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 * This is useful to dump out the page tables associated with
34 void show_pte(const char *lvl, struct mm_struct *mm, unsigned long addr)
41 printk("%spgd = %p\n", lvl, mm->pgd);
42 pgd = pgd_offset(mm, addr);
43 printk("%s[%08lx] *pgd=%08llx", lvl, addr, (long long)pgd_val(*pgd));
58 pud = pud_offset(pgd, addr);
59 if (PTRS_PER_PUD != 1)
60 pr_cont(", *pud=%08llx", (long long)pud_val(*pud));
70 pmd = pmd_offset(pud, addr);
71 if (PTRS_PER_PMD != 1)
72 pr_cont(", *pmd=%08llx", (long long)pmd_val(*pmd));
82 /* We must not map this if we have highmem enabled */
83 if (PageHighMem(pfn_to_page(pmd_val(*pmd) >> PAGE_SHIFT)))
86 pte = pte_offset_map(pmd, addr);
87 pr_cont(", *pte=%08llx", (long long)pte_val(*pte));
88 #ifndef CONFIG_ARM_LPAE
89 pr_cont(", *ppte=%08llx",
90 (long long)pte_val(pte[PTE_HWTABLE_PTRS]));
97 #else /* CONFIG_MMU */
98 void show_pte(const char *lvl, struct mm_struct *mm, unsigned long addr)
100 #endif /* CONFIG_MMU */
103 * Oops. The kernel tried to access some page that wasn't present.
106 __do_kernel_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
107 struct pt_regs *regs)
110 * Are we prepared to handle this kernel fault?
112 if (fixup_exception(regs))
116 * No handler, we'll have to terminate things with extreme prejudice.
119 pr_alert("8<--- cut here ---\n");
120 pr_alert("Unable to handle kernel %s at virtual address %08lx\n",
121 (addr < PAGE_SIZE) ? "NULL pointer dereference" :
122 "paging request", addr);
124 show_pte(KERN_ALERT, mm, addr);
125 die("Oops", regs, fsr);
131 * Something tried to access memory that isn't in our memory map..
132 * User mode accesses just cause a SIGSEGV
135 __do_user_fault(unsigned long addr, unsigned int fsr, unsigned int sig,
136 int code, struct pt_regs *regs)
138 struct task_struct *tsk = current;
140 if (addr > TASK_SIZE)
141 harden_branch_predictor();
143 #ifdef CONFIG_DEBUG_USER
144 if (((user_debug & UDBG_SEGV) && (sig == SIGSEGV)) ||
145 ((user_debug & UDBG_BUS) && (sig == SIGBUS))) {
146 pr_err("8<--- cut here ---\n");
147 pr_err("%s: unhandled page fault (%d) at 0x%08lx, code 0x%03x\n",
148 tsk->comm, sig, addr, fsr);
149 show_pte(KERN_ERR, tsk->mm, addr);
153 #ifndef CONFIG_KUSER_HELPERS
154 if ((sig == SIGSEGV) && ((addr & PAGE_MASK) == 0xffff0000))
155 printk_ratelimited(KERN_DEBUG
156 "%s: CONFIG_KUSER_HELPERS disabled at 0x%08lx\n",
160 tsk->thread.address = addr;
161 tsk->thread.error_code = fsr;
162 tsk->thread.trap_no = 14;
163 force_sig_fault(sig, code, (void __user *)addr);
166 void do_bad_area(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
168 struct task_struct *tsk = current;
169 struct mm_struct *mm = tsk->active_mm;
172 * If we are in kernel mode at this point, we
173 * have no context to handle this fault with.
176 __do_user_fault(addr, fsr, SIGSEGV, SEGV_MAPERR, regs);
178 __do_kernel_fault(mm, addr, fsr, regs);
182 #define VM_FAULT_BADMAP 0x010000
183 #define VM_FAULT_BADACCESS 0x020000
186 * Check that the permissions on the VMA allow for the fault which occurred.
187 * If we encountered a write fault, we must have write permission, otherwise
188 * we allow any permission.
190 static inline bool access_error(unsigned int fsr, struct vm_area_struct *vma)
192 unsigned int mask = VM_READ | VM_WRITE | VM_EXEC;
194 if ((fsr & FSR_WRITE) && !(fsr & FSR_CM))
196 if (fsr & FSR_LNX_PF)
199 return vma->vm_flags & mask ? false : true;
202 static vm_fault_t __kprobes
203 __do_page_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
204 unsigned int flags, struct task_struct *tsk)
206 struct vm_area_struct *vma;
209 vma = find_vma(mm, addr);
210 fault = VM_FAULT_BADMAP;
213 if (unlikely(vma->vm_start > addr))
217 * Ok, we have a good vm_area for this
218 * memory access, so we can handle it.
