2 * Copyright (C) 1995 Linus Torvalds
3 * Copyright (C) 2001, 2002 Andi Kleen, SuSE Labs.
4 * Copyright (C) 2008-2009, Red Hat Inc., Ingo Molnar
6 #include <linux/magic.h> /* STACK_END_MAGIC */
7 #include <linux/sched.h> /* test_thread_flag(), ... */
8 #include <linux/kdebug.h> /* oops_begin/end, ... */
9 #include <linux/module.h> /* search_exception_table */
10 #include <linux/bootmem.h> /* max_low_pfn */
11 #include <linux/kprobes.h> /* __kprobes, ... */
12 #include <linux/mmiotrace.h> /* kmmio_handler, ... */
13 #include <linux/perf_event.h> /* perf_sw_event */
15 #include <asm/traps.h> /* dotraplinkage, ... */
16 #include <asm/pgalloc.h> /* pgd_*(), ... */
17 #include <asm/kmemcheck.h> /* kmemcheck_*(), ... */
20 * Page fault error code bits:
22 * bit 0 == 0: no page found 1: protection fault
23 * bit 1 == 0: read access 1: write access
24 * bit 2 == 0: kernel-mode access 1: user-mode access
25 * bit 3 == 1: use of reserved bit detected
26 * bit 4 == 1: fault was an instruction fetch
28 enum x86_pf_error_code
{
38 * Returns 0 if mmiotrace is disabled, or if the fault is not
39 * handled by mmiotrace:
41 static inline int __kprobes
42 kmmio_fault(struct pt_regs
*regs
, unsigned long addr
)
44 if (unlikely(is_kmmio_active()))
45 if (kmmio_handler(regs
, addr
) == 1)
50 static inline int __kprobes
notify_page_fault(struct pt_regs
*regs
)
54 /* kprobe_running() needs smp_processor_id() */
55 if (kprobes_built_in() && !user_mode_vm(regs
)) {
57 if (kprobe_running() && kprobe_fault_handler(regs
, 14))
70 * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch.
71 * Check that here and ignore it.
75 * Sometimes the CPU reports invalid exceptions on prefetch.
76 * Check that here and ignore it.
78 * Opcode checker based on code by Richard Brunner.
81 check_prefetch_opcode(struct pt_regs
*regs
, unsigned char *instr
,
82 unsigned char opcode
, int *prefetch
)
84 unsigned char instr_hi
= opcode
& 0xf0;
85 unsigned char instr_lo
= opcode
& 0x0f;
91 * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes.
92 * In X86_64 long mode, the CPU will signal invalid
93 * opcode if some of these prefixes are present so
94 * X86_64 will never get here anyway
96 return ((instr_lo
& 7) == 0x6);
100 * In AMD64 long mode 0x40..0x4F are valid REX prefixes
101 * Need to figure out under what instruction mode the
102 * instruction was issued. Could check the LDT for lm,
103 * but for now it's good enough to assume that long
104 * mode only uses well known segments or kernel.
