1 #ifndef _ASM_POWERPC_BOOK3S_32_PGTABLE_H 2 #define _ASM_POWERPC_BOOK3S_32_PGTABLE_H 3 4 #include <asm-generic/pgtable-nopmd.h> 5 6 #include <asm/book3s/32/hash.h> 7 8 /* And here we include common definitions */ 9 #include <asm/pte-common.h> 10 11 /* 12 * The normal case is that PTEs are 32-bits and we have a 1-page 13 * 1024-entry pgdir pointing to 1-page 1024-entry PTE pages. -- paulus 14 * 15 * For any >32-bit physical address platform, we can use the following 16 * two level page table layout where the pgdir is 8KB and the MS 13 bits 17 * are an index to the second level table. The combined pgdir/pmd first 18 * level has 2048 entries and the second level has 512 64-bit PTE entries. 19 * -Matt 20 */ 21 /* PGDIR_SHIFT determines what a top-level page table entry can map */ 22 #define PGDIR_SHIFT (PAGE_SHIFT + PTE_SHIFT) 23 #define PGDIR_SIZE (1UL << PGDIR_SHIFT) 24 #define PGDIR_MASK (~(PGDIR_SIZE-1)) 25 26 #define PTRS_PER_PTE (1 << PTE_SHIFT) 27 #define PTRS_PER_PMD 1 28 #define PTRS_PER_PGD (1 << (32 - PGDIR_SHIFT)) 29 30 #define USER_PTRS_PER_PGD (TASK_SIZE / PGDIR_SIZE) 31 /* 32 * This is the bottom of the PKMAP area with HIGHMEM or an arbitrary 33 * value (for now) on others, from where we can start layout kernel 34 * virtual space that goes below PKMAP and FIXMAP 35 */ 36 #ifdef CONFIG_HIGHMEM 37 #define KVIRT_TOP PKMAP_BASE 38 #else 39 #define KVIRT_TOP (0xfe000000UL) /* for now, could be FIXMAP_BASE ? */ 40 #endif 41 42 /* 43 * ioremap_bot starts at that address. Early ioremaps move down from there, 44 * until mem_init() at which point this becomes the top of the vmalloc 45 * and ioremap space 46 */ 47 #ifdef CONFIG_NOT_COHERENT_CACHE 48 #define IOREMAP_TOP ((KVIRT_TOP - CONFIG_CONSISTENT_SIZE) & PAGE_MASK) 49 #else 50 #define IOREMAP_TOP KVIRT_TOP 51 #endif 52 53 /* 54 * Just any arbitrary offset to the start of the vmalloc VM area: the 55 * current 16MB value just means that there will be a 64MB "hole" after the 56 * physical memory until the kernel virtual memory starts. That means that 57 * any out-of-bounds memory accesses will hopefully be caught. 58 * The vmalloc() routines leaves a hole of 4kB between each vmalloced 59 * area for the same reason. ;) 60 * 61 * We no longer map larger than phys RAM with the BATs so we don't have 62 * to worry about the VMALLOC_OFFSET causing problems. We do have to worry 63 * about clashes between our early calls to ioremap() that start growing down 64 * from ioremap_base being run into the VM area allocations (growing upwards 65 * from VMALLOC_START). For this reason we have ioremap_bot to check when 66 * we actually run into our mappings setup in the early boot with the VM 67 * system. This really does become a problem for machines with good amounts 68 * of RAM. -- Cort 69 */ 70 #define VMALLOC_OFFSET (0x1000000) /* 16M */ 71 #ifdef PPC_PIN_SIZE 72 #define VMALLOC_START (((_ALIGN((long)high_memory, PPC_PIN_SIZE) + VMALLOC_OFFSET) & ~(VMALLOC_OFFSET-1))) 73 #else 74 #define VMALLOC_START ((((long)high_memory + VMALLOC_OFFSET) & ~(VMALLOC_OFFSET-1))) 75 #endif 76 #define VMALLOC_END ioremap_bot 77 78 #ifndef __ASSEMBLY__ 79 #include <linux/sched.h> 80 #include <linux/threads.h> 81 #include <asm/io.h> /* For sub-arch specific PPC_PIN_SIZE */ 82 83 extern unsigned long ioremap_bot; 84 85 /* 86 * entries per page directory level: our page-table tree is two-level, so 87 * we don't really have any PMD directory. 