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 inline bool is_translation_fault(unsigned int fsr)
110 int fs = fsr_fs(fsr);
111 #ifdef CONFIG_ARM_LPAE
112 if ((fs & FS_MMU_NOLL_MASK) == FS_TRANS_NOLL)
115 if (fs == FS_L1_TRANS || fs == FS_L2_TRANS)
121 static void die_kernel_fault(const char *msg, struct mm_struct *mm,
122 unsigned long addr, unsigned int fsr,
123 struct pt_regs *regs)
126 pr_alert("8<--- cut here ---\n");
127 pr_alert("Unable to handle kernel %s at virtual address %08lx when %s\n",
128 msg, addr, fsr & FSR_LNX_PF ? "execute" :
129 fsr & FSR_WRITE ? "write" : "read");
131 show_pte(KERN_ALERT, mm, addr);
132 die("Oops", regs, fsr);
134 make_task_dead(SIGKILL);
138 * Oops. The kernel tried to access some page that wasn't present.
141 __do_kernel_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
142 struct pt_regs *regs)
146 * Are we prepared to handle this kernel fault?
148 if (fixup_exception(regs))
152 * No handler, we'll have to terminate things with extreme prejudice.
154 if (addr < PAGE_SIZE) {
155 msg = "NULL pointer dereference";
157 if (is_translation_fault(fsr) &&
158 kfence_handle_page_fault(addr, is_write_fault(fsr), regs))
161 msg = "paging request";
164 die_kernel_fault(msg, mm, addr, fsr, regs);
168 * Something tried to access memory that isn't in our memory map..
169 * User mode accesses just cause a SIGSEGV
172 __do_user_fault(unsigned long addr, unsigned int fsr, unsigned int sig,
173 int code, struct pt_regs *regs)
175 struct task_struct *tsk = current;
177 if (addr > TASK_SIZE)
178 harden_branch_predictor();
180 #ifdef CONFIG_DEBUG_USER
181 if (((user_debug & UDBG_SEGV) && (sig == SIGSEGV)) ||
182 ((user_debug & UDBG_BUS) && (sig == SIGBUS))) {
183 pr_err("8<--- cut here ---\n");
184 pr_err("%s: unhandled page fault (%d) at 0x%08lx, code 0x%03x\n",
185 tsk->comm, sig, addr, fsr);
186 show_pte(KERN_ERR, tsk->mm, addr);
190 #ifndef CONFIG_KUSER_HELPERS
191 if ((sig == SIGSEGV) && ((addr & PAGE_MASK) == 0xffff0000))
192 printk_ratelimited(KERN_DEBUG
193 "%s: CONFIG_KUSER_HELPERS disabled at 0x%08lx\n",
197 tsk->thread.address = addr;
198 tsk->thread.error_code = fsr;
199 tsk->thread.trap_no = 14;
200 force_sig_fault(sig, code, (void __user *)addr);
203 void do_bad_area(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
205 struct task_struct *tsk = current;
206 struct mm_struct *mm = tsk->active_mm;
209 * If we are in kernel mode at this point, we
210 * have no context to handle this fault with.
213 __do_user_fault(addr, fsr, SIGSEGV, SEGV_MAPERR, regs);
215 __do_kernel_fault(mm, addr, fsr, regs);
219 #define VM_FAULT_BADMAP ((__force vm_fault_t)0x010000)
220 #define VM_FAULT_BADACCESS ((__force vm_fault_t)0x020000)
222 static inline bool is_permission_fault(unsigned int fsr)
224 int fs = fsr_fs(fsr);
225 #ifdef CONFIG_ARM_LPAE
226 if ((fs & FS_MMU_NOLL_MASK) == FS_PERM_NOLL)
229 if (fs == FS_L1_PERM || fs == FS_L2_PERM)
235 static vm_fault_t __kprobes
236 __do_page_fault(struct mm_struct *mm, unsigned long addr, unsigned int flags,
237 unsigned long vma_flags, struct pt_regs *regs)
239 struct vm_area_struct *vma = find_vma(mm, addr);
241 return VM_FAULT_BADMAP;
243 if (unlikely(vma->vm_start > addr)) {
244 if (!(vma->vm_flags & VM_GROWSDOWN))
245 return VM_FAULT_BADMAP;
246 if (addr < FIRST_USER_ADDRESS)
247 return VM_FAULT_BADMAP;
248 if (expand_stack(vma, addr))
249 return VM_FAULT_BADMAP;
253 * ok, we have a good vm_area for this memory access, check the
254 * permissions on the VMA allow for the fault which occurred.
