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
28 #ifdef CONFIG_MMU
29
copy_from_kernel_nofault_allowed(const void * unsafe_src,size_t size)30 bool copy_from_kernel_nofault_allowed(const void *unsafe_src, size_t size)
31 {
32 unsigned long addr = (unsigned long)unsafe_src;
33
34 return addr >= TASK_SIZE && ULONG_MAX - addr >= size;
35 }
36
37 /*
38 * This is useful to dump out the page tables associated with
39 * 'addr' in mm 'mm'.
40 */
show_pte(const char * lvl,struct mm_struct * mm,unsigned long addr)41 void 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, addr);
58 if (p4d_none(*p4d))
59 break;
60
61 if (p4d_bad(*p4d)) {
62 pr_cont("(bad)");
63 break;
64 }
65
66 pud = pud_offset(p4d, addr);
67 if (PTRS_PER_PUD != 1)
68 pr_cont(", *pud=%08llx", (long long)pud_val(*pud));
69
70 if (pud_none(*pud))
71 break;
72
73 if (pud_bad(*pud)) {
74 pr_cont("(bad)");
75 break;
76 }
77
78 pmd = pmd_offset(pud, addr);
79 if (PTRS_PER_PMD != 1)
80 pr_cont(", *pmd=%08llx", (long long)pmd_val(*pmd));
81
82 if (pmd_none(*pmd))
83 break;
84
85 if (pmd_bad(*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 */
show_pte(const char * lvl,struct mm_struct * mm,unsigned long addr)109 void show_pte(const char *lvl, struct mm_struct *mm, unsigned long addr)
110 { }
111 #endif /* CONFIG_MMU */
112
is_write_fault(unsigned int fsr)113 static inline bool is_write_fault(unsigned int fsr)
114 {
115 return (fsr & FSR_WRITE) && !(fsr & FSR_CM);
116 }
117
is_translation_fault(unsigned int fsr)118 static 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
die_kernel_fault(const char * msg,struct mm_struct * mm,unsigned long addr,unsigned int fsr,struct pt_regs * regs)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)
134 {
135 bust_spinlocks(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" : str_write_read(fsr & FSR_WRITE));
139
140 show_pte(KERN_ALERT, mm, addr);
141 die("Oops", regs, fsr);
142 bust_spinlocks(0);
143 make_task_dead(SIGKILL);
144 }
145
146 /*
147 * Oops. The kernel tried to access some page that wasn't present.
148 */
149 static void
__do_kernel_fault(struct mm_struct * mm,unsigned long addr,unsigned int fsr,struct pt_regs * regs)150 __do_kernel_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
151 struct pt_regs *regs)
152 {
153 const char *msg;
154 /*
155 * Are we prepared to handle this kernel fault?
156 */
157 if (fixup_exception(regs))
158 return;
159
160 /*
161 * No handler, we'll have to terminate things with extreme prejudice.
162 */
163 if (addr < PAGE_SIZE) {
164 msg = "NULL pointer dereference";
165 } else {
166 if (is_translation_fault(fsr) &&
167 kfence_handle_page_fault(addr, is_write_fault(fsr), regs))
168 return;
169
170 msg = "paging request";
171 }
172
173 die_kernel_fault(msg, mm, addr, fsr, regs);
174 }
175
176 /*
177 * Something tried to access memory that isn't in our memory map..
178 * User mode accesses just cause a SIGSEGV
179 */
180 static void
__do_user_fault(unsigned long addr,unsigned int fsr,unsigned int sig,int code,struct pt_regs * regs)181 __do_user_fault(unsigned long addr, unsigned int fsr, unsigned int sig,
182 int code, struct pt_regs *regs)
183 {
184 struct task_struct *tsk = current;
185
186 if (addr > TASK_SIZE)
187 harden_branch_predictor();
188
189 #ifdef CONFIG_DEBUG_USER
190 if (((user_debug & UDBG_SEGV) && (sig == SIGSEGV)) ||
191 ((user_debug & UDBG_BUS) && (sig == SIGBUS))) {
192 pr_err("8<--- cut here ---\n");
193 pr_err("%s: unhandled page fault (%d) at 0x%08lx, code 0x%03x\n",
194 tsk->comm, sig, addr, fsr);
195 show_pte(KERN_ERR, tsk->mm, addr);
196 show_regs(regs);
197 }
198 #endif
199 #ifndef CONFIG_KUSER_HELPERS
200 if ((sig == SIGSEGV) && ((addr & PAGE_MASK) == 0xffff0000))
201 printk_ratelimited(KERN_DEBUG
202 "%s: CONFIG_KUSER_HELPERS disabled at 0x%08lx\n",
203 tsk->comm, addr);
204 #endif
205
206 tsk->thread.address = addr;
207 tsk->thread.error_code = fsr;
208 tsk->thread.trap_no = 14;
209 force_sig_fault(sig, code, (void __user *)addr);
210 }
211
do_bad_area(unsigned long addr,unsigned int fsr,struct pt_regs * regs)212 void do_bad_area(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
213 {
214 struct task_struct *tsk = current;
215 struct mm_struct *mm = tsk->active_mm;
216
217 /*
218 * If we are in kernel mode at this point, we
219 * have no context to handle this fault with.
