1 /* SPDX-License-Identifier: GPL-2.0 */ 2 /* 3 * S390 version 4 * Copyright IBM Corp. 1999, 2000 5 * Author(s): Hartmut Penner (hp@de.ibm.com) 6 * Ulrich Weigand (weigand@de.ibm.com) 7 * Martin Schwidefsky (schwidefsky@de.ibm.com) 8 * 9 * Derived from "include/asm-i386/pgtable.h" 10 */ 11 12 #ifndef _ASM_S390_PGTABLE_H 13 #define _ASM_S390_PGTABLE_H 14 15 #include <linux/sched.h> 16 #include <linux/mm_types.h> 17 #include <linux/cpufeature.h> 18 #include <linux/page-flags.h> 19 #include <linux/radix-tree.h> 20 #include <linux/atomic.h> 21 #include <asm/ctlreg.h> 22 #include <asm/bug.h> 23 #include <asm/page.h> 24 #include <asm/uv.h> 25 26 extern pgd_t swapper_pg_dir[]; 27 extern pgd_t invalid_pg_dir[]; 28 extern void paging_init(void); 29 extern struct ctlreg s390_invalid_asce; 30 31 enum { 32 PG_DIRECT_MAP_4K = 0, 33 PG_DIRECT_MAP_1M, 34 PG_DIRECT_MAP_2G, 35 PG_DIRECT_MAP_MAX 36 }; 37 38 extern atomic_long_t direct_pages_count[PG_DIRECT_MAP_MAX]; 39 40 static inline void update_page_count(int level, long count) 41 { 42 if (IS_ENABLED(CONFIG_PROC_FS)) 43 atomic_long_add(count, &direct_pages_count[level]); 44 } 45 46 /* 47 * The S390 doesn't have any external MMU info: the kernel page 48 * tables contain all the necessary information. 49 */ 50 #define update_mmu_cache(vma, address, ptep) do { } while (0) 51 #define update_mmu_cache_range(vmf, vma, addr, ptep, nr) do { } while (0) 52 #define update_mmu_cache_pmd(vma, address, ptep) do { } while (0) 53 54 /* 55 * ZERO_PAGE is a global shared page that is always zero; used 56 * for zero-mapped memory areas etc.. 57 */ 58 59 extern unsigned long empty_zero_page; 60 extern unsigned long zero_page_mask; 61 62 #define ZERO_PAGE(vaddr) \ 63 (virt_to_page((void *)(empty_zero_page + \ 64 (((unsigned long)(vaddr)) &zero_page_mask)))) 65 #define __HAVE_COLOR_ZERO_PAGE 66 67 /* TODO: s390 cannot support io_remap_pfn_range... */ 68 69 #define pte_ERROR(e) \ 70 pr_err("%s:%d: bad pte %016lx.\n", __FILE__, __LINE__, pte_val(e)) 71 #define pmd_ERROR(e) \ 72 pr_err("%s:%d: bad pmd %016lx.\n", __FILE__, __LINE__, pmd_val(e)) 73 #define pud_ERROR(e) \ 74 pr_err("%s:%d: bad pud %016lx.\n", __FILE__, __LINE__, pud_val(e)) 75 #define p4d_ERROR(e) \ 76 pr_err("%s:%d: bad p4d %016lx.\n", __FILE__, __LINE__, p4d_val(e)) 77 #define pgd_ERROR(e) \ 78 pr_err("%s:%d: bad pgd %016lx.\n", __FILE__, __LINE__, pgd_val(e)) 79 80 /* 81 * The vmalloc and module area will always be on the topmost area of the 82 * kernel mapping. 512GB are reserved for vmalloc by default. 83 * At the top of the vmalloc area a 2GB area is reserved where modules 84 * will reside. That makes sure that inter module branches always 85 * happen without trampolines and in addition the placement within a 86 * 2GB frame is branch prediction unit friendly. 87 */ 88 extern unsigned long VMALLOC_START; 89 extern unsigned long VMALLOC_END; 90 #define VMALLOC_DEFAULT_SIZE ((512UL << 30) - MODULES_LEN) 91 extern struct page *vmemmap; 92 extern unsigned long vmemmap_size; 93 94 extern unsigned long MODULES_VADDR; 95 extern unsigned long MODULES_END; 96 #define MODULES_VADDR MODULES_VADDR 97 #define MODULES_END MODULES_END 98 #define MODULES_LEN (1UL << 31) 99 100 static inline int is_module_addr(void *addr) 101 { 102 BUILD_BUG_ON(MODULES_LEN > (1UL << 31)); 103 if (addr < (void *)MODULES_VADDR) 104 return 0; 105 if (addr > (void *)MODULES_END) 106 return 0; 107 return 1; 108 } 109 110 #ifdef CONFIG_KMSAN 111 #define KMSAN_VMALLOC_SIZE (VMALLOC_END - VMALLOC_START) 112 #define KMSAN_VMALLOC_SHADOW_START VMALLOC_END 113 #define KMSAN_VMALLOC_SHADOW_END (KMSAN_VMALLOC_SHADOW_START + KMSAN_VMALLOC_SIZE) 114 #define KMSAN_VMALLOC_ORIGIN_START KMSAN_VMALLOC_SHADOW_END 115 #define KMSAN_VMALLOC_ORIGIN_END (KMSAN_VMALLOC_ORIGIN_START + KMSAN_VMALLOC_SIZE) 116 #define KMSAN_MODULES_SHADOW_START KMSAN_VMALLOC_ORIGIN_END 117 #define KMSAN_MODULES_SHADOW_END (KMSAN_MODULES_SHADOW_START + MODULES_LEN) 118 #define KMSAN_MODULES_ORIGIN_START KMSAN_MODULES_SHADOW_END 119 #define KMSAN_MODULES_ORIGIN_END (KMSAN_MODULES_ORIGIN_START + MODULES_LEN) 120 #endif 121 122 #ifdef CONFIG_RANDOMIZE_BASE 123 #define KASLR_LEN (1UL << 31) 124 #else 125 #define KASLR_LEN 0UL 126 #endif 127 128 void setup_protection_map(void); 129 130 /* 131 * A 64 bit pagetable entry of S390 has following format: 132 * | PFRA |0IPC| OS | 133 * 0000000000111111111122222222223333333333444444444455555555556666 134 * 0123456789012345678901234567890123456789012345678901234567890123 135 * 136 * I Page-Invalid Bit: Page is not available for address-translation 137 * P Page-Protection Bit: Store access not possible for page 138 * C Change-bit override: HW is not required to set change bit 139 * 140 * A 64 bit segmenttable entry of S390 has following format: 141 * | P-table origin | TT 142 * 0000000000111111111122222222223333333333444444444455555555556666 143 * 0123456789012345678901234567890123456789012345678901234567890123 144 * 145 * I Segment-Invalid Bit: Segment is not available for address-translation 146 * C Common-Segment Bit: Segment is not private (PoP 3-30) 147 * P Page-Protection Bit: Store access not possible for page 148 * TT Type 00 149 * 150 * A 64 bit region table entry of S390 has following format: 151 * | S-table origin | TF TTTL 152 * 0000000000111111111122222222223333333333444444444455555555556666 153 * 0123456789012345678901234567890123456789012345678901234567890123 154 * 155 * I Segment-Invalid Bit: Segment is not available for address-translation 156 * TT Type 01 157 * TF 158 * TL Table length 159 * 160 * The 64 bit regiontable origin of S390 has following format: 161 * | region table origon | DTTL 162 * 0000000000111111111122222222223333333333444444444455555555556666 163 * 0123456789012345678901234567890123456789012345678901234567890123 164 * 165 * X Space-Switch event: 166 * G Segment-Invalid Bit: 167 * P Private-Space Bit: 168 * S Storage-Alteration: 169 * R Real space 170 * TL Table-Length: 171 * 172 * A storage key has the following format: 173 * | ACC |F|R|C|0| 174 * 0 3 4 5 6 7 175 * ACC: access key 176 * F : fetch protection bit 177 * R : referenced bit 178 * C : changed bit 179 */ 180 181 /* Hardware bits in the page table entry */ 182 #define _PAGE_NOEXEC 0x100 /* HW no-execute bit */ 183 #define _PAGE_PROTECT 0x200 /* HW read-only bit */ 184 #define _PAGE_INVALID 0x400 /* HW invalid bit */ 185 #define _PAGE_LARGE 0x800 /* Bit to mark a large pte */ 186 187 /* Software bits in the page table entry */ 188 #define _PAGE_PRESENT 0x001 /* SW pte present bit */ 189 #define _PAGE_YOUNG 0x004 /* SW pte young bit */ 190 #define _PAGE_DIRTY 0x008 /* SW pte dirty bit */ 191 #define _PAGE_READ 0x010 /* SW pte read bit */ 192 #define _PAGE_WRITE 0x020 /* SW pte write bit */ 193 #define _PAGE_SPECIAL 0x040 /* SW associated with special page */ 194 #define _PAGE_UNUSED 0x080 /* SW bit for pgste usage state */ 195 196 #ifdef CONFIG_MEM_SOFT_DIRTY 197 #define _PAGE_SOFT_DIRTY 0x002 /* SW pte soft dirty bit */ 198 #else 199 #define _PAGE_SOFT_DIRTY 0x000 200 #endif 201 202 #define _PAGE_SW_BITS 0xffUL /* All SW bits */ 203 204 #define _PAGE_SWP_EXCLUSIVE _PAGE_LARGE /* SW pte exclusive swap bit */ 205 206 /* Set of bits not changed in pte_modify */ 207 #define _PAGE_CHG_MASK (PAGE_MASK | _PAGE_SPECIAL | _PAGE_DIRTY | \ 208 _PAGE_YOUNG | _PAGE_SOFT_DIRTY) 209 210 /* 211 * Mask of bits that must not be changed with RDP. Allow only _PAGE_PROTECT 212 * HW bit and all SW bits. 213 */ 214 #define _PAGE_RDP_MASK ~(_PAGE_PROTECT | _PAGE_SW_BITS) 215 216 /* 217 * handle_pte_fault uses pte_present and pte_none to find out the pte type 218 * WITHOUT holding the page table lock. The _PAGE_PRESENT bit is used to 219 * distinguish present from not-present ptes. It is changed only with the page 220 * table lock held. 221 * 222 * The following table gives the different possible bit combinations for 223 * the pte hardware and software bits in the last 12 bits of a pte 224 * (. unassigned bit, x don't care, t swap type): 225 * 226 * 842100000000 227 * 000084210000 228 * 000000008421 229 * .IR.uswrdy.p 230 * empty .10.00000000 231 * swap .11..ttttt.0 232 * prot-none, clean, old .11.xx0000.1 233 * prot-none, clean, young .11.xx0001.1 234 * prot-none, dirty, old .11.xx0010.1 235 * prot-none, dirty, young .11.xx0011.1 236 * read-only, clean, old .11.xx0100.1 237 * read-only, clean, young .01.xx0101.1 238 * read-only, dirty, old .11.xx0110.1 239 * read-only, dirty, young .01.xx0111.1 240 * read-write, clean, old .11.xx1100.1 241 * read-write, clean, young .01.xx1101.1 242 * read-write, dirty, old .10.xx1110.1 243 * read-write, dirty, young .00.xx1111.1 244 * HW-bits: R read-only, I invalid 245 * SW-bits: p present, y young, d dirty, r read, w write, s special, 246 * u unused, l large 247 * 248 * pte_none is true for the bit pattern .10.00000000, pte == 0x400 249 * pte_swap is true for the bit pattern .11..ooooo.0, (pte & 0x201) == 0x200 250 * pte_present is true for the bit pattern .xx.xxxxxx.1, (pte & 0x001) == 0x001 251 */ 252 253 /* Bits in the segment/region table address-space-control-element */ 254 #define _ASCE_ORIGIN ~0xfffUL/* region/segment table origin */ 255 #define _ASCE_PRIVATE_SPACE 0x100 /* private space control */ 256 #define _ASCE_ALT_EVENT 0x80 /* storage alteration event control */ 257 #define _ASCE_SPACE_SWITCH 0x40 /* space switch event */ 258 #define _ASCE_REAL_SPACE 0x20 /* real space control */ 259 #define _ASCE_TYPE_MASK 0x0c /* asce table type mask */ 260 #define _ASCE_TYPE_REGION1 0x0c /* region first table type */ 261 #define _ASCE_TYPE_REGION2 0x08 /* region second table type */ 262 #define _ASCE_TYPE_REGION3 0x04 /* region third table type */ 263 #define _ASCE_TYPE_SEGMENT 0x00 /* segment table type */ 264 #define _ASCE_TABLE_LENGTH 0x03 /* region table length */ 265 266 /* Bits in the region table entry */ 267 #define _REGION_ENTRY_ORIGIN ~0xfffUL/* region/segment table origin */ 268 #define _REGION_ENTRY_PROTECT 0x200 /* region protection bit */ 269 #define _REGION_ENTRY_NOEXEC 0x100 /* region no-execute bit */ 270 #define _REGION_ENTRY_OFFSET 0xc0 /* region table offset */ 271 #define _REGION_ENTRY_INVALID 0x20 /* invalid region table entry */ 272 #define _REGION_ENTRY_TYPE_MASK 0x0c /* region table type mask */ 273 #define _REGION_ENTRY_TYPE_R1 0x0c /* region first table type */ 274 #define _REGION_ENTRY_TYPE_R2 0x08 /* region second table type */ 275 #define _REGION_ENTRY_TYPE_R3 0x04 /* region third table type */ 276 #define _REGION_ENTRY_LENGTH 0x03 /* region third length */ 277 278 #define _REGION1_ENTRY (_REGION_ENTRY_TYPE_R1 | _REGION_ENTRY_LENGTH) 279 #define _REGION1_ENTRY_EMPTY (_REGION_ENTRY_TYPE_R1 | _REGION_ENTRY_INVALID) 280 #define _REGION2_ENTRY (_REGION_ENTRY_TYPE_R2 | _REGION_ENTRY_LENGTH) 281 #define _REGION2_ENTRY_EMPTY (_REGION_ENTRY_TYPE_R2 | _REGION_ENTRY_INVALID) 282 #define _REGION3_ENTRY (_REGION_ENTRY_TYPE_R3 | _REGION_ENTRY_LENGTH | \ 283 _REGION3_ENTRY_PRESENT) 284 #define _REGION3_ENTRY_EMPTY (_REGION_ENTRY_TYPE_R3 | _REGION_ENTRY_INVALID) 285 286 #define _REGION3_ENTRY_HARDWARE_BITS 0xfffffffffffff6ffUL 287 #define _REGION3_ENTRY_HARDWARE_BITS_LARGE 0xffffffff8001073cUL 288 #define _REGION3_ENTRY_ORIGIN_LARGE ~0x7fffffffUL /* large page address */ 289 #define _REGION3_ENTRY_DIRTY 0x2000 /* SW region dirty bit */ 290 #define _REGION3_ENTRY_YOUNG 0x1000 /* SW region young bit */ 291 #define _REGION3_ENTRY_COMM 0x0010 /* Common-Region, marks swap entry */ 292 #define _REGION3_ENTRY_LARGE 0x0400 /* RTTE-format control, large page */ 293 #define _REGION3_ENTRY_WRITE 0x8000 /* SW region write bit */ 294 #define _REGION3_ENTRY_READ 0x4000 /* SW region read bit */ 295 296 #ifdef CONFIG_MEM_SOFT_DIRTY 297 #define _REGION3_ENTRY_SOFT_DIRTY 0x0002 /* SW region soft dirty bit */ 298 #else 299 #define _REGION3_ENTRY_SOFT_DIRTY 0x0000 /* SW region soft dirty bit */ 300 #endif 301 302 #define _REGION_ENTRY_BITS 0xfffffffffffff22fUL 303 304 /* 305 * SW region present bit. For non-leaf region-third-table entries, bits 62-63 306 * indicate the TABLE LENGTH and both must be set to 1. But such entries 307 * would always be considered as present, so it is safe to use bit 63 as 308 * PRESENT bit for PUD. 309 */ 310 #define _REGION3_ENTRY_PRESENT 0x0001 311 312 /* Bits in the segment table entry */ 313 #define _SEGMENT_ENTRY_BITS 0xfffffffffffffe3fUL 314 #define _SEGMENT_ENTRY_HARDWARE_BITS 0xfffffffffffffe3cUL 315 #define _SEGMENT_ENTRY_HARDWARE_BITS_LARGE 0xfffffffffff1073cUL 316 #define _SEGMENT_ENTRY_ORIGIN_LARGE ~0xfffffUL /* large page address */ 317 #define _SEGMENT_ENTRY_ORIGIN ~0x7ffUL/* page table origin */ 318 #define _SEGMENT_ENTRY_PROTECT 0x200 /* segment protection bit */ 319 #define _SEGMENT_ENTRY_NOEXEC 0x100 /* segment no-execute bit */ 320 #define _SEGMENT_ENTRY_INVALID 0x20 /* invalid segment table entry */ 321 #define _SEGMENT_ENTRY_TYPE_MASK 0x0c /* segment table type mask */ 322 323 #define _SEGMENT_ENTRY (_SEGMENT_ENTRY_PRESENT) 324 #define _SEGMENT_ENTRY_EMPTY (_SEGMENT_ENTRY_INVALID) 325 326 #define _SEGMENT_ENTRY_DIRTY 0x2000 /* SW segment dirty bit */ 327 #define _SEGMENT_ENTRY_YOUNG 0x1000 /* SW segment young bit */ 328 329 #define _SEGMENT_ENTRY_COMM 0x0010 /* Common-Segment, marks swap entry */ 330 #define _SEGMENT_ENTRY_LARGE 0x0400 /* STE-format control, large page */ 331 #define _SEGMENT_ENTRY_WRITE 0x8000 /* SW segment write bit */ 332 #define _SEGMENT_ENTRY_READ 0x4000 /* SW segment read bit */ 333 334 #ifdef CONFIG_MEM_SOFT_DIRTY 335 #define _SEGMENT_ENTRY_SOFT_DIRTY 0x0002 /* SW segment soft dirty bit */ 336 #else 337 #define _SEGMENT_ENTRY_SOFT_DIRTY 0x0000 /* SW segment soft dirty bit */ 338 #endif 339 340 #define _SEGMENT_ENTRY_PRESENT 0x0001 /* SW segment present bit */ 341 342 /* Common bits in region and segment table entries, for swap entries */ 343 #define _RST_ENTRY_COMM 0x0010 /* Common-Region/Segment, marks swap entry */ 344 #define _RST_ENTRY_INVALID 0x0020 /* invalid region/segment table entry */ 345 346 #define _CRST_ENTRIES 2048 /* number of region/segment table entries */ 347 #define _PAGE_ENTRIES 256 /* number of page table entries */ 348 349 #define _CRST_TABLE_SIZE (_CRST_ENTRIES * 8) 350 #define _PAGE_TABLE_SIZE (_PAGE_ENTRIES * 8) 351 352 #define _REGION1_SHIFT 53 353 #define _REGION2_SHIFT 42 354 #define _REGION3_SHIFT 31 355 #define _SEGMENT_SHIFT 20 356 357 #define _REGION1_INDEX (0x7ffUL << _REGION1_SHIFT) 358 #define _REGION2_INDEX (0x7ffUL << _REGION2_SHIFT) 359 #define _REGION3_INDEX (0x7ffUL << _REGION3_SHIFT) 360 #define _SEGMENT_INDEX (0x7ffUL << _SEGMENT_SHIFT) 361 #define _PAGE_INDEX (0xffUL << PAGE_SHIFT) 362 363 #define _REGION1_SIZE (1UL << _REGION1_SHIFT) 364 #define _REGION2_SIZE (1UL << _REGION2_SHIFT) 365 #define _REGION3_SIZE (1UL << _REGION3_SHIFT) 366 #define _SEGMENT_SIZE (1UL << _SEGMENT_SHIFT) 367 368 #define _REGION1_MASK (~(_REGION1_SIZE - 1)) 369 #define _REGION2_MASK (~(_REGION2_SIZE - 1)) 370 #define _REGION3_MASK (~(_REGION3_SIZE - 1)) 371 #define _SEGMENT_MASK (~(_SEGMENT_SIZE - 1)) 372 373 #define PMD_SHIFT _SEGMENT_SHIFT 374 #define PUD_SHIFT _REGION3_SHIFT 375 #define P4D_SHIFT _REGION2_SHIFT 376 #define PGDIR_SHIFT _REGION1_SHIFT 377 378 #define PMD_SIZE _SEGMENT_SIZE 379 #define PUD_SIZE _REGION3_SIZE 380 #define P4D_SIZE _REGION2_SIZE 381 #define PGDIR_SIZE _REGION1_SIZE 382 383 #define PMD_MASK _SEGMENT_MASK 384 #define PUD_MASK _REGION3_MASK 385 #define P4D_MASK _REGION2_MASK 386 #define PGDIR_MASK _REGION1_MASK 387 388 #define PTRS_PER_PTE _PAGE_ENTRIES 389 #define PTRS_PER_PMD _CRST_ENTRIES 390 #define PTRS_PER_PUD _CRST_ENTRIES 391 #define PTRS_PER_P4D _CRST_ENTRIES 392 #define PTRS_PER_PGD _CRST_ENTRIES 393 394 /* 395 * Segment table and region3 table entry encoding 396 * (R = read-only, I = invalid, y = young bit): 397 * dy..R...I...wr 398 * prot-none, clean, old 00..1...1...00 399 * prot-none, clean, young 01..1...1...00 400 * prot-none, dirty, old 10..1...1...00 401 * prot-none, dirty, young 11..1...1...00 402 * read-only, clean, old 00..1...1...01 403 * read-only, clean, young 01..1...0...01 404 * read-only, dirty, old 10..1...1...01 405 * read-only, dirty, young 11..1...0...01 406 * read-write, clean, old 00..1...1...11 407 * read-write, clean, young 01..1...0...11 408 * read-write, dirty, old 10..0...1...11 409 * read-write, dirty, young 11..0...0...11 410 * The segment table origin is used to distinguish empty (origin==0) from 411 * read-write, old segment table entries (origin!=0) 412 * HW-bits: R read-only, I invalid 413 * SW-bits: y young, d dirty, r read, w write 414 */ 415 416 /* Page status table bits for virtualization */ 417 #define PGSTE_ACC_BITS 0xf000000000000000UL 418 #define PGSTE_FP_BIT 0x0800000000000000UL 419 #define PGSTE_PCL_BIT 0x0080000000000000UL 420 #define PGSTE_HR_BIT 0x0040000000000000UL 421 #define PGSTE_HC_BIT 0x0020000000000000UL 422 #define PGSTE_GR_BIT 0x0004000000000000UL 423 #define PGSTE_GC_BIT 0x0002000000000000UL 424 #define PGSTE_ST2_MASK 0x0000ffff00000000UL 425 #define PGSTE_UC_BIT 0x0000000000008000UL /* user dirty (migration) */ 426 #define PGSTE_IN_BIT 0x0000000000004000UL /* IPTE notify bit */ 427 #define PGSTE_VSIE_BIT 0x0000000000002000UL /* ref'd in a shadow table */ 428 429 /* Guest Page State used for virtualization */ 430 #define _PGSTE_GPS_ZERO 0x0000000080000000UL 431 #define _PGSTE_GPS_NODAT 0x0000000040000000UL 432 #define _PGSTE_GPS_USAGE_MASK 0x0000000003000000UL 433 #define _PGSTE_GPS_USAGE_STABLE 0x0000000000000000UL 434 #define _PGSTE_GPS_USAGE_UNUSED 0x0000000001000000UL 435 #define _PGSTE_GPS_USAGE_POT_VOLATILE 0x0000000002000000UL 436 #define _PGSTE_GPS_USAGE_VOLATILE _PGSTE_GPS_USAGE_MASK 437 438 /* 439 * A user page table pointer has the space-switch-event bit, the 440 * private-space-control bit and the storage-alteration-event-control 441 * bit set. A kernel page table pointer doesn't need them. 442 */ 443 #define _ASCE_USER_BITS (_ASCE_SPACE_SWITCH | _ASCE_PRIVATE_SPACE | \ 444 _ASCE_ALT_EVENT) 445 446 /* 447 * Page protection definitions. 