88 */ 89 #define PTE_TABLE_SIZE (sizeof(pte_t) << PTE_SHIFT) 90 #define PGD_TABLE_SIZE (sizeof(pgd_t) << (32 - PGDIR_SHIFT)) 91 92 #define pte_ERROR(e) \ 93 pr_err("%s:%d: bad pte %llx.\n", __FILE__, __LINE__, \ 94 (unsigned long long)pte_val(e)) 95 #define pgd_ERROR(e) \ 96 pr_err("%s:%d: bad pgd %08lx.\n", __FILE__, __LINE__, pgd_val(e)) 97 /* 98 * Bits in a linux-style PTE. These match the bits in the 99 * (hardware-defined) PowerPC PTE as closely as possible. 100 */ 101 102 #define pte_clear(mm, addr, ptep) \ 103 do { pte_update(ptep, ~_PAGE_HASHPTE, 0); } while (0) 104 105 #define pmd_none(pmd) (!pmd_val(pmd)) 106 #define pmd_bad(pmd) (pmd_val(pmd) & _PMD_BAD) 107 #define pmd_present(pmd) (pmd_val(pmd) & _PMD_PRESENT_MASK) 108 static inline void pmd_clear(pmd_t *pmdp) 109 { 110 *pmdp = __pmd(0); 111 } 112 113 114 /* 115 * When flushing the tlb entry for a page, we also need to flush the hash 116 * table entry. flush_hash_pages is assembler (for speed) in hashtable.S. 117 */ 118 extern int flush_hash_pages(unsigned context, unsigned long va, 119 unsigned long pmdval, int count); 120 121 /* Add an HPTE to the hash table */ 122 extern void add_hash_page(unsigned context, unsigned long va, 123 unsigned long pmdval); 124 125 /* Flush an entry from the TLB/hash table */ 126 extern void flush_hash_entry(struct mm_struct *mm, pte_t *ptep, 127 unsigned long address); 128 129 /* 130 * PTE updates. This function is called whenever an existing 131 * valid PTE is updated. This does -not- include set_pte_at() 132 * which nowadays only sets a new PTE. 133 * 134 * Depending on the type of MMU, we may need to use atomic updates 135 * and the PTE may be either 32 or 64 bit wide. In the later case, 136 * when using atomic updates, only the low part of the PTE is 137 * accessed atomically. 138 * 139 * In addition, on 44x, we also maintain a global flag indicating 140 * that an executable user mapping was modified, which is needed 141 * to properly flush the virtually tagged instruction cache of 142 * those implementations. 143 */ 144 #ifndef CONFIG_PTE_64BIT 145 static inline unsigned long pte_update(pte_t *p, 146 unsigned long clr, 147 unsigned long set) 148 { 149 unsigned long old, tmp; 150 151 __asm__ __volatile__("\ 152 1: lwarx %0,0,%3\n\ 153 andc %1,%0,%4\n\ 154 or %1,%1,%5\n" 155 PPC405_ERR77(0,%3) 156 " stwcx. %1,0,%3\n\ 157 bne- 1b" 158 : "=&r" (old), "=&r" (tmp), "=m" (*p) 159 : "r" (p), "r" (clr), "r" (set), "m" (*p) 160 : "cc" ); 161 162 return old; 163 } 164 #else /* CONFIG_PTE_64BIT */ 165 static inline unsigned long long pte_update(pte_t *p, 166 unsigned long clr, 167 unsigned long set) 168 { 169 unsigned long long old; 170 unsigned long tmp; 171 172 __asm__ __volatile__("\ 173 1: lwarx %L0,0,%4\n\ 174 lwzx %0,0,%3\n\ 175 andc %1,%L0,%5\n\ 176 or %1,%1,%6\n" 177 PPC405_ERR77(0,%3) 178 " stwcx. %1,0,%4\n\ 179 bne- 1b" 180 : "=&r" (old), "=&r" (tmp), "=m" (*p) 181 : "r" (p), "r" ((unsigned long)(p) + 4), "r" (clr), "r" (set), "m" (*p) 182 : "cc" ); 183 184 return old; 185 } 186 #endif /* CONFIG_PTE_64BIT */ 187 188 /* 189 * 2.6 calls this without flushing the TLB entry; this is wrong 190 * for our hash-based implementation, we fix that up here. 191 */ 192 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG 193 static inline int __ptep_test_and_clear_young(unsigned int context, unsigned long addr, pte_t *ptep) 194 { 195 unsigned long old; 196 old = pte_update(ptep, _PAGE_ACCESSED, 0); 197 if (old & _PAGE_HASHPTE) { 198 unsigned long ptephys = __pa(ptep) & PAGE_MASK; 199 flush_hash_pages(context, addr, ptephys, 1); 200 } 201 return (old & _PAGE_ACCESSED) != 0; 202 } 203 #define ptep_test_and_clear_young(__vma, __addr, __ptep) \ 204 __ptep_test_and_clear_young((__vma)->vm_mm->context.id, __addr, __ptep) 205 206 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR 207 static inline pte_t ptep_get_and_clear(struct mm_struct *mm, unsigned long addr, 208 pte_t *ptep) 209 { 210 return __pte(pte_update(ptep, ~_PAGE_HASHPTE, 0)); 211 } 212 213 #define __HAVE_ARCH_PTEP_SET_WRPROTECT 214 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long addr, 215 pte_t *ptep) 216 { 217 pte_update(ptep, (_PAGE_RW | _PAGE_HWWRITE), _PAGE_RO); 218 } 219 static inline void huge_ptep_set_wrprotect(struct mm_struct *mm, 220 unsigned long addr, pte_t *ptep) 221 { 222 ptep_set_wrprotect(mm, addr, ptep); 223 } 224 225 226 static inline void __ptep_set_access_flags(struct mm_struct *mm, 227 pte_t *ptep, pte_t entry) 228 { 229 unsigned long set = pte_val(entry) & 230 (_PAGE_DIRTY | _PAGE_ACCESSED | _PAGE_RW | _PAGE_EXEC); 231 unsigned long clr = ~pte_val(entry) & _PAGE_RO; 232 233 pte_update(ptep, clr, set); 234 } 235 236 #define __HAVE_ARCH_PTE_SAME 237 #define pte_same(A,B) (((pte_val(A) ^ pte_val(B)) & ~_PAGE_HASHPTE) == 0) 238 239 /* 240 * Note that on Book E processors, the pmd contains the kernel virtual 241 * (lowmem) address of the pte page. The physical address is less useful 242 * because everything runs with translation enabled (even the TLB miss 243 * handler). On everything else the pmd contains the physical address 244 * of the pte page. -- paulus 245 */ 246 #ifndef CONFIG_BOOKE 247 #define pmd_page_vaddr(pmd) \ 248 ((unsigned long) __va(pmd_val(pmd) & PAGE_MASK)) 249 #define pmd_page(pmd) \ 250 pfn_to_page(pmd_val(pmd) >> PAGE_SHIFT) 251 #else 252 #define pmd_page_vaddr(pmd) \ 253 ((unsigned long) (pmd_val(pmd) & PAGE_MASK)) 254 #define pmd_page(pmd) \ 255 pfn_to_page((__pa(pmd_val(pmd)) >> PAGE_SHIFT)) 256 #endif 257 258 /* to find an entry in a kernel page-table-directory */ 259 #define pgd_offset_k(address) pgd_offset(&init_mm, address) 260 261 /* to find an entry in a page-table-directory */ 262 #define pgd_index(address) ((address) >> PGDIR_SHIFT) 263 #define pgd_offset(mm, address) ((mm)->pgd + pgd_index(address)) 264 265 /* Find an entry in the third-level page table.. */ 266 #define pte_index(address) \ 267 (((address) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1)) 268 #define pte_offset_kernel(dir, addr) \ 269 ((pte_t *) pmd_page_vaddr(*(dir)) + pte_index(addr)) 270 #define pte_offset_map(dir, addr) \ 271 ((pte_t *) kmap_atomic(pmd_page(*(dir))) + pte_index(addr)) 272 #define pte_unmap(pte) kunmap_atomic(pte) 273 274 /* 275 * Encode and decode a swap entry. 276 * Note that the bits we use in a PTE for representing a swap entry 277 * must not include the _PAGE_PRESENT bit or the _PAGE_HASHPTE bit (if used). 