106 return (!user_mode(regs
)) || (regs
->cs
== __USER_CS
);
109 /* 0x64 thru 0x67 are valid prefixes in all modes. */
110 return (instr_lo
& 0xC) == 0x4;
112 /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */
113 return !instr_lo
|| (instr_lo
>>1) == 1;
115 /* Prefetch instruction is 0x0F0D or 0x0F18 */
116 if (probe_kernel_address(instr
, opcode
))
119 *prefetch
= (instr_lo
== 0xF) &&
120 (opcode
== 0x0D || opcode
== 0x18);
128 is_prefetch(struct pt_regs
*regs
, unsigned long error_code
, unsigned long addr
)
130 unsigned char *max_instr
;
131 unsigned char *instr
;
135 * If it was a exec (instruction fetch) fault on NX page, then
136 * do not ignore the fault:
138 if (error_code
& PF_INSTR
)
141 instr
= (void *)convert_ip_to_linear(current
, regs
);
142 max_instr
= instr
+ 15;
144 if (user_mode(regs
) && instr
>= (unsigned char *)TASK_SIZE
)
147 while (instr
< max_instr
) {
148 unsigned char opcode
;
150 if (probe_kernel_address(instr
, opcode
))
155 if (!check_prefetch_opcode(regs
, instr
, opcode
, &prefetch
))
162 force_sig_info_fault(int si_signo
, int si_code
, unsigned long address
,
163 struct task_struct
*tsk
)
167 info
.si_signo
= si_signo
;
169 info
.si_code
= si_code
;
170 info
.si_addr
= (void __user
*)address
;
171 info
.si_addr_lsb
= si_code
== BUS_MCEERR_AR
? PAGE_SHIFT
: 0;
173 force_sig_info(si_signo
, &info
, tsk
);
176 DEFINE_SPINLOCK(pgd_lock
);
180 static inline pmd_t
*vmalloc_sync_one(pgd_t
*pgd
, unsigned long address
)
182 unsigned index
= pgd_index(address
);
188 pgd_k
= init_mm
.pgd
+ index
;
190 if (!pgd_present(*pgd_k
))
194 * set_pgd(pgd, *pgd_k); here would be useless on PAE
195 * and redundant with the set_pmd() on non-PAE. As would
198 pud
= pud_offset(pgd
, address
);
199 pud_k
= pud_offset(pgd_k
, address
);
200 if (!pud_present(*pud_k
))
203 pmd
= pmd_offset(pud
, address
);
204 pmd_k
= pmd_offset(pud_k
, address
);
205 if (!pmd_present(*pmd_k
))
208 if (!pmd_present(*pmd
))
209 set_pmd(pmd
, *pmd_k
);
211 BUG_ON(pmd_page(*pmd
) != pmd_page(*pmd_k
));
216 void vmalloc_sync_all(void)
218 unsigned long address
;
220 if (SHARED_KERNEL_PMD
)
223 for (address
= VMALLOC_START
& PMD_MASK
;
224 address
>= TASK_SIZE
&& address
< FIXADDR_TOP
;
225 address
+= PMD_SIZE
) {
230 spin_lock_irqsave(&pgd_lock
, flags
);
231 list_for_each_entry(page
, &pgd_list
, lru
) {
232 if (!vmalloc_sync_one(page_address(page
), address
))
235 spin_unlock_irqrestore(&pgd_lock
, flags
);
242 * Handle a fault on the vmalloc or module mapping area
244 static noinline __kprobes
int vmalloc_fault(unsigned long address
)
246 unsigned long pgd_paddr
;
250 /* Make sure we are in vmalloc area: */
251 if (!(address
>= VMALLOC_START
&& address
< VMALLOC_END
))
255 * Synchronize this task's top level page-table
256 * with the 'reference' page table.
258 * Do _not_ use "current" here. We might be inside
259 * an interrupt in the middle of a task switch..
261 pgd_paddr
= read_cr3();
262 pmd_k
= vmalloc_sync_one(__va(pgd_paddr
), address
);
266 pte_k
= pte_offset_kernel(pmd_k
, address
);
267 if (!pte_present(*pte_k
))
274 * Did it hit the DOS screen memory VA from vm86 mode?
277 check_v8086_mode(struct pt_regs
*regs
, unsigned long address
,
278 struct task_struct
*tsk
)
282 if (!v8086_mode(regs
))
285 bit
= (address
- 0xA0000) >> PAGE_SHIFT
;
287 tsk
->thread
.screen_bitmap
|= 1 << bit
;
290 static bool low_pfn(unsigned long pfn
)
292 return pfn
< max_low_pfn
;
295 static void dump_pagetable(unsigned long address
)
297 pgd_t
*base
= __va(read_cr3());
298 pgd_t
*pgd
= &base
[pgd_index(address
)];
302 #ifdef CONFIG_X86_PAE
303 printk("*pdpt = %016Lx ", pgd_val(*pgd
));
304 if (!low_pfn(pgd_val(*pgd
) >> PAGE_SHIFT
) || !pgd_present(*pgd
))
307 pmd
= pmd_offset(pud_offset(pgd
, address
), address
);
308 printk(KERN_CONT
"*pde = %0*Lx ", sizeof(*pmd
) * 2, (u64
)pmd_val(*pmd
));
311 * We must not directly access the pte in the highpte
312 * case if the page table is located in highmem.
