| 1 | /* |
| 2 | * This file contains ioremap and related functions for 64-bit machines. |
| 3 | * |
| 4 | * Derived from arch/ppc64/mm/init.c |
| 5 | * Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org) |
| 6 | * |
| 7 | * Modifications by Paul Mackerras (PowerMac) (paulus@samba.org) |
| 8 | * and Cort Dougan (PReP) (cort@cs.nmt.edu) |
| 9 | * Copyright (C) 1996 Paul Mackerras |
| 10 | * |
| 11 | * Derived from "arch/i386/mm/init.c" |
| 12 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds |
| 13 | * |
| 14 | * Dave Engebretsen <engebret@us.ibm.com> |
| 15 | * Rework for PPC64 port. |
| 16 | * |
| 17 | * This program is free software; you can redistribute it and/or |
| 18 | * modify it under the terms of the GNU General Public License |
| 19 | * as published by the Free Software Foundation; either version |
| 20 | * 2 of the License, or (at your option) any later version. |
| 21 | * |
| 22 | */ |
| 23 | |
| 24 | #include <linux/signal.h> |
| 25 | #include <linux/sched.h> |
| 26 | #include <linux/kernel.h> |
| 27 | #include <linux/errno.h> |
| 28 | #include <linux/string.h> |
| 29 | #include <linux/export.h> |
| 30 | #include <linux/types.h> |
| 31 | #include <linux/mman.h> |
| 32 | #include <linux/mm.h> |
| 33 | #include <linux/swap.h> |
| 34 | #include <linux/stddef.h> |
| 35 | #include <linux/vmalloc.h> |
| 36 | #include <linux/init.h> |
| 37 | #include <linux/bootmem.h> |
| 38 | #include <linux/memblock.h> |
| 39 | #include <linux/slab.h> |
| 40 | |
| 41 | #include <asm/pgalloc.h> |
| 42 | #include <asm/page.h> |
| 43 | #include <asm/prom.h> |
| 44 | #include <asm/io.h> |
| 45 | #include <asm/mmu_context.h> |
| 46 | #include <asm/pgtable.h> |
| 47 | #include <asm/mmu.h> |
| 48 | #include <asm/smp.h> |
| 49 | #include <asm/machdep.h> |
| 50 | #include <asm/tlb.h> |
| 51 | #include <asm/processor.h> |
| 52 | #include <asm/cputable.h> |
| 53 | #include <asm/sections.h> |
| 54 | #include <asm/firmware.h> |
| 55 | |
| 56 | #include "mmu_decl.h" |
| 57 | |
| 58 | /* Some sanity checking */ |
| 59 | #if TASK_SIZE_USER64 > PGTABLE_RANGE |
| 60 | #error TASK_SIZE_USER64 exceeds pagetable range |
| 61 | #endif |
| 62 | |
| 63 | #ifdef CONFIG_PPC_STD_MMU_64 |
| 64 | #if TASK_SIZE_USER64 > (1UL << (ESID_BITS + SID_SHIFT)) |
| 65 | #error TASK_SIZE_USER64 exceeds user VSID range |
| 66 | #endif |
| 67 | #endif |
| 68 | |
| 69 | unsigned long ioremap_bot = IOREMAP_BASE; |
| 70 | |
| 71 | #ifdef CONFIG_PPC_MMU_NOHASH |
| 72 | static void *early_alloc_pgtable(unsigned long size) |
| 73 | { |
| 74 | void *pt; |
| 75 | |
| 76 | if (init_bootmem_done) |
| 77 | pt = __alloc_bootmem(size, size, __pa(MAX_DMA_ADDRESS)); |
| 78 | else |
| 79 | pt = __va(memblock_alloc_base(size, size, |
| 80 | __pa(MAX_DMA_ADDRESS))); |
| 81 | memset(pt, 0, size); |
| 82 | |
| 83 | return pt; |
| 84 | } |
| 85 | #endif /* CONFIG_PPC_MMU_NOHASH */ |
| 86 | |
| 87 | /* |
| 88 | * map_kernel_page currently only called by __ioremap |
| 89 | * map_kernel_page adds an entry to the ioremap page table |