448 */ 449 #define __PAGE_NONE (_PAGE_PRESENT | _PAGE_INVALID | _PAGE_PROTECT) 450 #define __PAGE_RO (_PAGE_PRESENT | _PAGE_READ | \ 451 _PAGE_NOEXEC | _PAGE_INVALID | _PAGE_PROTECT) 452 #define __PAGE_RX (_PAGE_PRESENT | _PAGE_READ | \ 453 _PAGE_INVALID | _PAGE_PROTECT) 454 #define __PAGE_RW (_PAGE_PRESENT | _PAGE_READ | _PAGE_WRITE | \ 455 _PAGE_NOEXEC | _PAGE_INVALID | _PAGE_PROTECT) 456 #define __PAGE_RWX (_PAGE_PRESENT | _PAGE_READ | _PAGE_WRITE | \ 457 _PAGE_INVALID | _PAGE_PROTECT) 458 #define __PAGE_SHARED (_PAGE_PRESENT | _PAGE_READ | _PAGE_WRITE | \ 459 _PAGE_YOUNG | _PAGE_DIRTY | _PAGE_NOEXEC) 460 #define __PAGE_KERNEL (_PAGE_PRESENT | _PAGE_READ | _PAGE_WRITE | \ 461 _PAGE_YOUNG | _PAGE_DIRTY | _PAGE_NOEXEC) 462 #define __PAGE_KERNEL_RO (_PAGE_PRESENT | _PAGE_READ | _PAGE_YOUNG | \ 463 _PAGE_PROTECT | _PAGE_NOEXEC) 464 465 extern unsigned long page_noexec_mask; 466 467 #define __pgprot_page_mask(x) __pgprot((x) & page_noexec_mask) 468 469 #define PAGE_NONE __pgprot_page_mask(__PAGE_NONE) 470 #define PAGE_RO __pgprot_page_mask(__PAGE_RO) 471 #define PAGE_RX __pgprot_page_mask(__PAGE_RX) 472 #define PAGE_RW __pgprot_page_mask(__PAGE_RW) 473 #define PAGE_RWX __pgprot_page_mask(__PAGE_RWX) 474 #define PAGE_SHARED __pgprot_page_mask(__PAGE_SHARED) 475 #define PAGE_KERNEL __pgprot_page_mask(__PAGE_KERNEL) 476 #define PAGE_KERNEL_RO __pgprot_page_mask(__PAGE_KERNEL_RO) 477 478 /* 479 * Segment entry (large page) protection definitions. 480 */ 481 #define __SEGMENT_NONE (_SEGMENT_ENTRY_PRESENT | \ 482 _SEGMENT_ENTRY_INVALID | \ 483 _SEGMENT_ENTRY_PROTECT) 484 #define __SEGMENT_RO (_SEGMENT_ENTRY_PRESENT | \ 485 _SEGMENT_ENTRY_PROTECT | \ 486 _SEGMENT_ENTRY_READ | \ 487 _SEGMENT_ENTRY_NOEXEC) 488 #define __SEGMENT_RX (_SEGMENT_ENTRY_PRESENT | \ 489 _SEGMENT_ENTRY_PROTECT | \ 490 _SEGMENT_ENTRY_READ) 491 #define __SEGMENT_RW (_SEGMENT_ENTRY_PRESENT | \ 492 _SEGMENT_ENTRY_READ | \ 493 _SEGMENT_ENTRY_WRITE | \ 494 _SEGMENT_ENTRY_NOEXEC) 495 #define __SEGMENT_RWX (_SEGMENT_ENTRY_PRESENT | \ 496 _SEGMENT_ENTRY_READ | \ 497 _SEGMENT_ENTRY_WRITE) 498 #define __SEGMENT_KERNEL (_SEGMENT_ENTRY | \ 499 _SEGMENT_ENTRY_LARGE | \ 500 _SEGMENT_ENTRY_READ | \ 501 _SEGMENT_ENTRY_WRITE | \ 502 _SEGMENT_ENTRY_YOUNG | \ 503 _SEGMENT_ENTRY_DIRTY | \ 504 _SEGMENT_ENTRY_NOEXEC) 505 #define __SEGMENT_KERNEL_RO (_SEGMENT_ENTRY | \ 506 _SEGMENT_ENTRY_LARGE | \ 507 _SEGMENT_ENTRY_READ | \ 508 _SEGMENT_ENTRY_YOUNG | \ 509 _SEGMENT_ENTRY_PROTECT | \ 510 _SEGMENT_ENTRY_NOEXEC) 511 512 extern unsigned long segment_noexec_mask; 513 514 #define __pgprot_segment_mask(x) __pgprot((x) & segment_noexec_mask) 515 516 #define SEGMENT_NONE __pgprot_segment_mask(__SEGMENT_NONE) 517 #define SEGMENT_RO __pgprot_segment_mask(__SEGMENT_RO) 518 #define SEGMENT_RX __pgprot_segment_mask(__SEGMENT_RX) 519 #define SEGMENT_RW __pgprot_segment_mask(__SEGMENT_RW) 520 #define SEGMENT_RWX __pgprot_segment_mask(__SEGMENT_RWX) 521 #define SEGMENT_KERNEL __pgprot_segment_mask(__SEGMENT_KERNEL) 522 #define SEGMENT_KERNEL_RO __pgprot_segment_mask(__SEGMENT_KERNEL_RO) 523 524 /* 525 * Region3 entry (large page) protection definitions. 526 */ 527 528 #define __REGION3_KERNEL (_REGION_ENTRY_TYPE_R3 | \ 529 _REGION3_ENTRY_PRESENT | \ 530 _REGION3_ENTRY_LARGE | \ 531 _REGION3_ENTRY_READ | \ 532 _REGION3_ENTRY_WRITE | \ 533 _REGION3_ENTRY_YOUNG | \ 534 _REGION3_ENTRY_DIRTY | \ 535 _REGION_ENTRY_NOEXEC) 536 #define __REGION3_KERNEL_RO (_REGION_ENTRY_TYPE_R3 | \ 537 _REGION3_ENTRY_PRESENT | \ 538 _REGION3_ENTRY_LARGE | \ 539 _REGION3_ENTRY_READ | \ 540 _REGION3_ENTRY_YOUNG | \ 541 _REGION_ENTRY_PROTECT | \ 542 _REGION_ENTRY_NOEXEC) 543 544 extern unsigned long region_noexec_mask; 545 546 #define __pgprot_region_mask(x) __pgprot((x) & region_noexec_mask) 547 548 #define REGION3_KERNEL __pgprot_region_mask(__REGION3_KERNEL) 549 #define REGION3_KERNEL_RO __pgprot_region_mask(__REGION3_KERNEL_RO) 550 551 static inline bool mm_p4d_folded(struct mm_struct *mm) 552 { 553 return mm->context.asce_limit <= _REGION1_SIZE; 554 } 555 #define mm_p4d_folded(mm) mm_p4d_folded(mm) 556 557 static inline bool mm_pud_folded(struct mm_struct *mm) 558 { 559 return mm->context.asce_limit <= _REGION2_SIZE; 560 } 561 #define mm_pud_folded(mm) mm_pud_folded(mm) 562 563 static inline bool mm_pmd_folded(struct mm_struct *mm) 564 { 565 return mm->context.asce_limit <= _REGION3_SIZE; 566 } 567 #define mm_pmd_folded(mm) mm_pmd_folded(mm) 568 569 static inline int mm_has_pgste(struct mm_struct *mm) 570 { 571 #ifdef CONFIG_PGSTE 572 if (unlikely(mm->context.has_pgste)) 573 return 1; 574 #endif 575 return 0; 576 } 577 578 static inline int mm_is_protected(struct mm_struct *mm) 579 { 580 #ifdef CONFIG_PGSTE 581 if (unlikely(atomic_read(&mm->context.protected_count))) 582 return 1; 583 #endif 584 return 0; 585 } 586 587 static inline pgste_t clear_pgste_bit(pgste_t pgste, unsigned long mask) 588 { 589 return __pgste(pgste_val(pgste) & ~mask); 590 } 591 592 static inline pgste_t set_pgste_bit(pgste_t pgste, unsigned long mask) 593 { 594 return __pgste(pgste_val(pgste) | mask); 595 } 596 597 static inline pte_t clear_pte_bit(pte_t pte, pgprot_t prot) 598 { 599 return __pte(pte_val(pte) & ~pgprot_val(prot)); 600 } 601 602 static inline pte_t set_pte_bit(pte_t pte, pgprot_t prot) 603 { 604 return __pte(pte_val(pte) | pgprot_val(prot)); 605 } 606 607 static inline pmd_t clear_pmd_bit(pmd_t pmd, pgprot_t prot) 608 { 609 return __pmd(pmd_val(pmd) & ~pgprot_val(prot)); 610 } 611 612 static inline pmd_t set_pmd_bit(pmd_t pmd, pgprot_t prot) 613 { 614 return __pmd(pmd_val(pmd) | pgprot_val(prot)); 615 } 616 617 static inline pud_t clear_pud_bit(pud_t pud, pgprot_t prot) 618 { 619 return __pud(pud_val(pud) & ~pgprot_val(prot)); 620 } 621 622 static inline pud_t set_pud_bit(pud_t pud, pgprot_t prot) 623 { 624 return __pud(pud_val(pud) | pgprot_val(prot)); 625 } 626 627 /* 628 * As soon as the guest uses storage keys or enables PV, we deduplicate all 629 * mapped shared zeropages and prevent new shared zeropages from getting 630 * mapped. 631 */ 632 #define mm_forbids_zeropage mm_forbids_zeropage 633 static inline int mm_forbids_zeropage(struct mm_struct *mm) 634 { 635 #ifdef CONFIG_PGSTE 636 if (!mm->context.allow_cow_sharing) 637 return 1; 638 #endif 639 return 0; 640 } 641 642 static inline int mm_uses_skeys(struct mm_struct *mm) 643 { 644 #ifdef CONFIG_PGSTE 645 if (mm->context.uses_skeys) 646 return 1; 647 #endif 648 return 0; 649 } 650 651 /** 652 * cspg() - Compare and Swap and Purge (CSPG) 653 * @ptr: Pointer to the value to be exchanged 654 * @old: The expected old value 655 * @new: The new value 656 * 657 * Return: True if compare and swap was successful, otherwise false. 658 */ 659 static inline bool cspg(unsigned long *ptr, unsigned long old, unsigned long new) 660 { 661 union register_pair r1 = { .even = old, .odd = new, }; 662 unsigned long address = (unsigned long)ptr | 1; 663 664 asm volatile( 665 " cspg %[r1],%[address]" 666 : [r1] "+&d" (r1.pair), "+m" (*ptr) 667 : [address] "d" (address) 668 : "cc"); 669 return old == r1.even; 670 } 671 672 #define CRDTE_DTT_PAGE 0x00UL 673 #define CRDTE_DTT_SEGMENT 0x10UL 674 #define CRDTE_DTT_REGION3 0x14UL 675 #define CRDTE_DTT_REGION2 0x18UL 676 #define CRDTE_DTT_REGION1 0x1cUL 677 678 /** 679 * crdte() - Compare and Replace DAT Table Entry 680 * @old: The expected old value 681 * @new: The new value 682 * @table: Pointer to the value to be exchanged 683 * @dtt: Table type of the table to be exchanged 684 * @address: The address mapped by the entry to be replaced 685 * @asce: The ASCE of this entry 686 * 687 * Return: True if compare and replace was successful, otherwise false. 