Loading...
1/*
2 * linux/arch/arm/mm/fault.c
3 *
4 * Copyright (C) 1995 Linus Torvalds
5 * Modifications for ARM processor (c) 1995-2004 Russell King
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
10 */
11#include <linux/module.h>
12#include <linux/signal.h>
13#include <linux/mm.h>
14#include <linux/hardirq.h>
15#include <linux/init.h>
16#include <linux/kprobes.h>
17#include <linux/uaccess.h>
18#include <linux/page-flags.h>
19#include <linux/sched.h>
20#include <linux/highmem.h>
21#include <linux/perf_event.h>
22
23#include <asm/system.h>
24#include <asm/pgtable.h>
25#include <asm/tlbflush.h>
26
27#include "fault.h"
28
29/*
30 * Fault status register encodings. We steal bit 31 for our own purposes.
31 */
32#define FSR_LNX_PF (1 << 31)
33#define FSR_WRITE (1 << 11)
34#define FSR_FS4 (1 << 10)
35#define FSR_FS3_0 (15)
36
37static inline int fsr_fs(unsigned int fsr)
38{
39 return (fsr & FSR_FS3_0) | (fsr & FSR_FS4) >> 6;
40}
41
42#ifdef CONFIG_MMU
43
44#ifdef CONFIG_KPROBES
45static inline int notify_page_fault(struct pt_regs *regs, unsigned int fsr)
46{
47 int ret = 0;
48
49 if (!user_mode(regs)) {
50 /* kprobe_running() needs smp_processor_id() */
51 preempt_disable();
52 if (kprobe_running() && kprobe_fault_handler(regs, fsr))
53 ret = 1;
54 preempt_enable();
55 }
56
57 return ret;
58}
59#else
60static inline int notify_page_fault(struct pt_regs *regs, unsigned int fsr)
61{
62 return 0;
63}
64#endif
65
66/*
67 * This is useful to dump out the page tables associated with
68 * 'addr' in mm 'mm'.
69 */
70void show_pte(struct mm_struct *mm, unsigned long addr)
71{
72 pgd_t *pgd;
73
74 if (!mm)
75 mm = &init_mm;
76
77 printk(KERN_ALERT "pgd = %p\n", mm->pgd);
78 pgd = pgd_offset(mm, addr);
79 printk(KERN_ALERT "[%08lx] *pgd=%08llx",
80 addr, (long long)pgd_val(*pgd));
81
82 do {
83 pud_t *pud;
84 pmd_t *pmd;
85 pte_t *pte;
86
87 if (pgd_none(*pgd))
88 break;
89
90 if (pgd_bad(*pgd)) {
91 printk("(bad)");
92 break;
93 }
94
95 pud = pud_offset(pgd, addr);
96 if (PTRS_PER_PUD != 1)
97 printk(", *pud=%08llx", (long long)pud_val(*pud));
98
99 if (pud_none(*pud))
100 break;
101
102 if (pud_bad(*pud)) {
103 printk("(bad)");
104 break;
105 }
106
107 pmd = pmd_offset(pud, addr);
108 if (PTRS_PER_PMD != 1)
109 printk(", *pmd=%08llx", (long long)pmd_val(*pmd));
110
111 if (pmd_none(*pmd))
112 break;
113
114 if (pmd_bad(*pmd)) {
115 printk("(bad)");
116 break;
117 }
118
119 /* We must not map this if we have highmem enabled */
120 if (PageHighMem(pfn_to_page(pmd_val(*pmd) >> PAGE_SHIFT)))
121 break;
122
123 pte = pte_offset_map(pmd, addr);
124 printk(", *pte=%08llx", (long long)pte_val(*pte));
125 printk(", *ppte=%08llx",
126 (long long)pte_val(pte[PTE_HWTABLE_PTRS]));
127 pte_unmap(pte);
128 } while(0);
129
130 printk("\n");
131}
132#else /* CONFIG_MMU */
133void show_pte(struct mm_struct *mm, unsigned long addr)
134{ }
135#endif /* CONFIG_MMU */
136
137/*
138 * Oops. The kernel tried to access some page that wasn't present.
139 */
140static void
141__do_kernel_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
142 struct pt_regs *regs)
143{
144 /*
145 * Are we prepared to handle this kernel fault?
146 */
147 if (fixup_exception(regs))
148 return;
149
150 /*
151 * No handler, we'll have to terminate things with extreme prejudice.
152 */
153 bust_spinlocks(1);
154 printk(KERN_ALERT
155 "Unable to handle kernel %s at virtual address %08lx\n",
156 (addr < PAGE_SIZE) ? "NULL pointer dereference" :
157 "paging request", addr);
158
159 show_pte(mm, addr);
160 die("Oops", regs, fsr);
161 bust_spinlocks(0);
162 do_exit(SIGKILL);
163}
164
165/*
166 * Something tried to access memory that isn't in our memory map..