278 * -- paulus 279 */ 280 #define __swp_type(entry) ((entry).val & 0x1f) 281 #define __swp_offset(entry) ((entry).val >> 5) 282 #define __swp_entry(type, offset) ((swp_entry_t) { (type) | ((offset) << 5) }) 283 #define __pte_to_swp_entry(pte) ((swp_entry_t) { pte_val(pte) >> 3 }) 284 #define __swp_entry_to_pte(x) ((pte_t) { (x).val << 3 }) 285 286 #ifndef CONFIG_PPC_4K_PAGES 287 void pgtable_cache_init(void); 288 #else 289 /* 290 * No page table caches to initialise 291 */ 292 #define pgtable_cache_init() do { } while (0) 293 #endif 294 295 extern int get_pteptr(struct mm_struct *mm, unsigned long addr, pte_t **ptep, 296 pmd_t **pmdp); 297 298 /* Generic accessors to PTE bits */ 299 static inline int pte_write(pte_t pte) { return !!(pte_val(pte) & _PAGE_RW);} 300 static inline int pte_dirty(pte_t pte) { return !!(pte_val(pte) & _PAGE_DIRTY); } 301 static inline int pte_young(pte_t pte) { return !!(pte_val(pte) & _PAGE_ACCESSED); } 302 static inline int pte_special(pte_t pte) { return !!(pte_val(pte) & _PAGE_SPECIAL); } 303 static inline int pte_none(pte_t pte) { return (pte_val(pte) & ~_PTE_NONE_MASK) == 0; } 304 static inline pgprot_t pte_pgprot(pte_t pte) { return __pgprot(pte_val(pte) & PAGE_PROT_BITS); } 305 306 static inline int pte_present(pte_t pte) 307 { 308 return pte_val(pte) & _PAGE_PRESENT; 309 } 310 311 /* Conversion functions: convert a page and protection to a page entry, 312 * and a page entry and page directory to the page they refer to. 313 * 314 * Even if PTEs can be unsigned long long, a PFN is always an unsigned 315 * long for now. 316 */ 317 static inline pte_t pfn_pte(unsigned long pfn, pgprot_t pgprot) 318 { 319 return __pte(((pte_basic_t)(pfn) << PTE_RPN_SHIFT) | 320 pgprot_val(pgprot)); 321 } 322 323 static inline unsigned long pte_pfn(pte_t pte) 324 { 325 return pte_val(pte) >> PTE_RPN_SHIFT; 326 } 327 328 /* Generic modifiers for PTE bits */ 329 static inline pte_t pte_wrprotect(pte_t pte) 330 { 331 return __pte(pte_val(pte) & ~_PAGE_RW); 332 } 333 334 static inline pte_t pte_mkclean(pte_t pte) 335 { 336 return __pte(pte_val(pte) & ~_PAGE_DIRTY); 337 } 338 339 static inline pte_t pte_mkold(pte_t pte) 340 { 341 return __pte(pte_val(pte) & ~_PAGE_ACCESSED); 342 } 343 344 static inline pte_t pte_mkwrite(pte_t pte) 345 { 346 return __pte(pte_val(pte) | _PAGE_RW); 347 } 348 349 static inline pte_t pte_mkdirty(pte_t pte) 350 { 351 return __pte(pte_val(pte) | _PAGE_DIRTY); 352 } 353 354 static inline pte_t pte_mkyoung(pte_t pte) 355 { 356 return __pte(pte_val(pte) | _PAGE_ACCESSED); 357 } 358 359 static inline pte_t pte_mkspecial(pte_t pte) 360 { 361 return __pte(pte_val(pte) | _PAGE_SPECIAL); 362 } 363 364 static inline pte_t pte_mkhuge(pte_t pte) 365 { 366 return pte; 367 } 368 369 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot) 370 { 371 return __pte((pte_val(pte) & _PAGE_CHG_MASK) | pgprot_val(newprot)); 372 } 373 374 375 376 /* This low level function performs the actual PTE insertion 377 * Setting the PTE depends on the MMU type and other factors. It's 378 * an horrible mess that I'm not going to try to clean up now but 379 * I'm keeping it in one place rather than spread around 380 */ 381 static inline void __set_pte_at(struct mm_struct *mm, unsigned long addr, 382 pte_t *ptep, pte_t pte, int percpu) 383 { 384 #if defined(CONFIG_PPC_STD_MMU_32) && defined(CONFIG_SMP) && !defined(CONFIG_PTE_64BIT) 385 /* First case is 32-bit Hash MMU in SMP mode with 32-bit PTEs. We use the 