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>
28 bool copy_from_kernel_nofault_allowed(const void *unsafe_src, size_t size)
30 unsigned long addr = (unsigned long)unsafe_src;
32 return addr >= TASK_SIZE && ULONG_MAX - addr >= size;
38 * This is useful to dump out the page tables associated with
41 void show_pte(const char *lvl, struct mm_struct *mm, unsigned long addr)
48 pgd = pgd_offset(mm, addr);
49 printk("%s[%08lx] *pgd=%08llx", lvl, addr, (long long)pgd_val(*pgd));
57 p4d = p4d_offset(pgd, addr);
66 pud = pud_offset(p4d, addr);
67 if (PTRS_PER_PUD != 1)
68 pr_cont(", *pud=%08llx", (long long)pud_val(*pud));
78 pmd = pmd_offset(pud, addr);
79 if (PTRS_PER_PMD != 1)
80 pr_cont(", *pmd=%08llx", (long long)pmd_val(*pmd));
90 /* We must not map this if we have highmem enabled */
91 if (PageHighMem(pfn_to_page(pmd_val(*pmd) >> PAGE_SHIFT)))
94 pte = pte_offset_map(pmd, addr);
98 pr_cont(", *pte=%08llx", (long long)pte_val(*pte));
99 #ifndef CONFIG_ARM_LPAE
100 pr_cont(", *ppte=%08llx",
101 (long long)pte_val(pte[PTE_HWTABLE_PTRS]));
108 #else /* CONFIG_MMU */
109 void show_pte(const char *lvl, struct mm_struct *mm, unsigned long addr)
111 #endif /* CONFIG_MMU */
113 static inline bool is_write_fault(unsigned int fsr)
115 return (fsr & FSR_WRITE) && !(fsr & FSR_CM);
118 static inline bool is_translation_fault(unsigned int fsr)
120 int fs = fsr_fs(fsr);
121 #ifdef CONFIG_ARM_LPAE
122 if ((fs & FS_MMU_NOLL_MASK) == FS_TRANS_NOLL)
125 if (fs == FS_L1_TRANS || fs == FS_L2_TRANS)
131 static void die_kernel_fault(const char *msg, struct mm_struct *mm,
132 unsigned long addr, unsigned int fsr,
133 struct pt_regs *regs)
136 pr_alert("8<--- cut here ---\n");
137 pr_alert("Unable to handle kernel %s at virtual address %08lx when %s\n",
138 msg, addr, fsr & FSR_LNX_PF ? "execute" :
139 fsr & FSR_WRITE ? "write" : "read");
141 show_pte(KERN_ALERT, mm, addr);
142 die("Oops", regs, fsr);
144 make_task_dead(SIGKILL);
148 * Oops. The kernel tried to access some page that wasn't present.
151 __do_kernel_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
152 struct pt_regs *regs)
156 * Are we prepared to handle this kernel fault?
158 if (fixup_exception(regs))
162 * No handler, we'll have to terminate things with extreme prejudice.
164 if (addr < PAGE_SIZE) {
165 msg = "NULL pointer dereference";
167 if (is_translation_fault(fsr) &&
168 kfence_handle_page_fault(addr, is_write_fault(fsr), regs))
171 msg = "paging request";
174 die_kernel_fault(msg, mm, addr, fsr, regs);
178 * Something tried to access memory that isn't in our memory map..