167 * User mode accesses just cause a SIGSEGV
168 */
169static void
170__do_user_fault(struct task_struct *tsk, unsigned long addr,
171 unsigned int fsr, unsigned int sig, int code,
172 struct pt_regs *regs)
173{
174 struct siginfo si;
175
176#ifdef CONFIG_DEBUG_USER
177 if (user_debug & UDBG_SEGV) {
178 printk(KERN_DEBUG "%s: unhandled page fault (%d) at 0x%08lx, code 0x%03x\n",
179 tsk->comm, sig, addr, fsr);
180 show_pte(tsk->mm, addr);
181 show_regs(regs);
182 }
183#endif
184
185 tsk->thread.address = addr;
186 tsk->thread.error_code = fsr;
187 tsk->thread.trap_no = 14;
188 si.si_signo = sig;
189 si.si_errno = 0;
190 si.si_code = code;
191 si.si_addr = (void __user *)addr;
192 force_sig_info(sig, &si, tsk);
193}
194
195void do_bad_area(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
196{
197 struct task_struct *tsk = current;
198 struct mm_struct *mm = tsk->active_mm;
199
200 /*
201 * If we are in kernel mode at this point, we
202 * have no context to handle this fault with.
203 */
204 if (user_mode(regs))
205 __do_user_fault(tsk, addr, fsr, SIGSEGV, SEGV_MAPERR, regs);
206 else
207 __do_kernel_fault(mm, addr, fsr, regs);
208}
209
210#ifdef CONFIG_MMU
211#define VM_FAULT_BADMAP 0x010000
212#define VM_FAULT_BADACCESS 0x020000
213
214/*
215 * Check that the permissions on the VMA allow for the fault which occurred.
216 * If we encountered a write fault, we must have write permission, otherwise
217 * we allow any permission.
218 */
219static inline bool access_error(unsigned int fsr, struct vm_area_struct *vma)
220{
221 unsigned int mask = VM_READ | VM_WRITE | VM_EXEC;
222
223 if (fsr & FSR_WRITE)
224 mask = VM_WRITE;
225 if (fsr & FSR_LNX_PF)
226 mask = VM_EXEC;
227
228 return vma->vm_flags & mask ? false : true;
229}
230
231static int __kprobes
232__do_page_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
233 struct task_struct *tsk)
234{
235 struct vm_area_struct *vma;
236 int fault;
237
238 vma = find_vma(mm, addr);
239 fault = VM_FAULT_BADMAP;
240 if (unlikely(!vma))
241 goto out;
242 if (unlikely(vma->vm_start > addr))
243 goto check_stack;
244
245 /*
246 * Ok, we have a good vm_area for this
247 * memory access, so we can handle it.
248 */
249good_area:
250 if (access_error(fsr, vma)) {
251 fault = VM_FAULT_BADACCESS;
252 goto out;
253 }
254
255 /*
256 * If for any reason at all we couldn't handle the fault, make
257 * sure we exit gracefully rather than endlessly redo the fault.
258 */
259 fault = handle_mm_fault(mm, vma, addr & PAGE_MASK, (fsr & FSR_WRITE) ? FAULT_FLAG_WRITE : 0);
260 if (unlikely(fault & VM_FAULT_ERROR))
261 return fault;
262 if (fault & VM_FAULT_MAJOR)
263 tsk->maj_flt++;
264 else
265 tsk->min_flt++;
266 return fault;
267
268check_stack:
269 if (vma->vm_flags & VM_GROWSDOWN && !expand_stack(vma, addr))
270 goto good_area;
271out:
272 return fault;
273}
274
275static int __kprobes
276do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
277{
278 struct task_struct *tsk;
279 struct mm_struct *mm;
280 int fault, sig, code;
281
282 if (notify_page_fault(regs, fsr))
283 return 0;
284
285 tsk = current;
286 mm = tsk->mm;
287
288 /* Enable interrupts if they were enabled in the parent context. */
289 if (interrupts_enabled(regs))
290 local_irq_enable();
291
292 /*
293 * If we're in an interrupt or have no user
294 * context, we must not take the fault..
295 */
296 if (in_atomic() || !mm)
297 goto no_context;
298
299 /*
300 * As per x86, we may deadlock here. However, since the kernel only
301 * validly references user space from well defined areas of the code,
302 * we can bug out early if this is from code which shouldn't.