256 if (!(vma->vm_flags & vma_flags))
257 return VM_FAULT_BADACCESS;
259 return handle_mm_fault(vma, addr & PAGE_MASK, flags, regs);
263 do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
265 struct mm_struct *mm = current->mm;
268 unsigned int flags = FAULT_FLAG_DEFAULT;
269 unsigned long vm_flags = VM_ACCESS_FLAGS;
271 if (kprobe_page_fault(regs, fsr))
275 /* Enable interrupts if they were enabled in the parent context. */
276 if (interrupts_enabled(regs))
280 * If we're in an interrupt or have no user
281 * context, we must not take the fault..
283 if (faulthandler_disabled() || !mm)
287 flags |= FAULT_FLAG_USER;
289 if (is_write_fault(fsr)) {
290 flags |= FAULT_FLAG_WRITE;
294 if (fsr & FSR_LNX_PF) {
297 if (is_permission_fault(fsr) && !user_mode(regs))
298 die_kernel_fault("execution of memory",
299 mm, addr, fsr, regs);
302 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
305 * As per x86, we may deadlock here. However, since the kernel only
306 * validly references user space from well defined areas of the code,
307 * we can bug out early if this is from code which shouldn't.
309 if (!mmap_read_trylock(mm)) {
310 if (!user_mode(regs) && !search_exception_tables(regs->ARM_pc))
316 * The above down_read_trylock() might have succeeded in
317 * which case, we'll have missed the might_sleep() from
321 #ifdef CONFIG_DEBUG_VM
322 if (!user_mode(regs) &&
323 !search_exception_tables(regs->ARM_pc))
328 fault = __do_page_fault(mm, addr, flags, vm_flags, regs);
330 /* If we need to retry but a fatal signal is pending, handle the
331 * signal first. We do not need to release the mmap_lock because
332 * it would already be released in __lock_page_or_retry in
334 if (fault_signal_pending(fault, regs)) {
335 if (!user_mode(regs))
340 /* The fault is fully completed (including releasing mmap lock) */
341 if (fault & VM_FAULT_COMPLETED)
344 if (!(fault & VM_FAULT_ERROR)) {
345 if (fault & VM_FAULT_RETRY) {
346 flags |= FAULT_FLAG_TRIED;
351 mmap_read_unlock(mm);
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(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)
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 p4d = p4d_offset(pgd, addr);
448 p4d_k = p4d_offset(pgd_k, addr);
450 if (p4d_none(*p4d_k))
452 if (!p4d_present(*p4d))
453 set_p4d(p4d, *p4d_k);
455 pud = pud_offset(p4d, addr);
456 pud_k = pud_offset(p4d_k, addr);
458 if (pud_none(*pud_k))
460 if (!pud_present(*pud))
461 set_pud(pud, *pud_k);
463 pmd = pmd_offset(pud, addr);
464 pmd_k = pmd_offset(pud_k, addr);
466 #ifdef CONFIG_ARM_LPAE
468 * Only one hardware entry per PMD with LPAE.
473 * On ARM one Linux PGD entry contains two hardware entries (see page
474 * tables layout in pgtable.h). We normally guarantee that we always
475 * fill both L1 entries. But create_mapping() doesn't follow the rule.
476 * It can create inidividual L1 entries, so here we have to call
477 * pmd_none() check for the entry really corresponded to address, not
478 * for the first of pair.
