1 // SPDX-License-Identifier: GPL-2.0-only
3 * linux/arch/arm/mm/fault.c
5 * Copyright (C) 1995 Linus Torvalds
6 * Modifications for ARM processor (c) 1995-2004 Russell King
8 #include <linux/extable.h>
9 #include <linux/signal.h>
11 #include <linux/hardirq.h>
12 #include <linux/init.h>
13 #include <linux/kprobes.h>
14 #include <linux/uaccess.h>
15 #include <linux/page-flags.h>
16 #include <linux/sched/signal.h>
17 #include <linux/sched/debug.h>
18 #include <linux/highmem.h>
19 #include <linux/perf_event.h>
21 #include <asm/pgtable.h>
22 #include <asm/system_misc.h>
23 #include <asm/system_info.h>
24 #include <asm/tlbflush.h>
31 static inline int notify_page_fault(struct pt_regs *regs, unsigned int fsr)
35 if (!user_mode(regs)) {
36 /* kprobe_running() needs smp_processor_id() */
38 if (kprobe_running() && kprobe_fault_handler(regs, fsr))
46 static inline int notify_page_fault(struct pt_regs *regs, unsigned int fsr)
53 * This is useful to dump out the page tables associated with
56 void show_pte(const char *lvl, struct mm_struct *mm, unsigned long addr)
63 printk("%spgd = %p\n", lvl, mm->pgd);
64 pgd = pgd_offset(mm, addr);
65 printk("%s[%08lx] *pgd=%08llx", lvl, addr, (long long)pgd_val(*pgd));
80 pud = pud_offset(pgd, addr);
81 if (PTRS_PER_PUD != 1)
82 pr_cont(", *pud=%08llx", (long long)pud_val(*pud));
92 pmd = pmd_offset(pud, addr);
93 if (PTRS_PER_PMD != 1)
94 pr_cont(", *pmd=%08llx", (long long)pmd_val(*pmd));
104 /* We must not map this if we have highmem enabled */
105 if (PageHighMem(pfn_to_page(pmd_val(*pmd) >> PAGE_SHIFT)))
108 pte = pte_offset_map(pmd, addr);
109 pr_cont(", *pte=%08llx", (long long)pte_val(*pte));
110 #ifndef CONFIG_ARM_LPAE
111 pr_cont(", *ppte=%08llx",
112 (long long)pte_val(pte[PTE_HWTABLE_PTRS]));
119 #else /* CONFIG_MMU */
120 void show_pte(const char *lvl, struct mm_struct *mm, unsigned long addr)
122 #endif /* CONFIG_MMU */
125 * Oops. The kernel tried to access some page that wasn't present.
128 __do_kernel_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
129 struct pt_regs *regs)
132 * Are we prepared to handle this kernel fault?
134 if (fixup_exception(regs))
138 * No handler, we'll have to terminate things with extreme prejudice.
141 pr_alert("8<--- cut here ---\n");
142 pr_alert("Unable to handle kernel %s at virtual address %08lx\n",
143 (addr < PAGE_SIZE) ? "NULL pointer dereference" :
144 "paging request", addr);
146 show_pte(KERN_ALERT, mm, addr);
147 die("Oops", regs, fsr);
153 * Something tried to access memory that isn't in our memory map..
154 * User mode accesses just cause a SIGSEGV
157 __do_user_fault(struct task_struct *tsk, unsigned long addr,
158 unsigned int fsr, unsigned int sig, int code,
159 struct pt_regs *regs)
161 if (addr > TASK_SIZE)
162 harden_branch_predictor();
164 #ifdef CONFIG_DEBUG_USER
165 if (((user_debug & UDBG_SEGV) && (sig == SIGSEGV)) ||
166 ((user_debug & UDBG_BUS) && (sig == SIGBUS))) {
167 pr_err("8<--- cut here ---\n");
168 pr_err("%s: unhandled page fault (%d) at 0x%08lx, code 0x%03x\n",
169 tsk->comm, sig, addr, fsr);
170 show_pte(KERN_ERR, tsk->mm, addr);
174 #ifndef CONFIG_KUSER_HELPERS
175 if ((sig == SIGSEGV) && ((addr & PAGE_MASK) == 0xffff0000))
176 printk_ratelimited(KERN_DEBUG
177 "%s: CONFIG_KUSER_HELPERS disabled at 0x%08lx\n",
181 tsk->thread.address = addr;
182 tsk->thread.error_code = fsr;
183 tsk->thread.trap_no = 14;
184 force_sig_fault(sig, code, (void __user *)addr, tsk);
187 void do_bad_area(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
189 struct task_struct *tsk = current;
190 struct mm_struct *mm = tsk->active_mm;
193 * If we are in kernel mode at this point, we
194 * have no context to handle this fault with.
197 __do_user_fault(tsk, addr, fsr, SIGSEGV, SEGV_MAPERR, regs);
199 __do_kernel_fault(mm, addr, fsr, regs);
203 #define VM_FAULT_BADMAP 0x010000
204 #define VM_FAULT_BADACCESS 0x020000
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.