303 */
304 if (!down_read_trylock(&mm->mmap_sem)) {
305 if (!user_mode(regs) && !search_exception_tables(regs->ARM_pc))
306 goto no_context;
307 down_read(&mm->mmap_sem);
308 } else {
309 /*
310 * The above down_read_trylock() might have succeeded in
311 * which case, we'll have missed the might_sleep() from
312 * down_read()
313 */
314 might_sleep();
315#ifdef CONFIG_DEBUG_VM
316 if (!user_mode(regs) &&
317 !search_exception_tables(regs->ARM_pc))
318 goto no_context;
319#endif
320 }
321
322 fault = __do_page_fault(mm, addr, fsr, tsk);
323 up_read(&mm->mmap_sem);
324
325 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
326 if (fault & VM_FAULT_MAJOR)
327 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, addr);
328 else if (fault & VM_FAULT_MINOR)
329 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, addr);
330
331 /*
332 * Handle the "normal" case first - VM_FAULT_MAJOR / VM_FAULT_MINOR
333 */
334 if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP | VM_FAULT_BADACCESS))))
335 return 0;
336
337 if (fault & VM_FAULT_OOM) {
338 /*
339 * We ran out of memory, call the OOM killer, and return to
340 * userspace (which will retry the fault, or kill us if we
341 * got oom-killed)
342 */
343 pagefault_out_of_memory();
344 return 0;
345 }
346
347 /*
348 * If we are in kernel mode at this point, we
349 * have no context to handle this fault with.
350 */
351 if (!user_mode(regs))
352 goto no_context;
353
354 if (fault & VM_FAULT_SIGBUS) {
355 /*
356 * We had some memory, but were unable to
357 * successfully fix up this page fault.
358 */
359 sig = SIGBUS;
360 code = BUS_ADRERR;
361 } else {
362 /*
363 * Something tried to access memory that
364 * isn't in our memory map..
365 */
366 sig = SIGSEGV;
367 code = fault == VM_FAULT_BADACCESS ?
368 SEGV_ACCERR : SEGV_MAPERR;
369 }
370
371 __do_user_fault(tsk, addr, fsr, sig, code, regs);
372 return 0;
373
374no_context:
375 __do_kernel_fault(mm, addr, fsr, regs);
376 return 0;
377}
378#else /* CONFIG_MMU */
379static int
380do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
381{
382 return 0;
383}
384#endif /* CONFIG_MMU */
385
386/*
387 * First Level Translation Fault Handler
388 *
389 * We enter here because the first level page table doesn't contain
390 * a valid entry for the address.
391 *
392 * If the address is in kernel space (>= TASK_SIZE), then we are
393 * probably faulting in the vmalloc() area.
394 *
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.
398 *
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.
402 */
403#ifdef CONFIG_MMU
404static int __kprobes
405do_translation_fault(unsigned long addr, unsigned int fsr,
406 struct pt_regs *regs)
407{
408 unsigned int index;
409 pgd_t *pgd, *pgd_k;
410 pud_t *pud, *pud_k;
411 pmd_t *pmd, *pmd_k;
412
413 if (addr < TASK_SIZE)
414 return do_page_fault(addr, fsr, regs);
415
416 if (user_mode(regs))
417 goto bad_area;
418
419 index = pgd_index(addr);
420
421 /*
422 * FIXME: CP15 C1 is write only on ARMv3 architectures.
423 */
424 pgd = cpu_get_pgd() + index;
425 pgd_k = init_mm.pgd + index;
426
427 if (pgd_none(*pgd_k))
428 goto bad_area;
429 if (!pgd_present(*pgd))
430 set_pgd(pgd, *pgd_k);
431
432 pud = pud_offset(pgd, addr);
433 pud_k = pud_offset(pgd_k, addr);
434
435 if (pud_none(*pud_k))
436 goto bad_area;
437 if (!pud_present(*pud))
438 set_pud(pud, *pud_k);
439
440 pmd = pmd_offset(pud, addr);
441 pmd_k = pmd_offset(pud_k, addr);
442
443 /*
444 * On ARM one Linux PGD entry contains two hardware entries (see page
445 * tables layout in pgtable.h). We normally guarantee that we always
446 * fill both L1 entries. But create_mapping() doesn't follow the rule.
447 * It can create inidividual L1 entries, so here we have to call
448 * pmd_none() check for the entry really corresponded to address, not
449 * for the first of pair.
450 */
451 index = (addr >> SECTION_SHIFT) & 1;
452 if (pmd_none(pmd_k[index]))
453 goto bad_area;
454
455 copy_pmd(pmd, pmd_k);
456 return 0;
457
458bad_area:
459 do_bad_area(addr, fsr, regs);
460 return 0;
461}
462#else /* CONFIG_MMU */
463static int
464do_translation_fault(unsigned long addr, unsigned int fsr,
465 struct pt_regs *regs)
466{
467 return 0;
468}
469#endif /* CONFIG_MMU */
470
471/*
472 * Some section permission faults need to be handled gracefully.
473 * They can happen due to a __{get,put}_user during an oops.
474 */
475static int
476do_sect_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
477{
478 do_bad_area(addr, fsr, regs);
479 return 0;
480}
481
482/*
483 * This abort handler always returns "fault".