179 * User mode accesses just cause a SIGSEGV
182 __do_user_fault(unsigned long addr, unsigned int fsr, unsigned int sig,
183 int code, struct pt_regs *regs)
185 struct task_struct *tsk = current;
187 if (addr > TASK_SIZE)
188 harden_branch_predictor();
190 #ifdef CONFIG_DEBUG_USER
191 if (((user_debug & UDBG_SEGV) && (sig == SIGSEGV)) ||
192 ((user_debug & UDBG_BUS) && (sig == SIGBUS))) {
193 pr_err("8<--- cut here ---\n");
194 pr_err("%s: unhandled page fault (%d) at 0x%08lx, code 0x%03x\n",
195 tsk->comm, sig, addr, fsr);
196 show_pte(KERN_ERR, tsk->mm, addr);
200 #ifndef CONFIG_KUSER_HELPERS
201 if ((sig == SIGSEGV) && ((addr & PAGE_MASK) == 0xffff0000))
202 printk_ratelimited(KERN_DEBUG
203 "%s: CONFIG_KUSER_HELPERS disabled at 0x%08lx\n",
207 tsk->thread.address = addr;
208 tsk->thread.error_code = fsr;
209 tsk->thread.trap_no = 14;
210 force_sig_fault(sig, code, (void __user *)addr);
213 void do_bad_area(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
215 struct task_struct *tsk = current;
216 struct mm_struct *mm = tsk->active_mm;
219 * If we are in kernel mode at this point, we
220 * have no context to handle this fault with.
223 __do_user_fault(addr, fsr, SIGSEGV, SEGV_MAPERR, regs);
225 __do_kernel_fault(mm, addr, fsr, regs);
229 #define VM_FAULT_BADMAP ((__force vm_fault_t)0x010000)
230 #define VM_FAULT_BADACCESS ((__force vm_fault_t)0x020000)
232 static inline bool is_permission_fault(unsigned int fsr)
234 int fs = fsr_fs(fsr);
235 #ifdef CONFIG_ARM_LPAE
236 if ((fs & FS_MMU_NOLL_MASK) == FS_PERM_NOLL)
239 if (fs == FS_L1_PERM || fs == FS_L2_PERM)
246 do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
248 struct mm_struct *mm = current->mm;
249 struct vm_area_struct *vma;
252 unsigned int flags = FAULT_FLAG_DEFAULT;
253 unsigned long vm_flags = VM_ACCESS_FLAGS;
255 if (kprobe_page_fault(regs, fsr))
259 /* Enable interrupts if they were enabled in the parent context. */
260 if (interrupts_enabled(regs))
264 * If we're in an interrupt or have no user
265 * context, we must not take the fault..
267 if (faulthandler_disabled() || !mm)
271 flags |= FAULT_FLAG_USER;
273 if (is_write_fault(fsr)) {
274 flags |= FAULT_FLAG_WRITE;
278 if (fsr & FSR_LNX_PF) {
281 if (is_permission_fault(fsr) && !user_mode(regs))
282 die_kernel_fault("execution of memory",
283 mm, addr, fsr, regs);
286 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
288 if (!(flags & FAULT_FLAG_USER))
291 vma = lock_vma_under_rcu(mm, addr);
295 if (!(vma->vm_flags & vm_flags)) {
299 fault = handle_mm_fault(vma, addr, flags | FAULT_FLAG_VMA_LOCK, regs);
300 if (!(fault & (VM_FAULT_RETRY | VM_FAULT_COMPLETED)))
303 if (!(fault & VM_FAULT_RETRY)) {
304 count_vm_vma_lock_event(VMA_LOCK_SUCCESS);
307 count_vm_vma_lock_event(VMA_LOCK_RETRY);
308 if (fault & VM_FAULT_MAJOR)
309 flags |= FAULT_FLAG_TRIED;
311 /* Quick path to respond to signals */
312 if (fault_signal_pending(fault, regs)) {
313 if (!user_mode(regs))
320 vma = lock_mm_and_find_vma(mm, addr, regs);
321 if (unlikely(!vma)) {
322 fault = VM_FAULT_BADMAP;
327 * ok, we have a good vm_area for this memory access, check the
328 * permissions on the VMA allow for the fault which occurred.