211 static inline bool access_error(unsigned int fsr, struct vm_area_struct *vma)
213 unsigned int mask = VM_READ | VM_WRITE | VM_EXEC;
217 if (fsr & FSR_LNX_PF)
220 return vma->vm_flags & mask ? false : true;
223 static vm_fault_t __kprobes
224 __do_page_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
225 unsigned int flags, struct task_struct *tsk)
227 struct vm_area_struct *vma;
230 vma = find_vma(mm, addr);
231 fault = VM_FAULT_BADMAP;
234 if (unlikely(vma->vm_start > addr))
238 * Ok, we have a good vm_area for this
239 * memory access, so we can handle it.
242 if (access_error(fsr, vma)) {
243 fault = VM_FAULT_BADACCESS;
247 return handle_mm_fault(vma, addr & PAGE_MASK, flags);
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))
259 do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
261 struct task_struct *tsk;
262 struct mm_struct *mm;
265 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
267 if (notify_page_fault(regs, fsr))
273 /* Enable interrupts if they were enabled in the parent context. */
274 if (interrupts_enabled(regs))
278 * If we're in an interrupt or have no user
279 * context, we must not take the fault..
281 if (faulthandler_disabled() || !mm)
285 flags |= FAULT_FLAG_USER;
287 flags |= FAULT_FLAG_WRITE;
290 * As per x86, we may deadlock here. However, since the kernel only
291 * validly references user space from well defined areas of the code,
292 * we can bug out early if this is from code which shouldn't.
294 if (!down_read_trylock(&mm->mmap_sem)) {
295 if (!user_mode(regs) && !search_exception_tables(regs->ARM_pc))
298 down_read(&mm->mmap_sem);
301 * The above down_read_trylock() might have succeeded in
302 * which case, we'll have missed the might_sleep() from
306 #ifdef CONFIG_DEBUG_VM
307 if (!user_mode(regs) &&
308 !search_exception_tables(regs->ARM_pc))
313 fault = __do_page_fault(mm, addr, fsr, flags, tsk);
315 /* If we need to retry but a fatal signal is pending, handle the
316 * signal first. We do not need to release the mmap_sem because
317 * it would already be released in __lock_page_or_retry in
319 if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current)) {
320 if (!user_mode(regs))
326 * Major/minor page fault accounting is only done on the
327 * initial attempt. If we go through a retry, it is extremely
328 * likely that the page will be found in page cache at that point.
331 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
332 if (!(fault & VM_FAULT_ERROR) && flags & FAULT_FLAG_ALLOW_RETRY) {
333 if (fault & VM_FAULT_MAJOR) {
335 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
339 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
342 if (fault & VM_FAULT_RETRY) {
343 /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
345 flags &= ~FAULT_FLAG_ALLOW_RETRY;
346 flags |= FAULT_FLAG_TRIED;
351 up_read(&mm->mmap_sem);
354 * Handle the "normal" case first - VM_FAULT_MAJOR
356 if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP | VM_FAULT_BADACCESS))))
360 * If we are in kernel mode at this point, we
361 * have no context to handle this fault with.
363 if (!user_mode(regs))
366 if (fault & VM_FAULT_OOM) {
368 * We ran out of memory, call the OOM killer, and return to
369 * userspace (which will retry the fault, or kill us if we
372 pagefault_out_of_memory();
376 if (fault & VM_FAULT_SIGBUS) {
378 * We had some memory, but were unable to
379 * successfully fix up this page fault.
385 * Something tried to access memory that
386 * isn't in our memory map..
389 code = fault == VM_FAULT_BADACCESS ?
390 SEGV_ACCERR : SEGV_MAPERR;
393 __do_user_fault(tsk, addr, fsr, sig, code, regs);
397 __do_kernel_fault(mm, addr, fsr, regs);
400 #else /* CONFIG_MMU */
402 do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
406 #endif /* CONFIG_MMU */
409 * First Level Translation Fault Handler
411 * We enter here because the first level page table doesn't contain
412 * a valid entry for the address.
414 * If the address is in kernel space (>= TASK_SIZE), then we are
415 * probably faulting in the vmalloc() area.
417 * If the init_task's first level page tables contains the relevant
418 * entry, we copy the it to this task. If not, we send the process
419 * a signal, fixup the exception, or oops the kernel.
421 * NOTE! We MUST NOT take any locks for this case. We may be in an
422 * interrupt or a critical region, and should only copy the information
423 * from the master page table, nothing more.
427 do_translation_fault(unsigned long addr, unsigned int fsr,
428 struct pt_regs *regs)
435 if (addr < TASK_SIZE)
436 return do_page_fault(addr, fsr, regs);
441 index = pgd_index(addr);
443 pgd = cpu_get_pgd() + index;
444 pgd_k = init_mm.pgd + index;
446 if (pgd_none(*pgd_k))
448 if (!pgd_present(*pgd))
449 set_pgd(pgd, *pgd_k);
451 pud = pud_offset(pgd, addr);
452 pud_k = pud_offset(pgd_k, addr);
454 if (pud_none(*pud_k))
456 if (!pud_present(*pud))
457 set_pud(pud, *pud_k);
459 pmd = pmd_offset(pud, addr);
460 pmd_k = pmd_offset(pud_k, addr);
462 #ifdef CONFIG_ARM_LPAE
464 * Only one hardware entry per PMD with LPAE.