484 */
485static int
486do_bad(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
487{
488 return 1;
489}
490
491static struct fsr_info {
492 int (*fn)(unsigned long addr, unsigned int fsr, struct pt_regs *regs);
493 int sig;
494 int code;
495 const char *name;
496} fsr_info[] = {
497 /*
498 * The following are the standard ARMv3 and ARMv4 aborts. ARMv5
499 * defines these to be "precise" aborts.
500 */
501 { do_bad, SIGSEGV, 0, "vector exception" },
502 { do_bad, SIGBUS, BUS_ADRALN, "alignment exception" },
503 { do_bad, SIGKILL, 0, "terminal exception" },
504 { do_bad, SIGBUS, BUS_ADRALN, "alignment exception" },
505 { do_bad, SIGBUS, 0, "external abort on linefetch" },
506 { do_translation_fault, SIGSEGV, SEGV_MAPERR, "section translation fault" },
507 { do_bad, SIGBUS, 0, "external abort on linefetch" },
508 { do_page_fault, SIGSEGV, SEGV_MAPERR, "page translation fault" },
509 { do_bad, SIGBUS, 0, "external abort on non-linefetch" },
510 { do_bad, SIGSEGV, SEGV_ACCERR, "section domain fault" },
511 { do_bad, SIGBUS, 0, "external abort on non-linefetch" },
512 { do_bad, SIGSEGV, SEGV_ACCERR, "page domain fault" },
513 { do_bad, SIGBUS, 0, "external abort on translation" },
514 { do_sect_fault, SIGSEGV, SEGV_ACCERR, "section permission fault" },
515 { do_bad, SIGBUS, 0, "external abort on translation" },
516 { do_page_fault, SIGSEGV, SEGV_ACCERR, "page permission fault" },
517 /*
518 * The following are "imprecise" aborts, which are signalled by bit
519 * 10 of the FSR, and may not be recoverable. These are only
520 * supported if the CPU abort handler supports bit 10.
521 */
522 { do_bad, SIGBUS, 0, "unknown 16" },
523 { do_bad, SIGBUS, 0, "unknown 17" },
524 { do_bad, SIGBUS, 0, "unknown 18" },
525 { do_bad, SIGBUS, 0, "unknown 19" },
526 { do_bad, SIGBUS, 0, "lock abort" }, /* xscale */
527 { do_bad, SIGBUS, 0, "unknown 21" },
528 { do_bad, SIGBUS, BUS_OBJERR, "imprecise external abort" }, /* xscale */
529 { do_bad, SIGBUS, 0, "unknown 23" },
530 { do_bad, SIGBUS, 0, "dcache parity error" }, /* xscale */
531 { do_bad, SIGBUS, 0, "unknown 25" },
532 { do_bad, SIGBUS, 0, "unknown 26" },
533 { do_bad, SIGBUS, 0, "unknown 27" },
534 { do_bad, SIGBUS, 0, "unknown 28" },
535 { do_bad, SIGBUS, 0, "unknown 29" },
536 { do_bad, SIGBUS, 0, "unknown 30" },
537 { do_bad, SIGBUS, 0, "unknown 31" }
538};
539
540void __init
541hook_fault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
542 int sig, int code, const char *name)
543{
544 if (nr < 0 || nr >= ARRAY_SIZE(fsr_info))
545 BUG();
546
547 fsr_info[nr].fn = fn;
548 fsr_info[nr].sig = sig;
549 fsr_info[nr].code = code;
550 fsr_info[nr].name = name;
551}
552
553/*
554 * Dispatch a data abort to the relevant handler.
555 */
556asmlinkage void __exception
557do_DataAbort(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
558{
559 const struct fsr_info *inf = fsr_info + fsr_fs(fsr);
560 struct siginfo info;
561
562 if (!inf->fn(addr, fsr & ~FSR_LNX_PF, regs))
563 return;
564
565 printk(KERN_ALERT "Unhandled fault: %s (0x%03x) at 0x%08lx\n",
566 inf->name, fsr, addr);
567
568 info.si_signo = inf->sig;
569 info.si_errno = 0;
570 info.si_code = inf->code;
571 info.si_addr = (void __user *)addr;
572 arm_notify_die("", regs, &info, fsr, 0);
573}
574
575