220 */
221 if (user_mode(regs))
222 __do_user_fault(addr, fsr, SIGSEGV, SEGV_MAPERR, regs);
223 else
224 __do_kernel_fault(mm, addr, fsr, regs);
225 }
226
227 #ifdef CONFIG_MMU
is_permission_fault(unsigned int fsr)228 static inline bool is_permission_fault(unsigned int fsr)
229 {
230 int fs = fsr_fs(fsr);
231 #ifdef CONFIG_ARM_LPAE
232 if ((fs & FS_MMU_NOLL_MASK) == FS_PERM_NOLL)
233 return true;
234 #else
235 if (fs == FS_L1_PERM || fs == FS_L2_PERM)
236 return true;
237 #endif
238 return false;
239 }
240
241 #ifdef CONFIG_CPU_TTBR0_PAN
ttbr0_usermode_access_allowed(struct pt_regs * regs)242 static inline bool ttbr0_usermode_access_allowed(struct pt_regs *regs)
243 {
244 struct svc_pt_regs *svcregs;
245
246 /* If we are in user mode: permission granted */
247 if (user_mode(regs))
248 return true;
249
250 /* uaccess state saved above pt_regs on SVC exception entry */
251 svcregs = to_svc_pt_regs(regs);
252
253 return !(svcregs->ttbcr & TTBCR_EPD0);
254 }
255 #else
ttbr0_usermode_access_allowed(struct pt_regs * regs)256 static inline bool ttbr0_usermode_access_allowed(struct pt_regs *regs)
257 {
258 return true;
259 }
260 #endif
261
262 static int __kprobes
do_page_fault(unsigned long addr,unsigned int fsr,struct pt_regs * regs)263 do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
264 {
265 struct mm_struct *mm = current->mm;
266 struct vm_area_struct *vma;
267 int sig, code;
268 vm_fault_t fault;
269 unsigned int flags = FAULT_FLAG_DEFAULT;
270 vm_flags_t vm_flags = VM_ACCESS_FLAGS;
271
272 if (kprobe_page_fault(regs, fsr))
273 return 0;
274
275
276 /* Enable interrupts if they were enabled in the parent context. */
277 if (interrupts_enabled(regs))
278 local_irq_enable();
279
280 /*
281 * If we're in an interrupt or have no user
282 * context, we must not take the fault..
283 */
284 if (faulthandler_disabled() || !mm)
285 goto no_context;
286
287 if (user_mode(regs))
288 flags |= FAULT_FLAG_USER;
289
290 if (is_write_fault(fsr)) {
291 flags |= FAULT_FLAG_WRITE;
292 vm_flags = VM_WRITE;
293 }
294
295 if (fsr & FSR_LNX_PF) {
296 vm_flags = VM_EXEC;
297
298 if (is_permission_fault(fsr) && !user_mode(regs))
299 die_kernel_fault("execution of memory",
300 mm, addr, fsr, regs);
301 }
302
303 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
304
305 /*
306 * Privileged access aborts with CONFIG_CPU_TTBR0_PAN enabled are
307 * routed via the translation fault mechanism. Check whether uaccess
308 * is disabled while in kernel mode.
