1 /* 2 * linux/arch/arm/mm/fault.c 3 * 4 * Copyright (C) 1995 Linus Torvalds 5 * Modifications for ARM processor (c) 1995-2004 Russell King 6 * 7 * This program is free software; you can redistribute it and/or modify 8 * it under the terms of the GNU General Public License version 2 as 9 * published by the Free Software Foundation. 10 */ 11 #include <linux/module.h> 12 #include <linux/signal.h> 13 #include <linux/mm.h> 14 #include <linux/hardirq.h> 15 #include <linux/init.h> 16 #include <linux/kprobes.h> 17 #include <linux/uaccess.h> 18 #include <linux/page-flags.h> 19 #include <linux/sched.h> 20 #include <linux/highmem.h> 21 #include <linux/perf_event.h> 22 23 #include <asm/exception.h> 24 #include <asm/pgtable.h> 25 #include <asm/system_misc.h> 26 #include <asm/system_info.h> 27 #include <asm/tlbflush.h> 28 29 #include "fault.h" 30 31 #ifdef CONFIG_MMU 32 33 #ifdef CONFIG_KPROBES 34 static inline int notify_page_fault(struct pt_regs *regs, unsigned int fsr) 35 { 36 int ret = 0; 37 38 if (!user_mode(regs)) { 39 /* kprobe_running() needs smp_processor_id() */ 40 preempt_disable(); 41 if (kprobe_running() && kprobe_fault_handler(regs, fsr)) 42 ret = 1; 43 preempt_enable(); 44 } 45 46 return ret; 47 } 48 #else 49 static inline int notify_page_fault(struct pt_regs *regs, unsigned int fsr) 50 { 51 return 0; 52 } 53 #endif 54 55 /* 56 * This is useful to dump out the page tables associated with 57 * 'addr' in mm 'mm'. 58 */ 59 void show_pte(struct mm_struct *mm, unsigned long addr) 60 { 61 pgd_t *pgd; 62 63 if (!mm) 64 mm = &init_mm; 65 66 printk(KERN_ALERT "pgd = %p\n", mm->pgd); 67 pgd = pgd_offset(mm, addr); 68 printk(KERN_ALERT "[%08lx] *pgd=%08llx", 69 addr, (long long)pgd_val(*pgd)); 70 71 do { 72 pud_t *pud; 73 pmd_t *pmd; 74 pte_t *pte; 75 76 if (pgd_none(*pgd)) 77 break; 78 79 if (pgd_bad(*pgd)) { 80 printk("(bad)"); 81 break; 82 } 83 84 pud = pud_offset(pgd, addr); 85 if (PTRS_PER_PUD != 1) 86 printk(", *pud=%08llx", (long long)pud_val(*pud)); 87 88 if (pud_none(*pud)) 89 break; 90 91 if (pud_bad(*pud)) { 92 printk("(bad)"); 93 break; 94 } 95 96 pmd = pmd_offset(pud, addr); 97 if (PTRS_PER_PMD != 1) 98 printk(", *pmd=%08llx", (long long)pmd_val(*pmd)); 99 100 if (pmd_none(*pmd)) 101 break; 102 103 if (pmd_bad(*pmd)) { 104 printk("(bad)"); 105 break; 106 } 107 108 /* We must not map this if we have highmem enabled */ 109 if (PageHighMem(pfn_to_page(pmd_val(*pmd) >> PAGE_SHIFT))) 110 break; 111 112 pte = pte_offset_map(pmd, addr); 113 printk(", *pte=%08llx", (long long)pte_val(*pte)); 114 #ifndef CONFIG_ARM_LPAE 115 printk(", *ppte=%08llx", 116 (long long)pte_val(pte[PTE_HWTABLE_PTRS])); 117 #endif 118 pte_unmap(pte); 119 } while(0); 120 121 printk("\n"); 122 } 123 #else /* CONFIG_MMU */ 124 void show_pte(struct mm_struct *mm, unsigned long addr) 125 { } 126 #endif /* CONFIG_MMU */ 127 128 /* 129 * Oops. The kernel tried to access some page that wasn't present. 130 */ 131 static void 132 __do_kernel_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr, 133 struct pt_regs *regs) 134 { 135 /* 136 * Are we prepared to handle this kernel fault? 137 */ 138 if (fixup_exception(regs)) 139 return; 140 141 /* 142 * No handler, we'll have to terminate things with extreme prejudice. 