221 if (access_error(fsr, vma)) {
222 fault = VM_FAULT_BADACCESS;
226 return handle_mm_fault(vma, addr & PAGE_MASK, flags);
229 /* Don't allow expansion below FIRST_USER_ADDRESS */
230 if (vma->vm_flags & VM_GROWSDOWN &&
231 addr >= FIRST_USER_ADDRESS && !expand_stack(vma, addr))
238 do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
240 struct task_struct *tsk;
241 struct mm_struct *mm;
244 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
246 if (kprobe_page_fault(regs, fsr))
252 /* Enable interrupts if they were enabled in the parent context. */
253 if (interrupts_enabled(regs))
257 * If we're in an interrupt or have no user
258 * context, we must not take the fault..
260 if (faulthandler_disabled() || !mm)
264 flags |= FAULT_FLAG_USER;
265 if ((fsr & FSR_WRITE) && !(fsr & FSR_CM))
266 flags |= FAULT_FLAG_WRITE;
269 * As per x86, we may deadlock here. However, since the kernel only
270 * validly references user space from well defined areas of the code,
271 * we can bug out early if this is from code which shouldn't.
273 if (!down_read_trylock(&mm->mmap_sem)) {
274 if (!user_mode(regs) && !search_exception_tables(regs->ARM_pc))
277 down_read(&mm->mmap_sem);
280 * The above down_read_trylock() might have succeeded in
281 * which case, we'll have missed the might_sleep() from
285 #ifdef CONFIG_DEBUG_VM
286 if (!user_mode(regs) &&
287 !search_exception_tables(regs->ARM_pc))
292 fault = __do_page_fault(mm, addr, fsr, flags, tsk);
294 /* If we need to retry but a fatal signal is pending, handle the
295 * signal first. We do not need to release the mmap_sem because
296 * it would already be released in __lock_page_or_retry in
298 if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current)) {
299 if (!user_mode(regs))
305 * Major/minor page fault accounting is only done on the
306 * initial attempt. If we go through a retry, it is extremely
307 * likely that the page will be found in page cache at that point.
310 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
311 if (!(fault & VM_FAULT_ERROR) && flags & FAULT_FLAG_ALLOW_RETRY) {
312 if (fault & VM_FAULT_MAJOR) {
314 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
318 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
321 if (fault & VM_FAULT_RETRY) {
322 /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
324 flags &= ~FAULT_FLAG_ALLOW_RETRY;
325 flags |= FAULT_FLAG_TRIED;
330 up_read(&mm->mmap_sem);
333 * Handle the "normal" case first - VM_FAULT_MAJOR
335 if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP | VM_FAULT_BADACCESS))))
339 * If we are in kernel mode at this point, we
340 * have no context to handle this fault with.
342 if (!user_mode(regs))
345 if (fault & VM_FAULT_OOM) {
347 * We ran out of memory, call the OOM killer, and return to
348 * userspace (which will retry the fault, or kill us if we
351 pagefault_out_of_memory();
355 if (fault & VM_FAULT_SIGBUS) {
357 * We had some memory, but were unable to
358 * successfully fix up this page fault.
364 * Something tried to access memory that
365 * isn't in our memory map..
368 code = fault == VM_FAULT_BADACCESS ?
369 SEGV_ACCERR : SEGV_MAPERR;
372 __do_user_fault(addr, fsr, sig, code, regs);
376 __do_kernel_fault(mm, addr, fsr, regs);
379 #else /* CONFIG_MMU */
381 do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
385 #endif /* CONFIG_MMU */
388 * First Level Translation Fault Handler
390 * We enter here because the first level page table doesn't contain
391 * a valid entry for the address.
393 * If the address is in kernel space (>= TASK_SIZE), then we are
394 * probably faulting in the vmalloc() area.
396 * If the init_task's first level page tables contains the relevant
397 * entry, we copy the it to this task. If not, we send the process
398 * a signal, fixup the exception, or oops the kernel.
400 * NOTE! We MUST NOT take any locks for this case. We may be in an
401 * interrupt or a critical region, and should only copy the information
402 * from the master page table, nothing more.
406 do_translation_fault(unsigned long addr, unsigned int fsr,
407 struct pt_regs *regs)
414 if (addr < TASK_SIZE)
415 return do_page_fault(addr, fsr, regs);
420 index = pgd_index(addr);
422 pgd = cpu_get_pgd() + index;
423 pgd_k = init_mm.pgd + index;
425 if (pgd_none(*pgd_k))
427 if (!pgd_present(*pgd))
428 set_pgd(pgd, *pgd_k);
430 pud = pud_offset(pgd, addr);
431 pud_k = pud_offset(pgd_k, addr);
433 if (pud_none(*pud_k))
435 if (!pud_present(*pud))
436 set_pud(pud, *pud_k);
438 pmd = pmd_offset(pud, addr);
439 pmd_k = pmd_offset(pud_k, addr);
441 #ifdef CONFIG_ARM_LPAE
443 * Only one hardware entry per PMD with LPAE.