469 * On ARM one Linux PGD entry contains two hardware entries (see page
470 * tables layout in pgtable.h). We normally guarantee that we always
471 * fill both L1 entries. But create_mapping() doesn't follow the rule.
472 * It can create inidividual L1 entries, so here we have to call
473 * pmd_none() check for the entry really corresponded to address, not
474 * for the first of pair.
476 index = (addr >> SECTION_SHIFT) & 1;
478 if (pmd_none(pmd_k[index]))
481 copy_pmd(pmd, pmd_k);
485 do_bad_area(addr, fsr, regs);
488 #else /* CONFIG_MMU */
490 do_translation_fault(unsigned long addr, unsigned int fsr,
491 struct pt_regs *regs)
495 #endif /* CONFIG_MMU */
498 * Some section permission faults need to be handled gracefully.
499 * They can happen due to a __{get,put}_user during an oops.
501 #ifndef CONFIG_ARM_LPAE
503 do_sect_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
505 do_bad_area(addr, fsr, regs);
508 #endif /* CONFIG_ARM_LPAE */
511 * This abort handler always returns "fault".
514 do_bad(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
520 int (*fn)(unsigned long addr, unsigned int fsr, struct pt_regs *regs);
527 #ifdef CONFIG_ARM_LPAE
528 #include "fsr-3level.c"
530 #include "fsr-2level.c"
534 hook_fault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
535 int sig, int code, const char *name)
537 if (nr < 0 || nr >= ARRAY_SIZE(fsr_info))
540 fsr_info[nr].fn = fn;
541 fsr_info[nr].sig = sig;
542 fsr_info[nr].code = code;
543 fsr_info[nr].name = name;
547 * Dispatch a data abort to the relevant handler.
550 do_DataAbort(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
552 const struct fsr_info *inf = fsr_info + fsr_fs(fsr);
554 if (!inf->fn(addr, fsr & ~FSR_LNX_PF, regs))
557 pr_alert("8<--- cut here ---\n");
558 pr_alert("Unhandled fault: %s (0x%03x) at 0x%08lx\n",
559 inf->name, fsr, addr);
560 show_pte(KERN_ALERT, current->mm, addr);
562 arm_notify_die("", regs, inf->sig, inf->code, (void __user *)addr,
567 hook_ifault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
568 int sig, int code, const char *name)
570 if (nr < 0 || nr >= ARRAY_SIZE(ifsr_info))
573 ifsr_info[nr].fn = fn;
574 ifsr_info[nr].sig = sig;
575 ifsr_info[nr].code = code;
576 ifsr_info[nr].name = name;
580 do_PrefetchAbort(unsigned long addr, unsigned int ifsr, struct pt_regs *regs)
582 const struct fsr_info *inf = ifsr_info + fsr_fs(ifsr);
584 if (!inf->fn(addr, ifsr | FSR_LNX_PF, regs))
587 pr_alert("Unhandled prefetch abort: %s (0x%03x) at 0x%08lx\n",
588 inf->name, ifsr, addr);
590 arm_notify_die("", regs, inf->sig, inf->code, (void __user *)addr,
595 * Abort handler to be used only during first unmasking of asynchronous aborts
596 * on the boot CPU. This makes sure that the machine will not die if the
597 * firmware/bootloader left an imprecise abort pending for us to trip over.
599 static int __init early_abort_handler(unsigned long addr, unsigned int fsr,
600 struct pt_regs *regs)
602 pr_warn("Hit pending asynchronous external abort (FSR=0x%08x) during "
603 "first unmask, this is most likely caused by a "
604 "firmware/bootloader bug.\n", fsr);
609 void __init early_abt_enable(void)
611 fsr_info[FSR_FS_AEA].fn = early_abort_handler;
613 fsr_info[FSR_FS_AEA].fn = do_bad;
616 #ifndef CONFIG_ARM_LPAE
617 static int __init exceptions_init(void)
619 if (cpu_architecture() >= CPU_ARCH_ARMv6) {
620 hook_fault_code(4, do_translation_fault, SIGSEGV, SEGV_MAPERR,
621 "I-cache maintenance fault");
624 if (cpu_architecture() >= CPU_ARCH_ARMv7) {
626 * TODO: Access flag faults introduced in ARMv6K.
627 * Runtime check for 'K' extension is needed
629 hook_fault_code(3, do_bad, SIGSEGV, SEGV_MAPERR,
630 "section access flag fault");
631 hook_fault_code(6, do_bad, SIGSEGV, SEGV_MAPERR,
632 "section access flag fault");
638 arch_initcall(exceptions_init);