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>
20 #include <linux/kfence.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 pgd = pgd_offset(mm, addr);
42 printk("%s[%08lx] *pgd=%08llx", lvl, addr, (long long)pgd_val(*pgd));
50 p4d = p4d_offset(pgd, addr);
59 pud = pud_offset(p4d, addr);
60 if (PTRS_PER_PUD != 1)
61 pr_cont(", *pud=%08llx", (long long)pud_val(*pud));
71 pmd = pmd_offset(pud, addr);
72 if (PTRS_PER_PMD != 1)
73 pr_cont(", *pmd=%08llx", (long long)pmd_val(*pmd));
83 /* We must not map this if we have highmem enabled */
84 if (PageHighMem(pfn_to_page(pmd_val(*pmd) >> PAGE_SHIFT)))
87 pte = pte_offset_map(pmd, addr);
88 pr_cont(", *pte=%08llx", (long long)pte_val(*pte));
89 #ifndef CONFIG_ARM_LPAE
90 pr_cont(", *ppte=%08llx",
91 (long long)pte_val(pte[PTE_HWTABLE_PTRS]));
98 #else /* CONFIG_MMU */
99 void show_pte(const char *lvl, struct mm_struct *mm, unsigned long addr)
101 #endif /* CONFIG_MMU */
103 static inline bool is_write_fault(unsigned int fsr)
105 return (fsr & FSR_WRITE) && !(fsr & FSR_CM);
108 static void die_kernel_fault(const char *msg, struct mm_struct *mm,
109 unsigned long addr, unsigned int fsr,
110 struct pt_regs *regs)
113 pr_alert("8<--- cut here ---\n");
114 pr_alert("Unable to handle kernel %s at virtual address %08lx\n",
117 show_pte(KERN_ALERT, mm, addr);
118 die("Oops", regs, fsr);
124 * Oops. The kernel tried to access some page that wasn't present.
127 __do_kernel_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
128 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.
140 if (addr < PAGE_SIZE) {
141 msg = "NULL pointer dereference";
143 if (kfence_handle_page_fault(addr, is_write_fault(fsr), regs))
146 msg = "paging request";
149 die_kernel_fault(msg, mm, addr, fsr, regs);
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(unsigned long addr, unsigned int fsr, unsigned int sig,
158 int code, struct pt_regs *regs)
160 struct task_struct *tsk = current;
162 if (addr > TASK_SIZE)
163 harden_branch_predictor();
165 #ifdef CONFIG_DEBUG_USER
166 if (((user_debug & UDBG_SEGV) && (sig == SIGSEGV)) ||
167 ((user_debug & UDBG_BUS) && (sig == SIGBUS))) {
168 pr_err("8<--- cut here ---\n");
169 pr_err("%s: unhandled page fault (%d) at 0x%08lx, code 0x%03x\n",
170 tsk->comm, sig, addr, fsr);
171 show_pte(KERN_ERR, tsk->mm, addr);
175 #ifndef CONFIG_KUSER_HELPERS
176 if ((sig == SIGSEGV) && ((addr & PAGE_MASK) == 0xffff0000))
177 printk_ratelimited(KERN_DEBUG
178 "%s: CONFIG_KUSER_HELPERS disabled at 0x%08lx\n",
182 tsk->thread.address = addr;
183 tsk->thread.error_code = fsr;
184 tsk->thread.trap_no = 14;
185 force_sig_fault(sig, code, (void __user *)addr);
188 void do_bad_area(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
190 struct task_struct *tsk = current;
191 struct mm_struct *mm = tsk->active_mm;
194 * If we are in kernel mode at this point, we
195 * have no context to handle this fault with.
198 __do_user_fault(addr, fsr, SIGSEGV, SEGV_MAPERR, regs);
200 __do_kernel_fault(mm, addr, fsr, regs);
204 #define VM_FAULT_BADMAP ((__force vm_fault_t)0x010000)
205 #define VM_FAULT_BADACCESS ((__force vm_fault_t)0x020000)
207 static inline bool is_permission_fault(unsigned int fsr)
209 int fs = fsr_fs(fsr);
210 #ifdef CONFIG_ARM_LPAE
211 if ((fs & FS_PERM_NOLL_MASK) == FS_PERM_NOLL)
214 if (fs == FS_L1_PERM || fs == FS_L2_PERM)
220 static vm_fault_t __kprobes
221 __do_page_fault(struct mm_struct *mm, unsigned long addr, unsigned int flags,
222 unsigned long vma_flags, struct pt_regs *regs)
224 struct vm_area_struct *vma = find_vma(mm, addr);
226 return VM_FAULT_BADMAP;
228 if (unlikely(vma->vm_start > addr)) {
229 if (!(vma->vm_flags & VM_GROWSDOWN))
230 return VM_FAULT_BADMAP;
231 if (addr < FIRST_USER_ADDRESS)
232 return VM_FAULT_BADMAP;
233 if (expand_stack(vma, addr))
234 return VM_FAULT_BADMAP;
238 * ok, we have a good vm_area for this memory access, check the
239 * permissions on the VMA allow for the fault which occurred.
