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));
51 p4d = p4d_offset(pgd, addr);
60 pud = pud_offset(p4d, addr);
61 if (PTRS_PER_PUD != 1)
62 pr_cont(", *pud=%08llx", (long long)pud_val(*pud));
72 pmd = pmd_offset(pud, addr);
73 if (PTRS_PER_PMD != 1)
74 pr_cont(", *pmd=%08llx", (long long)pmd_val(*pmd));
84 /* We must not map this if we have highmem enabled */
85 if (PageHighMem(pfn_to_page(pmd_val(*pmd) >> PAGE_SHIFT)))
88 pte = pte_offset_map(pmd, addr);
89 pr_cont(", *pte=%08llx", (long long)pte_val(*pte));
90 #ifndef CONFIG_ARM_LPAE
91 pr_cont(", *ppte=%08llx",
92 (long long)pte_val(pte[PTE_HWTABLE_PTRS]));
99 #else /* CONFIG_MMU */
100 void show_pte(const char *lvl, struct mm_struct *mm, unsigned long addr)
102 #endif /* CONFIG_MMU */
105 * Oops. The kernel tried to access some page that wasn't present.
108 __do_kernel_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
109 struct pt_regs *regs)
112 * Are we prepared to handle this kernel fault?
114 if (fixup_exception(regs))
118 * No handler, we'll have to terminate things with extreme prejudice.
121 pr_alert("8<--- cut here ---\n");
122 pr_alert("Unable to handle kernel %s at virtual address %08lx\n",
123 (addr < PAGE_SIZE) ? "NULL pointer dereference" :
124 "paging request", addr);
126 show_pte(KERN_ALERT, mm, addr);
127 die("Oops", regs, fsr);
133 * Something tried to access memory that isn't in our memory map..
134 * User mode accesses just cause a SIGSEGV
137 __do_user_fault(unsigned long addr, unsigned int fsr, unsigned int sig,
138 int code, struct pt_regs *regs)
140 struct task_struct *tsk = current;
142 if (addr > TASK_SIZE)
143 harden_branch_predictor();
145 #ifdef CONFIG_DEBUG_USER
146 if (((user_debug & UDBG_SEGV) && (sig == SIGSEGV)) ||
147 ((user_debug & UDBG_BUS) && (sig == SIGBUS))) {
148 pr_err("8<--- cut here ---\n");
149 pr_err("%s: unhandled page fault (%d) at 0x%08lx, code 0x%03x\n",
150 tsk->comm, sig, addr, fsr);
151 show_pte(KERN_ERR, tsk->mm, addr);
155 #ifndef CONFIG_KUSER_HELPERS
156 if ((sig == SIGSEGV) && ((addr & PAGE_MASK) == 0xffff0000))
157 printk_ratelimited(KERN_DEBUG
158 "%s: CONFIG_KUSER_HELPERS disabled at 0x%08lx\n",
162 tsk->thread.address = addr;
163 tsk->thread.error_code = fsr;
164 tsk->thread.trap_no = 14;
165 force_sig_fault(sig, code, (void __user *)addr);
168 void do_bad_area(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
170 struct task_struct *tsk = current;
171 struct mm_struct *mm = tsk->active_mm;
174 * If we are in kernel mode at this point, we
175 * have no context to handle this fault with.
178 __do_user_fault(addr, fsr, SIGSEGV, SEGV_MAPERR, regs);
180 __do_kernel_fault(mm, addr, fsr, regs);
184 #define VM_FAULT_BADMAP 0x010000
185 #define VM_FAULT_BADACCESS 0x020000
188 * Check that the permissions on the VMA allow for the fault which occurred.
189 * If we encountered a write fault, we must have write permission, otherwise
190 * we allow any permission.
192 static inline bool access_error(unsigned int fsr, struct vm_area_struct *vma)
194 unsigned int mask = VM_ACCESS_FLAGS;
196 if ((fsr & FSR_WRITE) && !(fsr & FSR_CM))
198 if (fsr & FSR_LNX_PF)
201 return vma->vm_flags & mask ? false : true;
204 static vm_fault_t __kprobes
205 __do_page_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
206 unsigned int flags, struct task_struct *tsk)
208 struct vm_area_struct *vma;
211 vma = find_vma(mm, addr);
212 fault = VM_FAULT_BADMAP;
215 if (unlikely(vma->vm_start > addr))
219 * Ok, we have a good vm_area for this
220 * memory access, so we can handle it.
