PER_PTE; i++) { 140 set_pte(ptep, __pte(pte_addr | prot)); 141 pte_addr += PAGE_SIZE; 142 ptep++; 143 } 144 new = __pmd(__pa(pt_dir) | _SEGMENT_ENTRY); 145 pgt_set((unsigned long *)pmdp, pmd_val(new), addr, CRDTE_DTT_SEGMENT); 146 update_page_count(PG_DIRECT_MAP_4K, PTRS_PER_PTE); 147 update_page_count(PG_DIRECT_MAP_1M, -1); 148 return 0; 149 } 150 151 static void modify_pmd_page(pmd_t *pmdp, unsigned long addr, 152 unsigned long flags) 153 { 154 pmd_t new = *pmdp; 155 156 if (flags & SET_MEMORY_RO) 157 new = pmd_wrprotect(new); 158 else if (flags & SET_MEMORY_RW) 159 new = pmd_mkwrite_novma(pmd_mkdirty(new)); 160 if (flags & SET_MEMORY_NX) 161 new = set_pmd_bit(new, __pgprot(_SEGMENT_ENTRY_NOEXEC)); 162 else if (flags & SET_MEMORY_X) 163 new = clear_pmd_bit(new, __pgprot(_SEGMENT_ENTRY_NOEXEC)); 164 if (flags & SET_MEMORY_INV) { 165 new = set_pmd_bit(new, __pgprot(_SEGMENT_ENTRY_INVALID)); 166 } else if (flags & SET_MEMORY_DEF) { 167 new = __pmd(pmd_val(new) & PMD_MASK); 168 new = set_pmd_bit(new, SEGMENT_KERNEL); 169 } 170 pgt_set((unsigned long *)pmdp, pmd_val(new), addr, CRDTE_DTT_SEGMENT); 171 } 172 173 static int walk_pmd_level(pud_t *pudp, unsigned long addr, unsigned long end, 174 unsigned long flags) 175 { 176 unsigned long next; 177 int need_split; 178 pmd_t *pmdp; 179 int rc = 0; 180 181 pmdp = pmd_offset(pudp, addr); 182 do { 183 if (pmd_none(*pmdp)) 184 return -EINVAL; 185 next = pmd_addr_end(addr, end); 186 if (pmd_leaf(*pmdp)) { 187 need_split = !!(flags & SET_MEMORY_4K); 188 need_split |= !!(addr & ~PMD_MASK); 189 need_split |= !!(addr + PMD_SIZE > next); 190 if (need_split) { 191 rc = split_pmd_page(pmdp, addr); 192 if (rc) 193 return rc; 194 continue; 195 } 196 modify_pmd_page(pmdp, addr, flags); 197 } else { 198 rc = walk_pte_level(pmdp, addr, next, flags); 199 if (rc) 200 return rc; 201 } 202 pmdp++; 203 addr = next; 204 cond_resched(); 205 } while (addr < end); 206 return rc; 207 } 208 209 static int split_pud_page(pud_t *pudp, unsigned long addr) 210 { 211 unsigned long pmd_addr, prot; 212 pmd_t *pm_dir, *pmdp; 213 pud_t new; 214 int i, ro, nx; 215 216 pm_dir = vmem_crst_alloc(_SEGMENT_ENTRY_EMPTY); 217 if (!pm_dir) 218 return -ENOMEM; 219 pmd_addr = pud_pfn(*pudp) << PAGE_SHIFT; 220 ro = !!(pud_val(*pudp) & _REGION_ENTRY_PROTECT); 221 nx = !!(pud_val(*pudp) & _REGION_ENTRY_NOEXEC); 222 prot = pgprot_val(ro ? SEGMENT_KERNEL_RO : SEGMENT_KERNEL); 223 if (!nx) 224 prot &= ~_SEGMENT_ENTRY_NOEXEC; 225 pmdp = pm_dir; 226 for (i = 0; i < PTRS_PER_PMD; i++) { 227 set_pmd(pmdp, __pmd(pmd_addr | prot)); 228 pmd_addr += PMD_SIZE; 229 pmdp++; 230 } 231 new = __pud(__pa(pm_dir) | _REGION3_ENTRY); 232 pgt_set((unsigned long *)pudp, pud_val(new), addr, CRDTE_DTT_REGION3); 233 update_page_count(PG_DIRECT_MAP_1M, PTRS_PER_PMD); 234 update_page_count(PG_DIRECT_MAP_2G, -1); 235 return 0; 236 } 237 238 static void modify_pud_page(pud_t *pudp, unsigned long addr, 239 unsigned long flags) 240 { 241 pud_t new = *pudp; 242 243 if (flags & SET_MEMORY_RO) 244 new = pud_wrprotect(new); 245 else if (flags & SET_MEMORY_RW) 246 new = pud_mkwrite(pud_mkdirty(new)); 247 if (flags & SET_MEMORY_NX) 248 new = set_pud_bit(new, __pgprot(_REGION_ENTRY_NOEXEC)); 249 else if (flags & SET_MEMORY_X) 250 new = clear_pud_bit(new, __pgprot(_REGION_ENTRY_NOEXEC)); 251 if (flags & SET_MEMORY_INV) { 252 new = set_pud_bit(new, __pgprot(_REGION_ENTRY_INVALID)); 253 } else if (flags & SET_MEMORY_DEF) { 254 new = __pud(pud_val(new) & PUD_MASK); 255 new = set_pud_bit(new, REGION3_KERNEL); 256 } 257 pgt_set((unsigned long *)pudp, pud_val(new), addr, CRDTE_DTT_REGION3); 258 } 259 260 static int walk_pud_level(p4d_t *p4d, unsigned long addr, unsigned long end, 261 unsigned long flags) 262 { 263 unsigned long next; 264 int need_split; 265 pud_t *pudp; 266 int rc = 0; 267 268 pudp = pud_offset(p4d, addr); 269 do { 270 if (pud_none(*pudp)) 271 return -EINVAL; 272 next = pud_addr_end(addr, end); 273 if (pud_leaf(*pudp)) { 274 need_split = !!(flags & SET_MEMORY_4K); 275 need_split |= !!(addr & ~PUD_MASK); 276 need_split |= !!(addr + PUD_SIZE > next); 277 if (need_split) { 278 rc = split_pud_page(pudp, addr); 279 if (rc) 280 break; 281 continue; 282 } 283 modify_pud_page(pudp, addr, flags); 284 } else { 285 rc = walk_pmd_level(pudp, addr, next, flags); 286 } 287 pudp++; 288 addr = next; 289 cond_resched(); 290 } while (addr < end && !rc); 291 return rc; 292 } 293 294 static int walk_p4d_level(pgd_t *pgd, unsigned long addr, unsigned long end, 295 unsigned long flags) 296 { 297 unsigned long next; 298 p4d_t *p4dp; 299 int rc = 0; 300 301 p4dp = p4d_offset(pgd, addr); 302 do { 303 if (p4d_none(*p4dp)) 304 return -EINVAL; 305 next = p4d_addr_end(addr, end); 306 rc = walk_pud_level(p4dp, addr, next, flags); 307 p4dp++; 308 addr = next; 309 cond_resched(); 310 } while (addr < end && !rc); 311 return rc; 312 } 313 314 DEFINE_MUTEX(cpa_mutex); 315 316 static int change_page_attr(unsigned long addr, unsigned long end, 317 unsigned long flags) 318 { 319 unsigned long next; 320 int rc = -EINVAL; 321 pgd_t *pgdp; 322 323 pgdp = pgd_offset_k(addr); 324 do { 325 if (pgd_none(*pgdp)) 326 break; 327 next = pgd_addr_end(addr, end); 328 rc = walk_p4d_level(pgdp, addr, next, flags); 329 if (rc) 330 break; 331 cond_resched(); 332 } while (pgdp++, addr = next, addr < end && !rc); 333 return rc; 334 } 335 336 static int change_page_attr_alias(unsigned long addr, unsigned long end, 337 unsigned long flags) 338 { 339 unsigned long alias, offset, va_start, va_end; 340 struct vm_struct *area; 341 int rc = 0; 342 343 /* 344 * Changes to read-only permissions on kernel VA mappings are also 345 * applied to the kernel direct mapping. Execute permissions are 346 * intentionally not transferred to keep all allocated pages within 347 * the direct mapping non-executable. 348 */ 349 flags &= SET_MEMORY_RO | SET_MEMORY_RW; 350 if (!flags) 351 return 0; 352 area = NULL; 353 while (addr < end) { 354 if (!area) 355 area = find_vm_area((void *)addr); 356 if (!area || !