241 if (!(vma->vm_flags & vma_flags))
242 return VM_FAULT_BADACCESS;
244 return handle_mm_fault(vma, addr & PAGE_MASK, flags, regs);
248 do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
250 struct mm_struct *mm = current->mm;
253 unsigned int flags = FAULT_FLAG_DEFAULT;
254 unsigned long vm_flags = VM_ACCESS_FLAGS;
256 if (kprobe_page_fault(regs, fsr))
260 /* Enable interrupts if they were enabled in the parent context. */
261 if (interrupts_enabled(regs))
265 * If we're in an interrupt or have no user
266 * context, we must not take the fault..
268 if (faulthandler_disabled() || !mm)
272 flags |= FAULT_FLAG_USER;
274 if (is_write_fault(fsr)) {
275 flags |= FAULT_FLAG_WRITE;
279 if (fsr & FSR_LNX_PF) {
282 if (is_permission_fault(fsr) && !user_mode(regs))
283 die_kernel_fault("execution of memory",
284 mm, addr, fsr, regs);
287 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
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 (!mmap_read_trylock(mm)) {
295 if (!user_mode(regs) && !search_exception_tables(regs->ARM_pc))
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, flags, vm_flags, regs);
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_lock because
317 * it would already be released in __lock_page_or_retry in
319 if (fault_signal_pending(fault, regs)) {
320 if (!user_mode(regs))
325 if (!(fault & VM_FAULT_ERROR)) {
326 if (fault & VM_FAULT_RETRY) {
327 flags |= FAULT_FLAG_TRIED;
332 mmap_read_unlock(mm);
335 * Handle the "normal" case first - VM_FAULT_MAJOR
337 if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP | VM_FAULT_BADACCESS))))
341 * If we are in kernel mode at this point, we
342 * have no context to handle this fault with.
344 if (!user_mode(regs))
347 if (fault & VM_FAULT_OOM) {
349 * We ran out of memory, call the OOM killer, and return to
350 * userspace (which will retry the fault, or kill us if we
353 pagefault_out_of_memory();
357 if (fault & VM_FAULT_SIGBUS) {
359 * We had some memory, but were unable to
360 * successfully fix up this page fault.
366 * Something tried to access memory that
367 * isn't in our memory map..
370 code = fault == VM_FAULT_BADACCESS ?
371 SEGV_ACCERR : SEGV_MAPERR;
374 __do_user_fault(addr, fsr, sig, code, regs);
378 __do_kernel_fault(mm, addr, fsr, regs);
381 #else /* CONFIG_MMU */
383 do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
387 #endif /* CONFIG_MMU */
390 * First Level Translation Fault Handler
392 * We enter here because the first level page table doesn't contain
393 * a valid entry for the address.
395 * If the address is in kernel space (>= TASK_SIZE), then we are
396 * probably faulting in the vmalloc() area.
398 * If the init_task's first level page tables contains the relevant
399 * entry, we copy the it to this task. If not, we send the process
400 * a signal, fixup the exception, or oops the kernel.
402 * NOTE! We MUST NOT take any locks for this case. We may be in an
403 * interrupt or a critical region, and should only copy the information
404 * from the master page table, nothing more.
408 do_translation_fault(unsigned long addr, unsigned int fsr,
409 struct pt_regs *regs)
417 if (addr < TASK_SIZE)
418 return do_page_fault(addr, fsr, regs);
423 index = pgd_index(addr);
425 pgd = cpu_get_pgd() + index;
426 pgd_k = init_mm.pgd + index;
428 p4d = p4d_offset(pgd, addr);
429 p4d_k = p4d_offset(pgd_k, addr);
431 if (p4d_none(*p4d_k))
433 if (!p4d_present(*p4d))
434 set_p4d(p4d, *p4d_k);
436 pud = pud_offset(p4d, addr);
437 pud_k = pud_offset(p4d_k, addr);
439 if (pud_none(*pud_k))
441 if (!pud_present(*pud))
442 set_pud(pud, *pud_k);
444 pmd = pmd_offset(pud, addr);
445 pmd_k = pmd_offset(pud_k, addr);
447 #ifdef CONFIG_ARM_LPAE
449 * Only one hardware entry per PMD with LPAE.
