2 * linux/arch/arm/mm/fault.c
4 * Copyright (C) 1995 Linus Torvalds
5 * Modifications for ARM processor (c) 1995-2004 Russell King
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.
11 #include <linux/module.h>
12 #include <linux/signal.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>
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>
34 static inline int notify_page_fault(struct pt_regs *regs, unsigned int fsr)
38 if (!user_mode(regs)) {
39 /* kprobe_running() needs smp_processor_id() */
41 if (kprobe_running() && kprobe_fault_handler(regs, fsr))
49 static inline int notify_page_fault(struct pt_regs *regs, unsigned int fsr)
56 * This is useful to dump out the page tables associated with
59 void show_pte(struct mm_struct *mm, unsigned long addr)
66 pr_alert("pgd = %p\n", mm->pgd);
67 pgd = pgd_offset(mm, addr);
68 pr_alert("[%08lx] *pgd=%08llx",
69 addr, (long long)pgd_val(*pgd));
84 pud = pud_offset(pgd, addr);
85 if (PTRS_PER_PUD != 1)
86 pr_cont(", *pud=%08llx", (long long)pud_val(*pud));
96 pmd = pmd_offset(pud, addr);
97 if (PTRS_PER_PMD != 1)
98 pr_cont(", *pmd=%08llx", (long long)pmd_val(*pmd));
108 /* We must not map this if we have highmem enabled */
109 if (PageHighMem(pfn_to_page(pmd_val(*pmd) >> PAGE_SHIFT)))
112 pte = pte_offset_map(pmd, addr);
113 pr_cont(", *pte=%08llx", (long long)pte_val(*pte));
114 #ifndef CONFIG_ARM_LPAE
115 pr_cont(", *ppte=%08llx",
116 (long long)pte_val(pte[PTE_HWTABLE_PTRS]));
123 #else /* CONFIG_MMU */
124 void show_pte(struct mm_struct *mm, unsigned long addr)
126 #endif /* CONFIG_MMU */
129 * Oops. The kernel tried to access some page that wasn't present.
132 __do_kernel_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
133 struct pt_regs *regs)
136 * Are we prepared to handle this kernel fault?
138 if (fixup_exception(regs))
142 * No handler, we'll have to terminate things with extreme prejudice.
145 pr_alert("Unable to handle kernel %s at virtual address %08lx\n",
146 (addr < PAGE_SIZE) ? "NULL pointer dereference" :
147 "paging request", addr);
150 die("Oops", regs, fsr);
156 * Something tried to access memory that isn't in our memory map..
157 * User mode accesses just cause a SIGSEGV
160 __do_user_fault(struct task_struct *tsk, unsigned long addr,
161 unsigned int fsr, unsigned int sig, int code,
162 struct pt_regs *regs)
166 if (addr > TASK_SIZE)
167 harden_branch_predictor();
169 #ifdef CONFIG_DEBUG_USER
170 if (((user_debug & UDBG_SEGV) && (sig == SIGSEGV)) ||
171 ((user_debug & UDBG_BUS) && (sig == SIGBUS))) {
172 printk(KERN_DEBUG "%s: unhandled page fault (%d) at 0x%08lx, code 0x%03x\n",
173 tsk->comm, sig, addr, fsr);
174 show_pte(tsk->mm, addr);
179 tsk->thread.address = addr;
180 tsk->thread.error_code = fsr;
181 tsk->thread.trap_no = 14;
185 si.si_addr = (void __user *)addr;
186 force_sig_info(sig, &si, tsk);
189 void do_bad_area(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
191 struct task_struct *tsk = current;
192 struct mm_struct *mm = tsk->active_mm;
195 * If we are in kernel mode at this point, we
196 * have no context to handle this fault with.
199 __do_user_fault(tsk, addr, fsr, SIGSEGV, SEGV_MAPERR, regs);
201 __do_kernel_fault(mm, addr, fsr, regs);
205 #define VM_FAULT_BADMAP 0x010000
206 #define VM_FAULT_BADACCESS 0x020000
209 * Check that the permissions on the VMA allow for the fault which occurred.
210 * If we encountered a write fault, we must have write permission, otherwise
211 * we allow any permission.
213 static inline bool access_error(unsigned int fsr, struct vm_area_struct *vma)
215 unsigned int mask = VM_READ | VM_WRITE | VM_EXEC;
217 if ((fsr & FSR_WRITE) && !(fsr & FSR_CM))
219 if (fsr & FSR_LNX_PF)
222 return vma->vm_flags & mask ? false : true;
226 __do_page_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
227 unsigned int flags, struct task_struct *tsk)
229 struct vm_area_struct *vma;
232 vma = find_vma(mm, addr);
233 fault = VM_FAULT_BADMAP;
236 if (unlikely(vma->vm_start > addr))
240 * Ok, we have a good vm_area for this
241 * memory access, so we can handle it.