330 if (!(vma->vm_flags & vm_flags))
331 fault = VM_FAULT_BADACCESS;
333 fault = handle_mm_fault(vma, addr & PAGE_MASK, flags, regs);
335 /* If we need to retry but a fatal signal is pending, handle the
336 * signal first. We do not need to release the mmap_lock because
337 * it would already be released in __lock_page_or_retry in
339 if (fault_signal_pending(fault, regs)) {
340 if (!user_mode(regs))
345 /* The fault is fully completed (including releasing mmap lock) */
346 if (fault & VM_FAULT_COMPLETED)
349 if (!(fault & VM_FAULT_ERROR)) {
350 if (fault & VM_FAULT_RETRY) {
351 flags |= FAULT_FLAG_TRIED;
356 mmap_read_unlock(mm);
360 * Handle the "normal" case first - VM_FAULT_MAJOR
362 if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP | VM_FAULT_BADACCESS))))
367 * If we are in kernel mode at this point, we
368 * have no context to handle this fault with.
370 if (!user_mode(regs))
373 if (fault & VM_FAULT_OOM) {
375 * We ran out of memory, call the OOM killer, and return to
376 * userspace (which will retry the fault, or kill us if we
379 pagefault_out_of_memory();
383 if (fault & VM_FAULT_SIGBUS) {
385 * We had some memory, but were unable to
386 * successfully fix up this page fault.
392 * Something tried to access memory that
393 * isn't in our memory map..
396 code = fault == VM_FAULT_BADACCESS ?
397 SEGV_ACCERR : SEGV_MAPERR;
400 __do_user_fault(addr, fsr, sig, code, regs);
404 __do_kernel_fault(mm, addr, fsr, regs);
407 #else /* CONFIG_MMU */
409 do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
413 #endif /* CONFIG_MMU */
416 * First Level Translation Fault Handler
418 * We enter here because the first level page table doesn't contain
419 * a valid entry for the address.
421 * If the address is in kernel space (>= TASK_SIZE), then we are
422 * probably faulting in the vmalloc() area.
424 * If the init_task's first level page tables contains the relevant
425 * entry, we copy the it to this task. If not, we send the process
426 * a signal, fixup the exception, or oops the kernel.
428 * NOTE! We MUST NOT take any locks for this case. We may be in an
429 * interrupt or a critical region, and should only copy the information
430 * from the master page table, nothing more.
434 do_translation_fault(unsigned long addr, unsigned int fsr,
435 struct pt_regs *regs)
443 if (addr < TASK_SIZE)
444 return do_page_fault(addr, fsr, regs);
449 index = pgd_index(addr);
451 pgd = cpu_get_pgd() + index;
452 pgd_k = init_mm.pgd + index;
454 p4d = p4d_offset(pgd, addr);
455 p4d_k = p4d_offset(pgd_k, addr);
457 if (p4d_none(*p4d_k))
459 if (!p4d_present(*p4d))
460 set_p4d(p4d, *p4d_k);
462 pud = pud_offset(p4d, addr);
463 pud_k = pud_offset(p4d_k, addr);
465 if (pud_none(*pud_k))
467 if (!pud_present(*pud))
468 set_pud(pud, *pud_k);
470 pmd = pmd_offset(pud, addr);
471 pmd_k = pmd_offset(pud_k, addr);
473 #ifdef CONFIG_ARM_LPAE
475 * Only one hardware entry per PMD with LPAE.
480 * On ARM one Linux PGD entry contains two hardware entries (see page
481 * tables layout in pgtable.h). We normally guarantee that we always
482 * fill both L1 entries. But create_mapping() doesn't follow the rule.
483 * It can create inidividual L1 entries, so here we have to call
484 * pmd_none() check for the entry really corresponded to address, not
485 * for the first of pair.
487 index = (addr >> SECTION_SHIFT) & 1;
489 if (pmd_none(pmd_k[index]))
492 copy_pmd(pmd, pmd_k);
496 do_bad_area(addr, fsr, regs);
499 #else /* CONFIG_MMU */
501 do_translation_fault(unsigned long addr, unsigned int fsr,
502 struct pt_regs *regs)
506 #endif /* CONFIG_MMU */
509 * Some section permission faults need to be handled gracefully.