688 */ 689 static inline bool crdte(unsigned long old, unsigned long new, 690 unsigned long *table, unsigned long dtt, 691 unsigned long address, unsigned long asce) 692 { 693 union register_pair r1 = { .even = old, .odd = new, }; 694 union register_pair r2 = { .even = __pa(table) | dtt, .odd = address, }; 695 696 asm volatile(".insn rrf,0xb98f0000,%[r1],%[r2],%[asce],0" 697 : [r1] "+&d" (r1.pair) 698 : [r2] "d" (r2.pair), [asce] "a" (asce) 699 : "memory", "cc"); 700 return old == r1.even; 701 } 702 703 /* 704 * pgd/p4d/pud/pmd/pte query functions 705 */ 706 static inline int pgd_folded(pgd_t pgd) 707 { 708 return (pgd_val(pgd) & _REGION_ENTRY_TYPE_MASK) < _REGION_ENTRY_TYPE_R1; 709 } 710 711 static inline int pgd_present(pgd_t pgd) 712 { 713 if (pgd_folded(pgd)) 714 return 1; 715 return (pgd_val(pgd) & _REGION_ENTRY_ORIGIN) != 0UL; 716 } 717 718 static inline int pgd_none(pgd_t pgd) 719 { 720 if (pgd_folded(pgd)) 721 return 0; 722 return (pgd_val(pgd) & _REGION_ENTRY_INVALID) != 0UL; 723 } 724 725 static inline int pgd_bad(pgd_t pgd) 726 { 727 if ((pgd_val(pgd) & _REGION_ENTRY_TYPE_MASK) < _REGION_ENTRY_TYPE_R1) 728 return 0; 729 return (pgd_val(pgd) & ~_REGION_ENTRY_BITS) != 0; 730 } 731 732 static inline unsigned long pgd_pfn(pgd_t pgd) 733 { 734 unsigned long origin_mask; 735 736 origin_mask = _REGION_ENTRY_ORIGIN; 737 return (pgd_val(pgd) & origin_mask) >> PAGE_SHIFT; 738 } 739 740 static inline int p4d_folded(p4d_t p4d) 741 { 742 return (p4d_val(p4d) & _REGION_ENTRY_TYPE_MASK) < _REGION_ENTRY_TYPE_R2; 743 } 744 745 static inline int p4d_present(p4d_t p4d) 746 { 747 if (p4d_folded(p4d)) 748 return 1; 749 return (p4d_val(p4d) & _REGION_ENTRY_ORIGIN) != 0UL; 750 } 751 752 static inline int p4d_none(p4d_t p4d) 753 { 754 if (p4d_folded(p4d)) 755 return 0; 756 return p4d_val(p4d) == _REGION2_ENTRY_EMPTY; 757 } 758 759 static inline unsigned long p4d_pfn(p4d_t p4d) 760 { 761 unsigned long origin_mask; 762 763 origin_mask = _REGION_ENTRY_ORIGIN; 764 return (p4d_val(p4d) & origin_mask) >> PAGE_SHIFT; 765 } 766 767 static inline int pud_folded(pud_t pud) 768 { 769 return (pud_val(pud) & _REGION_ENTRY_TYPE_MASK) < _REGION_ENTRY_TYPE_R3; 770 } 771 772 static inline int pud_present(pud_t pud) 773 { 774 if (pud_folded(pud)) 775 return 1; 776 return (pud_val(pud) & _REGION3_ENTRY_PRESENT) != 0; 777 } 778 779 static inline int pud_none(pud_t pud) 780 { 781 if (pud_folded(pud)) 782 return 0; 783 return pud_val(pud) == _REGION3_ENTRY_EMPTY; 784 } 785 786 #define pud_leaf pud_leaf 787 static inline bool pud_leaf(pud_t pud) 788 { 789 if ((pud_val(pud) & _REGION_ENTRY_TYPE_MASK) != _REGION_ENTRY_TYPE_R3) 790 return 0; 791 return (pud_present(pud) && (pud_val(pud) & _REGION3_ENTRY_LARGE) != 0); 792 } 793 794 static inline int pmd_present(pmd_t pmd) 795 { 796 return (pmd_val(pmd) & _SEGMENT_ENTRY_PRESENT) != 0; 797 } 798 799 #define pmd_leaf pmd_leaf 800 static inline bool pmd_leaf(pmd_t pmd) 801 { 802 return (pmd_present(pmd) && (pmd_val(pmd) & _SEGMENT_ENTRY_LARGE) != 0); 803 } 804 805 static inline int pmd_bad(pmd_t pmd) 806 { 807 if ((pmd_val(pmd) & _SEGMENT_ENTRY_TYPE_MASK) > 0 || pmd_leaf(pmd)) 808 return 1; 809 return (pmd_val(pmd) & ~_SEGMENT_ENTRY_BITS) != 0; 810 } 811 812 static inline int pud_bad(pud_t pud) 813 { 814 unsigned long type = pud_val(pud) & _REGION_ENTRY_TYPE_MASK; 815 816 if (type > _REGION_ENTRY_TYPE_R3 || pud_leaf(pud)) 817 return 1; 818 if (type < _REGION_ENTRY_TYPE_R3) 819 return 0; 820 return (pud_val(pud) & ~_REGION_ENTRY_BITS) != 0; 821 } 822 823 static inline int p4d_bad(p4d_t p4d) 824 { 825 unsigned long type = p4d_val(p4d) & _REGION_ENTRY_TYPE_MASK; 826 827 if (type > _REGION_ENTRY_TYPE_R2) 828 return 1; 829 if (type < _REGION_ENTRY_TYPE_R2) 830 return 0; 831 return (p4d_val(p4d) & ~_REGION_ENTRY_BITS) != 0; 832 } 833 834 static inline int pmd_none(pmd_t pmd) 835 { 836 return pmd_val(pmd) == _SEGMENT_ENTRY_EMPTY; 837 } 838 839 #define pmd_write pmd_write 840 static inline int pmd_write(pmd_t pmd) 841 { 842 return (pmd_val(pmd) & _SEGMENT_ENTRY_WRITE) != 0; 843 } 844 845 #define pud_write pud_write 846 static inline int pud_write(pud_t pud) 847 { 848 return (pud_val(pud) & _REGION3_ENTRY_WRITE) != 0; 849 } 850 851 #define pmd_dirty pmd_dirty 852 static inline int pmd_dirty(pmd_t pmd) 853 { 854 return (pmd_val(pmd) & _SEGMENT_ENTRY_DIRTY) != 0; 855 } 856 857 #define pmd_young pmd_young 858 static inline int pmd_young(pmd_t pmd) 859 { 860 return (pmd_val(pmd) & _SEGMENT_ENTRY_YOUNG) != 0; 861 } 862 863 static inline int pte_present(pte_t pte) 864 { 865 /* Bit pattern: (pte & 0x001) == 0x001 */ 866 return (pte_val(pte) & _PAGE_PRESENT) != 0; 867 } 868 869 static inline int pte_none(pte_t pte) 870 { 871 /* Bit pattern: pte == 0x400 */ 872 return pte_val(pte) == _PAGE_INVALID; 873 } 874 875 static inline int pte_swap(pte_t pte) 876 { 877 /* Bit pattern: (pte & 0x201) == 0x200 */ 878 return (pte_val(pte) & (_PAGE_PROTECT | _PAGE_PRESENT)) 879 == _PAGE_PROTECT; 880 } 881 882 static inline int pte_special(pte_t pte) 883 { 884 return (pte_val(pte) & _PAGE_SPECIAL); 885 } 886 887 #define __HAVE_ARCH_PTE_SAME 888 static inline int pte_same(pte_t a, pte_t b) 889 { 890 return pte_val(a) == pte_val(b); 891 } 892 893 #ifdef CONFIG_NUMA_BALANCING 894 static inline int pte_protnone(pte_t pte) 895 { 896 return pte_present(pte) && !(pte_val(pte) & _PAGE_READ); 897 } 898 899 static inline int pmd_protnone(pmd_t pmd) 900 { 901 /* pmd_leaf(pmd) implies pmd_present(pmd) */ 902 return pmd_leaf(pmd) && !(pmd_val(pmd) & _SEGMENT_ENTRY_READ); 903 } 904 #endif 905 906 static inline bool pte_swp_exclusive(pte_t pte) 907 { 908 return pte_val(pte) & _PAGE_SWP_EXCLUSIVE; 909 } 910 911 static inline pte_t pte_swp_mkexclusive(pte_t pte) 912 { 913 return set_pte_bit(pte, __pgprot(_PAGE_SWP_EXCLUSIVE)); 914 } 915 916 static inline pte_t pte_swp_clear_exclusive(pte_t pte) 917 { 918 return clear_pte_bit(pte, __pgprot(_PAGE_SWP_EXCLUSIVE)); 919 } 920 921 static inline int pte_soft_dirty(pte_t pte) 922 { 923 return pte_val(pte) & _PAGE_SOFT_DIRTY; 924 } 925 #define pte_swp_soft_dirty pte_soft_dirty 926 927 static inline pte_t pte_mksoft_dirty(pte_t pte) 928 { 929 return set_pte_bit(pte, __pgprot(_PAGE_SOFT_DIRTY)); 930 } 931 #define pte_swp_mksoft_dirty pte_mksoft_dirty 932 933 static inline pte_t pte_clear_soft_dirty(pte_t pte) 934 { 935 return clear_pte_bit(pte, __pgprot(_PAGE_SOFT_DIRTY)); 936 } 937 #define pte_swp_clear_soft_dirty pte_clear_soft_dirty 938 939 static inline int pmd_soft_dirty(pmd_t pmd) 940 { 941 return pmd_val(pmd) & _SEGMENT_ENTRY_SOFT_DIRTY; 942 } 943 944 static inline pmd_t pmd_mksoft_dirty(pmd_t pmd) 945 { 946 return set_pmd_bit(pmd, __pgprot(_SEGMENT_ENTRY_SOFT_DIRTY)); 947 } 948 949 static inline pmd_t pmd_clear_soft_dirty(pmd_t pmd) 950 { 951 return clear_pmd_bit(pmd, __pgprot(_SEGMENT_ENTRY_SOFT_DIRTY)); 952 } 953 954 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION 955 #define pmd_swp_soft_dirty(pmd) pmd_soft_dirty(pmd) 956 #define pmd_swp_mksoft_dirty(pmd) pmd_mksoft_dirty(pmd) 957 #define pmd_swp_clear_soft_dirty(pmd) pmd_clear_soft_dirty(pmd) 958 #endif 959 960 /* 961 * query functions pte_write/pte_dirty/pte_young only work if 962 * pte_present() is true. Undefined behaviour if not.. 963 */ 964 static inline int pte_write(pte_t pte) 965 { 966 return (pte_val(pte) & _PAGE_WRITE) != 0; 967 } 968 969 static inline int pte_dirty(pte_t pte) 970 { 971 return (pte_val(pte) & _PAGE_DIRTY) != 0; 972 } 973 974 static inline int pte_young(pte_t pte) 975 { 976 return (pte_val(pte) & _PAGE_YOUNG) != 0; 977 } 978 979 #define __HAVE_ARCH_PTE_UNUSED 980 static inline int pte_unused(pte_t pte) 981 { 982 return pte_val(pte) & _PAGE_UNUSED; 983 } 984 985 /* 986 * Extract the pgprot value from the given pte while at the same time making it 987 * usable for kernel address space mappings where fault driven dirty and 988 * young/old accounting is not supported, i.e _PAGE_PROTECT and _PAGE_INVALID 989 * must not be set. 990 */ 991 #define pte_pgprot pte_pgprot 992 static inline pgprot_t pte_pgprot(pte_t pte) 993 { 994 unsigned long pte_flags = pte_val(pte) & _PAGE_CHG_MASK; 995 996 if (pte_write(pte)) 997 pte_flags |= pgprot_val(PAGE_KERNEL); 998 else 999 pte_flags |= pgprot_val(PAGE_KERNEL_RO); 1000 pte_flags |= pte_val(pte) & mio_wb_bit_mask; 1001 1002 return __pgprot(pte_flags); 1003 } 1004 1005 /* 1006 * pgd/pmd/pte modification functions 1007 */ 1008 1009 static inline void set_pgd(pgd_t *pgdp, pgd_t pgd) 1010 { 1011 WRITE_ONCE(*pgdp, pgd); 1012 } 1013 1014 static inline void set_p4d(p4d_t *p4dp, p4d_t p4d) 1015 { 1016 WRITE_ONCE(*p4dp, p4d); 1017 } 1018 1019 static inline void set_pud(pud_t *pudp, pud_t pud) 1020 { 1021 WRITE_ONCE(*pudp, pud); 1022 } 1023 1024 static inline void set_pmd(pmd_t *pmdp, pmd_t pmd) 1025 { 1026 WRITE_ONCE(*pmdp, pmd); 1027 } 1028 1029 static inline void set_pte(pte_t *ptep, pte_t pte) 1030 { 1031 WRITE_ONCE(*ptep, pte); 1032 } 1033 1034 static inline void pgd_clear(pgd_t *pgd) 1035 { 1036 if ((pgd_val(*pgd) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R1) 1037 set_pgd(pgd, __pgd(_REGION1_ENTRY_EMPTY)); 1038 } 1039 1040 static inline void p4d_clear(p4d_t *p4d) 1041 { 1042 if ((p4d_val(*p4d) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R2) 1043 set_p4d(p4d, __p4d(_REGION2_ENTRY_EMPTY)); 1044 } 1045 1046 static inline void pud_clear(pud_t *pud) 1047 { 1048 if ((pud_val(*pud) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R3) 1049 set_pud(pud, __pud(_REGION3_ENTRY_EMPTY)); 1050 } 1051 1052 static inline void pmd_clear(pmd_t *pmdp) 1053 { 1054 