244 if (access_error(fsr, vma)) {
245 fault = VM_FAULT_BADACCESS;
249 return handle_mm_fault(vma, addr & PAGE_MASK, flags);
252 /* Don't allow expansion below FIRST_USER_ADDRESS */
253 if (vma->vm_flags & VM_GROWSDOWN &&
254 addr >= FIRST_USER_ADDRESS && !expand_stack(vma, addr))
261 do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
263 struct task_struct *tsk;
264 struct mm_struct *mm;
265 int fault, sig, code;
266 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
268 if (notify_page_fault(regs, fsr))
274 /* Enable interrupts if they were enabled in the parent context. */
275 if (interrupts_enabled(regs))
279 * If we're in an interrupt or have no user
280 * context, we must not take the fault..
282 if (faulthandler_disabled() || !mm)
286 flags |= FAULT_FLAG_USER;
287 if ((fsr & FSR_WRITE) && !(fsr & FSR_CM))
288 flags |= FAULT_FLAG_WRITE;
291 * As per x86, we may deadlock here. However, since the kernel only
292 * validly references user space from well defined areas of the code,
293 * we can bug out early if this is from code which shouldn't.
295 if (!down_read_trylock(&mm->mmap_sem)) {
296 if (!user_mode(regs) && !search_exception_tables(regs->ARM_pc))
299 down_read(&mm->mmap_sem);
302 * The above down_read_trylock() might have succeeded in
303 * which case, we'll have missed the might_sleep() from
307 #ifdef CONFIG_DEBUG_VM
308 if (!user_mode(regs) &&
309 !search_exception_tables(regs->ARM_pc))
314 fault = __do_page_fault(mm, addr, fsr, flags, tsk);
316 /* If we need to retry but a fatal signal is pending, handle the
317 * signal first. We do not need to release the mmap_sem because
318 * it would already be released in __lock_page_or_retry in
320 if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current)) {
321 if (!user_mode(regs))
327 * Major/minor page fault accounting is only done on the
328 * initial attempt. If we go through a retry, it is extremely
329 * likely that the page will be found in page cache at that point.
332 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
333 if (!(fault & VM_FAULT_ERROR) && flags & FAULT_FLAG_ALLOW_RETRY) {
334 if (fault & VM_FAULT_MAJOR) {
336 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
340 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
343 if (fault & VM_FAULT_RETRY) {
344 /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
346 flags &= ~FAULT_FLAG_ALLOW_RETRY;
347 flags |= FAULT_FLAG_TRIED;
352 up_read(&mm->mmap_sem);
355 * Handle the "normal" case first - VM_FAULT_MAJOR
357 if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP | VM_FAULT_BADACCESS))))
361 * If we are in kernel mode at this point, we
362 * have no context to handle this fault with.
364 if (!user_mode(regs))
367 if (fault & VM_FAULT_OOM) {
369 * We ran out of memory, call the OOM killer, and return to
370 * userspace (which will retry the fault, or kill us if we
373 pagefault_out_of_memory();
377 if (fault & VM_FAULT_SIGBUS) {
379 * We had some memory, but were unable to
380 * successfully fix up this page fault.
386 * Something tried to access memory that
387 * isn't in our memory map..
390 code = fault == VM_FAULT_BADACCESS ?
391 SEGV_ACCERR : SEGV_MAPERR;
394 __do_user_fault(tsk, addr, fsr, sig, code, regs);
398 __do_kernel_fault(mm, addr, fsr, regs);
401 #else /* CONFIG_MMU */
403 do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
407 #endif /* CONFIG_MMU */
410 * First Level Translation Fault Handler
412 * We enter here because the first level page table doesn't contain
413 * a valid entry for the address.
415 * If the address is in kernel space (>= TASK_SIZE), then we are
416 * probably faulting in the vmalloc() area.
418 * If the init_task's first level page tables contains the relevant
419 * entry, we copy the it to this task. If not, we send the process
420 * a signal, fixup the exception, or oops the kernel.
422 * NOTE! We MUST NOT take any locks for this case. We may be in an
423 * interrupt or a critical region, and should only copy the information
424 * from the master page table, nothing more.
428 do_translation_fault(unsigned long addr, unsigned int fsr,
429 struct pt_regs *regs)
436 if (addr < TASK_SIZE)
437 return do_page_fault(addr, fsr, regs);
442 index = pgd_index(addr);
444 pgd = cpu_get_pgd() + index;
445 pgd_k = init_mm.pgd + index;
447 if (pgd_none(*pgd_k))
449 if (!pgd_present(*pgd))
450 set_pgd(pgd, *pgd_k);
452 pud = pud_offset(pgd, addr);
453 pud_k = pud_offset(pgd_k, addr);
455 if (pud_none(*pud_k))
457 if (!pud_present(*pud))
458 set_pud(pud, *pud_k);
460 pmd = pmd_offset(pud, addr);
461 pmd_k = pmd_offset(pud_k, addr);
463 #ifdef CONFIG_ARM_LPAE
465 * Only one hardware entry per PMD with LPAE.
