1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/arch/arm/mm/fault.c
4 *
5 * Copyright (C) 1995 Linus Torvalds
6 * Modifications for ARM processor (c) 1995-2004 Russell King
7 */
8#include <linux/extable.h>
9#include <linux/signal.h>
10#include <linux/mm.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>
21
22#include <asm/system_misc.h>
23#include <asm/system_info.h>
24#include <asm/tlbflush.h>
25
26#include "fault.h"
27
28bool copy_from_kernel_nofault_allowed(const void *unsafe_src, size_t size)
29{
30 unsigned long addr = (unsigned long)unsafe_src;
31
32 return addr >= TASK_SIZE && ULONG_MAX - addr >= size;
33}
34
35#ifdef CONFIG_MMU
36
37/*
38 * This is useful to dump out the page tables associated with
39 * 'addr' in mm 'mm'.
40 */
41void show_pte(const char *lvl, struct mm_struct *mm, unsigned long addr)
42{
43 pgd_t *pgd;
44
45 if (!mm)
46 mm = &init_mm;
47
48 pgd = pgd_offset(mm, addr);
49 printk("%s[%08lx] *pgd=%08llx", lvl, addr, (long long)pgd_val(*pgd));
50
51 do {
52 p4d_t *p4d;
53 pud_t *pud;
54 pmd_t *pmd;
55 pte_t *pte;
56
57 p4d = p4d_offset(pgd, address: addr);
58 if (p4d_none(p4d: *p4d))
59 break;
60
61 if (p4d_bad(p4d: *p4d)) {
62 pr_cont("(bad)");
63 break;
64 }
65
66 pud = pud_offset(p4d, address: addr);
67 if (PTRS_PER_PUD != 1)
68 pr_cont(", *pud=%08llx", (long long)pud_val(*pud));
69
70 if (pud_none(pud: *pud))
71 break;
72
73 if (pud_bad(pud: *pud)) {
74 pr_cont("(bad)");
75 break;
76 }
77
78 pmd = pmd_offset(pud, address: addr);
79 if (PTRS_PER_PMD != 1)
80 pr_cont(", *pmd=%08llx", (long long)pmd_val(*pmd));
81
82 if (pmd_none(pmd: *pmd))
83 break;
84
85 if (pmd_bad(pmd: *pmd)) {
86 pr_cont("(bad)");
87 break;
88 }
89
90 /* We must not map this if we have highmem enabled */
91 if (PageHighMem(pfn_to_page(pmd_val(*pmd) >> PAGE_SHIFT)))
92 break;
93
94 pte = pte_offset_map(pmd, addr);
95 if (!pte)
96 break;
97
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]));
102#endif
103 pte_unmap(pte);
104 } while(0);
105
106 pr_cont("\n");
107}
108#else /* CONFIG_MMU */
109void show_pte(const char *lvl, struct mm_struct *mm, unsigned long addr)
110{ }
111#endif /* CONFIG_MMU */
112
113static inline bool is_write_fault(unsigned int fsr)
114{
115 return (fsr & FSR_WRITE) && !(fsr & FSR_CM);
116}
117
118static inline bool is_translation_fault(unsigned int fsr)
119{
120 int fs = fsr_fs(fsr);
121#ifdef CONFIG_ARM_LPAE
122 if ((fs & FS_MMU_NOLL_MASK) == FS_TRANS_NOLL)
123 return true;
124#else
125 if (fs == FS_L1_TRANS || fs == FS_L2_TRANS)
126 return true;
127#endif
128 return false;
129}
130
131static void die_kernel_fault(const char *msg, struct mm_struct *mm,
132 unsigned long addr, unsigned int fsr,
133 struct pt_regs *regs)
134{
135 bust_spinlocks(yes: 1);
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");
140
141 show_pte(KERN_ALERT, mm, addr);
142 die("Oops", regs, fsr);
143 bust_spinlocks(yes: 0);
144 make_task_dead(SIGKILL);
145}
146
147/*
148 * Oops. The kernel tried to access some page that wasn't present.
149 */
150static void
151__do_kernel_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
152 struct pt_regs *regs)
153{
154 const char *msg;
155 /*
156 * Are we prepared to handle this kernel fault?
157 */
158 if (fixup_exception(regs))
159 return;
160
161 /*
162 * No handler, we'll have to terminate things with extreme prejudice.