143 */ 144 bust_spinlocks(1); 145 printk(KERN_ALERT 146 "Unable to handle kernel %s at virtual address %08lx\n", 147 (addr < PAGE_SIZE) ? "NULL pointer dereference" : 148 "paging request", addr); 149 150 show_pte(mm, addr); 151 die("Oops", regs, fsr); 152 bust_spinlocks(0); 153 do_exit(SIGKILL); 154 } 155 156 /* 157 * Something tried to access memory that isn't in our memory map.. 158 * User mode accesses just cause a SIGSEGV 159 */ 160 static void 161 __do_user_fault(struct task_struct *tsk, unsigned long addr, 162 unsigned int fsr, unsigned int sig, int code, 163 struct pt_regs *regs) 164 { 165 struct siginfo si; 166 167 #ifdef CONFIG_DEBUG_USER 168 if (((user_debug & UDBG_SEGV) && (sig == SIGSEGV)) || 169 ((user_debug & UDBG_BUS) && (sig == SIGBUS))) { 170 printk(KERN_DEBUG "%s: unhandled page fault (%d) at 0x%08lx, code 0x%03x\n", 171 tsk->comm, sig, addr, fsr); 172 show_pte(tsk->mm, addr); 173 show_regs(regs); 174 } 175 #endif 176 177 tsk->thread.address = addr; 178 tsk->thread.error_code = fsr; 179 tsk->thread.trap_no = 14; 180 si.si_signo = sig; 181 si.si_errno = 0; 182 si.si_code = code; 183 si.si_addr = (void __user *)addr; 184 force_sig_info(sig, &si, tsk); 185 } 186 187 void do_bad_area(unsigned long addr, unsigned int fsr, struct pt_regs *regs) 188 { 189 struct task_struct *tsk = current; 190 struct mm_struct *mm = tsk->active_mm; 191 192 /* 193 * If we are in kernel mode at this point, we 194 * have no context to handle this fault with. 195 */ 196 if (user_mode(regs)) 197 __do_user_fault(tsk, addr, fsr, SIGSEGV, SEGV_MAPERR, regs); 198 else 199 __do_kernel_fault(mm, addr, fsr, regs); 200 } 201 202 #ifdef CONFIG_MMU 203 #define VM_FAULT_BADMAP 0x010000 204 #define VM_FAULT_BADACCESS 0x020000 205 206 /* 207 * Check that the permissions on the VMA allow for the fault which occurred. 208 * If we encountered a write fault, we must have write permission, otherwise 209 * we allow any permission. 210 */ 211 static inline bool access_error(unsigned int fsr, struct vm_area_struct *vma) 212 { 213 unsigned int mask = VM_READ | VM_WRITE | VM_EXEC; 214 215 if (fsr & FSR_WRITE) 216 mask = VM_WRITE; 217 if (fsr & FSR_LNX_PF) 218 mask = VM_EXEC; 219 220 return vma->vm_flags & mask ? false : true; 221 } 222 223 static int __kprobes 224 __do_page_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr, 225 unsigned int flags, struct task_struct *tsk) 226 { 227 struct vm_area_struct *vma; 228 int fault; 229 230 vma = find_vma(mm, addr); 231 fault = VM_FAULT_BADMAP; 232 if (unlikely(!vma)) 233 goto out; 234 if (unlikely(vma->vm_start > addr)) 235 goto check_stack; 236 237 /* 238 * Ok, we have a good vm_area for this 239 * memory access, so we can handle it. 240 */ 241 good_area: 242 if (access_error(fsr, vma)) { 243 fault = VM_FAULT_BADACCESS; 244 goto out; 245 } 246 247 return handle_mm_fault(mm, vma, addr & PAGE_MASK, flags); 248 249 check_stack: 250 /* Don't allow expansion below FIRST_USER_ADDRESS */ 251 if (vma->vm_flags & VM_GROWSDOWN && 252 addr >= FIRST_USER_ADDRESS && !expand_stack(vma, addr)) 253 goto good_area; 254 out: 255 return fault; 256 } 257 258 static int __kprobes 259 