470 * On ARM one Linux PGD entry contains two hardware entries (see page
471 * tables layout in pgtable.h). We normally guarantee that we always
472 * fill both L1 entries. But create_mapping() doesn't follow the rule.
473 * It can create inidividual L1 entries, so here we have to call
474 * pmd_none() check for the entry really corresponded to address, not
475 * for the first of pair.
477 index = (addr >> SECTION_SHIFT) & 1;
479 if (pmd_none(pmd_k[index]))
482 copy_pmd(pmd, pmd_k);
486 do_bad_area(addr, fsr, regs);
489 #else /* CONFIG_MMU */
491 do_translation_fault(unsigned long addr, unsigned int fsr,
492 struct pt_regs *regs)
496 #endif /* CONFIG_MMU */
499 * Some section permission faults need to be handled gracefully.
500 * They can happen due to a __{get,put}_user during an oops.
502 #ifndef CONFIG_ARM_LPAE
504 do_sect_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
506 do_bad_area(addr, fsr, regs);
509 #endif /* CONFIG_ARM_LPAE */
512 * This abort handler always returns "fault".
515 do_bad(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
521 int (*fn)(unsigned long addr, unsigned int fsr, struct pt_regs *regs);
528 #ifdef CONFIG_ARM_LPAE
529 #include "fsr-3level.c"
531 #include "fsr-2level.c"
535 hook_fault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
536 int sig, int code, const char *name)
538 if (nr < 0 || nr >= ARRAY_SIZE(fsr_info))
541 fsr_info[nr].fn = fn;
542 fsr_info[nr].sig = sig;
543 fsr_info[nr].code = code;
544 fsr_info[nr].name = name;
548 * Dispatch a data abort to the relevant handler.
550 asmlinkage void __exception
551 do_DataAbort(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
553 const struct fsr_info *inf = fsr_info + fsr_fs(fsr);
556 if (!inf->fn(addr, fsr & ~FSR_LNX_PF, regs))
559 pr_alert("Unhandled fault: %s (0x%03x) at 0x%08lx\n",
560 inf->name, fsr, addr);
561 show_pte(current->mm, addr);
563 info.si_signo = inf->sig;
565 info.si_code = inf->code;
566 info.si_addr = (void __user *)addr;
567 arm_notify_die("", regs, &info, fsr, 0);
571 hook_ifault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
572 int sig, int code, const char *name)
574 if (nr < 0 || nr >= ARRAY_SIZE(ifsr_info))
577 ifsr_info[nr].fn = fn;
578 ifsr_info[nr].sig = sig;
579 ifsr_info[nr].code = code;
580 ifsr_info[nr].name = name;
583 asmlinkage void __exception
584 do_PrefetchAbort(unsigned long addr, unsigned int ifsr, struct pt_regs *regs)
586 const struct fsr_info *inf = ifsr_info + fsr_fs(ifsr);
589 if (!inf->fn(addr, ifsr | FSR_LNX_PF, regs))
592 pr_alert("Unhandled prefetch abort: %s (0x%03x) at 0x%08lx\n",
593 inf->name, ifsr, addr);
595 info.si_signo = inf->sig;
597 info.si_code = inf->code;
598 info.si_addr = (void __user *)addr;
599 arm_notify_die("", regs, &info, ifsr, 0);
603 * Abort handler to be used only during first unmasking of asynchronous aborts
604 * on the boot CPU. This makes sure that the machine will not die if the
605 * firmware/bootloader left an imprecise abort pending for us to trip over.
607 static int __init early_abort_handler(unsigned long addr, unsigned int fsr,
608 struct pt_regs *regs)
610 pr_warn("Hit pending asynchronous external abort (FSR=0x%08x) during "
611 "first unmask, this is most likely caused by a "
612 "firmware/bootloader bug.\n", fsr);
617 void __init early_abt_enable(void)
619 fsr_info[FSR_FS_AEA].fn = early_abort_handler;
621 fsr_info[FSR_FS_AEA].fn = do_bad;
624 #ifndef CONFIG_ARM_LPAE
625 static int __init exceptions_init(void)
627 if (cpu_architecture() >= CPU_ARCH_ARMv6) {
628 hook_fault_code(4, do_translation_fault, SIGSEGV, SEGV_MAPERR,
629 "I-cache maintenance fault");
632 if (cpu_architecture() >= CPU_ARCH_ARMv7) {
634 * TODO: Access flag faults introduced in ARMv6K.
635 * Runtime check for 'K' extension is needed
637 hook_fault_code(3, do_bad, SIGSEGV, SEGV_MAPERR,
638 "section access flag fault");
639 hook_fault_code(6, do_bad, SIGSEGV, SEGV_MAPERR,
640 "section access flag fault");
646 arch_initcall(exceptions_init);