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);
91 pr_cont(", *pte=%08llx", (long long)pte_val(*pte));
92 #ifndef CONFIG_ARM_LPAE
93 pr_cont(", *ppte=%08llx",
94 (long long)pte_val(pte[PTE_HWTABLE_PTRS]));
101 #else /* CONFIG_MMU */
102 void show_pte(const char *lvl, struct mm_struct *mm, unsigned long addr)
104 #endif /* CONFIG_MMU */
106 static inline bool is_write_fault(unsigned int fsr)
108 return (fsr & FSR_WRITE) && !(fsr & FSR_CM);
111 static inline bool is_translation_fault(unsigned int fsr)
113 int fs = fsr_fs(fsr);
114 #ifdef CONFIG_ARM_LPAE
115 if ((fs & FS_MMU_NOLL_MASK) == FS_TRANS_NOLL)
118 if (fs == FS_L1_TRANS || fs == FS_L2_TRANS)
124 static void die_kernel_fault(const char *msg, struct mm_struct *mm,
125 unsigned long addr, unsigned int fsr,
126 struct pt_regs *regs)
129 pr_alert("8<--- cut here ---\n");
130 pr_alert("Unable to handle kernel %s at virtual address %08lx when %s\n",
131 msg, addr, fsr & FSR_LNX_PF ? "execute" :
132 fsr & FSR_WRITE ? "write" : "read");
134 show_pte(KERN_ALERT, mm, addr);
135 die("Oops", regs, fsr);
137 make_task_dead(SIGKILL);
141 * Oops. The kernel tried to access some page that wasn't present.
144 __do_kernel_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
145 struct pt_regs *regs)
149 * Are we prepared to handle this kernel fault?
151 if (fixup_exception(regs))
155 * No handler, we'll have to terminate things with extreme prejudice.
157 if (addr < PAGE_SIZE) {
158 msg = "NULL pointer dereference";
160 if (is_translation_fault(fsr) &&
161 kfence_handle_page_fault(addr, is_write_fault(fsr), regs))
164 msg = "paging request";
167 die_kernel_fault(msg, mm, addr, fsr, regs);
171 * Something tried to access memory that isn't in our memory map..
172 * User mode accesses just cause a SIGSEGV
175 __do_user_fault(unsigned long addr, unsigned int fsr, unsigned int sig,
176 int code, struct pt_regs *regs)
178 struct task_struct *tsk = current;
180 if (addr > TASK_SIZE)
181 harden_branch_predictor();
183 #ifdef CONFIG_DEBUG_USER
184 if (((user_debug & UDBG_SEGV) && (sig == SIGSEGV)) ||
185 ((user_debug & UDBG_BUS) && (sig == SIGBUS))) {
186 pr_err("8<--- cut here ---\n");
187 pr_err("%s: unhandled page fault (%d) at 0x%08lx, code 0x%03x\n",
188 tsk->comm, sig, addr, fsr);
189 show_pte(KERN_ERR, tsk->mm, addr);
193 #ifndef CONFIG_KUSER_HELPERS
194 if ((sig == SIGSEGV) && ((addr & PAGE_MASK) == 0xffff0000))
195 printk_ratelimited(KERN_DEBUG
196 "%s: CONFIG_KUSER_HELPERS disabled at 0x%08lx\n",
200 tsk->thread.address = addr;
201 tsk->thread.error_code = fsr;
202 tsk->thread.trap_no = 14;
203 force_sig_fault(sig, code, (void __user *)addr);
206 void do_bad_area(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
208 struct task_struct *tsk = current;
209 struct mm_struct *mm = tsk->active_mm;
212 * If we are in kernel mode at this point, we
213 * have no context to handle this fault with.