223 if (access_error(fsr, vma)) {
224 fault = VM_FAULT_BADACCESS;
228 return handle_mm_fault(vma, addr & PAGE_MASK, flags);
231 /* Don't allow expansion below FIRST_USER_ADDRESS */
232 if (vma->vm_flags & VM_GROWSDOWN &&
233 addr >= FIRST_USER_ADDRESS && !expand_stack(vma, addr))
240 do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
242 struct task_struct *tsk;
243 struct mm_struct *mm;
246 unsigned int flags = FAULT_FLAG_DEFAULT;
248 if (kprobe_page_fault(regs, fsr))
254 /* Enable interrupts if they were enabled in the parent context. */
255 if (interrupts_enabled(regs))
259 * If we're in an interrupt or have no user
260 * context, we must not take the fault..
262 if (faulthandler_disabled() || !mm)
266 flags |= FAULT_FLAG_USER;
267 if ((fsr & FSR_WRITE) && !(fsr & FSR_CM))
268 flags |= FAULT_FLAG_WRITE;
271 * As per x86, we may deadlock here. However, since the kernel only
272 * validly references user space from well defined areas of the code,
273 * we can bug out early if this is from code which shouldn't.
275 if (!down_read_trylock(&mm->mmap_sem)) {
276 if (!user_mode(regs) && !search_exception_tables(regs->ARM_pc))
279 down_read(&mm->mmap_sem);
282 * The above down_read_trylock() might have succeeded in
283 * which case, we'll have missed the might_sleep() from
287 #ifdef CONFIG_DEBUG_VM
288 if (!user_mode(regs) &&
289 !search_exception_tables(regs->ARM_pc))
294 fault = __do_page_fault(mm, addr, fsr, flags, tsk);
296 /* If we need to retry but a fatal signal is pending, handle the
297 * signal first. We do not need to release the mmap_sem because
298 * it would already be released in __lock_page_or_retry in
300 if (fault_signal_pending(fault, regs)) {
301 if (!user_mode(regs))
307 * Major/minor page fault accounting is only done on the
308 * initial attempt. If we go through a retry, it is extremely
309 * likely that the page will be found in page cache at that point.
312 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
313 if (!(fault & VM_FAULT_ERROR) && flags & FAULT_FLAG_ALLOW_RETRY) {
314 if (fault & VM_FAULT_MAJOR) {
316 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
320 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
323 if (fault & VM_FAULT_RETRY) {
324 flags |= FAULT_FLAG_TRIED;
329 up_read(&mm->mmap_sem);
332 * Handle the "normal" case first - VM_FAULT_MAJOR
334 if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP | VM_FAULT_BADACCESS))))
338 * If we are in kernel mode at this point, we
339 * have no context to handle this fault with.
341 if (!user_mode(regs))
344 if (fault & VM_FAULT_OOM) {
346 * We ran out of memory, call the OOM killer, and return to
347 * userspace (which will retry the fault, or kill us if we
350 pagefault_out_of_memory();
354 if (fault & VM_FAULT_SIGBUS) {
356 * We had some memory, but were unable to
357 * successfully fix up this page fault.
363 * Something tried to access memory that
364 * isn't in our memory map..
367 code = fault == VM_FAULT_BADACCESS ?
368 SEGV_ACCERR : SEGV_MAPERR;
371 __do_user_fault(addr, fsr, sig, code, regs);
375 __do_kernel_fault(mm, addr, fsr, regs);
378 #else /* CONFIG_MMU */
380 do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
384 #endif /* CONFIG_MMU */
387 * First Level Translation Fault Handler
389 * We enter here because the first level page table doesn't contain
390 * a valid entry for the address.
392 * If the address is in kernel space (>= TASK_SIZE), then we are
393 * probably faulting in the vmalloc() area.
395 * If the init_task's first level page tables contains the relevant
396 * entry, we copy the it to this task. If not, we send the process
397 * a signal, fixup the exception, or oops the kernel.
399 * NOTE! We MUST NOT take any locks for this case. We may be in an
400 * interrupt or a critical region, and should only copy the information
401 * from the master page table, nothing more.
405 do_translation_fault(unsigned long addr, unsigned int fsr,
406 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 p4d = p4d_offset(pgd, addr);
426 p4d_k = p4d_offset(pgd_k, addr);
428 if (p4d_none(*p4d_k))
430 if (!p4d_present(*p4d))
431 set_p4d(p4d, *p4d_k);
433 pud = pud_offset(p4d, addr);
434 pud_k = pud_offset(p4d_k, addr);
436 if (pud_none(*pud_k))
438 if (!pud_present(*pud))
439 set_pud(pud, *pud_k);
441 pmd = pmd_offset(pud, addr);
442 pmd_k = pmd_offset(pud_k, addr);
444 #ifdef CONFIG_ARM_LPAE
446 * Only one hardware entry per PMD with LPAE.