480 index = (addr >> SECTION_SHIFT) & 1;
482 if (pmd_none(pmd_k[index]))
485 copy_pmd(pmd, pmd_k);
489 do_bad_area(addr, fsr, regs);
492 #else /* CONFIG_MMU */
494 do_translation_fault(unsigned long addr, unsigned int fsr,
495 struct pt_regs *regs)
499 #endif /* CONFIG_MMU */
502 * Some section permission faults need to be handled gracefully.
503 * They can happen due to a __{get,put}_user during an oops.
505 #ifndef CONFIG_ARM_LPAE
507 do_sect_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
509 do_bad_area(addr, fsr, regs);
512 #endif /* CONFIG_ARM_LPAE */
515 * This abort handler always returns "fault".
518 do_bad(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
524 int (*fn)(unsigned long addr, unsigned int fsr, struct pt_regs *regs);
531 #ifdef CONFIG_ARM_LPAE
532 #include "fsr-3level.c"
534 #include "fsr-2level.c"
538 hook_fault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
539 int sig, int code, const char *name)
541 if (nr < 0 || nr >= ARRAY_SIZE(fsr_info))
544 fsr_info[nr].fn = fn;
545 fsr_info[nr].sig = sig;
546 fsr_info[nr].code = code;
547 fsr_info[nr].name = name;
551 * Dispatch a data abort to the relevant handler.
554 do_DataAbort(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
556 const struct fsr_info *inf = fsr_info + fsr_fs(fsr);
558 if (!inf->fn(addr, fsr & ~FSR_LNX_PF, regs))
561 pr_alert("8<--- cut here ---\n");
562 pr_alert("Unhandled fault: %s (0x%03x) at 0x%08lx\n",
563 inf->name, fsr, addr);
564 show_pte(KERN_ALERT, current->mm, addr);
566 arm_notify_die("", regs, inf->sig, inf->code, (void __user *)addr,
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;
584 do_PrefetchAbort(unsigned long addr, unsigned int ifsr, struct pt_regs *regs)
586 const struct fsr_info *inf = ifsr_info + fsr_fs(ifsr);
588 if (!inf->fn(addr, ifsr | FSR_LNX_PF, regs))
591 pr_alert("Unhandled prefetch abort: %s (0x%03x) at 0x%08lx\n",
592 inf->name, ifsr, addr);
594 arm_notify_die("", regs, inf->sig, inf->code, (void __user *)addr,
599 * Abort handler to be used only during first unmasking of asynchronous aborts
600 * on the boot CPU. This makes sure that the machine will not die if the
601 * firmware/bootloader left an imprecise abort pending for us to trip over.
603 static int __init early_abort_handler(unsigned long addr, unsigned int fsr,
604 struct pt_regs *regs)
606 pr_warn("Hit pending asynchronous external abort (FSR=0x%08x) during "
607 "first unmask, this is most likely caused by a "
608 "firmware/bootloader bug.\n", fsr);
613 void __init early_abt_enable(void)
615 fsr_info[FSR_FS_AEA].fn = early_abort_handler;
617 fsr_info[FSR_FS_AEA].fn = do_bad;
620 #ifndef CONFIG_ARM_LPAE
621 static int __init exceptions_init(void)
623 if (cpu_architecture() >= CPU_ARCH_ARMv6) {
624 hook_fault_code(4, do_translation_fault, SIGSEGV, SEGV_MAPERR,
625 "I-cache maintenance fault");
628 if (cpu_architecture() >= CPU_ARCH_ARMv7) {
630 * TODO: Access flag faults introduced in ARMv6K.
631 * Runtime check for 'K' extension is needed
633 hook_fault_code(3, do_bad, SIGSEGV, SEGV_MAPERR,
634 "section access flag fault");
635 hook_fault_code(6, do_bad, SIGSEGV, SEGV_MAPERR,
636 "section access flag fault");
642 arch_initcall(exceptions_init);