309 */
310 if (!ttbr0_usermode_access_allowed(regs))
311 goto no_context;
312
313 if (!(flags & FAULT_FLAG_USER))
314 goto lock_mmap;
315
316 vma = lock_vma_under_rcu(mm, addr);
317 if (!vma)
318 goto lock_mmap;
319
320 if (!(vma->vm_flags & vm_flags)) {
321 vma_end_read(vma);
322 count_vm_vma_lock_event(VMA_LOCK_SUCCESS);
323 fault = 0;
324 code = SEGV_ACCERR;
325 goto bad_area;
326 }
327 fault = handle_mm_fault(vma, addr, flags | FAULT_FLAG_VMA_LOCK, regs);
328 if (!(fault & (VM_FAULT_RETRY | VM_FAULT_COMPLETED)))
329 vma_end_read(vma);
330
331 if (!(fault & VM_FAULT_RETRY)) {
332 count_vm_vma_lock_event(VMA_LOCK_SUCCESS);
333 goto done;
334 }
335 count_vm_vma_lock_event(VMA_LOCK_RETRY);
336 if (fault & VM_FAULT_MAJOR)
337 flags |= FAULT_FLAG_TRIED;
338
339 /* Quick path to respond to signals */
340 if (fault_signal_pending(fault, regs)) {
341 if (!user_mode(regs))
342 goto no_context;
343 return 0;
344 }
345 lock_mmap:
346
347 retry:
348 vma = lock_mm_and_find_vma(mm, addr, regs);
349 if (unlikely(!vma)) {
350 fault = 0;
351 code = SEGV_MAPERR;
352 goto bad_area;
353 }
354
355 /*
356 * ok, we have a good vm_area for this memory access, check the
357 * permissions on the VMA allow for the fault which occurred.
358 */
359 if (!(vma->vm_flags & vm_flags)) {
360 mmap_read_unlock(mm);
361 fault = 0;
362 code = SEGV_ACCERR;
363 goto bad_area;
364 }
365
366 fault = handle_mm_fault(vma, addr & PAGE_MASK, flags, regs);
367
368 /* If we need to retry but a fatal signal is pending, handle the
369 * signal first. We do not need to release the mmap_lock because
370 * it would already be released in __lock_page_or_retry in
371 * mm/filemap.c. */
372 if (fault_signal_pending(fault, regs)) {
373 if (!user_mode(regs))
374 goto no_context;
375 return 0;
376 }
377
378 /* The fault is fully completed (including releasing mmap lock) */
379 if (fault & VM_FAULT_COMPLETED)
380 return 0;
381
382 if (!(fault & VM_FAULT_ERROR)) {
383 if (fault & VM_FAULT_RETRY) {
384 flags |= FAULT_FLAG_TRIED;
385 goto retry;
386 }
387 }
388
389 mmap_read_unlock(mm);
390 done:
391
392 /* Handle the "normal" case first */
393 if (likely(!(fault & VM_FAULT_ERROR)))
394 return 0;
395
396 code = SEGV_MAPERR;
397 bad_area:
398 /*
399 * If we are in kernel mode at this point, we
400 * have no context to handle this fault with.
401 */
402 if (!user_mode(regs))
403 goto no_context;
404
405 if (fault & VM_FAULT_OOM) {
406 /*
407 * We ran out of memory, call the OOM killer, and return to
408 * userspace (which will retry the fault, or kill us if we
409 * got oom-killed)
410 */
411 pagefault_out_of_memory();
412 return 0;
413 }
414
415 if (fault & VM_FAULT_SIGBUS) {
416 /*
417 * We had some memory, but were unable to
418 * successfully fix up this page fault.
419 */
420 sig = SIGBUS;
421 code = BUS_ADRERR;
422 } else {
423 /*
424 * Something tried to access memory that
425 * isn't in our memory map..
426 */
427 sig = SIGSEGV;
428 }
429
430 __do_user_fault(addr, fsr, sig, code, regs);
431 return 0;
432
433 no_context:
434 __do_kernel_fault(mm, addr, fsr, regs);
435 return 0;
436 }
437 #else /* CONFIG_MMU */
438 static int
do_page_fault(unsigned long addr,unsigned int fsr,struct pt_regs * regs)439 do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
440 {
441 return 0;
442 }
443 #endif /* CONFIG_MMU */
444
445 /*
446 * First Level Translation Fault Handler
447 *
448 * We enter here because the first level page table doesn't contain
449 * a valid entry for the address.
450 *
451 * If the address is in kernel space (>= TASK_SIZE), then we are
452 * probably faulting in the vmalloc() area.