454 * On ARM one Linux PGD entry contains two hardware entries (see page
455 * tables layout in pgtable.h). We normally guarantee that we always
456 * fill both L1 entries. But create_mapping() doesn't follow the rule.
457 * It can create inidividual L1 entries, so here we have to call
458 * pmd_none() check for the entry really corresponded to address, not
459 * for the first of pair.
461 index = (addr >> SECTION_SHIFT) & 1;
463 if (pmd_none(pmd_k[index]))
466 copy_pmd(pmd, pmd_k);
470 do_bad_area(addr, fsr, regs);
473 #else /* CONFIG_MMU */
475 do_translation_fault(unsigned long addr, unsigned int fsr,
476 struct pt_regs *regs)
480 #endif /* CONFIG_MMU */
483 * Some section permission faults need to be handled gracefully.
484 * They can happen due to a __{get,put}_user during an oops.
486 #ifndef CONFIG_ARM_LPAE
488 do_sect_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
490 do_bad_area(addr, fsr, regs);
493 #endif /* CONFIG_ARM_LPAE */
496 * This abort handler always returns "fault".
499 do_bad(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
505 int (*fn)(unsigned long addr, unsigned int fsr, struct pt_regs *regs);
512 #ifdef CONFIG_ARM_LPAE
513 #include "fsr-3level.c"
515 #include "fsr-2level.c"
519 hook_fault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
520 int sig, int code, const char *name)
522 if (nr < 0 || nr >= ARRAY_SIZE(fsr_info))
525 fsr_info[nr].fn = fn;
526 fsr_info[nr].sig = sig;
527 fsr_info[nr].code = code;
528 fsr_info[nr].name = name;
532 * Dispatch a data abort to the relevant handler.
535 do_DataAbort(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
537 const struct fsr_info *inf = fsr_info + fsr_fs(fsr);
539 if (!inf->fn(addr, fsr & ~FSR_LNX_PF, regs))
542 pr_alert("8<--- cut here ---\n");
543 pr_alert("Unhandled fault: %s (0x%03x) at 0x%08lx\n",
544 inf->name, fsr, addr);
545 show_pte(KERN_ALERT, current->mm, addr);
547 arm_notify_die("", regs, inf->sig, inf->code, (void __user *)addr,
552 hook_ifault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
553 int sig, int code, const char *name)
555 if (nr < 0 || nr >= ARRAY_SIZE(ifsr_info))
558 ifsr_info[nr].fn = fn;
559 ifsr_info[nr].sig = sig;
560 ifsr_info[nr].code = code;
561 ifsr_info[nr].name = name;
565 do_PrefetchAbort(unsigned long addr, unsigned int ifsr, struct pt_regs *regs)
567 const struct fsr_info *inf = ifsr_info + fsr_fs(ifsr);
569 if (!inf->fn(addr, ifsr | FSR_LNX_PF, regs))
572 pr_alert("Unhandled prefetch abort: %s (0x%03x) at 0x%08lx\n",
573 inf->name, ifsr, addr);
575 arm_notify_die("", regs, inf->sig, inf->code, (void __user *)addr,
580 * Abort handler to be used only during first unmasking of asynchronous aborts
581 * on the boot CPU. This makes sure that the machine will not die if the
582 * firmware/bootloader left an imprecise abort pending for us to trip over.
584 static int __init early_abort_handler(unsigned long addr, unsigned int fsr,
585 struct pt_regs *regs)
587 pr_warn("Hit pending asynchronous external abort (FSR=0x%08x) during "
588 "first unmask, this is most likely caused by a "
589 "firmware/bootloader bug.\n", fsr);
594 void __init early_abt_enable(void)
596 fsr_info[FSR_FS_AEA].fn = early_abort_handler;
598 fsr_info[FSR_FS_AEA].fn = do_bad;
601 #ifndef CONFIG_ARM_LPAE
602 static int __init exceptions_init(void)
604 if (cpu_architecture() >= CPU_ARCH_ARMv6) {
605 hook_fault_code(4, do_translation_fault, SIGSEGV, SEGV_MAPERR,
606 "I-cache maintenance fault");
609 if (cpu_architecture() >= CPU_ARCH_ARMv7) {
611 * TODO: Access flag faults introduced in ARMv6K.
612 * Runtime check for 'K' extension is needed
614 hook_fault_code(3, do_bad, SIGSEGV, SEGV_MAPERR,
615 "section access flag fault");
616 hook_fault_code(6, do_bad, SIGSEGV, SEGV_MAPERR,
617 "section access flag fault");
623 arch_initcall(exceptions_init);