216 __do_user_fault(addr, fsr, SIGSEGV, SEGV_MAPERR, regs);
218 __do_kernel_fault(mm, addr, fsr, regs);
222 #define VM_FAULT_BADMAP ((__force vm_fault_t)0x010000)
223 #define VM_FAULT_BADACCESS ((__force vm_fault_t)0x020000)
225 static inline bool is_permission_fault(unsigned int fsr)
227 int fs = fsr_fs(fsr);
228 #ifdef CONFIG_ARM_LPAE
229 if ((fs & FS_MMU_NOLL_MASK) == FS_PERM_NOLL)
232 if (fs == FS_L1_PERM || fs == FS_L2_PERM)
239 do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
241 struct mm_struct *mm = current->mm;
242 struct vm_area_struct *vma;
245 unsigned int flags = FAULT_FLAG_DEFAULT;
246 unsigned long vm_flags = VM_ACCESS_FLAGS;
248 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;
266 if (is_write_fault(fsr)) {
267 flags |= FAULT_FLAG_WRITE;
271 if (fsr & FSR_LNX_PF) {
274 if (is_permission_fault(fsr) && !user_mode(regs))
275 die_kernel_fault("execution of memory",
276 mm, addr, fsr, regs);
279 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
282 vma = lock_mm_and_find_vma(mm, addr, regs);
283 if (unlikely(!vma)) {
284 fault = VM_FAULT_BADMAP;
289 * ok, we have a good vm_area for this memory access, check the
290 * permissions on the VMA allow for the fault which occurred.
292 if (!(vma->vm_flags & vm_flags))
293 fault = VM_FAULT_BADACCESS;
295 fault = handle_mm_fault(vma, addr & PAGE_MASK, flags, regs);
297 /* If we need to retry but a fatal signal is pending, handle the
298 * signal first. We do not need to release the mmap_lock because
299 * it would already be released in __lock_page_or_retry in
301 if (fault_signal_pending(fault, regs)) {
302 if (!user_mode(regs))
307 /* The fault is fully completed (including releasing mmap lock) */
308 if (fault & VM_FAULT_COMPLETED)
311 if (!(fault & VM_FAULT_ERROR)) {
312 if (fault & VM_FAULT_RETRY) {
313 flags |= FAULT_FLAG_TRIED;
318 mmap_read_unlock(mm);
321 * Handle the "normal" case first - VM_FAULT_MAJOR
323 if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP | VM_FAULT_BADACCESS))))
328 * If we are in kernel mode at this point, we
329 * have no context to handle this fault with.
331 if (!user_mode(regs))
334 if (fault & VM_FAULT_OOM) {
336 * We ran out of memory, call the OOM killer, and return to
337 * userspace (which will retry the fault, or kill us if we
340 pagefault_out_of_memory();
344 if (fault & VM_FAULT_SIGBUS) {
346 * We had some memory, but were unable to
347 * successfully fix up this page fault.
353 * Something tried to access memory that
354 * isn't in our memory map..
357 code = fault == VM_FAULT_BADACCESS ?
358 SEGV_ACCERR : SEGV_MAPERR;
361 __do_user_fault(addr, fsr, sig, code, regs);
365 __do_kernel_fault(mm, addr, fsr, regs);
368 #else /* CONFIG_MMU */
370 do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
374 #endif /* CONFIG_MMU */
377 * First Level Translation Fault Handler
379 * We enter here because the first level page table doesn't contain
380 * a valid entry for the address.
382 * If the address is in kernel space (>= TASK_SIZE), then we are
383 * probably faulting in the vmalloc() area.
385 * If the init_task's first level page tables contains the relevant
386 * entry, we copy the it to this task. If not, we send the process
387 * a signal, fixup the exception, or oops the kernel.
389 * NOTE! We MUST NOT take any locks for this case. We may be in an
390 * interrupt or a critical region, and should only copy the information
391 * from the master page table, nothing more.
395 do_translation_fault(unsigned long addr, unsigned int fsr,
396 struct pt_regs *regs)
404 if (addr < TASK_SIZE)
405 return do_page_fault(addr, fsr, regs);
410 index = pgd_index(addr);
412 pgd = cpu_get_pgd() + index;
413 pgd_k = init_mm.pgd + index;
415 p4d = p4d_offset(pgd, addr);
416 p4d_k = p4d_offset(pgd_k, addr);
418 if (p4d_none(*p4d_k))
420 if (!p4d_present(*p4d))
421 set_p4d(p4d, *p4d_k);
423 pud = pud_offset(p4d, addr);
424 pud_k = pud_offset(p4d_k, addr);
426 if (pud_none(*pud_k))
428 if (!pud_present(*pud))
429 set_pud(pud, *pud_k);
431 pmd = pmd_offset(pud, addr);
432 pmd_k = pmd_offset(pud_k, addr);
434 #ifdef CONFIG_ARM_LPAE
436 * Only one hardware entry per PMD with LPAE.
441 * On ARM one Linux PGD entry contains two hardware entries (see page
442 * tables layout in pgtable.h). We normally guarantee that we always
443 * fill both L1 entries. But create_mapping() doesn't follow the rule.