576static struct fsr_info ifsr_info[] = {
577 { do_bad, SIGBUS, 0, "unknown 0" },
578 { do_bad, SIGBUS, 0, "unknown 1" },
579 { do_bad, SIGBUS, 0, "debug event" },
580 { do_bad, SIGSEGV, SEGV_ACCERR, "section access flag fault" },
581 { do_bad, SIGBUS, 0, "unknown 4" },
582 { do_translation_fault, SIGSEGV, SEGV_MAPERR, "section translation fault" },
583 { do_bad, SIGSEGV, SEGV_ACCERR, "page access flag fault" },
584 { do_page_fault, SIGSEGV, SEGV_MAPERR, "page translation fault" },
585 { do_bad, SIGBUS, 0, "external abort on non-linefetch" },
586 { do_bad, SIGSEGV, SEGV_ACCERR, "section domain fault" },
587 { do_bad, SIGBUS, 0, "unknown 10" },
588 { do_bad, SIGSEGV, SEGV_ACCERR, "page domain fault" },
589 { do_bad, SIGBUS, 0, "external abort on translation" },
590 { do_sect_fault, SIGSEGV, SEGV_ACCERR, "section permission fault" },
591 { do_bad, SIGBUS, 0, "external abort on translation" },
592 { do_page_fault, SIGSEGV, SEGV_ACCERR, "page permission fault" },
593 { do_bad, SIGBUS, 0, "unknown 16" },
594 { do_bad, SIGBUS, 0, "unknown 17" },
595 { do_bad, SIGBUS, 0, "unknown 18" },
596 { do_bad, SIGBUS, 0, "unknown 19" },
597 { do_bad, SIGBUS, 0, "unknown 20" },
598 { do_bad, SIGBUS, 0, "unknown 21" },
599 { do_bad, SIGBUS, 0, "unknown 22" },
600 { do_bad, SIGBUS, 0, "unknown 23" },
601 { do_bad, SIGBUS, 0, "unknown 24" },
602 { do_bad, SIGBUS, 0, "unknown 25" },
603 { do_bad, SIGBUS, 0, "unknown 26" },
604 { do_bad, SIGBUS, 0, "unknown 27" },
605 { do_bad, SIGBUS, 0, "unknown 28" },
606 { do_bad, SIGBUS, 0, "unknown 29" },
607 { do_bad, SIGBUS, 0, "unknown 30" },
608 { do_bad, SIGBUS, 0, "unknown 31" },
609};
610
611void __init
612hook_ifault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
613 int sig, int code, const char *name)
614{
615 if (nr < 0 || nr >= ARRAY_SIZE(ifsr_info))
616 BUG();
617
618 ifsr_info[nr].fn = fn;
619 ifsr_info[nr].sig = sig;
620 ifsr_info[nr].code = code;
621 ifsr_info[nr].name = name;
622}
623
624asmlinkage void __exception
625do_PrefetchAbort(unsigned long addr, unsigned int ifsr, struct pt_regs *regs)
626{
627 const struct fsr_info *inf = ifsr_info + fsr_fs(ifsr);
628 struct siginfo info;
629
630 if (!inf->fn(addr, ifsr | FSR_LNX_PF, regs))
631 return;
632
633 printk(KERN_ALERT "Unhandled prefetch abort: %s (0x%03x) at 0x%08lx\n",
634 inf->name, ifsr, addr);
635
636 info.si_signo = inf->sig;
637 info.si_errno = 0;
638 info.si_code = inf->code;
639 info.si_addr = (void __user *)addr;
640 arm_notify_die("", regs, &info, ifsr, 0);
641}
642
643static int __init exceptions_init(void)
644{
645 if (cpu_architecture() >= CPU_ARCH_ARMv6) {
646 hook_fault_code(4, do_translation_fault, SIGSEGV, SEGV_MAPERR,
647 "I-cache maintenance fault");
648 }
649
650 if (cpu_architecture() >= CPU_ARCH_ARMv7) {
651 /*
652 * TODO: Access flag faults introduced in ARMv6K.
653 * Runtime check for 'K' extension is needed
654 */
655 hook_fault_code(3, do_bad, SIGSEGV, SEGV_MAPERR,
656 "section access flag fault");
657 hook_fault_code(6, do_bad, SIGSEGV, SEGV_MAPERR,
658 "section access flag fault");
659 }
660
661 return 0;
662}
663
664arch_initcall(exceptions_init);
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
30/*
31 * This is useful to dump out the page tables associated with
32 * 'addr' in mm 'mm'.