448 * On ARM one Linux PGD entry contains two hardware entries (see page
449 * tables layout in pgtable.h). We normally guarantee that we always
450 * fill both L1 entries. But create_mapping() doesn't follow the rule.
451 * It can create inidividual L1 entries, so here we have to call
452 * pmd_none() check for the entry really corresponded to address, not
453 * for the first of pair.
455 index = (addr >> SECTION_SHIFT) & 1;
457 if (pmd_none(pmd_k[index]))
460 copy_pmd(pmd, pmd_k);
464 do_bad_area(addr, fsr, regs);
467 #else /* CONFIG_MMU */
469 do_translation_fault(unsigned long addr, unsigned int fsr,
470 struct pt_regs *regs)
474 #endif /* CONFIG_MMU */
477 * Some section permission faults need to be handled gracefully.
478 * They can happen due to a __{get,put}_user during an oops.
480 #ifndef CONFIG_ARM_LPAE
482 do_sect_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
484 do_bad_area(addr, fsr, regs);
487 #endif /* CONFIG_ARM_LPAE */
490 * This abort handler always returns "fault".
493 do_bad(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
499 int (*fn)(unsigned long addr, unsigned int fsr, struct pt_regs *regs);
506 #ifdef CONFIG_ARM_LPAE
507 #include "fsr-3level.c"
509 #include "fsr-2level.c"
513 hook_fault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
514 int sig, int code, const char *name)
516 if (nr < 0 || nr >= ARRAY_SIZE(fsr_info))
519 fsr_info[nr].fn = fn;
520 fsr_info[nr].sig = sig;
521 fsr_info[nr].code = code;
522 fsr_info[nr].name = name;
526 * Dispatch a data abort to the relevant handler.
529 do_DataAbort(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
531 const struct fsr_info *inf = fsr_info + fsr_fs(fsr);
533 if (!inf->fn(addr, fsr & ~FSR_LNX_PF, regs))
536 pr_alert("8<--- cut here ---\n");
537 pr_alert("Unhandled fault: %s (0x%03x) at 0x%08lx\n",
538 inf->name, fsr, addr);
539 show_pte(KERN_ALERT, current->mm, addr);
541 arm_notify_die("", regs, inf->sig, inf->code, (void __user *)addr,
546 hook_ifault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
547 int sig, int code, const char *name)
549 if (nr < 0 || nr >= ARRAY_SIZE(ifsr_info))
552 ifsr_info[nr].fn = fn;
553 ifsr_info[nr].sig = sig;
554 ifsr_info[nr].code = code;
555 ifsr_info[nr].name = name;
559 do_PrefetchAbort(unsigned long addr, unsigned int ifsr, struct pt_regs *regs)
561 const struct fsr_info *inf = ifsr_info + fsr_fs(ifsr);
563 if (!inf->fn(addr, ifsr | FSR_LNX_PF, regs))
566 pr_alert("Unhandled prefetch abort: %s (0x%03x) at 0x%08lx\n",
567 inf->name, ifsr, addr);
569 arm_notify_die("", regs, inf->sig, inf->code, (void __user *)addr,
574 * Abort handler to be used only during first unmasking of asynchronous aborts
575 * on the boot CPU. This makes sure that the machine will not die if the
576 * firmware/bootloader left an imprecise abort pending for us to trip over.
578 static int __init early_abort_handler(unsigned long addr, unsigned int fsr,
579 struct pt_regs *regs)
581 pr_warn("Hit pending asynchronous external abort (FSR=0x%08x) during "
582 "first unmask, this is most likely caused by a "
583 "firmware/bootloader bug.\n", fsr);
588 void __init early_abt_enable(void)
590 fsr_info[FSR_FS_AEA].fn = early_abort_handler;
592 fsr_info[FSR_FS_AEA].fn = do_bad;
595 #ifndef CONFIG_ARM_LPAE
596 static int __init exceptions_init(void)
598 if (cpu_architecture() >= CPU_ARCH_ARMv6) {
599 hook_fault_code(4, do_translation_fault, SIGSEGV, SEGV_MAPERR,
600 "I-cache maintenance fault");
603 if (cpu_architecture() >= CPU_ARCH_ARMv7) {
605 * TODO: Access flag faults introduced in ARMv6K.
606 * Runtime check for 'K' extension is needed
608 hook_fault_code(3, do_bad, SIGSEGV, SEGV_MAPERR,
609 "section access flag fault");
610 hook_fault_code(6, do_bad, SIGSEGV, SEGV_MAPERR,
611 "section access flag fault");
617 arch_initcall(exceptions_init);