386 * helper pte_update() which does an atomic update. We need to do that 387 * because a concurrent invalidation can clear _PAGE_HASHPTE. If it's a 388 * per-CPU PTE such as a kmap_atomic, we do a simple update preserving 389 * the hash bits instead (ie, same as the non-SMP case) 390 */ 391 if (percpu) 392 *ptep = __pte((pte_val(*ptep) & _PAGE_HASHPTE) 393 | (pte_val(pte) & ~_PAGE_HASHPTE)); 394 else 395 pte_update(ptep, ~_PAGE_HASHPTE, pte_val(pte)); 396 397 #elif defined(CONFIG_PPC32) && defined(CONFIG_PTE_64BIT) 398 /* Second case is 32-bit with 64-bit PTE. In this case, we 399 * can just store as long as we do the two halves in the right order 400 * with a barrier in between. This is possible because we take care, 401 * in the hash code, to pre-invalidate if the PTE was already hashed, 402 * which synchronizes us with any concurrent invalidation. 403 * In the percpu case, we also fallback to the simple update preserving 404 * the hash bits 405 */ 406 if (percpu) { 407 *ptep = __pte((pte_val(*ptep) & _PAGE_HASHPTE) 408 | (pte_val(pte) & ~_PAGE_HASHPTE)); 409 return; 410 } 411 if (pte_val(*ptep) & _PAGE_HASHPTE) 412 flush_hash_entry(mm, ptep, addr); 413 __asm__ __volatile__("\ 414 stw%U0%X0 %2,%0\n\ 415 eieio\n\ 416 stw%U0%X0 %L2,%1" 417 : "=m" (*ptep), "=m" (*((unsigned char *)ptep+4)) 418 : "r" (pte) : "memory"); 419 420 #elif defined(CONFIG_PPC_STD_MMU_32) 421 /* Third case is 32-bit hash table in UP mode, we need to preserve 422 * the _PAGE_HASHPTE bit since we may not have invalidated the previous 423 * translation in the hash yet (done in a subsequent flush_tlb_xxx()) 424 * and see we need to keep track that this PTE needs invalidating 425 */ 426 *ptep = __pte((pte_val(*ptep) & _PAGE_HASHPTE) 427 | (pte_val(pte) & ~_PAGE_HASHPTE)); 428 429 #else 430 #error "Not supported " 431 #endif 432 } 433 434 /* 435 * Macro to mark a page protection value as "uncacheable". 436 */ 437 438 #define _PAGE_CACHE_CTL (_PAGE_COHERENT | _PAGE_GUARDED | _PAGE_NO_CACHE | \ 439 _PAGE_WRITETHRU) 440 441 #define pgprot_noncached pgprot_noncached 442 static inline pgprot_t pgprot_noncached(pgprot_t prot) 443 { 444 return __pgprot((pgprot_val(prot) & ~_PAGE_CACHE_CTL) | 445 _PAGE_NO_CACHE | _PAGE_GUARDED); 446 } 447 448 #define pgprot_noncached_wc pgprot_noncached_wc 449 static inline pgprot_t pgprot_noncached_wc(pgprot_t prot) 450 { 451 return __pgprot((pgprot_val(prot) & ~_PAGE_CACHE_CTL) | 452 _PAGE_NO_CACHE); 453 } 454 455 #define pgprot_cached pgprot_cached 456 static inline pgprot_t pgprot_cached(pgprot_t prot) 457 { 458 return __pgprot((pgprot_val(prot) & ~_PAGE_CACHE_CTL) | 459 _PAGE_COHERENT); 460 } 461 462 #define pgprot_cached_wthru pgprot_cached_wthru 463 static inline pgprot_t pgprot_cached_wthru(pgprot_t prot) 464 { 465 return __pgprot((pgprot_val(prot) & ~_PAGE_CACHE_CTL) | 466 _PAGE_COHERENT | _PAGE_WRITETHRU); 467 } 468 469 #define pgprot_cached_noncoherent pgprot_cached_noncoherent 470 static inline pgprot_t pgprot_cached_noncoherent(pgprot_t prot) 471 { 472 return __pgprot(pgprot_val(prot) & ~_PAGE_CACHE_CTL); 473 } 474 475 #define pgprot_writecombine pgprot_writecombine 476 static inline pgprot_t pgprot_writecombine(pgprot_t prot) 477 { 478 return pgprot_noncached_wc(prot); 479 } 480 481 #endif /* !__ASSEMBLY__ */ 482 483 #endif /* _ASM_POWERPC_BOOK3S_32_PGTABLE_H */ 484