451 * On ARM one Linux PGD entry contains two hardware entries (see page
452 * tables layout in pgtable.h). We normally guarantee that we always
453 * fill both L1 entries. But create_mapping() doesn't follow the rule.
454 * It can create inidividual L1 entries, so here we have to call
455 * pmd_none() check for the entry really corresponded to address, not
456 * for the first of pair.
458 index = (addr >> SECTION_SHIFT) & 1;
460 if (pmd_none(pmd_k[index]))
463 copy_pmd(pmd, pmd_k);
467 do_bad_area(addr, fsr, regs);
470 #else /* CONFIG_MMU */
472 do_translation_fault(unsigned long addr, unsigned int fsr,
473 struct pt_regs *regs)
477 #endif /* CONFIG_MMU */
480 * Some section permission faults need to be handled gracefully.
481 * They can happen due to a __{get,put}_user during an oops.
483 #ifndef CONFIG_ARM_LPAE
485 do_sect_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
487 do_bad_area(addr, fsr, regs);
490 #endif /* CONFIG_ARM_LPAE */
493 * This abort handler always returns "fault".
496 do_bad(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
502 int (*fn)(unsigned long addr, unsigned int fsr, struct pt_regs *regs);
509 #ifdef CONFIG_ARM_LPAE
510 #include "fsr-3level.c"
512 #include "fsr-2level.c"
516 hook_fault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
517 int sig, int code, const char *name)
519 if (nr < 0 || nr >= ARRAY_SIZE(fsr_info))
522 fsr_info[nr].fn = fn;
523 fsr_info[nr].sig = sig;
524 fsr_info[nr].code = code;
525 fsr_info[nr].name = name;
529 * Dispatch a data abort to the relevant handler.
532 do_DataAbort(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
534 const struct fsr_info *inf = fsr_info + fsr_fs(fsr);
536 if (!inf->fn(addr, fsr & ~FSR_LNX_PF, regs))
539 pr_alert("8<--- cut here ---\n");
540 pr_alert("Unhandled fault: %s (0x%03x) at 0x%08lx\n",
541 inf->name, fsr, addr);
542 show_pte(KERN_ALERT, current->mm, addr);
544 arm_notify_die("", regs, inf->sig, inf->code, (void __user *)addr,
549 hook_ifault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
550 int sig, int code, const char *name)
552 if (nr < 0 || nr >= ARRAY_SIZE(ifsr_info))
555 ifsr_info[nr].fn = fn;
556 ifsr_info[nr].sig = sig;
557 ifsr_info[nr].code = code;
558 ifsr_info[nr].name = name;
562 do_PrefetchAbort(unsigned long addr, unsigned int ifsr, struct pt_regs *regs)
564 const struct fsr_info *inf = ifsr_info + fsr_fs(ifsr);
566 if (!inf->fn(addr, ifsr | FSR_LNX_PF, regs))
569 pr_alert("Unhandled prefetch abort: %s (0x%03x) at 0x%08lx\n",
570 inf->name, ifsr, addr);
572 arm_notify_die("", regs, inf->sig, inf->code, (void __user *)addr,
577 * Abort handler to be used only during first unmasking of asynchronous aborts
578 * on the boot CPU. This makes sure that the machine will not die if the
579 * firmware/bootloader left an imprecise abort pending for us to trip over.
581 static int __init early_abort_handler(unsigned long addr, unsigned int fsr,
582 struct pt_regs *regs)
584 pr_warn("Hit pending asynchronous external abort (FSR=0x%08x) during "
585 "first unmask, this is most likely caused by a "
586 "firmware/bootloader bug.\n", fsr);
591 void __init early_abt_enable(void)
593 fsr_info[FSR_FS_AEA].fn = early_abort_handler;
595 fsr_info[FSR_FS_AEA].fn = do_bad;
598 #ifndef CONFIG_ARM_LPAE
599 static int __init exceptions_init(void)
601 if (cpu_architecture() >= CPU_ARCH_ARMv6) {
602 hook_fault_code(4, do_translation_fault, SIGSEGV, SEGV_MAPERR,
603 "I-cache maintenance fault");
606 if (cpu_architecture() >= CPU_ARCH_ARMv7) {
608 * TODO: Access flag faults introduced in ARMv6K.
609 * Runtime check for 'K' extension is needed
611 hook_fault_code(3, do_bad, SIGSEGV, SEGV_MAPERR,
612 "section access flag fault");
613 hook_fault_code(6, do_bad, SIGSEGV, SEGV_MAPERR,
614 "section access flag fault");
620 arch_initcall(exceptions_init);