313 * And let's rather not kmap-atomic the pte, just in case
314 * it's allocated already:
316 if (!low_pfn(pmd_pfn(*pmd
)) || !pmd_present(*pmd
) || pmd_large(*pmd
))
319 pte
= pte_offset_kernel(pmd
, address
);
320 printk("*pte = %0*Lx ", sizeof(*pte
) * 2, (u64
)pte_val(*pte
));
325 #else /* CONFIG_X86_64: */
327 void vmalloc_sync_all(void)
329 unsigned long address
;
331 for (address
= VMALLOC_START
& PGDIR_MASK
; address
<= VMALLOC_END
;
332 address
+= PGDIR_SIZE
) {
334 const pgd_t
*pgd_ref
= pgd_offset_k(address
);
338 if (pgd_none(*pgd_ref
))
341 spin_lock_irqsave(&pgd_lock
, flags
);
342 list_for_each_entry(page
, &pgd_list
, lru
) {
344 pgd
= (pgd_t
*)page_address(page
) + pgd_index(address
);
346 set_pgd(pgd
, *pgd_ref
);
348 BUG_ON(pgd_page_vaddr(*pgd
) != pgd_page_vaddr(*pgd_ref
));
350 spin_unlock_irqrestore(&pgd_lock
, flags
);
357 * Handle a fault on the vmalloc area
359 * This assumes no large pages in there.
361 static noinline __kprobes
int vmalloc_fault(unsigned long address
)
363 pgd_t
*pgd
, *pgd_ref
;
364 pud_t
*pud
, *pud_ref
;
365 pmd_t
*pmd
, *pmd_ref
;
366 pte_t
*pte
, *pte_ref
;
368 /* Make sure we are in vmalloc area: */
369 if (!(address
>= VMALLOC_START
&& address
< VMALLOC_END
))
373 * Copy kernel mappings over when needed. This can also
374 * happen within a race in page table update. In the later
377 pgd
= pgd_offset(current
->active_mm
, address
);
378 pgd_ref
= pgd_offset_k(address
);
379 if (pgd_none(*pgd_ref
))
383 set_pgd(pgd
, *pgd_ref
);
385 BUG_ON(pgd_page_vaddr(*pgd
) != pgd_page_vaddr(*pgd_ref
));
388 * Below here mismatches are bugs because these lower tables
392 pud
= pud_offset(pgd
, address
);
393 pud_ref
= pud_offset(pgd_ref
, address
);
394 if (pud_none(*pud_ref
))
397 if (pud_none(*pud
) || pud_page_vaddr(*pud
) != pud_page_vaddr(*pud_ref
))
400 pmd
= pmd_offset(pud
, address
);
401 pmd_ref
= pmd_offset(pud_ref
, address
);
402 if (pmd_none(*pmd_ref
))
405 if (pmd_none(*pmd
) || pmd_page(*pmd
) != pmd_page(*pmd_ref
))
408 pte_ref
= pte_offset_kernel(pmd_ref
, address
);
409 if (!pte_present(*pte_ref
))
412 pte
= pte_offset_kernel(pmd
, address
);
415 * Don't use pte_page here, because the mappings can point
416 * outside mem_map, and the NUMA hash lookup cannot handle
419 if (!pte_present(*pte
) || pte_pfn(*pte
) != pte_pfn(*pte_ref
))
425 static const char errata93_warning
[] =
427 "******* Your BIOS seems to not contain a fix for K8 errata #93\n"
428 "******* Working around it, but it may cause SEGVs or burn power.\n"
429 "******* Please consider a BIOS update.\n"
430 "******* Disabling USB legacy in the BIOS may also help.\n";
433 * No vm86 mode in 64-bit mode:
436 check_v8086_mode(struct pt_regs
*regs
, unsigned long address
,
437 struct task_struct
*tsk
)
441 static int bad_address(void *p
)
445 return probe_kernel_address((unsigned long *)p
, dummy
);
448 static void dump_pagetable(unsigned long address
)
450 pgd_t
*base