do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs) 260 { 261 struct task_struct *tsk; 262 struct mm_struct *mm; 263 int fault, sig, code; 264 int write = fsr & FSR_WRITE; 265 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE | 266 (write ? FAULT_FLAG_WRITE : 0); 267 268 if (notify_page_fault(regs, fsr)) 269 return 0; 270 271 tsk = current; 272 mm = tsk->mm; 273 274 /* Enable interrupts if they were enabled in the parent context. */ 275 if (interrupts_enabled(regs)) 276 local_irq_enable(); 277 278 /* 279 * If we're in an interrupt or have no user 280 * context, we must not take the fault.. 281 */ 282 if (in_atomic() || !mm) 283 goto no_context; 284 285 /* 286 * As per x86, we may deadlock here. However, since the kernel only 287 * validly references user space from well defined areas of the code, 288 * we can bug out early if this is from code which shouldn't. 289 */ 290 if (!down_read_trylock(&mm->mmap_sem)) { 291 if (!user_mode(regs) && !search_exception_tables(regs->ARM_pc)) 292 goto no_context; 293 retry: 294 down_read(&mm->mmap_sem); 295 } else { 296 /* 297 * The above down_read_trylock() might have succeeded in 298 * which case, we'll have missed the might_sleep() from 299 * down_read() 300 */ 301 might_sleep(); 302 #ifdef CONFIG_DEBUG_VM 303 if (!user_mode(regs) && 304 !search_exception_tables(regs->ARM_pc)) 305 goto no_context; 306 #endif 307 } 308 309 fault = __do_page_fault(mm, addr, fsr, flags, tsk); 310 311 /* If we need to retry but a fatal signal is pending, handle the 312 * signal first. We do not need to release the mmap_sem because 313 * it would already be released in __lock_page_or_retry in 314 * mm/filemap.c. */ 315 if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current)) 316 return 0; 317 318 /* 319 * Major/minor page fault accounting is only done on the 320 * initial attempt. If we go through a retry, it is extremely 321 * likely that the page will be found in page cache at that point. 322 */ 323 324 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr); 325 if (!(fault & VM_FAULT_ERROR) && flags & FAULT_FLAG_ALLOW_RETRY) { 326 if (fault & VM_FAULT_MAJOR) { 327 tsk->maj_flt++; 328 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, 329 regs, addr); 330 } else { 331 tsk->min_flt++; 332 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, 333 regs, addr); 334 } 335 if (fault & VM_FAULT_RETRY) { 336 /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk 337 * of starvation. */ 338 flags &= ~FAULT_FLAG_ALLOW_RETRY; 339 flags |= FAULT_FLAG_TRIED; 340 goto retry; 341 } 342 } 343 344 up_read(&mm->mmap_sem); 345 346 /* 347 * Handle the "normal" case first - VM_FAULT_MAJOR / VM_FAULT_MINOR 348 */ 349 if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP | VM_FAULT_BADACCESS)))) 350 return 0; 351 352 if (fault & VM_FAULT_OOM) { 353 /* 354 * We ran out of memory, call the OOM killer, and return to 355 * userspace (which will retry the fault, or kill us if we 356 * got oom-killed) 357 */ 358 pagefault_out_of_memory(); 359 return 0; 360 } 361 362 /* 363 * If we are in kernel mode at this point, we 364 * have no context to handle this fault with. 365 */ 366 if (!user_mode(regs)) 367 goto no_context; 368 369 if (fault & VM_FAULT_SIGBUS) { 370 /* 371 * We had some memory, but were unable to 372 * successfully fix up this page fault. 