453 *
454 * If the init_task's first level page tables contains the relevant
455 * entry, we copy the it to this task. If not, we send the process
456 * a signal, fixup the exception, or oops the kernel.
457 *
458 * NOTE! We MUST NOT take any locks for this case. We may be in an
459 * interrupt or a critical region, and should only copy the information
460 * from the master page table, nothing more.
461 */
462 #ifdef CONFIG_MMU
463 static int __kprobes
do_translation_fault(unsigned long addr,unsigned int fsr,struct pt_regs * regs)464 do_translation_fault(unsigned long addr, unsigned int fsr,
465 struct pt_regs *regs)
466 {
467 unsigned int index;
468 pgd_t *pgd, *pgd_k;
469 p4d_t *p4d, *p4d_k;
470 pud_t *pud, *pud_k;
471 pmd_t *pmd, *pmd_k;
472
473 if (addr < TASK_SIZE)
474 return do_page_fault(addr, fsr, regs);
475
476 if (user_mode(regs))
477 goto bad_area;
478
479 index = pgd_index(addr);
480
481 pgd = cpu_get_pgd() + index;
482 pgd_k = init_mm.pgd + index;
483
484 p4d = p4d_offset(pgd, addr);
485 p4d_k = p4d_offset(pgd_k, addr);
486
487 if (p4d_none(*p4d_k))
488 goto bad_area;
489 if (!p4d_present(*p4d))
490 set_p4d(p4d, *p4d_k);
491
492 pud = pud_offset(p4d, addr);
493 pud_k = pud_offset(p4d_k, addr);
494
495 if (pud_none(*pud_k))
496 goto bad_area;
497 if (!pud_present(*pud))
498 set_pud(pud, *pud_k);
499
500 pmd = pmd_offset(pud, addr);
501 pmd_k = pmd_offset(pud_k, addr);
502
503 #ifdef CONFIG_ARM_LPAE
504 /*
505 * Only one hardware entry per PMD with LPAE.
506 */
507 index = 0;
508 #else
509 /*
510 * On ARM one Linux PGD entry contains two hardware entries (see page
511 * tables layout in pgtable.h). We normally guarantee that we always
512 * fill both L1 entries. But create_mapping() doesn't follow the rule.
513 * It can create inidividual L1 entries, so here we have to call
514 * pmd_none() check for the entry really corresponded to address, not
515 * for the first of pair.
516 */
517 index = (addr >> SECTION_SHIFT) & 1;
518 #endif
519 if (pmd_none(pmd_k[index]))
520 goto bad_area;
521
522 copy_pmd(pmd, pmd_k);
523 return 0;
524
525 bad_area:
526 do_bad_area(addr, fsr, regs);
527 return 0;
528 }
529 #else /* CONFIG_MMU */
530 static int
do_translation_fault(unsigned long addr,unsigned int fsr,struct pt_regs * regs)531 do_translation_fault(unsigned long addr, unsigned int fsr,
532 struct pt_regs *regs)
533 {
534 return 0;
535 }
536 #endif /* CONFIG_MMU */
537
538 /*
539 * Some section permission faults need to be handled gracefully.
540 * They can happen due to a __{get,put}_user during an oops.
541 */
542 #ifndef CONFIG_ARM_LPAE
543 static int
do_sect_fault(unsigned long addr,unsigned int fsr,struct pt_regs * regs)544 do_sect_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
545 {
546 do_bad_area(addr, fsr, regs);
547 return 0;
548 }
549 #endif /* CONFIG_ARM_LPAE */
550
551 /*
552 * This abort handler always returns "fault".
553 */
554 static int
do_bad(unsigned long addr,unsigned int fsr,struct pt_regs * regs)555 do_bad(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
556 {
557 return 1;
558 }
559
560 struct fsr_info {
561 int (*fn)(unsigned long addr, unsigned int fsr, struct pt_regs *regs);
562 int sig;
563 int code;
564 const char *name;
565 };
566
567 /* FSR definition */
568 #ifdef CONFIG_ARM_LPAE
569 #include "fsr-3level.c"
570 #else
571 #include "fsr-2level.c"
572 #endif
573
574 void __init
hook_fault_code(int nr,int (* fn)(unsigned long,unsigned int,struct pt_regs *),int sig,int code,const char * name)575 hook_fault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
576 int sig, int code, const char *name)
577 {
578 if (nr < 0 || nr >= ARRAY_SIZE(fsr_info))
579 BUG();
580
581 fsr_info[nr].fn = fn;
582 fsr_info[nr].sig = sig;
583 fsr_info[nr].code = code;
584 fsr_info[nr].name = name;
585 }
586
587 /*
588 * Dispatch a data abort to the relevant handler.