= __va(read_cr3() & PHYSICAL_PAGE_MASK
);
451 pgd_t
*pgd
= base
+ pgd_index(address
);
456 if (bad_address(pgd
))
459 printk("PGD %lx ", pgd_val(*pgd
));
461 if (!pgd_present(*pgd
))
464 pud
= pud_offset(pgd
, address
);
465 if (bad_address(pud
))
468 printk("PUD %lx ", pud_val(*pud
));
469 if (!pud_present(*pud
) || pud_large(*pud
))
472 pmd
= pmd_offset(pud
, address
);
473 if (bad_address(pmd
))
476 printk("PMD %lx ", pmd_val(*pmd
));
477 if (!pmd_present(*pmd
) || pmd_large(*pmd
))
480 pte
= pte_offset_kernel(pmd
, address
);
481 if (bad_address(pte
))
484 printk("PTE %lx", pte_val(*pte
));
492 #endif /* CONFIG_X86_64 */
494 static int is_errata93(struct pt_regs
*regs
, unsigned long address
)
497 if (address
!= regs
->ip
)
500 if ((address
>> 32) != 0)
503 address
|= 0xffffffffUL
<< 32;
504 if ((address
>= (u64
)_stext
&& address
<= (u64
)_etext
) ||
505 (address
>= MODULES_VADDR
&& address
<= MODULES_END
)) {
506 printk_once(errata93_warning
);
514 static int is_errata100(struct pt_regs
*regs
, unsigned long address
)
517 if ((regs
->cs
== __USER32_CS
|| (regs
->cs
& (1<<2))) && (address
>> 32))
523 static int is_f00f_bug(struct pt_regs
*regs
, unsigned long address
)
525 #ifdef CONFIG_X86_F00F_BUG
528 if (boot_cpu_data
.f00f_bug
) {
529 nr
= (address
- idt_descr
.address
) >> 3;
532 do_invalid_op(regs
, 0);
540 static const char nx_warning
[] = KERN_CRIT
541 "kernel tried to execute NX-protected page - exploit attempt? (uid: %d)\n";
544 show_fault_oops(struct pt_regs
*regs
, unsigned long error_code
,
545 unsigned long address
)
547 if (!oops_may_print())
550 if (error_code
& PF_INSTR
) {
553 pte_t
*pte
= lookup_address(address
, &level
);
555 if (pte
&& pte_present(*pte
) && !pte_exec(*pte
))
556 printk(nx_warning
, current_uid());
559 printk(KERN_ALERT
"BUG: unable to handle kernel ");
560 if (address
< PAGE_SIZE
)
561 printk(KERN_CONT
"NULL pointer dereference");
563 printk(KERN_CONT
"paging request");
565 printk(KERN_CONT
" at %p\n", (void *) address
);
566 printk(KERN_ALERT
"IP:");
567 printk_address(regs
->ip
, 1);
569 dump_pagetable(address
);
573 pgtable_bad(struct pt_regs
*regs
, unsigned long error_code
,
574 unsigned long address
)
576 struct task_struct
*tsk
;
580 flags
= oops_begin();
584 printk(KERN_ALERT
"%s: Corrupted page table at address %lx\n",
586 dump_pagetable(address
);
588 tsk
->thread
.cr2
= address
;
589 tsk
->thread
.trap_no
= 14;
590 tsk
->thread
.error_code
= error_code
;
592 if (__die("Bad pagetable", regs
, error_code
))
595 oops_end(flags
, regs
, sig
);
599 no_context(struct pt_regs
*regs
, unsigned long error_code
,
600 unsigned long address
)
602 struct task_struct
*tsk
= current
;
603 unsigned long *stackend
;
607 /* Are we prepared to handle this kernel fault? */
608 if (fixup_exception(regs
))
614 * Valid to do another page fault here, because if this fault
615 * had been triggered by is_prefetch fixup_exception would have
620 * Hall of shame of CPU/BIOS bugs.