163 */
164 if (addr < PAGE_SIZE) {
165 msg = "NULL pointer dereference";
166 } else {
167 if (is_translation_fault(fsr) &&
168 kfence_handle_page_fault(addr, is_write: is_write_fault(fsr), regs))
169 return;
170
171 msg = "paging request";
172 }
173
174 die_kernel_fault(msg, mm, addr, fsr, regs);
175}
176
177/*
178 * Something tried to access memory that isn't in our memory map..
179 * User mode accesses just cause a SIGSEGV
180 */
181static void
182__do_user_fault(unsigned long addr, unsigned int fsr, unsigned int sig,
183 int code, struct pt_regs *regs)
184{
185 struct task_struct *tsk = current;
186
187 if (addr > TASK_SIZE)
188 harden_branch_predictor();
189
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);
197 show_regs(regs);
198 }
199#endif
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",
204 tsk->comm, addr);
205#endif
206
207 tsk->thread.address = addr;
208 tsk->thread.error_code = fsr;
209 tsk->thread.trap_no = 14;
210 force_sig_fault(sig, code, addr: (void __user *)addr);
211}
212
213void do_bad_area(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
214{
215 struct task_struct *tsk = current;
216 struct mm_struct *mm = tsk->active_mm;
217
218 /*
219 * If we are in kernel mode at this point, we
220 * have no context to handle this fault with.
221 */
222 if (user_mode(regs))
223 __do_user_fault(addr, fsr, SIGSEGV, SEGV_MAPERR, regs);
224 else
225 __do_kernel_fault(mm, addr, fsr, regs);
226}
227
228#ifdef CONFIG_MMU
229#define VM_FAULT_BADMAP ((__force vm_fault_t)0x010000)
230#define VM_FAULT_BADACCESS ((__force vm_fault_t)0x020000)
231
232static inline bool is_permission_fault(unsigned int fsr)
233{
234 int fs = fsr_fs(fsr);
235#ifdef CONFIG_ARM_LPAE
236 if ((fs & FS_MMU_NOLL_MASK) == FS_PERM_NOLL)
237 return true;
238#else
239 if (fs == FS_L1_PERM || fs == FS_L2_PERM)
240 return true;
241#endif
242 return false;
243}
244
245static int __kprobes
246do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
247{
248 struct mm_struct *mm = current->mm;
249 struct vm_area_struct *vma;
250 int sig, code;
251 vm_fault_t fault;
252 unsigned int flags = FAULT_FLAG_DEFAULT;
253 unsigned long vm_flags = VM_ACCESS_FLAGS;
254
255 if (kprobe_page_fault(regs, trap: fsr))
256 return 0;
257
258
259 /* Enable interrupts if they were enabled in the parent context. */
260 if (interrupts_enabled(regs))
261 local_irq_enable();
262
263 /*
264 * If we're in an interrupt or have no user
265 * context, we must not take the fault..
266 */
267 if (faulthandler_disabled() || !mm)
268 goto no_context;
269
270 if (user_mode(regs))
271 flags |= FAULT_FLAG_USER;
272
273 if (is_write_fault(fsr)) {
274 flags |= FAULT_FLAG_WRITE;
275 vm_flags = VM_WRITE;
276 }
277
278 if (fsr & FSR_LNX_PF) {
279 vm_flags = VM_EXEC;
280
281 if (is_permission_fault(fsr) && !user_mode(regs))
282 die_kernel_fault(msg: "execution of memory",
283 mm, addr, fsr, regs);
284 }
285
286 perf_sw_event(event_id: PERF_COUNT_SW_PAGE_FAULTS, nr: 1, regs, addr);
287
288 if (!(flags & FAULT_FLAG_USER))
289 goto lock_mmap;
290
291 vma = lock_vma_under_rcu(mm, address: addr);
292 if (!vma)
293 goto lock_mmap;
294
295 if (!(vma->vm_flags & vm_flags)) {
296 vma_end_read(vma);
297 goto lock_mmap;
298 }
299 fault = handle_mm_fault(vma, address: addr, flags: flags | FAULT_FLAG_VMA_LOCK, regs);
300 if (!(fault & (VM_FAULT_RETRY | VM_FAULT_COMPLETED)))
301 vma_end_read(vma);
302
303 if (!(fault & VM_FAULT_RETRY)) {
304 count_vm_vma_lock_event(VMA_LOCK_SUCCESS);
305 goto done;
306 }
307 count_vm_vma_lock_event(VMA_LOCK_RETRY);
308 if (fault & VM_FAULT_MAJOR)
309 flags |= FAULT_FLAG_TRIED;
310
311 /* Quick path to respond to signals */
312 if (fault_signal_pending(fault_flags: fault, regs)) {
313 if (!user_mode(regs))
314 goto no_context;
315 return 0;
316 }
317lock_mmap:
318
319retry:
320 vma = lock_mm_and_find_vma(mm, address: addr, regs);
321 if (unlikely(!vma)) {
322 fault = VM_FAULT_BADMAP;
323 goto bad_area;
324 }
325
326 /*
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.