444 * It can create inidividual L1 entries, so here we have to call
445 * pmd_none() check for the entry really corresponded to address, not
446 * for the first of pair.
448 index = (addr >> SECTION_SHIFT) & 1;
450 if (pmd_none(pmd_k[index]))
453 copy_pmd(pmd, pmd_k);
457 do_bad_area(addr, fsr, regs);
460 #else /* CONFIG_MMU */
462 do_translation_fault(unsigned long addr, unsigned int fsr,
463 struct pt_regs *regs)
467 #endif /* CONFIG_MMU */
470 * Some section permission faults need to be handled gracefully.
471 * They can happen due to a __{get,put}_user during an oops.
473 #ifndef CONFIG_ARM_LPAE
475 do_sect_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
477 do_bad_area(addr, fsr, regs);
480 #endif /* CONFIG_ARM_LPAE */
483 * This abort handler always returns "fault".
486 do_bad(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
492 int (*fn)(unsigned long addr, unsigned int fsr, struct pt_regs *regs);
499 #ifdef CONFIG_ARM_LPAE
500 #include "fsr-3level.c"
502 #include "fsr-2level.c"
506 hook_fault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
507 int sig, int code, const char *name)
509 if (nr < 0 || nr >= ARRAY_SIZE(fsr_info))
512 fsr_info[nr].fn = fn;
513 fsr_info[nr].sig = sig;
514 fsr_info[nr].code = code;
515 fsr_info[nr].name = name;
519 * Dispatch a data abort to the relevant handler.
522 do_DataAbort(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
524 const struct fsr_info *inf = fsr_info + fsr_fs(fsr);
526 if (!inf->fn(addr, fsr & ~FSR_LNX_PF, regs))
529 pr_alert("8<--- cut here ---\n");
530 pr_alert("Unhandled fault: %s (0x%03x) at 0x%08lx\n",
531 inf->name, fsr, addr);
532 show_pte(KERN_ALERT, current->mm, addr);
534 arm_notify_die("", regs, inf->sig, inf->code, (void __user *)addr,
539 hook_ifault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
540 int sig, int code, const char *name)
542 if (nr < 0 || nr >= ARRAY_SIZE(ifsr_info))
545 ifsr_info[nr].fn = fn;
546 ifsr_info[nr].sig = sig;
547 ifsr_info[nr].code = code;
548 ifsr_info[nr].name = name;
552 do_PrefetchAbort(unsigned long addr, unsigned int ifsr, struct pt_regs *regs)
554 const struct fsr_info *inf = ifsr_info + fsr_fs(ifsr);
556 if (!inf->fn(addr, ifsr | FSR_LNX_PF, regs))
559 pr_alert("Unhandled prefetch abort: %s (0x%03x) at 0x%08lx\n",
560 inf->name, ifsr, addr);
562 arm_notify_die("", regs, inf->sig, inf->code, (void __user *)addr,
567 * Abort handler to be used only during first unmasking of asynchronous aborts
568 * on the boot CPU. This makes sure that the machine will not die if the
569 * firmware/bootloader left an imprecise abort pending for us to trip over.
571 static int __init early_abort_handler(unsigned long addr, unsigned int fsr,
572 struct pt_regs *regs)
574 pr_warn("Hit pending asynchronous external abort (FSR=0x%08x) during "
575 "first unmask, this is most likely caused by a "
576 "firmware/bootloader bug.\n", fsr);
581 void __init early_abt_enable(void)
583 fsr_info[FSR_FS_AEA].fn = early_abort_handler;
585 fsr_info[FSR_FS_AEA].fn = do_bad;
588 #ifndef CONFIG_ARM_LPAE
589 static int __init exceptions_init(void)
591 if (cpu_architecture() >= CPU_ARCH_ARMv6) {
592 hook_fault_code(4, do_translation_fault, SIGSEGV, SEGV_MAPERR,
593 "I-cache maintenance fault");
596 if (cpu_architecture() >= CPU_ARCH_ARMv7) {
598 * TODO: Access flag faults introduced in ARMv6K.
599 * Runtime check for 'K' extension is needed
601 hook_fault_code(3, do_bad, SIGSEGV, SEGV_MAPERR,
602 "section access flag fault");
603 hook_fault_code(6, do_bad, SIGSEGV, SEGV_MAPERR,
604 "section access flag fault");
610 arch_initcall(exceptions_init);