373 */ 374 sig = SIGBUS; 375 code = BUS_ADRERR; 376 } else { 377 /* 378 * Something tried to access memory that 379 * isn't in our memory map.. 380 */ 381 sig = SIGSEGV; 382 code = fault == VM_FAULT_BADACCESS ? 383 SEGV_ACCERR : SEGV_MAPERR; 384 } 385 386 __do_user_fault(tsk, addr, fsr, sig, code, regs); 387 return 0; 388 389 no_context: 390 __do_kernel_fault(mm, addr, fsr, regs); 391 return 0; 392 } 393 #else /* CONFIG_MMU */ 394 static int 395 do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs) 396 { 397 return 0; 398 } 399 #endif /* CONFIG_MMU */ 400 401 /* 402 * First Level Translation Fault Handler 403 * 404 * We enter here because the first level page table doesn't contain 405 * a valid entry for the address. 406 * 407 * If the address is in kernel space (>= TASK_SIZE), then we are 408 * probably faulting in the vmalloc() area. 409 * 410 * If the init_task's first level page tables contains the relevant 411 * entry, we copy the it to this task. If not, we send the process 412 * a signal, fixup the exception, or oops the kernel. 413 * 414 * NOTE! We MUST NOT take any locks for this case. We may be in an 415 * interrupt or a critical region, and should only copy the information 416 * from the master page table, nothing more. 417 */ 418 #ifdef CONFIG_MMU 419 static int __kprobes 420 do_translation_fault(unsigned long addr, unsigned int fsr, 421 struct pt_regs *regs) 422 { 423 unsigned int index; 424 pgd_t *pgd, *pgd_k; 425 pud_t *pud, *pud_k; 426 pmd_t *pmd, *pmd_k; 427 428 if (addr < TASK_SIZE) 429 return do_page_fault(addr, fsr, regs); 430 431 if (user_mode(regs)) 432 goto bad_area; 433 434 index = pgd_index(addr); 435 436 pgd = cpu_get_pgd() + index; 437 pgd_k = init_mm.pgd + index; 438 439 if (pgd_none(*pgd_k)) 440 goto bad_area; 441 if (!pgd_present(*pgd)) 442 set_pgd(pgd, *pgd_k); 443 444 pud = pud_offset(pgd, addr); 445 pud_k = pud_offset(pgd_k, addr); 446 447 if (pud_none(*pud_k)) 448 goto bad_area; 449 if (!pud_present(*pud)) 450 set_pud(pud, *pud_k); 451 452 pmd = pmd_offset(pud, addr); 453 pmd_k = pmd_offset(pud_k, addr); 454 455 #ifdef CONFIG_ARM_LPAE 456 /* 457 * Only one hardware entry per PMD with LPAE. 458 */ 459 index = 0; 460 #else 461 /* 462 * On ARM one Linux PGD entry contains two hardware entries (see page 463 * tables layout in pgtable.h). We normally guarantee that we always 464 * fill both L1 entries. But create_mapping() doesn't follow the rule. 465 * It can create inidividual L1 entries, so here we have to call 466 * pmd_none() check for the entry really corresponded to address, not 467 * for the first of pair. 468 */ 469 index = (addr >> SECTION_SHIFT) & 1; 470 #endif 471 if (pmd_none(pmd_k[index])) 472 goto bad_area; 473 474 copy_pmd(pmd, pmd_k); 475 return 0; 476 477 bad_area: 478 do_bad_area(addr, fsr, regs); 479 return 0; 480 } 481 #else /* CONFIG_MMU */ 482 static int 483 do_translation_fault(unsigned long addr, unsigned int fsr, 484 struct pt_regs *regs) 485 { 486 return 0; 487 } 488 #endif /* CONFIG_MMU */ 489 490 /* 491 * Some section permission faults need to be handled gracefully. 