(area->flags & VM_ALLOC)) 357 return 0; 358 va_start = (unsigned long)area->addr; 359 va_end = va_start + area->nr_pages * PAGE_SIZE; 360 offset = (addr - va_start) >> PAGE_SHIFT; 361 alias = (unsigned long)page_address(area->pages[offset]); 362 rc = change_page_attr(alias, alias + PAGE_SIZE, flags); 363 if (rc) 364 break; 365 addr += PAGE_SIZE; 366 if (addr >= va_end) 367 area = NULL; 368 } 369 return rc; 370 } 371 372 int __set_memory(unsigned long addr, unsigned long numpages, unsigned long flags) 373 { 374 unsigned long end; 375 int rc; 376 377 if (!cpu_has_nx()) 378 flags &= ~(SET_MEMORY_NX | SET_MEMORY_X); 379 if (!flags) 380 return 0; 381 if (!numpages) 382 return 0; 383 addr &= PAGE_MASK; 384 end = addr + numpages * PAGE_SIZE; 385 mutex_lock(&cpa_mutex); 386 rc = change_page_attr(addr, end, flags); 387 if (rc) 388 goto out; 389 rc = change_page_attr_alias(addr, end, flags); 390 out: 391 mutex_unlock(&cpa_mutex); 392 return rc; 393 } 394 395 int set_direct_map_invalid_noflush(struct page *page) 396 { 397 return __set_memory((unsigned long)page_to_virt(page), 1, SET_MEMORY_INV); 398 } 399 400 int set_direct_map_default_noflush(struct page *page) 401 { 402 return __set_memory((unsigned long)page_to_virt(page), 1, SET_MEMORY_DEF); 403 } 404 405 int set_direct_map_valid_noflush(struct page *page, unsigned nr, bool valid) 406 { 407 unsigned long flags; 408 409 if (valid) 410 flags = SET_MEMORY_DEF; 411 else 412 flags = SET_MEMORY_INV; 413 414 return __set_memory((unsigned long)page_to_virt(page), nr, flags); 415 } 416 417 bool kernel_page_present(struct page *page) 418 { 419 unsigned long addr; 420 unsigned int cc; 421 422 addr = (unsigned long)page_address(page); 423 asm volatile( 424 " lra %[addr],0(%[addr])\n" 425 CC_IPM(cc) 426 : CC_OUT(cc, cc), [addr] "+a" (addr) 427 : 428 : CC_CLOBBER); 429 return CC_TRANSFORM(cc) == 0; 430 } 431 432 #if defined(CONFIG_DEBUG_PAGEALLOC) || defined(CONFIG_KFENCE) 433 434 static void ipte_range(pte_t *pte, unsigned long address, int nr) 435 { 436 int i; 437 438 if (test_facility(13)) { 439 __ptep_ipte_range(address, nr - 1, pte, IPTE_GLOBAL); 440 return; 441 } 442 for (i = 0; i < nr; i++) { 443 __ptep_ipte(address, pte, 0, 0, IPTE_GLOBAL); 444 address += PAGE_SIZE; 445 pte++; 446 } 447 } 448 449 void __kernel_map_pages(struct page *page, int numpages, int enable) 450 { 451 unsigned long address; 452 pte_t *ptep, pte; 453 int nr, i, j; 454 455 for (i = 0; i < numpages;) { 456 address = (unsigned long)page_to_virt(page + i); 457 ptep = virt_to_kpte(address); 458 nr = (unsigned long)ptep >> ilog2(sizeof(long)); 459 nr = PTRS_PER_PTE - (nr & (PTRS_PER_PTE - 1)); 460 nr = min(numpages - i, nr); 461 if (enable) { 462 for (j = 0; j < nr; j++) { 463 pte = clear_pte_bit(*ptep, __pgprot(_PAGE_INVALID)); 464 set_pte(ptep, pte); 465 address += PAGE_SIZE; 466 ptep++; 467 } 468 } else { 469 ipte_range(ptep, address, nr); 470 } 471 i += nr; 472 } 473 } 474 475 #endif /* CONFIG_DEBUG_PAGEALLOC */ 476