| 90 | * and adds an entry to the HPT, possibly bolting it |
| 91 | */ |
| 92 | int map_kernel_page(unsigned long ea, unsigned long pa, int flags) |
| 93 | { |
| 94 | pgd_t *pgdp; |
| 95 | pud_t *pudp; |
| 96 | pmd_t *pmdp; |
| 97 | pte_t *ptep; |
| 98 | |
| 99 | if (slab_is_available()) { |
| 100 | pgdp = pgd_offset_k(ea); |
| 101 | pudp = pud_alloc(&init_mm, pgdp, ea); |
| 102 | if (!pudp) |
| 103 | return -ENOMEM; |
| 104 | pmdp = pmd_alloc(&init_mm, pudp, ea); |
| 105 | if (!pmdp) |
| 106 | return -ENOMEM; |
| 107 | ptep = pte_alloc_kernel(pmdp, ea); |
| 108 | if (!ptep) |
| 109 | return -ENOMEM; |
| 110 | set_pte_at(&init_mm, ea, ptep, pfn_pte(pa >> PAGE_SHIFT, |
| 111 | __pgprot(flags))); |
| 112 | } else { |
| 113 | #ifdef CONFIG_PPC_MMU_NOHASH |
| 114 | /* Warning ! This will blow up if bootmem is not initialized |
| 115 | * which our ppc64 code is keen to do that, we'll need to |
| 116 | * fix it and/or be more careful |
| 117 | */ |
| 118 | pgdp = pgd_offset_k(ea); |
| 119 | #ifdef PUD_TABLE_SIZE |
| 120 | if (pgd_none(*pgdp)) { |
| 121 | pudp = early_alloc_pgtable(PUD_TABLE_SIZE); |
| 122 | BUG_ON(pudp == NULL); |
| 123 | pgd_populate(&init_mm, pgdp, pudp); |
| 124 | } |
| 125 | #endif /* PUD_TABLE_SIZE */ |
| 126 | pudp = pud_offset(pgdp, ea); |
| 127 | if (pud_none(*pudp)) { |
| 128 | pmdp = early_alloc_pgtable(PMD_TABLE_SIZE); |
| 129 | BUG_ON(pmdp == NULL); |
| 130 | pud_populate(&init_mm, pudp, pmdp); |
| 131 | } |
| 132 | pmdp = pmd_offset(pudp, ea); |
| 133 | if (!pmd_present(*pmdp)) { |
| 134 | ptep = early_alloc_pgtable(PAGE_SIZE); |
| 135 | BUG_ON(ptep == NULL); |
| 136 | pmd_populate_kernel(&init_mm, pmdp, ptep); |
| 137 | } |
| 138 | ptep = pte_offset_kernel(pmdp, ea); |
| 139 | set_pte_at(&init_mm, ea, ptep, pfn_pte(pa >> PAGE_SHIFT, |
| 140 | __pgprot(flags))); |
| 141 | #else /* CONFIG_PPC_MMU_NOHASH */ |
| 142 | /* |
| 143 | * If the mm subsystem is not fully up, we cannot create a |
| 144 | * linux page table entry for this mapping. Simply bolt an |
| 145 | * entry in the hardware page table. |
| 146 | * |
| 147 | */ |
| 148 | if (htab_bolt_mapping(ea, ea + PAGE_SIZE, pa, flags, |
| 149 | mmu_io_psize, mmu_kernel_ssize)) { |
| 150 | printk(KERN_ERR "Failed to do bolted mapping IO " |
| 151 | "memory at %016lx !\n", pa); |
| 152 | return -ENOMEM; |
| 153 | } |
| 154 | #endif /* !CONFIG_PPC_MMU_NOHASH */ |
| 155 | } |
| 156 | return 0; |
| 157 | } |
| 158 | |
| 159 | |
| 160 | /** |
| 161 | * __ioremap_at - Low level function to establish the page tables |
| 162 | * for an IO mapping |
| 163 | */ |
| 164 | void __iomem * __ioremap_at(phys_addr_t pa, void *ea, unsigned long size, |
| 165 | unsigned long flags) |
| 166 | { |
| 167 | unsigned long i; |
| 168 | |
| 169 | /* Make sure we have the base flags */ |
| 170 | if ((flags & _PAGE_PRESENT) == 0) |
| 171 | flags |= pgprot_val(PAGE_KERNEL); |
| 172 | |
| 173 | /* Non-cacheable page cannot be coherent */ |
| 174 | if (flags & _PAGE_NO_CACHE) |
| 175 | flags &= ~_PAGE_COHERENT; |
| 176 | |