329 */
330 if (!(vma->vm_flags & vm_flags))
331 fault = VM_FAULT_BADACCESS;
332 else
333 fault = handle_mm_fault(vma, address: addr & PAGE_MASK, flags, regs);
334
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
338 * mm/filemap.c. */
339 if (fault_signal_pending(fault_flags: fault, regs)) {
340 if (!user_mode(regs))
341 goto no_context;
342 return 0;
343 }
344
345 /* The fault is fully completed (including releasing mmap lock) */
346 if (fault & VM_FAULT_COMPLETED)
347 return 0;
348
349 if (!(fault & VM_FAULT_ERROR)) {
350 if (fault & VM_FAULT_RETRY) {
351 flags |= FAULT_FLAG_TRIED;
352 goto retry;
353 }
354 }
355
356 mmap_read_unlock(mm);
357done:
358
359 /*
360 * Handle the "normal" case first - VM_FAULT_MAJOR
361 */
362 if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP | VM_FAULT_BADACCESS))))
363 return 0;
364
365bad_area:
366 /*
367 * If we are in kernel mode at this point, we
368 * have no context to handle this fault with.
369 */
370 if (!user_mode(regs))
371 goto no_context;
372
373 if (fault & VM_FAULT_OOM) {
374 /*
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
377 * got oom-killed)
378 */
379 pagefault_out_of_memory();
380 return 0;
381 }
382
383 if (fault & VM_FAULT_SIGBUS) {
384 /*
385 * We had some memory, but were unable to
386 * successfully fix up this page fault.
387 */
388 sig = SIGBUS;
389 code = BUS_ADRERR;
390 } else {
391 /*
392 * Something tried to access memory that
393 * isn't in our memory map..
394 */
395 sig = SIGSEGV;
396 code = fault == VM_FAULT_BADACCESS ?
397 SEGV_ACCERR : SEGV_MAPERR;
398 }
399
400 __do_user_fault(addr, fsr, sig, code, regs);
401 return 0;
402
403no_context:
404 __do_kernel_fault(mm, addr, fsr, regs);
405 return 0;
406}
407#else /* CONFIG_MMU */
408static int
409do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
410{
411 return 0;
412}
413#endif /* CONFIG_MMU */
414
415/*
416 * First Level Translation Fault Handler
417 *
418 * We enter here because the first level page table doesn't contain
419 * a valid entry for the address.
420 *
421 * If the address is in kernel space (>= TASK_SIZE), then we are
422 * probably faulting in the vmalloc() area.
423 *
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.
427 *
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.
431 */
432#ifdef CONFIG_MMU
433static int __kprobes
434do_translation_fault(unsigned long addr, unsigned int fsr,
435 struct pt_regs *regs)
436{
437 unsigned int index;
438 pgd_t *pgd, *pgd_k;
439 p4d_t *p4d, *p4d_k;
440 pud_t *pud, *pud_k;
441 pmd_t *pmd, *pmd_k;
442
443 if (addr < TASK_SIZE)
444 return do_page_fault(addr, fsr, regs);
445
446 if (user_mode(regs))
447 goto bad_area;
448
449 index = pgd_index(addr);
450
451 pgd = cpu_get_pgd() + index;
452 pgd_k = init_mm.pgd + index;
453
454 p4d = p4d_offset(pgd, address: addr);
455 p4d_k = p4d_offset(pgd: pgd_k, address: addr);
456
457 if (p4d_none(p4d: *p4d_k))
458 goto bad_area;
459 if (!p4d_present(p4d: *p4d))
460 set_p4d(p4dp: p4d, p4d: *p4d_k);
461
462 pud = pud_offset(p4d, address: addr);
463 pud_k = pud_offset(p4d: p4d_k, address: addr);
464
465 if (pud_none(pud: *pud_k))
466 goto bad_area;
467 if (!pud_present(pud: *pud))
468 set_pud(pudp: pud, pud: *pud_k);
469
470 pmd = pmd_offset(pud, address: addr);
471 pmd_k = pmd_offset(pud: pud_k, address: addr);
472
473#ifdef CONFIG_ARM_LPAE
474 /*
475 * Only one hardware entry per PMD with LPAE.
476 */
477 index = 0;
478#else
479 /*
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.