set_pmd(pmdp, __pmd(_SEGMENT_ENTRY_EMPTY)); 1055 } 1056 1057 static inline void pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep) 1058 { 1059 set_pte(ptep, __pte(_PAGE_INVALID)); 1060 } 1061 1062 /* 1063 * The following pte modification functions only work if 1064 * pte_present() is true. Undefined behaviour if not.. 1065 */ 1066 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot) 1067 { 1068 pte = clear_pte_bit(pte, __pgprot(~_PAGE_CHG_MASK)); 1069 pte = set_pte_bit(pte, newprot); 1070 /* 1071 * newprot for PAGE_NONE, PAGE_RO, PAGE_RX, PAGE_RW and PAGE_RWX 1072 * has the invalid bit set, clear it again for readable, young pages 1073 */ 1074 if ((pte_val(pte) & _PAGE_YOUNG) && (pte_val(pte) & _PAGE_READ)) 1075 pte = clear_pte_bit(pte, __pgprot(_PAGE_INVALID)); 1076 /* 1077 * newprot for PAGE_RO, PAGE_RX, PAGE_RW and PAGE_RWX has the page 1078 * protection bit set, clear it again for writable, dirty pages 1079 */ 1080 if ((pte_val(pte) & _PAGE_DIRTY) && (pte_val(pte) & _PAGE_WRITE)) 1081 pte = clear_pte_bit(pte, __pgprot(_PAGE_PROTECT)); 1082 return pte; 1083 } 1084 1085 static inline pte_t pte_wrprotect(pte_t pte) 1086 { 1087 pte = clear_pte_bit(pte, __pgprot(_PAGE_WRITE)); 1088 return set_pte_bit(pte, __pgprot(_PAGE_PROTECT)); 1089 } 1090 1091 static inline pte_t pte_mkwrite_novma(pte_t pte) 1092 { 1093 pte = set_pte_bit(pte, __pgprot(_PAGE_WRITE)); 1094 if (pte_val(pte) & _PAGE_DIRTY) 1095 pte = clear_pte_bit(pte, __pgprot(_PAGE_PROTECT)); 1096 return pte; 1097 } 1098 1099 static inline pte_t pte_mkclean(pte_t pte) 1100 { 1101 pte = clear_pte_bit(pte, __pgprot(_PAGE_DIRTY)); 1102 return set_pte_bit(pte, __pgprot(_PAGE_PROTECT)); 1103 } 1104 1105 static inline pte_t pte_mkdirty(pte_t pte) 1106 { 1107 pte = set_pte_bit(pte, __pgprot(_PAGE_DIRTY | _PAGE_SOFT_DIRTY)); 1108 if (pte_val(pte) & _PAGE_WRITE) 1109 pte = clear_pte_bit(pte, __pgprot(_PAGE_PROTECT)); 1110 return pte; 1111 } 1112 1113 static inline pte_t pte_mkold(pte_t pte) 1114 { 1115 pte = clear_pte_bit(pte, __pgprot(_PAGE_YOUNG)); 1116 return set_pte_bit(pte, __pgprot(_PAGE_INVALID)); 1117 } 1118 1119 static inline pte_t pte_mkyoung(pte_t pte) 1120 { 1121 pte = set_pte_bit(pte, __pgprot(_PAGE_YOUNG)); 1122 if (pte_val(pte) & _PAGE_READ) 1123 pte = clear_pte_bit(pte, __pgprot(_PAGE_INVALID)); 1124 return pte; 1125 } 1126 1127 static inline pte_t pte_mkspecial(pte_t pte) 1128 { 1129 return set_pte_bit(pte, __pgprot(_PAGE_SPECIAL)); 1130 } 1131 1132 #ifdef CONFIG_HUGETLB_PAGE 1133 static inline pte_t pte_mkhuge(pte_t pte) 1134 { 1135 return set_pte_bit(pte, __pgprot(_PAGE_LARGE)); 1136 } 1137 #endif 1138 1139 #define IPTE_GLOBAL 0 1140 #define IPTE_LOCAL 1 1141 1142 #define IPTE_NODAT 0x400 1143 #define IPTE_GUEST_ASCE 0x800 1144 1145 static __always_inline void __ptep_rdp(unsigned long addr, pte_t *ptep, int local) 1146 { 1147 unsigned long pto; 1148 1149 pto = __pa(ptep) & ~(PTRS_PER_PTE * sizeof(pte_t) - 1); 1150 asm volatile(".insn rrf,0xb98b0000,%[r1],%[r2],%%r0,%[m4]" 1151 : "+m" (*ptep) 1152 : [r1] "a" (pto), [r2] "a" (addr & PAGE_MASK), 1153 [m4] "i" (local)); 1154 } 1155 1156 static __always_inline void __ptep_ipte(unsigned long address, pte_t *ptep, 1157 unsigned long opt, unsigned long asce, 1158 int local) 1159 { 1160 unsigned long pto = __pa(ptep); 1161 1162 if (__builtin_constant_p(opt) && opt == 0) { 1163 /* Invalidation + TLB flush for the pte */ 1164 asm volatile( 1165 " ipte %[r1],%[r2],0,%[m4]" 1166 : "+m" (*ptep) : [r1] "a" (pto), [r2] "a" (address), 1167 [m4] "i" (local)); 1168 return; 1169 } 1170 1171 /* Invalidate ptes with options + TLB flush of the ptes */ 1172 opt = opt | (asce & _ASCE_ORIGIN); 1173 asm volatile( 1174 " ipte %[r1],%[r2],%[r3],%[m4]" 1175 : [r2] "+a" (address), [r3] "+a" (opt) 1176 : [r1] "a" (pto), [m4] "i" (local) : "memory"); 1177 } 1178 1179 static __always_inline void __ptep_ipte_range(unsigned long address, int nr, 1180 pte_t *ptep, int local) 1181 { 1182 unsigned long pto = __pa(ptep); 1183 1184 /* Invalidate a range of ptes + TLB flush of the ptes */ 1185 do { 1186 asm volatile( 1187 " ipte %[r1],%[r2],%[r3],%[m4]" 1188 : [r2] "+a" (address), [r3] "+a" (nr) 1189 : [r1] "a" (pto), [m4] "i" (local) : "memory"); 1190 } while (nr != 255); 1191 } 1192 1193 /* 1194 * This is hard to understand. ptep_get_and_clear and ptep_clear_flush 1195 * both clear the TLB for the unmapped pte. The reason is that 1196 * ptep_get_and_clear is used in common code (e.g. change_pte_range) 1197 * to modify an active pte. The sequence is 1198 * 1) ptep_get_and_clear 1199 * 2) set_pte_at 1200 * 3) flush_tlb_range 1201 * On s390 the tlb needs to get flushed with the modification of the pte 1202 * if the pte is active. The only way how this can be implemented is to 1203 * have ptep_get_and_clear do the tlb flush. In exchange flush_tlb_range 1204 * is a nop. 1205 */ 1206 pte_t ptep_xchg_direct(struct mm_struct *, unsigned long, pte_t *, pte_t); 1207 pte_t ptep_xchg_lazy(struct mm_struct *, unsigned long, pte_t *, pte_t); 1208 1209 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG 1210 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma, 1211 unsigned long addr, pte_t *ptep) 1212 { 1213 pte_t pte = *ptep; 1214 1215 pte = ptep_xchg_direct(vma->vm_mm, addr, ptep, pte_mkold(pte)); 1216 return pte_young(pte); 1217 } 1218 1219 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH 1220 static inline int ptep_clear_flush_young(struct vm_area_struct *vma, 1221 unsigned long address, pte_t *ptep) 1222 { 1223 return ptep_test_and_clear_young(vma, address, ptep); 1224 } 1225 1226 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR 1227 static inline pte_t ptep_get_and_clear(struct mm_struct *mm, 1228 unsigned long addr, pte_t *ptep) 1229 { 1230 pte_t res; 1231 1232 res = ptep_xchg_lazy(mm, addr, ptep, __pte(_PAGE_INVALID)); 1233 /* At this point the reference through the mapping is still present */ 1234 if (mm_is_protected(mm) && pte_present(res)) 1235 uv_convert_from_secure_pte(res); 1236 return res; 1237 } 1238 1239 #define __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION 1240 pte_t ptep_modify_prot_start(struct vm_area_struct *, unsigned long, pte_t *); 1241 void ptep_modify_prot_commit(struct vm_area_struct *, unsigned long, 1242 pte_t *, pte_t, pte_t); 1243 1244 #define __HAVE_ARCH_PTEP_CLEAR_FLUSH 1245 static inline pte_t ptep_clear_flush(struct vm_area_struct *vma, 1246 unsigned long addr, pte_t *ptep) 1247 { 1248 pte_t res; 1249 1250 res = ptep_xchg_direct(vma->vm_mm, addr, ptep, __pte(_PAGE_INVALID)); 1251 /* At this point the reference through the mapping is still present */ 1252 if (mm_is_protected(vma->vm_mm) && pte_present(res)) 1253 uv_convert_from_secure_pte(res); 1254 return res; 1255 } 1256 1257 /* 1258 * The batched pte unmap code uses ptep_get_and_clear_full to clear the 1259 * ptes. Here an optimization is possible. tlb_gather_mmu flushes all 1260 * tlbs of an mm if it can guarantee that the ptes of the mm_struct 1261 * cannot be accessed while the batched unmap is running. In this case 1262 * full==1 and a simple pte_clear is enough. See tlb.h. 1263 */ 1264 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL 1265 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm, 1266 unsigned long addr, 1267 pte_t *ptep, int full) 1268 { 1269 pte_t res; 1270 1271 if (full) { 1272 res = *ptep; 1273 set_pte(ptep, __pte(_PAGE_INVALID)); 1274 } else { 1275 res = ptep_xchg_lazy(mm, addr, ptep, __pte(_PAGE_INVALID)); 1276 } 1277 /* Nothing to do */ 1278 if (!mm_is_protected(mm) || !pte_present(res)) 1279 return res; 1280 /* 1281 * At this point the reference through the mapping is still present. 1282 * The notifier should have destroyed all protected vCPUs at this 1283 * point, so the destroy should be successful. 1284 */ 1285 if (full && !uv_destroy_pte(res)) 1286 return res; 1287 /* 1288 * If something went wrong and the page could not be destroyed, or 1289 * if this is not a mm teardown, the slower export is used as 1290 * fallback instead. 1291 */ 1292 uv_convert_from_secure_pte(res); 1293 return res; 1294 } 1295 1296 #define __HAVE_ARCH_PTEP_SET_WRPROTECT 1297 static inline void ptep_set_wrprotect(struct mm_struct *mm, 1298 unsigned long addr, pte_t *ptep) 1299 { 1300 pte_t pte = *ptep; 1301 1302 if (pte_write(pte)) 1303 ptep_xchg_lazy(mm, addr, ptep, pte_wrprotect(pte)); 1304 } 1305 1306 /* 1307 * Check if PTEs only differ in _PAGE_PROTECT HW bit, but also allow SW PTE 1308 * bits in the comparison. Those might change e.g. because of dirty and young 1309 * tracking. 1310 */ 1311 static inline int pte_allow_rdp(pte_t old, pte_t new) 1312 { 1313 /* 1314 * Only allow changes from RO to RW 1315 */ 1316 if (!