589 */
590 asmlinkage void
do_DataAbort(unsigned long addr,unsigned int fsr,struct pt_regs * regs)591 do_DataAbort(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
592 {
593 const struct fsr_info *inf = fsr_info + fsr_fs(fsr);
594
595 if (!inf->fn(addr, fsr & ~FSR_LNX_PF, regs))
596 return;
597
598 pr_alert("8<--- cut here ---\n");
599 pr_alert("Unhandled fault: %s (0x%03x) at 0x%08lx\n",
600 inf->name, fsr, addr);
601 show_pte(KERN_ALERT, current->mm, addr);
602
603 arm_notify_die("", regs, inf->sig, inf->code, (void __user *)addr,
604 fsr, 0);
605 }
606
607 void __init
hook_ifault_code(int nr,int (* fn)(unsigned long,unsigned int,struct pt_regs *),int sig,int code,const char * name)608 hook_ifault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
609 int sig, int code, const char *name)
610 {
611 if (nr < 0 || nr >= ARRAY_SIZE(ifsr_info))
612 BUG();
613
614 ifsr_info[nr].fn = fn;
615 ifsr_info[nr].sig = sig;
616 ifsr_info[nr].code = code;
617 ifsr_info[nr].name = name;
618 }
619
620 asmlinkage void
do_PrefetchAbort(unsigned long addr,unsigned int ifsr,struct pt_regs * regs)621 do_PrefetchAbort(unsigned long addr, unsigned int ifsr, struct pt_regs *regs)
622 {
623 const struct fsr_info *inf = ifsr_info + fsr_fs(ifsr);
624
625 if (!inf->fn(addr, ifsr | FSR_LNX_PF, regs))
626 return;
627
628 pr_alert("8<--- cut here ---\n");
629 pr_alert("Unhandled prefetch abort: %s (0x%03x) at 0x%08lx\n",
630 inf->name, ifsr, addr);
631
632 arm_notify_die("", regs, inf->sig, inf->code, (void __user *)addr,
633 ifsr, 0);
634 }
635
636 /*
637 * Abort handler to be used only during first unmasking of asynchronous aborts
638 * on the boot CPU. This makes sure that the machine will not die if the
639 * firmware/bootloader left an imprecise abort pending for us to trip over.
640 */
early_abort_handler(unsigned long addr,unsigned int fsr,struct pt_regs * regs)641 static int __init early_abort_handler(unsigned long addr, unsigned int fsr,
642 struct pt_regs *regs)
643 {
644 pr_warn("Hit pending asynchronous external abort (FSR=0x%08x) during "
645 "first unmask, this is most likely caused by a "
646 "firmware/bootloader bug.\n", fsr);
647
648 return 0;
649 }
650
early_abt_enable(void)651 void __init early_abt_enable(void)
652 {
653 fsr_info[FSR_FS_AEA].fn = early_abort_handler;
654 local_abt_enable();
655 fsr_info[FSR_FS_AEA].fn = do_bad;
656 }
657
658 #ifndef CONFIG_ARM_LPAE
exceptions_init(void)659 static int __init exceptions_init(void)
660 {
661 if (cpu_architecture() >= CPU_ARCH_ARMv6) {
662 hook_fault_code(4, do_translation_fault, SIGSEGV, SEGV_MAPERR,
663 "I-cache maintenance fault");
664 }
665
666 if (cpu_architecture() >= CPU_ARCH_ARMv7) {
667 /*
668 * TODO: Access flag faults introduced in ARMv6K.
669 * Runtime check for 'K' extension is needed
670 */
671 hook_fault_code(3, do_bad, SIGSEGV, SEGV_MAPERR,
672 "section access flag fault");
673 hook_fault_code(6, do_bad, SIGSEGV, SEGV_MAPERR,
674 "section access flag fault");
675 }
676
677 return 0;
678 }
679
680 arch_initcall(exceptions_init);
681 #endif
682