486 */
487 index = (addr >> SECTION_SHIFT) & 1;
488#endif
489 if (pmd_none(pmd: pmd_k[index]))
490 goto bad_area;
491
492 copy_pmd(pmd, pmd_k);
493 return 0;
494
495bad_area:
496 do_bad_area(addr, fsr, regs);
497 return 0;
498}
499#else /* CONFIG_MMU */
500static int
501do_translation_fault(unsigned long addr, unsigned int fsr,
502 struct pt_regs *regs)
503{
504 return 0;
505}
506#endif /* CONFIG_MMU */
507
508/*
509 * Some section permission faults need to be handled gracefully.
510 * They can happen due to a __{get,put}_user during an oops.
511 */
512#ifndef CONFIG_ARM_LPAE
513static int
514do_sect_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
515{
516 do_bad_area(addr, fsr, regs);
517 return 0;
518}
519#endif /* CONFIG_ARM_LPAE */
520
521/*
522 * This abort handler always returns "fault".
523 */
524static int
525do_bad(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
526{
527 return 1;
528}
529
530struct fsr_info {
531 int (*fn)(unsigned long addr, unsigned int fsr, struct pt_regs *regs);
532 int sig;
533 int code;
534 const char *name;
535};
536
537/* FSR definition */
538#ifdef CONFIG_ARM_LPAE
539#include "fsr-3level.c"
540#else
541#include "fsr-2level.c"
542#endif
543
544void __init
545hook_fault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
546 int sig, int code, const char *name)
547{
548 if (nr < 0 || nr >= ARRAY_SIZE(fsr_info))
549 BUG();
550
551 fsr_info[nr].fn = fn;
552 fsr_info[nr].sig = sig;
553 fsr_info[nr].code = code;
554 fsr_info[nr].name = name;
555}
556
557/*
558 * Dispatch a data abort to the relevant handler.
559 */
560asmlinkage void
561do_DataAbort(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
562{
563 const struct fsr_info *inf = fsr_info + fsr_fs(fsr);
564
565 if (!inf->fn(addr, fsr & ~FSR_LNX_PF, regs))
566 return;
567
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);
572
573 arm_notify_die("", regs, inf->sig, inf->code, (void __user *)addr,
574 fsr, 0);
575}
576
577void __init
578hook_ifault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
579 int sig, int code, const char *name)
580{
581 if (nr < 0 || nr >= ARRAY_SIZE(ifsr_info))
582 BUG();
583
584 ifsr_info[nr].fn = fn;
585 ifsr_info[nr].sig = sig;
586 ifsr_info[nr].code = code;
587 ifsr_info[nr].name = name;
588}
589
590asmlinkage void
591do_PrefetchAbort(unsigned long addr, unsigned int ifsr, struct pt_regs *regs)
592{
593 const struct fsr_info *inf = ifsr_info + fsr_fs(fsr: ifsr);
594
595 if (!inf->fn(addr, ifsr | FSR_LNX_PF, regs))
596 return;
597
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);
601
602 arm_notify_die("", regs, inf->sig, inf->code, (void __user *)addr,
603 ifsr, 0);
604}
605
606/*
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.
610 */
611static int __init early_abort_handler(unsigned long addr, unsigned int fsr,
612 struct pt_regs *regs)
613{
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);
617
618 return 0;
619}
620
621void __init early_abt_enable(void)
622{
623 fsr_info[FSR_FS_AEA].fn = early_abort_handler;
624 local_abt_enable();
625 fsr_info[FSR_FS_AEA].fn = do_bad;
626}
627
628#ifndef CONFIG_ARM_LPAE
629static int __init exceptions_init(void)
630{
631 if (cpu_architecture() >= CPU_ARCH_ARMv6) {
632 hook_fault_code(nr: 4, fn: do_translation_fault, SIGSEGV, SEGV_MAPERR,
633 name: "I-cache maintenance fault");
634 }
635
636 if (cpu_architecture() >= CPU_ARCH_ARMv7) {
637 /*
638 * TODO: Access flag faults introduced in ARMv6K.
639 * Runtime check for 'K' extension is needed
640 */
641 hook_fault_code(nr: 3, fn: do_bad, SIGSEGV, SEGV_MAPERR,
642 name: "section access flag fault");
643 hook_fault_code(nr: 6, fn: do_bad, SIGSEGV, SEGV_MAPERR,
644 name: "section access flag fault");
645 }
646
647 return 0;
648}
649
650arch_initcall(exceptions_init);
651#endif
652

source code of linux/arch/arm/mm/fault.c