(pte_val(old) & _PAGE_PROTECT) || pte_val(new) & _PAGE_PROTECT) 1317 return 0; 1318 1319 return (pte_val(old) & _PAGE_RDP_MASK) == (pte_val(new) & _PAGE_RDP_MASK); 1320 } 1321 1322 static inline void flush_tlb_fix_spurious_fault(struct vm_area_struct *vma, 1323 unsigned long address, 1324 pte_t *ptep) 1325 { 1326 /* 1327 * RDP might not have propagated the PTE protection reset to all CPUs, 1328 * so there could be spurious TLB protection faults. 1329 * NOTE: This will also be called when a racing pagetable update on 1330 * another thread already installed the correct PTE. Both cases cannot 1331 * really be distinguished. 1332 * Therefore, only do the local TLB flush when RDP can be used, and the 1333 * PTE does not have _PAGE_PROTECT set, to avoid unnecessary overhead. 1334 * A local RDP can be used to do the flush. 1335 */ 1336 if (cpu_has_rdp() && !(pte_val(*ptep) & _PAGE_PROTECT)) 1337 __ptep_rdp(address, ptep, 1); 1338 } 1339 #define flush_tlb_fix_spurious_fault flush_tlb_fix_spurious_fault 1340 1341 void ptep_reset_dat_prot(struct mm_struct *mm, unsigned long addr, pte_t *ptep, 1342 pte_t new); 1343 1344 #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS 1345 static inline int ptep_set_access_flags(struct vm_area_struct *vma, 1346 unsigned long addr, pte_t *ptep, 1347 pte_t entry, int dirty) 1348 { 1349 if (pte_same(*ptep, entry)) 1350 return 0; 1351 if (cpu_has_rdp() && !mm_has_pgste(vma->vm_mm) && pte_allow_rdp(*ptep, entry)) 1352 ptep_reset_dat_prot(vma->vm_mm, addr, ptep, entry); 1353 else 1354 ptep_xchg_direct(vma->vm_mm, addr, ptep, entry); 1355 return 1; 1356 } 1357 1358 /* 1359 * Additional functions to handle KVM guest page tables 1360 */ 1361 void ptep_set_pte_at(struct mm_struct *mm, unsigned long addr, 1362 pte_t *ptep, pte_t entry); 1363 void ptep_set_notify(struct mm_struct *mm, unsigned long addr, pte_t *ptep); 1364 void ptep_notify(struct mm_struct *mm, unsigned long addr, 1365 pte_t *ptep, unsigned long bits); 1366 int ptep_force_prot(struct mm_struct *mm, unsigned long gaddr, 1367 pte_t *ptep, int prot, unsigned long bit); 1368 void ptep_zap_unused(struct mm_struct *mm, unsigned long addr, 1369 pte_t *ptep , int reset); 1370 void ptep_zap_key(struct mm_struct *mm, unsigned long addr, pte_t *ptep); 1371 int ptep_shadow_pte(struct mm_struct *mm, unsigned long saddr, 1372 pte_t *sptep, pte_t *tptep, pte_t pte); 1373 void ptep_unshadow_pte(struct mm_struct *mm, unsigned long saddr, pte_t *ptep); 1374 1375 bool ptep_test_and_clear_uc(struct mm_struct *mm, unsigned long address, 1376 pte_t *ptep); 1377 int set_guest_storage_key(struct mm_struct *mm, unsigned long addr, 1378 unsigned char key, bool nq); 1379 int cond_set_guest_storage_key(struct mm_struct *mm, unsigned long addr, 1380 unsigned char key, unsigned char *oldkey, 1381 bool nq, bool mr, bool mc); 1382 int reset_guest_reference_bit(struct mm_struct *mm, unsigned long addr); 1383 int get_guest_storage_key(struct mm_struct *mm, unsigned long addr, 1384 unsigned char *key); 1385 1386 int set_pgste_bits(struct mm_struct *mm, unsigned long addr, 1387 unsigned long bits, unsigned long value); 1388 int get_pgste(struct mm_struct *mm, unsigned long hva, unsigned long *pgstep); 1389 int pgste_perform_essa(struct mm_struct *mm, unsigned long hva, int orc, 1390 unsigned long *oldpte, unsigned long *oldpgste); 1391 void gmap_pmdp_invalidate(struct mm_struct *mm, unsigned long vmaddr); 1392 void gmap_pmdp_idte_local(struct mm_struct *mm, unsigned long vmaddr); 1393 void gmap_pmdp_idte_global(struct mm_struct *mm, unsigned long vmaddr); 1394 1395 #define pgprot_writecombine pgprot_writecombine 1396 pgprot_t pgprot_writecombine(pgprot_t prot); 1397 1398 #define PFN_PTE_SHIFT PAGE_SHIFT 1399 1400 /* 1401 * Set multiple PTEs to consecutive pages with a single call. All PTEs 1402 * are within the same folio, PMD and VMA. 1403 */ 1404 static inline void set_ptes(struct mm_struct *mm, unsigned long addr, 1405 pte_t *ptep, pte_t entry, unsigned int nr) 1406 { 1407 if (pte_present(entry)) 1408 entry = clear_pte_bit(entry, __pgprot(_PAGE_UNUSED)); 1409 if (mm_has_pgste(mm)) { 1410 for (;;) { 1411 ptep_set_pte_at(mm, addr, ptep, entry); 1412 if (--nr == 0) 1413 break; 1414 ptep++; 1415 entry = __pte(pte_val(entry) + PAGE_SIZE); 1416 addr += PAGE_SIZE; 1417 } 1418 } else { 1419 for (;;) { 1420 set_pte(ptep, entry); 1421 if (--nr == 0) 1422 break; 1423 ptep++; 1424 entry = __pte(pte_val(entry) + PAGE_SIZE); 1425 } 1426 } 1427 } 1428 #define set_ptes set_ptes 1429 1430 /* 1431 * Conversion functions: convert a page and protection to a page entry, 1432 * and a page entry and page directory to the page they refer to. 1433 */ 1434 static inline pte_t mk_pte_phys(unsigned long physpage, pgprot_t pgprot) 1435 { 1436 pte_t __pte; 1437 1438 __pte = __pte(physpage | pgprot_val(pgprot)); 1439 return pte_mkyoung(__pte); 1440 } 1441 1442 #define pgd_index(address) (((address) >> PGDIR_SHIFT) & (PTRS_PER_PGD-1)) 1443 #define p4d_index(address) (((address) >> P4D_SHIFT) & (PTRS_PER_P4D-1)) 1444 #define pud_index(address) (((address) >> PUD_SHIFT) & (PTRS_PER_PUD-1)) 1445 #define pmd_index(address) (((address) >> PMD_SHIFT) & (PTRS_PER_PMD-1)) 1446 1447 #define p4d_deref(pud) ((unsigned long)__va(p4d_val(pud) & _REGION_ENTRY_ORIGIN)) 1448 #define pgd_deref(pgd) ((unsigned long)__va(pgd_val(pgd) & _REGION_ENTRY_ORIGIN)) 1449 1450 static inline unsigned long pmd_deref(pmd_t pmd) 1451 { 1452 unsigned long origin_mask; 1453 1454 origin_mask = _SEGMENT_ENTRY_ORIGIN; 1455 if (pmd_leaf(pmd)) 1456 origin_mask = _SEGMENT_ENTRY_ORIGIN_LARGE; 1457 return (unsigned long)__va(pmd_val(pmd) & origin_mask); 1458 } 1459 1460 static inline unsigned long pmd_pfn(pmd_t pmd) 1461 { 1462 return __pa(pmd_deref(pmd)) >> PAGE_SHIFT; 1463 } 1464 1465 static inline unsigned long pud_deref(pud_t pud) 1466 { 1467 unsigned long origin_mask; 1468 1469 origin_mask = _REGION_ENTRY_ORIGIN; 1470 if (pud_leaf(pud)) 1471 origin_mask = _REGION3_ENTRY_ORIGIN_LARGE; 1472 return (unsigned long)__va(pud_val(pud) & origin_mask); 1473 } 1474 1475 #define pud_pfn pud_pfn 1476 static inline unsigned long pud_pfn(pud_t pud) 1477 { 1478 return __pa(pud_deref(pud)) >> PAGE_SHIFT; 1479 } 1480 1481 /* 1482 * The pgd_offset function *always* adds the index for the top-level 1483 * region/segment table. This is done to get a sequence like the 1484 * following to work: 1485 * pgdp = pgd_offset(current->mm, addr); 1486 * pgd = READ_ONCE(*pgdp); 1487 * p4dp = p4d_offset(&pgd, addr); 1488 * ... 1489 * The subsequent p4d_offset, pud_offset and pmd_offset functions 1490 * only add an index if they dereferenced the pointer. 1491 */ 1492 static inline pgd_t *pgd_offset_raw(pgd_t *pgd, unsigned long address) 1493 { 1494 unsigned long rste; 1495 unsigned int shift; 1496 1497 /* Get the first entry of the top level table */ 1498 rste = pgd_val(*pgd); 1499 /* Pick up the shift from the table type of the first entry */ 1500 shift = ((rste & _REGION_ENTRY_TYPE_MASK) >> 2) * 11 + 20; 1501 return pgd + ((address >> shift) & (PTRS_PER_PGD - 1)); 1502 } 1503 1504 #define pgd_offset(mm, address) pgd_offset_raw(READ_ONCE((mm)->pgd), address) 1505 1506 static inline p4d_t *p4d_offset_lockless(pgd_t *pgdp, pgd_t pgd, unsigned long address) 1507 { 1508 if ((pgd_val(pgd) & _REGION_ENTRY_TYPE_MASK) >= _REGION_ENTRY_TYPE_R1) 1509 return (p4d_t *) pgd_deref(pgd) + p4d_index(address); 1510 return (p4d_t *) pgdp; 1511 } 1512 #define p4d_offset_lockless p4d_offset_lockless 1513 1514 static inline p4d_t *p4d_offset(pgd_t *pgdp, unsigned long address) 1515 { 1516 return p4d_offset_lockless(pgdp, *pgdp, address); 1517 } 1518 1519 static inline pud_t *pud_offset_lockless(p4d_t *p4dp, p4d_t p4d, unsigned long address) 1520 { 1521 if ((p4d_val(p4d) & _REGION_ENTRY_TYPE_MASK) >= _REGION_ENTRY_TYPE_R2) 1522 return (pud_t *) p4d_deref(p4d) + pud_index(address); 1523 return (pud_t *) p4dp; 1524 } 1525 #define pud_offset_lockless pud_offset_lockless 1526 1527 static inline pud_t *pud_offset(p4d_t *p4dp, unsigned long address) 1528 { 1529 return pud_offset_lockless(p4dp, *p4dp, address); 1530 } 1531 #define pud_offset pud_offset 1532 1533 static inline pmd_t *pmd_offset_lockless(pud_t *pudp, pud_t pud, unsigned long address) 1534 { 1535 if ((pud_val(pud) & _REGION_ENTRY_TYPE_MASK) >= _REGION_ENTRY_TYPE_R3) 1536 return (pmd_t *) pud_deref(pud) + pmd_index(address); 1537 return (pmd_t *) pudp; 1538 } 1539 #define pmd_offset_lockless pmd_offset_lockless 1540 1541 static inline pmd_t *pmd_offset(pud_t *pudp, unsigned long address) 1542 { 1543 return pmd_offset_lockless(pudp, *pudp, address); 1544 } 1545 #define pmd_offset pmd_offset 1546 1547 static inline unsigned long pmd_page_vaddr(pmd_t pmd) 1548 { 1549 return (unsigned long) pmd_deref(pmd); 1550 } 1551 1552 static inline bool gup_fast_permitted(unsigned long start, unsigned long end) 1553 { 1554 return end <= current->mm->context.asce_limit; 1555 } 1556 #define gup_fast_permitted gup_fast_permitted 