33 */
34void show_pte(const char *lvl, struct mm_struct *mm, unsigned long addr)
35{
36 pgd_t *pgd;
37
38 if (!mm)
39 mm = &init_mm;
40
41 pgd = pgd_offset(mm, addr);
42 printk("%s[%08lx] *pgd=%08llx", lvl, addr, (long long)pgd_val(*pgd));
43
44 do {
45 p4d_t *p4d;
46 pud_t *pud;
47 pmd_t *pmd;
48 pte_t *pte;
49
50 p4d = p4d_offset(pgd, addr);
51 if (p4d_none(*p4d))
52 break;
53
54 if (p4d_bad(*p4d)) {
55 pr_cont("(bad)");
56 break;
57 }
58
59 pud = pud_offset(p4d, addr);
60 if (PTRS_PER_PUD != 1)
61 pr_cont(", *pud=%08llx", (long long)pud_val(*pud));
62
63 if (pud_none(*pud))
64 break;
65
66 if (pud_bad(*pud)) {
67 pr_cont("(bad)");
68 break;
69 }
70
71 pmd = pmd_offset(pud, addr);
72 if (PTRS_PER_PMD != 1)
73 pr_cont(", *pmd=%08llx", (long long)pmd_val(*pmd));
74
75 if (pmd_none(*pmd))
76 break;
77
78 if (pmd_bad(*pmd)) {
79 pr_cont("(bad)");
80 break;
81 }
82
83 /* We must not map this if we have highmem enabled */
84 if (PageHighMem(pfn_to_page(pmd_val(*pmd) >> PAGE_SHIFT)))
85 break;
86
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]));
92#endif
93 pte_unmap(pte);
94 } while(0);
95
96 pr_cont("\n");
97}
98#else /* CONFIG_MMU */
99void show_pte(const char *lvl, struct mm_struct *mm, unsigned long addr)
100{ }
101#endif /* CONFIG_MMU */
102
103static inline bool is_write_fault(unsigned int fsr)
104{
105 return (fsr & FSR_WRITE) && !(fsr & FSR_CM);
106}
107
108static inline bool is_translation_fault(unsigned int fsr)
109{
110 int fs = fsr_fs(fsr);
111#ifdef CONFIG_ARM_LPAE
112 if ((fs & FS_MMU_NOLL_MASK) == FS_TRANS_NOLL)
113 return true;
114#else
115 if (fs == FS_L1_TRANS || fs == FS_L2_TRANS)
116 return true;
117#endif
118 return false;
119}
120
121static void die_kernel_fault(const char *msg, struct mm_struct *mm,
122 unsigned long addr, unsigned int fsr,
123 struct pt_regs *regs)
124{
125 bust_spinlocks(1);
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");
130
131 show_pte(KERN_ALERT, mm, addr);
132 die("Oops", regs, fsr);
133 bust_spinlocks(0);
134 make_task_dead(SIGKILL);
135}
136
137/*
138 * Oops. The kernel tried to access some page that wasn't present.
139 */
140static void
141__do_kernel_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
142 struct pt_regs *regs)
143{
144 const char *msg;
145 /*
146 * Are we prepared to handle this kernel fault?
147 */
148 if (fixup_exception(regs))
149 return;
150
151 /*
152 * No handler, we'll have to terminate things with extreme prejudice.
153 */
154 if (addr < PAGE_SIZE) {
155 msg = "NULL pointer dereference";
156 } else {
157 if (is_translation_fault(fsr) &&
158 kfence_handle_page_fault(addr, is_write_fault(fsr), regs))
159 return;
160
161 msg = "paging request";
162 }
163
164 die_kernel_fault(msg, mm, addr, fsr, regs);
165}
166
167/*
168 * Something tried to access memory that isn't in our memory map..
169 * User mode accesses just cause a SIGSEGV
170 */
171static void
172__do_user_fault(unsigned long addr, unsigned int fsr, unsigned int sig,
173 int code, struct pt_regs *regs)
174{
175 struct task_struct *tsk = current;
176
177 if (addr > TASK_SIZE)
178 harden_branch_predictor();
179
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);
187 show_regs(regs);
188 }
189#endif
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",
194 tsk->comm, addr);
195#endif
196
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);
201}
202
203void do_bad_area(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
204{
205 struct task_struct *tsk = current;
206 struct mm_struct *mm = tsk->active_mm;
207
208 /*
209 * If we are in kernel mode at this point, we
210 * have no context to handle this fault with.
211 */
212 if (user_mode(regs))
213 __do_user_fault(addr, fsr, SIGSEGV, SEGV_MAPERR, regs);
214 else
215 __do_kernel_fault(mm, addr, fsr, regs);
216}
217
218#ifdef CONFIG_MMU
219#define VM_FAULT_BADMAP ((__force vm_fault_t)0x010000)
220#define VM_FAULT_BADACCESS ((__force vm_fault_t)0x020000)
221
222static inline bool is_permission_fault(unsigned int fsr)
223{
224 int fs = fsr_fs(fsr);
225#ifdef CONFIG_ARM_LPAE
226 if ((fs & FS_MMU_NOLL_MASK) == FS_PERM_NOLL)
227 return true;
228#else
229 if (fs == FS_L1_PERM || fs == FS_L2_PERM)
230 return true;
231#endif
232 return false;
233}
234
235static 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)
238{
239 struct vm_area_struct *vma = find_vma(mm, addr);
240 if (unlikely(!vma))
241 return VM_FAULT_BADMAP;
242
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;
250 }
251
252 /*
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.
255 */
256 if (!(vma->vm_flags & vma_flags))
257 return VM_FAULT_BADACCESS;
258
259 return handle_mm_fault(vma, addr & PAGE_MASK, flags, regs);
260}
261
262static int __kprobes
263do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
264{
265 struct mm_struct *mm = current->mm;
266 int sig, code;
267 vm_fault_t fault;
268 unsigned int flags = FAULT_FLAG_DEFAULT;
269 unsigned long vm_flags = VM_ACCESS_FLAGS;
270
271 if (kprobe_page_fault(regs, fsr))
272 return 0;
273
274
275 /* Enable interrupts if they were enabled in the parent context. */
276 if (interrupts_enabled(regs))
277 local_irq_enable();
278
279 /*
280 * If we're in an interrupt or have no user
281 * context, we must not take the fault..