| 177 | /* We don't support the 4K PFN hack with ioremap */ |
| 178 | if (flags & _PAGE_4K_PFN) |
| 179 | return NULL; |
| 180 | |
| 181 | WARN_ON(pa & ~PAGE_MASK); |
| 182 | WARN_ON(((unsigned long)ea) & ~PAGE_MASK); |
| 183 | WARN_ON(size & ~PAGE_MASK); |
| 184 | |
| 185 | for (i = 0; i < size; i += PAGE_SIZE) |
| 186 | if (map_kernel_page((unsigned long)ea+i, pa+i, flags)) |
| 187 | return NULL; |
| 188 | |
| 189 | return (void __iomem *)ea; |
| 190 | } |
| 191 | |
| 192 | /** |
| 193 | * __iounmap_from - Low level function to tear down the page tables |
| 194 | * for an IO mapping. This is used for mappings that |
| 195 | * are manipulated manually, like partial unmapping of |
| 196 | * PCI IOs or ISA space. |
| 197 | */ |
| 198 | void __iounmap_at(void *ea, unsigned long size) |
| 199 | { |
| 200 | WARN_ON(((unsigned long)ea) & ~PAGE_MASK); |
| 201 | WARN_ON(size & ~PAGE_MASK); |
| 202 | |
| 203 | unmap_kernel_range((unsigned long)ea, size); |
| 204 | } |
| 205 | |
| 206 | void __iomem * __ioremap_caller(phys_addr_t addr, unsigned long size, |
| 207 | unsigned long flags, void *caller) |
| 208 | { |
| 209 | phys_addr_t paligned; |
| 210 | void __iomem *ret; |
| 211 | |
| 212 | /* |
| 213 | * Choose an address to map it to. |
| 214 | * Once the imalloc system is running, we use it. |
| 215 | * Before that, we map using addresses going |
| 216 | * up from ioremap_bot. imalloc will use |
| 217 | * the addresses from ioremap_bot through |
| 218 | * IMALLOC_END |
| 219 | * |
| 220 | */ |
| 221 | paligned = addr & PAGE_MASK; |
| 222 | size = PAGE_ALIGN(addr + size) - paligned; |
| 223 | |
| 224 | if ((size == 0) || (paligned == 0)) |
| 225 | return NULL; |
| 226 | |
| 227 | if (mem_init_done) { |
| 228 | struct vm_struct *area; |
| 229 | |
| 230 | area = __get_vm_area_caller(size, VM_IOREMAP, |
| 231 | ioremap_bot, IOREMAP_END, |
| 232 | caller); |
| 233 | if (area == NULL) |
| 234 | return NULL; |
| 235 | |
| 236 | area->phys_addr = paligned; |
| 237 | ret = __ioremap_at(paligned, area->addr, size, flags); |
| 238 | if (!ret) |
| 239 | vunmap(area->addr); |
| 240 | } else { |
| 241 | ret = __ioremap_at(paligned, (void *)ioremap_bot, size, flags); |
| 242 | if (ret) |
| 243 | ioremap_bot += size; |
| 244 | } |
| 245 | |
| 246 | if (ret) |
| 247 | ret += addr & ~PAGE_MASK; |
| 248 | return ret; |
| 249 | } |
| 250 | |
| 251 | void __iomem * __ioremap(phys_addr_t addr, unsigned long size, |
| 252 | unsigned long flags) |
| 253 | { |
| 254 | return __ioremap_caller(addr, size, flags, __builtin_return_address(0)); |
| 255 | } |
| 256 | |
| 257 | void __iomem * ioremap(phys_addr_t addr, unsigned long size) |
| 258 | { |
| 259 | unsigned long flags = _PAGE_NO_CACHE | _PAGE_GUARDED; |
| 260 | void *caller = __builtin_return_address(0); |
| 261 | |
| 262 | if (ppc_md.ioremap) |
| 263 | return ppc_md.ioremap(addr, size, flags, caller); |
| 264 | return __ioremap_caller(addr, size, flags, caller); |
| 265 | } |
| 266 | |
| 267 | void __iomem * ioremap_wc(phys_addr_t addr, unsigned long size) |
| 268 | { |