1557 1558 #define pfn_pte(pfn, pgprot) mk_pte_phys(((pfn) << PAGE_SHIFT), (pgprot)) 1559 #define pte_pfn(x) (pte_val(x) >> PAGE_SHIFT) 1560 #define pte_page(x) pfn_to_page(pte_pfn(x)) 1561 1562 #define pmd_page(pmd) pfn_to_page(pmd_pfn(pmd)) 1563 #define pud_page(pud) pfn_to_page(pud_pfn(pud)) 1564 #define p4d_page(p4d) pfn_to_page(p4d_pfn(p4d)) 1565 #define pgd_page(pgd) pfn_to_page(pgd_pfn(pgd)) 1566 1567 static inline pmd_t pmd_wrprotect(pmd_t pmd) 1568 { 1569 pmd = clear_pmd_bit(pmd, __pgprot(_SEGMENT_ENTRY_WRITE)); 1570 return set_pmd_bit(pmd, __pgprot(_SEGMENT_ENTRY_PROTECT)); 1571 } 1572 1573 static inline pmd_t pmd_mkwrite_novma(pmd_t pmd) 1574 { 1575 pmd = set_pmd_bit(pmd, __pgprot(_SEGMENT_ENTRY_WRITE)); 1576 if (pmd_val(pmd) & _SEGMENT_ENTRY_DIRTY) 1577 pmd = clear_pmd_bit(pmd, __pgprot(_SEGMENT_ENTRY_PROTECT)); 1578 return pmd; 1579 } 1580 1581 static inline pmd_t pmd_mkclean(pmd_t pmd) 1582 { 1583 pmd = clear_pmd_bit(pmd, __pgprot(_SEGMENT_ENTRY_DIRTY)); 1584 return set_pmd_bit(pmd, __pgprot(_SEGMENT_ENTRY_PROTECT)); 1585 } 1586 1587 static inline pmd_t pmd_mkdirty(pmd_t pmd) 1588 { 1589 pmd = set_pmd_bit(pmd, __pgprot(_SEGMENT_ENTRY_DIRTY | _SEGMENT_ENTRY_SOFT_DIRTY)); 1590 if (pmd_val(pmd) & _SEGMENT_ENTRY_WRITE) 1591 pmd = clear_pmd_bit(pmd, __pgprot(_SEGMENT_ENTRY_PROTECT)); 1592 return pmd; 1593 } 1594 1595 static inline pud_t pud_wrprotect(pud_t pud) 1596 { 1597 pud = clear_pud_bit(pud, __pgprot(_REGION3_ENTRY_WRITE)); 1598 return set_pud_bit(pud, __pgprot(_REGION_ENTRY_PROTECT)); 1599 } 1600 1601 static inline pud_t pud_mkwrite(pud_t pud) 1602 { 1603 pud = set_pud_bit(pud, __pgprot(_REGION3_ENTRY_WRITE)); 1604 if (pud_val(pud) & _REGION3_ENTRY_DIRTY) 1605 pud = clear_pud_bit(pud, __pgprot(_REGION_ENTRY_PROTECT)); 1606 return pud; 1607 } 1608 1609 static inline pud_t pud_mkclean(pud_t pud) 1610 { 1611 pud = clear_pud_bit(pud, __pgprot(_REGION3_ENTRY_DIRTY)); 1612 return set_pud_bit(pud, __pgprot(_REGION_ENTRY_PROTECT)); 1613 } 1614 1615 static inline pud_t pud_mkdirty(pud_t pud) 1616 { 1617 pud = set_pud_bit(pud, __pgprot(_REGION3_ENTRY_DIRTY | _REGION3_ENTRY_SOFT_DIRTY)); 1618 if (pud_val(pud) & _REGION3_ENTRY_WRITE) 1619 pud = clear_pud_bit(pud, __pgprot(_REGION_ENTRY_PROTECT)); 1620 return pud; 1621 } 1622 1623 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLB_PAGE) 1624 static inline unsigned long massage_pgprot_pmd(pgprot_t pgprot) 1625 { 1626 /* 1627 * pgprot is PAGE_NONE, PAGE_RO, PAGE_RX, PAGE_RW or PAGE_RWX 1628 * (see __Pxxx / __Sxxx). Convert to segment table entry format. 1629 */ 1630 if (pgprot_val(pgprot) == pgprot_val(PAGE_NONE)) 1631 return pgprot_val(SEGMENT_NONE); 1632 if (pgprot_val(pgprot) == pgprot_val(PAGE_RO)) 1633 return pgprot_val(SEGMENT_RO); 1634 if (pgprot_val(pgprot) == pgprot_val(PAGE_RX)) 1635 return pgprot_val(SEGMENT_RX); 1636 if (pgprot_val(pgprot) == pgprot_val(PAGE_RW)) 1637 return pgprot_val(SEGMENT_RW); 1638 return pgprot_val(SEGMENT_RWX); 1639 } 1640 1641 static inline pmd_t pmd_mkyoung(pmd_t pmd) 1642 { 1643 pmd = set_pmd_bit(pmd, __pgprot(_SEGMENT_ENTRY_YOUNG)); 1644 if (pmd_val(pmd) & _SEGMENT_ENTRY_READ) 1645 pmd = clear_pmd_bit(pmd, __pgprot(_SEGMENT_ENTRY_INVALID)); 1646 return pmd; 1647 } 1648 1649 static inline pmd_t pmd_mkold(pmd_t pmd) 1650 { 1651 pmd = clear_pmd_bit(pmd, __pgprot(_SEGMENT_ENTRY_YOUNG)); 1652 return set_pmd_bit(pmd, __pgprot(_SEGMENT_ENTRY_INVALID)); 1653 } 1654 1655 static inline pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot) 1656 { 1657 unsigned long mask; 1658 1659 mask = _SEGMENT_ENTRY_ORIGIN_LARGE; 1660 mask |= _SEGMENT_ENTRY_DIRTY; 1661 mask |= _SEGMENT_ENTRY_YOUNG; 1662 mask |= _SEGMENT_ENTRY_LARGE; 1663 mask |= _SEGMENT_ENTRY_SOFT_DIRTY; 1664 pmd = __pmd(pmd_val(pmd) & mask); 1665 pmd = set_pmd_bit(pmd, __pgprot(massage_pgprot_pmd(newprot))); 1666 if (!(pmd_val(pmd) & _SEGMENT_ENTRY_DIRTY)) 1667 pmd = set_pmd_bit(pmd, __pgprot(_SEGMENT_ENTRY_PROTECT)); 1668 if (!(pmd_val(pmd) & _SEGMENT_ENTRY_YOUNG)) 1669 pmd = set_pmd_bit(pmd, __pgprot(_SEGMENT_ENTRY_INVALID)); 1670 return pmd; 1671 } 1672 1673 static inline pmd_t mk_pmd_phys(unsigned long physpage, pgprot_t pgprot) 1674 { 1675 return __pmd(physpage + massage_pgprot_pmd(pgprot)); 1676 } 1677 1678 #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLB_PAGE */ 1679 1680 static inline void __pmdp_cspg(pmd_t *pmdp) 1681 { 1682 cspg((unsigned long *)pmdp, pmd_val(*pmdp), 1683 pmd_val(*pmdp) | _SEGMENT_ENTRY_INVALID); 1684 } 1685 1686 #define IDTE_GLOBAL 0 1687 #define IDTE_LOCAL 1 1688 1689 #define IDTE_PTOA 0x0800 1690 #define IDTE_NODAT 0x1000 1691 #define IDTE_GUEST_ASCE 0x2000 1692 1693 static __always_inline void __pmdp_idte(unsigned long addr, pmd_t *pmdp, 1694 unsigned long opt, unsigned long asce, 1695 int local) 1696 { 1697 unsigned long sto; 1698 1699 sto = __pa(pmdp) - pmd_index(addr) * sizeof(pmd_t); 1700 if (__builtin_constant_p(opt) && opt == 0) { 1701 /* flush without guest asce */ 1702 asm volatile( 1703 " idte %[r1],0,%[r2],%[m4]" 1704 : "+m" (*pmdp) 1705 : [r1] "a" (sto), [r2] "a" ((addr & HPAGE_MASK)), 1706 [m4] "i" (local) 1707 : "cc" ); 1708 } else { 1709 /* flush with guest asce */ 1710 asm volatile( 1711 " idte %[r1],%[r3],%[r2],%[m4]" 1712 : "+m" (*pmdp) 1713 : [r1] "a" (sto), [r2] "a" ((addr & HPAGE_MASK) | opt), 1714 [r3] "a" (asce), [m4] "i" (local) 1715 : "cc" ); 1716 } 1717 } 1718 1719 static __always_inline void __pudp_idte(unsigned long addr, pud_t *pudp, 1720 unsigned long opt, unsigned long asce, 1721 int local) 1722 { 1723 unsigned long r3o; 1724 1725 r3o = __pa(pudp) - pud_index(addr) * sizeof(pud_t); 1726 r3o |= _ASCE_TYPE_REGION3; 1727 if (__builtin_constant_p(opt) && opt == 0) { 1728 /* flush without guest asce */ 1729 asm volatile( 1730 " idte %[r1],0,%[r2],%[m4]" 1731 : "+m" (*pudp) 1732 : [r1] "a" (r3o), [r2] "a" ((addr & PUD_MASK)), 1733 [m4] "i" (local) 1734 : "cc"); 1735 } else { 1736 /* flush with guest asce */ 1737 asm volatile( 1738 " idte %[r1],%[r3],%[r2],%[m4]" 1739 : "+m" (*pudp) 1740 : [r1] "a" (r3o), [r2] "a" ((addr & PUD_MASK) | opt), 1741 [r3] "a" (asce), [m4] "i" (local) 1742 : "cc" ); 1743 } 1744 } 1745 1746 pmd_t pmdp_xchg_direct(struct mm_struct *, unsigned long, pmd_t *, pmd_t); 1747 pmd_t pmdp_xchg_lazy(struct mm_struct *, unsigned long, pmd_t *, pmd_t); 1748 pud_t pudp_xchg_direct(struct mm_struct *, unsigned long, pud_t *, pud_t); 1749 1750 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 1751 1752 #define __HAVE_ARCH_PGTABLE_DEPOSIT 1753 void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp, 1754 pgtable_t pgtable); 1755 1756 #define __HAVE_ARCH_PGTABLE_WITHDRAW 1757 pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp); 1758 1759 #define __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS 1760 static inline int pmdp_set_access_flags(struct vm_area_struct *vma, 1761 unsigned long addr, pmd_t *pmdp, 1762 pmd_t entry, int dirty) 1763 { 1764 VM_BUG_ON(addr & ~HPAGE_MASK); 1765 1766 entry = pmd_mkyoung(entry); 1767 if (dirty) 1768 entry = pmd_mkdirty(entry); 1769 if (pmd_val(*pmdp) == pmd_val(entry)) 1770 return 0; 1771 pmdp_xchg_direct(vma->vm_mm, addr, pmdp, entry); 1772 return 1; 1773 } 1774 1775 #define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG 1776 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma, 1777 unsigned long addr, pmd_t *pmdp) 1778 { 1779 pmd_t pmd = *pmdp; 1780 1781 pmd = pmdp_xchg_direct(vma->vm_mm, addr, pmdp, pmd_mkold(pmd)); 1782 return pmd_young(pmd); 1783 } 1784 1785 #define __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH 1786 static inline int pmdp_clear_flush_young(struct vm_area_struct *vma, 1787 unsigned long addr, pmd_t *pmdp) 1788 { 1789 VM_BUG_ON(addr & ~HPAGE_MASK); 1790 return pmdp_test_and_clear_young(vma, addr, pmdp); 1791 } 1792 1793 static inline void set_pmd_at(struct mm_struct *mm, unsigned long addr, 1794 pmd_t *pmdp, pmd_t entry) 1795 { 1796 set_pmd(pmdp, entry); 1797 } 1798 1799 static inline pmd_t pmd_mkhuge(pmd_t pmd) 1800 { 1801 pmd = set_pmd_bit(pmd, __pgprot(_SEGMENT_ENTRY_LARGE)); 1802 pmd = set_pmd_bit(pmd, __pgprot(_SEGMENT_ENTRY_YOUNG)); 1803 return set_pmd_bit(pmd, __pgprot(_SEGMENT_ENTRY_PROTECT)); 1804 } 1805 1806 #define __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR 1807 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm, 1808 unsigned long addr, pmd_t *pmdp) 1809 { 1810 return pmdp_xchg_direct(mm, addr, pmdp, __pmd(_SEGMENT_ENTRY_EMPTY)); 1811 } 1812 1813 #define __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR_FULL 1814 static inline pmd_t pmdp_huge_get_and_clear_full(struct vm_area_struct *vma, 1815 unsigned long addr, 1816 pmd_t *pmdp, int full) 1817 { 1818 if (full) { 1819 pmd_t pmd = *pmdp; 1820 set_pmd(pmdp, __pmd(_SEGMENT_ENTRY_EMPTY)); 1821 return pmd; 1822 } 1823 return