282 */
283 if (faulthandler_disabled() || !mm)
284 goto no_context;
285
286 if (user_mode(regs))
287 flags |= FAULT_FLAG_USER;
288
289 if (is_write_fault(fsr)) {
290 flags |= FAULT_FLAG_WRITE;
291 vm_flags = VM_WRITE;
292 }
293
294 if (fsr & FSR_LNX_PF) {
295 vm_flags = VM_EXEC;
296
297 if (is_permission_fault(fsr) && !user_mode(regs))
298 die_kernel_fault("execution of memory",
299 mm, addr, fsr, regs);
300 }
301
302 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
303
304 /*
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.
308 */
309 if (!mmap_read_trylock(mm)) {
310 if (!user_mode(regs) && !search_exception_tables(regs->ARM_pc))
311 goto no_context;
312retry:
313 mmap_read_lock(mm);
314 } else {
315 /*
316 * The above down_read_trylock() might have succeeded in
317 * which case, we'll have missed the might_sleep() from
318 * down_read()
319 */
320 might_sleep();
321#ifdef CONFIG_DEBUG_VM
322 if (!user_mode(regs) &&
323 !search_exception_tables(regs->ARM_pc))
324 goto no_context;
325#endif
326 }
327
328 fault = __do_page_fault(mm, addr, flags, vm_flags, regs);
329
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
333 * mm/filemap.c. */
334 if (fault_signal_pending(fault, regs)) {
335 if (!user_mode(regs))
336 goto no_context;
337 return 0;
338 }
339
340 /* The fault is fully completed (including releasing mmap lock) */
341 if (fault & VM_FAULT_COMPLETED)
342 return 0;
343
344 if (!(fault & VM_FAULT_ERROR)) {
345 if (fault & VM_FAULT_RETRY) {
346 flags |= FAULT_FLAG_TRIED;
347 goto retry;
348 }
349 }
350
351 mmap_read_unlock(mm);
352
353 /*
354 * Handle the "normal" case first - VM_FAULT_MAJOR
355 */
356 if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP | VM_FAULT_BADACCESS))))
357 return 0;
358
359 /*
360 * If we are in kernel mode at this point, we
361 * have no context to handle this fault with.
362 */
363 if (!user_mode(regs))
364 goto no_context;
365
366 if (fault & VM_FAULT_OOM) {
367 /*
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
370 * got oom-killed)
371 */
372 pagefault_out_of_memory();
373 return 0;
374 }
375
376 if (fault & VM_FAULT_SIGBUS) {
377 /*
378 * We had some memory, but were unable to
379 * successfully fix up this page fault.
380 */
381 sig = SIGBUS;
382 code = BUS_ADRERR;
383 } else {
384 /*
385 * Something tried to access memory that
386 * isn't in our memory map..
387 */
388 sig = SIGSEGV;
389 code = fault == VM_FAULT_BADACCESS ?
390 SEGV_ACCERR : SEGV_MAPERR;
391 }
392
393 __do_user_fault(addr, fsr, sig, code, regs);
394 return 0;
395
396no_context:
397 __do_kernel_fault(mm, addr, fsr, regs);
398 return 0;
399}
400#else /* CONFIG_MMU */
401static int
402do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
403{
404 return 0;
405}
406#endif /* CONFIG_MMU */
407
408/*
409 * First Level Translation Fault Handler
410 *
411 * We enter here because the first level page table doesn't contain
412 * a valid entry for the address.
413 *
414 * If the address is in kernel space (>= TASK_SIZE), then we are
415 * probably faulting in the vmalloc() area.
416 *
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.
420 *
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.
424 */
425#ifdef CONFIG_MMU
426static int __kprobes
427do_translation_fault(unsigned long addr, unsigned int fsr,
428 struct pt_regs *regs)
429{
430 unsigned int index;
431 pgd_t *pgd, *pgd_k;
432 p4d_t *p4d, *p4d_k;
433 pud_t *pud, *pud_k;
434 pmd_t *pmd, *pmd_k;
435
436 if (addr < TASK_SIZE)
437 return do_page_fault(addr, fsr, regs);
438
439 if (user_mode(regs))
440 goto bad_area;
441
442 index = pgd_index(addr);
443
444 pgd = cpu_get_pgd() + index;
445 pgd_k = init_mm.pgd + index;
446
447 p4d = p4d_offset(pgd, addr);
448 p4d_k = p4d_offset(pgd_k, addr);
449
450 if (p4d_none(*p4d_k))
451 goto bad_area;
452 if (!p4d_present(*p4d))
453 set_p4d(p4d, *p4d_k);
454
455 pud = pud_offset(p4d, addr);
456 pud_k = pud_offset(p4d_k, addr);
457
458 if (pud_none(*pud_k))
459 goto bad_area;
460 if (!pud_present(*pud))
461 set_pud(pud, *pud_k);
462
463 pmd = pmd_offset(pud, addr);
464 pmd_k = pmd_offset(pud_k, addr);
465
466#ifdef CONFIG_ARM_LPAE
467 /*
468 * Only one hardware entry per PMD with LPAE.