| 269 | unsigned long flags = _PAGE_NO_CACHE; |
| 270 | void *caller = __builtin_return_address(0); |
| 271 | |
| 272 | if (ppc_md.ioremap) |
| 273 | return ppc_md.ioremap(addr, size, flags, caller); |
| 274 | return __ioremap_caller(addr, size, flags, caller); |
| 275 | } |
| 276 | |
| 277 | void __iomem * ioremap_prot(phys_addr_t addr, unsigned long size, |
| 278 | unsigned long flags) |
| 279 | { |
| 280 | void *caller = __builtin_return_address(0); |
| 281 | |
| 282 | /* writeable implies dirty for kernel addresses */ |
| 283 | if (flags & _PAGE_RW) |
| 284 | flags |= _PAGE_DIRTY; |
| 285 | |
| 286 | /* we don't want to let _PAGE_USER and _PAGE_EXEC leak out */ |
| 287 | flags &= ~(_PAGE_USER | _PAGE_EXEC); |
| 288 | |
| 289 | #ifdef _PAGE_BAP_SR |
| 290 | /* _PAGE_USER contains _PAGE_BAP_SR on BookE using the new PTE format |
| 291 | * which means that we just cleared supervisor access... oops ;-) This |
| 292 | * restores it |
| 293 | */ |
| 294 | flags |= _PAGE_BAP_SR; |
| 295 | #endif |
| 296 | |
| 297 | if (ppc_md.ioremap) |
| 298 | return ppc_md.ioremap(addr, size, flags, caller); |
| 299 | return __ioremap_caller(addr, size, flags, caller); |
| 300 | } |
| 301 | |
| 302 | |
| 303 | /* |
| 304 | * Unmap an IO region and remove it from imalloc'd list. |
| 305 | * Access to IO memory should be serialized by driver. |
| 306 | */ |
| 307 | void __iounmap(volatile void __iomem *token) |
| 308 | { |
| 309 | void *addr; |
| 310 | |
| 311 | if (!mem_init_done) |
| 312 | return; |
| 313 | |
| 314 | addr = (void *) ((unsigned long __force) |
| 315 | PCI_FIX_ADDR(token) & PAGE_MASK); |
| 316 | if ((unsigned long)addr < ioremap_bot) { |
| 317 | printk(KERN_WARNING "Attempt to iounmap early bolted mapping" |
| 318 | " at 0x%p\n", addr); |
| 319 | return; |
| 320 | } |
| 321 | vunmap(addr); |
| 322 | } |
| 323 | |
| 324 | void iounmap(volatile void __iomem *token) |
| 325 | { |
| 326 | if (ppc_md.iounmap) |
| 327 | ppc_md.iounmap(token); |
| 328 | else |
| 329 | __iounmap(token); |
| 330 | } |
| 331 | |
| 332 | EXPORT_SYMBOL(ioremap); |
| 333 | EXPORT_SYMBOL(ioremap_wc); |
| 334 | EXPORT_SYMBOL(ioremap_prot); |
| 335 | EXPORT_SYMBOL(__ioremap); |
| 336 | EXPORT_SYMBOL(__ioremap_at); |
| 337 | EXPORT_SYMBOL(iounmap); |
| 338 | EXPORT_SYMBOL(__iounmap); |
| 339 | EXPORT_SYMBOL(__iounmap_at); |
| 340 | |
| 341 | #ifdef CONFIG_PPC_64K_PAGES |
| 342 | static pte_t *get_from_cache(struct mm_struct *mm) |
| 343 | { |
| 344 | void *pte_frag, *ret; |
| 345 | |
| 346 | spin_lock(&mm->page_table_lock); |
| 347 | ret = mm->context.pte_frag; |
| 348 | if (ret) { |
| 349 | pte_frag = ret + PTE_FRAG_SIZE; |
| 350 | /* |
| 351 | * If we have taken up all the fragments mark PTE page NULL |
| 352 | */ |
| 353 | if (((unsigned long)pte_frag & ~PAGE_MASK) == 0) |
| 354 | pte_frag = NULL; |
| 355 | mm->context.pte_frag = pte_frag; |
| 356 | } |
| 357 | spin_unlock(&mm->page_table_lock); |
| 358 | return (pte_t *)ret; |
| 359 | } |
| 360 | |
| 361 | static pte_t *__alloc_for_cache(struct mm_struct *mm, int kernel) |