492 * They can happen due to a __{get,put}_user during an oops. 493 */ 494 static int 495 do_sect_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs) 496 { 497 do_bad_area(addr, fsr, regs); 498 return 0; 499 } 500 501 /* 502 * This abort handler always returns "fault". 503 */ 504 static int 505 do_bad(unsigned long addr, unsigned int fsr, struct pt_regs *regs) 506 { 507 return 1; 508 } 509 510 struct fsr_info { 511 int (*fn)(unsigned long addr, unsigned int fsr, struct pt_regs *regs); 512 int sig; 513 int code; 514 const char *name; 515 }; 516 517 /* FSR definition */ 518 #ifdef CONFIG_ARM_LPAE 519 #include "fsr-3level.c" 520 #else 521 #include "fsr-2level.c" 522 #endif 523 524 void __init 525 hook_fault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *), 526 int sig, int code, const char *name) 527 { 528 if (nr < 0 || nr >= ARRAY_SIZE(fsr_info)) 529 BUG(); 530 531 fsr_info[nr].fn = fn; 532 fsr_info[nr].sig = sig; 533 fsr_info[nr].code = code; 534 fsr_info[nr].name = name; 535 } 536 537 /* 538 * Dispatch a data abort to the relevant handler. 539 */ 540 asmlinkage void __exception 541 do_DataAbort(unsigned long addr, unsigned int fsr, struct pt_regs *regs) 542 { 543 const struct fsr_info *inf = fsr_info + fsr_fs(fsr); 544 struct siginfo info; 545 546 if (!inf->fn(addr, fsr & ~FSR_LNX_PF, regs)) 547 return; 548 549 printk(KERN_ALERT "Unhandled fault: %s (0x%03x) at 0x%08lx\n", 550 inf->name, fsr, addr); 551 552 info.si_signo = inf->sig; 553 info.si_errno = 0; 554 info.si_code = inf->code; 555 info.si_addr = (void __user *)addr; 556 arm_notify_die("", regs, &info, fsr, 0); 557 } 558 559 void __init 560 hook_ifault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *), 561 int sig, int code, const char *name) 562 { 563 if (nr < 0 || nr >= ARRAY_SIZE(ifsr_info)) 564 BUG(); 565 566 ifsr_info[nr].fn = fn; 567 ifsr_info[nr].sig = sig; 568 ifsr_info[nr].code = code; 569 ifsr_info[nr].name = name; 570 } 571 572 asmlinkage void __exception 573 do_PrefetchAbort(unsigned long addr, unsigned int ifsr, struct pt_regs *regs) 574 { 575 const struct fsr_info *inf = ifsr_info + fsr_fs(ifsr); 576 struct siginfo info; 577 578 if (!inf->fn(addr, ifsr | FSR_LNX_PF, regs)) 579 return; 580 581 printk(KERN_ALERT "Unhandled prefetch abort: %s (0x%03x) at 0x%08lx\n", 582 inf->name, ifsr, addr); 583 584 info.si_signo = inf->sig; 585 info.si_errno = 0; 586 info.si_code = inf->code; 587 info.si_addr = (void __user *)addr; 588 arm_notify_die("", regs, &info, ifsr, 0); 589 } 590 591 #ifndef CONFIG_ARM_LPAE 592 static int __init exceptions_init(void) 593 { 594 if (cpu_architecture() >= CPU_ARCH_ARMv6) { 595 hook_fault_code(4, do_translation_fault, SIGSEGV, SEGV_MAPERR, 596 "I-cache maintenance fault"); 597 } 598 599 if (cpu_architecture() >= CPU_ARCH_ARMv7) { 600 /* 601 * TODO: Access flag faults introduced in ARMv6K. 602 * Runtime check for 'K' extension is needed 603 */ 604 hook_fault_code(3, do_bad, SIGSEGV, SEGV_MAPERR, 605 "section access flag fault"); 606 hook_fault_code(6, do_bad, SIGSEGV, SEGV_MAPERR, 607 "section access flag fault"); 608 } 609 610 return 0; 611 } 612 613 arch_initcall(exceptions_init); 614 #endif 615