510 * They can happen due to a __{get,put}_user during an oops.
512 #ifndef CONFIG_ARM_LPAE
514 do_sect_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
516 do_bad_area(addr, fsr, regs);
519 #endif /* CONFIG_ARM_LPAE */
522 * This abort handler always returns "fault".
525 do_bad(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
531 int (*fn)(unsigned long addr, unsigned int fsr, struct pt_regs *regs);
538 #ifdef CONFIG_ARM_LPAE
539 #include "fsr-3level.c"
541 #include "fsr-2level.c"
545 hook_fault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
546 int sig, int code, const char *name)
548 if (nr < 0 || nr >= ARRAY_SIZE(fsr_info))
551 fsr_info[nr].fn = fn;
552 fsr_info[nr].sig = sig;
553 fsr_info[nr].code = code;
554 fsr_info[nr].name = name;
558 * Dispatch a data abort to the relevant handler.
561 do_DataAbort(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
563 const struct fsr_info *inf = fsr_info + fsr_fs(fsr);
565 if (!inf->fn(addr, fsr & ~FSR_LNX_PF, regs))
568 pr_alert("8<--- cut here ---\n");
569 pr_alert("Unhandled fault: %s (0x%03x) at 0x%08lx\n",
570 inf->name, fsr, addr);
571 show_pte(KERN_ALERT, current->mm, addr);
573 arm_notify_die("", regs, inf->sig, inf->code, (void __user *)addr,
578 hook_ifault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
579 int sig, int code, const char *name)
581 if (nr < 0 || nr >= ARRAY_SIZE(ifsr_info))
584 ifsr_info[nr].fn = fn;
585 ifsr_info[nr].sig = sig;
586 ifsr_info[nr].code = code;
587 ifsr_info[nr].name = name;
591 do_PrefetchAbort(unsigned long addr, unsigned int ifsr, struct pt_regs *regs)
593 const struct fsr_info *inf = ifsr_info + fsr_fs(ifsr);
595 if (!inf->fn(addr, ifsr | FSR_LNX_PF, regs))
598 pr_alert("8<--- cut here ---\n");
599 pr_alert("Unhandled prefetch abort: %s (0x%03x) at 0x%08lx\n",
600 inf->name, ifsr, addr);
602 arm_notify_die("", regs, inf->sig, inf->code, (void __user *)addr,
607 * Abort handler to be used only during first unmasking of asynchronous aborts
608 * on the boot CPU. This makes sure that the machine will not die if the
609 * firmware/bootloader left an imprecise abort pending for us to trip over.
611 static int __init early_abort_handler(unsigned long addr, unsigned int fsr,
612 struct pt_regs *regs)
614 pr_warn("Hit pending asynchronous external abort (FSR=0x%08x) during "
615 "first unmask, this is most likely caused by a "
616 "firmware/bootloader bug.\n", fsr);
621 void __init early_abt_enable(void)
623 fsr_info[FSR_FS_AEA].fn = early_abort_handler;
625 fsr_info[FSR_FS_AEA].fn = do_bad;
628 #ifndef CONFIG_ARM_LPAE
629 static int __init exceptions_init(void)
631 if (cpu_architecture() >= CPU_ARCH_ARMv6) {
632 hook_fault_code(4, do_translation_fault, SIGSEGV, SEGV_MAPERR,
633 "I-cache maintenance fault");
636 if (cpu_architecture() >= CPU_ARCH_ARMv7) {
638 * TODO: Access flag faults introduced in ARMv6K.
639 * Runtime check for 'K' extension is needed
641 hook_fault_code(3, do_bad, SIGSEGV, SEGV_MAPERR,
642 "section access flag fault");
643 hook_fault_code(6, do_bad, SIGSEGV, SEGV_MAPERR,
644 "section access flag fault");
650 arch_initcall(exceptions_init);