| 362 | { |
| 363 | void *ret = NULL; |
| 364 | struct page *page = alloc_page(GFP_KERNEL | __GFP_NOTRACK | |
| 365 | __GFP_REPEAT | __GFP_ZERO); |
| 366 | if (!page) |
| 367 | return NULL; |
| 368 | |
| 369 | ret = page_address(page); |
| 370 | spin_lock(&mm->page_table_lock); |
| 371 | /* |
| 372 | * If we find pgtable_page set, we return |
| 373 | * the allocated page with single fragement |
| 374 | * count. |
| 375 | */ |
| 376 | if (likely(!mm->context.pte_frag)) { |
| 377 | atomic_set(&page->_count, PTE_FRAG_NR); |
| 378 | mm->context.pte_frag = ret + PTE_FRAG_SIZE; |
| 379 | } |
| 380 | spin_unlock(&mm->page_table_lock); |
| 381 | |
| 382 | if (!kernel) |
| 383 | pgtable_page_ctor(page); |
| 384 | |
| 385 | return (pte_t *)ret; |
| 386 | } |
| 387 | |
| 388 | pte_t *page_table_alloc(struct mm_struct *mm, unsigned long vmaddr, int kernel) |
| 389 | { |
| 390 | pte_t *pte; |
| 391 | |
| 392 | pte = get_from_cache(mm); |
| 393 | if (pte) |
| 394 | return pte; |
| 395 | |
| 396 | return __alloc_for_cache(mm, kernel); |
| 397 | } |
| 398 | |
| 399 | void page_table_free(struct mm_struct *mm, unsigned long *table, int kernel) |
| 400 | { |
| 401 | struct page *page = virt_to_page(table); |
| 402 | if (put_page_testzero(page)) { |
| 403 | if (!kernel) |
| 404 | pgtable_page_dtor(page); |
| 405 | free_hot_cold_page(page, 0); |
| 406 | } |
| 407 | } |
| 408 | |
| 409 | #ifdef CONFIG_SMP |
| 410 | static void page_table_free_rcu(void *table) |
| 411 | { |
| 412 | struct page *page = virt_to_page(table); |
| 413 | if (put_page_testzero(page)) { |
| 414 | pgtable_page_dtor(page); |
| 415 | free_hot_cold_page(page, 0); |
| 416 | } |
| 417 | } |
| 418 | |
| 419 | void pgtable_free_tlb(struct mmu_gather *tlb, void *table, int shift) |
| 420 | { |
| 421 | unsigned long pgf = (unsigned long)table; |
| 422 | |
| 423 | BUG_ON(shift > MAX_PGTABLE_INDEX_SIZE); |
| 424 | pgf |= shift; |
| 425 | tlb_remove_table(tlb, (void *)pgf); |
| 426 | } |
| 427 | |
| 428 | void __tlb_remove_table(void *_table) |
| 429 | { |
| 430 | void *table = (void *)((unsigned long)_table & ~MAX_PGTABLE_INDEX_SIZE); |
| 431 | unsigned shift = (unsigned long)_table & MAX_PGTABLE_INDEX_SIZE; |
| 432 | |
| 433 | if (!shift) |
| 434 | /* PTE page needs special handling */ |
| 435 | page_table_free_rcu(table); |
| 436 | else { |
| 437 | BUG_ON(shift > MAX_PGTABLE_INDEX_SIZE); |
| 438 | kmem_cache_free(PGT_CACHE(shift), table); |
| 439 | } |
| 440 | } |
| 441 | #else |
| 442 | void pgtable_free_tlb(struct mmu_gather *tlb, void *table, int shift) |
| 443 | { |
| 444 | if (!shift) { |
| 445 | /* PTE page needs special handling */ |
| 446 | struct page *page = virt_to_page(table); |
| 447 | if (put_page_testzero(page)) { |
| 448 | pgtable_page_dtor(page); |
| 449 | free_hot_cold_page(page, 0); |
| 450 | } |
| 451 | } else { |
| 452 | BUG_ON(shift > MAX_PGTABLE_INDEX_SIZE); |
| 453 | kmem_cache_free(PGT_CACHE(shift), table); |
| 454 | } |
| 455 | } |
| 456 | #endif |
| 457 | #endif /* CONFIG_PPC_64K_PAGES */ |