pmdp_xchg_lazy(vma->vm_mm, addr, pmdp, __pmd(_SEGMENT_ENTRY_EMPTY)); 1824 } 1825 1826 #define __HAVE_ARCH_PMDP_HUGE_CLEAR_FLUSH 1827 static inline pmd_t pmdp_huge_clear_flush(struct vm_area_struct *vma, 1828 unsigned long addr, pmd_t *pmdp) 1829 { 1830 return pmdp_huge_get_and_clear(vma->vm_mm, addr, pmdp); 1831 } 1832 1833 #define __HAVE_ARCH_PMDP_INVALIDATE 1834 static inline pmd_t pmdp_invalidate(struct vm_area_struct *vma, 1835 unsigned long addr, pmd_t *pmdp) 1836 { 1837 pmd_t pmd; 1838 1839 VM_WARN_ON_ONCE(!pmd_present(*pmdp)); 1840 pmd = __pmd(pmd_val(*pmdp) | _SEGMENT_ENTRY_INVALID); 1841 return pmdp_xchg_direct(vma->vm_mm, addr, pmdp, pmd); 1842 } 1843 1844 #define __HAVE_ARCH_PMDP_SET_WRPROTECT 1845 static inline void pmdp_set_wrprotect(struct mm_struct *mm, 1846 unsigned long addr, pmd_t *pmdp) 1847 { 1848 pmd_t pmd = *pmdp; 1849 1850 if (pmd_write(pmd)) 1851 pmd = pmdp_xchg_lazy(mm, addr, pmdp, pmd_wrprotect(pmd)); 1852 } 1853 1854 static inline pmd_t pmdp_collapse_flush(struct vm_area_struct *vma, 1855 unsigned long address, 1856 pmd_t *pmdp) 1857 { 1858 return pmdp_huge_get_and_clear(vma->vm_mm, address, pmdp); 1859 } 1860 #define pmdp_collapse_flush pmdp_collapse_flush 1861 1862 #define pfn_pmd(pfn, pgprot) mk_pmd_phys(((pfn) << PAGE_SHIFT), (pgprot)) 1863 1864 static inline int pmd_trans_huge(pmd_t pmd) 1865 { 1866 return pmd_leaf(pmd); 1867 } 1868 1869 #define has_transparent_hugepage has_transparent_hugepage 1870 static inline int has_transparent_hugepage(void) 1871 { 1872 return cpu_has_edat1() ? 1 : 0; 1873 } 1874 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ 1875 1876 /* 1877 * 64 bit swap entry format: 1878 * A page-table entry has some bits we have to treat in a special way. 1879 * Bits 54 and 63 are used to indicate the page type. Bit 53 marks the pte 1880 * as invalid. 1881 * A swap pte is indicated by bit pattern (pte & 0x201) == 0x200 1882 * | offset |E11XX|type |S0| 1883 * |0000000000111111111122222222223333333333444444444455|55555|55566|66| 1884 * |0123456789012345678901234567890123456789012345678901|23456|78901|23| 1885 * 1886 * Bits 0-51 store the offset. 1887 * Bit 52 (E) is used to remember PG_anon_exclusive. 1888 * Bits 57-61 store the type. 1889 * Bit 62 (S) is used for softdirty tracking. 1890 * Bits 55 and 56 (X) are unused. 1891 */ 1892 1893 #define __SWP_OFFSET_MASK ((1UL << 52) - 1) 1894 #define __SWP_OFFSET_SHIFT 12 1895 #define __SWP_TYPE_MASK ((1UL << 5) - 1) 1896 #define __SWP_TYPE_SHIFT 2 1897 1898 static inline pte_t mk_swap_pte(unsigned long type, unsigned long offset) 1899 { 1900 unsigned long pteval; 1901 1902 pteval = _PAGE_INVALID | _PAGE_PROTECT; 1903 pteval |= (offset & __SWP_OFFSET_MASK) << __SWP_OFFSET_SHIFT; 1904 pteval |= (type & __SWP_TYPE_MASK) << __SWP_TYPE_SHIFT; 1905 return __pte(pteval); 1906 } 1907 1908 static inline unsigned long __swp_type(swp_entry_t entry) 1909 { 1910 return (entry.val >> __SWP_TYPE_SHIFT) & __SWP_TYPE_MASK; 1911 } 1912 1913 static inline unsigned long __swp_offset(swp_entry_t entry) 1914 { 1915 return (entry.val >> __SWP_OFFSET_SHIFT) & __SWP_OFFSET_MASK; 1916 } 1917 1918 static inline swp_entry_t __swp_entry(unsigned long type, unsigned long offset) 1919 { 1920 return (swp_entry_t) { pte_val(mk_swap_pte(type, offset)) }; 1921 } 1922 1923 #define __pte_to_swp_entry(pte) ((swp_entry_t) { pte_val(pte) }) 1924 #define __swp_entry_to_pte(x) ((pte_t) { (x).val }) 1925 1926 /* 1927 * 64 bit swap entry format for REGION3 and SEGMENT table entries (RSTE) 1928 * Bits 59 and 63 are used to indicate the swap entry. Bit 58 marks the rste 1929 * as invalid. 1930 * A swap entry is indicated by bit pattern (rste & 0x011) == 0x010 1931 * | offset |Xtype |11TT|S0| 1932 * |0000000000111111111122222222223333333333444444444455|555555|5566|66| 1933 * |0123456789012345678901234567890123456789012345678901|234567|8901|23| 1934 * 1935 * Bits 0-51 store the offset. 1936 * Bits 53-57 store the type. 1937 * Bit 62 (S) is used for softdirty tracking. 1938 * Bits 60-61 (TT) indicate the table type: 0x01 for REGION3 and 0x00 for SEGMENT. 1939 * Bit 52 (X) is unused. 1940 */ 1941 1942 #define __SWP_OFFSET_MASK_RSTE ((1UL << 52) - 1) 1943 #define __SWP_OFFSET_SHIFT_RSTE 12 1944 #define __SWP_TYPE_MASK_RSTE ((1UL << 5) - 1) 1945 #define __SWP_TYPE_SHIFT_RSTE 6 1946 1947 /* 1948 * TT bits set to 0x00 == SEGMENT. For REGION3 entries, caller must add R3 1949 * bits 0x01. See also __set_huge_pte_at(). 1950 */ 1951 static inline unsigned long mk_swap_rste(unsigned long type, unsigned long offset) 1952 { 1953 unsigned long rste; 1954 1955 rste = _RST_ENTRY_INVALID | _RST_ENTRY_COMM; 1956 rste |= (offset & __SWP_OFFSET_MASK_RSTE) << __SWP_OFFSET_SHIFT_RSTE; 1957 rste |= (type & __SWP_TYPE_MASK_RSTE) << __SWP_TYPE_SHIFT_RSTE; 1958 return rste; 1959 } 1960 1961 static inline unsigned long __swp_type_rste(swp_entry_t entry) 1962 { 1963 return (entry.val >> __SWP_TYPE_SHIFT_RSTE) & __SWP_TYPE_MASK_RSTE; 1964 } 1965 1966 static inline unsigned long __swp_offset_rste(swp_entry_t entry) 1967 { 1968 return (entry.val >> __SWP_OFFSET_SHIFT_RSTE) & __SWP_OFFSET_MASK_RSTE; 1969 } 1970 1971 #define __rste_to_swp_entry(rste) ((swp_entry_t) { rste }) 1972 1973 /* 1974 * s390 has different layout for PTE and region / segment table entries (RSTE). 1975 * This is also true for swap entries, and their swap type and offset encoding. 1976 * For hugetlbfs PTE_MARKER support, s390 has internal __swp_type_rste() and 1977 * __swp_offset_rste() helpers to correctly handle RSTE swap entries. 1978 * 1979 * But common swap code does not know about this difference, and only uses 1980 * __swp_type(), __swp_offset() and __swp_entry() helpers for conversion between 1981 * arch-dependent and arch-independent representation of swp_entry_t for all 1982 * pagetable levels. On s390, those helpers only work for PTE swap entries. 1983 * 1984 * Therefore, implement __pmd_to_swp_entry() to build a fake PTE swap entry 1985 * and return the arch-dependent representation of that. Correspondingly, 1986 * implement __swp_entry_to_pmd() to convert that into a proper PMD swap 1987 * entry again. With this, the arch-dependent swp_entry_t representation will 1988 * always look like a PTE swap entry in common code. 1989 * 1990 * This is somewhat similar to fake PTEs in hugetlbfs code for s390, but only 1991 * requires conversion of the swap type and offset, and not all the possible 1992 * PTE bits. 1993 */ 1994 static inline swp_entry_t __pmd_to_swp_entry(pmd_t pmd) 1995 { 1996 swp_entry_t arch_entry; 1997 pte_t pte; 1998 1999 arch_entry = __rste_to_swp_entry(pmd_val(pmd)); 2000 pte = mk_swap_pte(__swp_type_rste(arch_entry), __swp_offset_rste(arch_entry)); 2001 return __pte_to_swp_entry(pte); 2002 } 2003 2004 static inline pmd_t __swp_entry_to_pmd(swp_entry_t arch_entry) 2005 { 2006 pmd_t pmd; 2007 2008 pmd = __pmd(mk_swap_rste(__swp_type(arch_entry), __swp_offset(arch_entry))); 2009 return pmd; 2010 } 2011 2012 extern int vmem_add_mapping(unsigned long start, unsigned long size); 2013 extern void vmem_remove_mapping(unsigned long start, unsigned long size); 2014 extern int __vmem_map_4k_page(unsigned long addr, unsigned long phys, pgprot_t prot, bool alloc); 2015 extern int vmem_map_4k_page(unsigned long addr, unsigned long phys, pgprot_t prot); 2016 extern void vmem_unmap_4k_page(unsigned long addr); 2017 extern pte_t *vmem_get_alloc_pte(unsigned long addr, bool alloc); 2018 extern int s390_enable_sie(void); 2019 extern int s390_enable_skey(void); 2020 extern void s390_reset_cmma(struct mm_struct *mm); 2021 2022 /* s390 has a private copy of get unmapped area to deal with cache synonyms */ 2023 #define HAVE_ARCH_UNMAPPED_AREA 2024 #define HAVE_ARCH_UNMAPPED_AREA_TOPDOWN 2025 2026 #define pmd_pgtable(pmd) \ 2027 ((pgtable_t)__va(pmd_val(pmd) & -sizeof(pte_t)*PTRS_PER_PTE)) 2028 2029 static inline unsigned long gmap_pgste_get_pgt_addr(unsigned long *pgt) 2030 { 2031 unsigned long *pgstes, res; 2032 2033 pgstes = pgt + _PAGE_ENTRIES; 2034 2035 res = (pgstes[0] & PGSTE_ST2_MASK) << 16; 2036 res |= pgstes[1] & PGSTE_ST2_MASK; 2037 res |= (pgstes[2] & PGSTE_ST2_MASK) >> 16; 2038 res |= (pgstes[3] & PGSTE_ST2_MASK) >> 32; 2039 2040 return res; 2041 } 2042 2043 static inline pgste_t pgste_get_lock(pte_t *ptep) 2044 { 2045 unsigned long value = 0; 2046 #ifdef CONFIG_PGSTE 2047 unsigned long *ptr = (unsigned long *)(ptep + PTRS_PER_PTE); 2048 2049 do { 2050 value = __atomic64_or_barrier(PGSTE_PCL_BIT, ptr); 2051 } while (value & PGSTE_PCL_BIT); 2052 value |= PGSTE_PCL_BIT; 2053 #endif 2054 return __pgste(value); 2055 } 2056 2057 static inline void pgste_set_unlock(pte_t *ptep, pgste_t pgste) 2058 { 2059 #ifdef CONFIG_PGSTE 2060 barrier(); 2061 WRITE_ONCE(*(unsigned long *)(ptep + PTRS_PER_PTE), pgste_val(pgste) & ~PGSTE_PCL_BIT); 2062 #endif 2063 } 2064 2065 #endif /* _S390_PAGE_H */ 2066