469 */
470 index = 0;
471#else
472 /*
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.
479 */
480 index = (addr >> SECTION_SHIFT) & 1;
481#endif
482 if (pmd_none(pmd_k[index]))
483 goto bad_area;
484
485 copy_pmd(pmd, pmd_k);
486 return 0;
487
488bad_area:
489 do_bad_area(addr, fsr, regs);
490 return 0;
491}
492#else /* CONFIG_MMU */
493static int
494do_translation_fault(unsigned long addr, unsigned int fsr,
495 struct pt_regs *regs)
496{
497 return 0;
498}
499#endif /* CONFIG_MMU */
500
501/*
502 * Some section permission faults need to be handled gracefully.
503 * They can happen due to a __{get,put}_user during an oops.
504 */
505#ifndef CONFIG_ARM_LPAE
506static int
507do_sect_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
508{
509 do_bad_area(addr, fsr, regs);
510 return 0;
511}
512#endif /* CONFIG_ARM_LPAE */
513
514/*
515 * This abort handler always returns "fault".
516 */
517static int
518do_bad(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
519{
520 return 1;
521}
522
523struct fsr_info {
524 int (*fn)(unsigned long addr, unsigned int fsr, struct pt_regs *regs);
525 int sig;
526 int code;
527 const char *name;
528};
529
530/* FSR definition */
531#ifdef CONFIG_ARM_LPAE
532#include "fsr-3level.c"
533#else
534#include "fsr-2level.c"
535#endif
536
537void __init
538hook_fault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
539 int sig, int code, const char *name)
540{
541 if (nr < 0 || nr >= ARRAY_SIZE(fsr_info))
542 BUG();
543
544 fsr_info[nr].fn = fn;
545 fsr_info[nr].sig = sig;
546 fsr_info[nr].code = code;
547 fsr_info[nr].name = name;
548}
549
550/*
551 * Dispatch a data abort to the relevant handler.
552 */
553asmlinkage void
554do_DataAbort(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
555{
556 const struct fsr_info *inf = fsr_info + fsr_fs(fsr);
557
558 if (!inf->fn(addr, fsr & ~FSR_LNX_PF, regs))
559 return;
560
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);
565
566 arm_notify_die("", regs, inf->sig, inf->code, (void __user *)addr,
567 fsr, 0);
568}
569
570void __init
571hook_ifault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
572 int sig, int code, const char *name)
573{
574 if (nr < 0 || nr >= ARRAY_SIZE(ifsr_info))
575 BUG();
576
577 ifsr_info[nr].fn = fn;
578 ifsr_info[nr].sig = sig;
579 ifsr_info[nr].code = code;
580 ifsr_info[nr].name = name;
581}
582
583asmlinkage void
584do_PrefetchAbort(unsigned long addr, unsigned int ifsr, struct pt_regs *regs)
585{
586 const struct fsr_info *inf = ifsr_info + fsr_fs(ifsr);
587
588 if (!inf->fn(addr, ifsr | FSR_LNX_PF, regs))
589 return;
590
591 pr_alert("Unhandled prefetch abort: %s (0x%03x) at 0x%08lx\n",
592 inf->name, ifsr, addr);
593
594 arm_notify_die("", regs, inf->sig, inf->code, (void __user *)addr,
595 ifsr, 0);
596}
597
598/*
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.
602 */
603static int __init early_abort_handler(unsigned long addr, unsigned int fsr,
604 struct pt_regs *regs)
605{
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);
609
610 return 0;
611}
612
613void __init early_abt_enable(void)
614{
615 fsr_info[FSR_FS_AEA].fn = early_abort_handler;
616 local_abt_enable();
617 fsr_info[FSR_FS_AEA].fn = do_bad;
618}
619
620#ifndef CONFIG_ARM_LPAE
621static int __init exceptions_init(void)
622{
623 if (cpu_architecture() >= CPU_ARCH_ARMv6) {
624 hook_fault_code(4, do_translation_fault, SIGSEGV, SEGV_MAPERR,
625 "I-cache maintenance fault");
626 }
627
628 if (cpu_architecture() >= CPU_ARCH_ARMv7) {
629 /*
630 * TODO: Access flag faults introduced in ARMv6K.
631 * Runtime check for 'K' extension is needed
632 */
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");
637 }
638
639 return 0;
640}
641
642arch_initcall(exceptions_init);
643#endif