622 if (is_prefetch(regs
, error_code
, address
))
625 if (is_errata93(regs
, address
))
629 * Oops. The kernel tried to access some bad page. We'll have to
630 * terminate things with extreme prejudice:
632 flags
= oops_begin();
634 show_fault_oops(regs
, error_code
, address
);
636 stackend
= end_of_stack(tsk
);
637 if (tsk
!= &init_task
&& *stackend
!= STACK_END_MAGIC
)
638 printk(KERN_ALERT
"Thread overran stack, or stack corrupted\n");
640 tsk
->thread
.cr2
= address
;
641 tsk
->thread
.trap_no
= 14;
642 tsk
->thread
.error_code
= error_code
;
645 if (__die("Oops", regs
, error_code
))
648 /* Executive summary in case the body of the oops scrolled away */
649 printk(KERN_EMERG
"CR2: %016lx\n", address
);
651 oops_end(flags
, regs
, sig
);
655 * Print out info about fatal segfaults, if the show_unhandled_signals
659 show_signal_msg(struct pt_regs
*regs
, unsigned long error_code
,
660 unsigned long address
, struct task_struct
*tsk
)
662 if (!unhandled_signal(tsk
, SIGSEGV
))
665 if (!printk_ratelimit())
668 printk("%s%s[%d]: segfault at %lx ip %p sp %p error %lx",
669 task_pid_nr(tsk
) > 1 ? KERN_INFO
: KERN_EMERG
,
670 tsk
->comm
, task_pid_nr(tsk
), address
,
671 (void *)regs
->ip
, (void *)regs
->sp
, error_code
);
673 print_vma_addr(KERN_CONT
" in ", regs
->ip
);
675 printk(KERN_CONT
"\n");
679 __bad_area_nosemaphore(struct pt_regs
*regs
, unsigned long error_code
,
680 unsigned long address
, int si_code
)
682 struct task_struct
*tsk
= current
;
684 /* User mode accesses just cause a SIGSEGV */
685 if (error_code
& PF_USER
) {
687 * It's possible to have interrupts off here:
692 * Valid to do another page fault here because this one came
695 if (is_prefetch(regs
, error_code
, address
))
698 if (is_errata100(regs
, address
))
701 if (unlikely(show_unhandled_signals
))
702 show_signal_msg(regs
, error_code
, address
, tsk
);
704 /* Kernel addresses are always protection faults: */
705 tsk
->thread
.cr2
= address
;
706 tsk
->thread
.error_code
= error_code
| (address
>= TASK_SIZE
);
707 tsk
->thread
.trap_no
= 14;
709 force_sig_info_fault(SIGSEGV
, si_code
, address
, tsk
);
714 if (is_f00f_bug(regs
, address
))
717 no_context(regs
, error_code
, address
);
721 bad_area_nosemaphore(struct pt_regs
*regs
, unsigned long error_code
,
722 unsigned long address
)
724 __bad_area_nosemaphore(regs
, error_code
, address
, SEGV_MAPERR
);
728 __bad_area(struct pt_regs
*regs
, unsigned long error_code
,
729 unsigned long address
, int si_code
)
731 struct mm_struct
*mm
= current
->mm
;
734 * Something tried to access memory that isn't in our memory map..
735 * Fix it, but check if it's kernel or user first..
737 up_read(&mm
->mmap_sem
);
739 __bad_area_nosemaphore(regs
, error_code
, address
, si_code
);
743 bad_area(struct pt_regs
*regs
, unsigned long error_code
, unsigned long address
)
745 __bad_area(regs
, error_code
, address
, SEGV_MAPERR
);
749 bad_area_access_error(struct pt_regs
*regs
, unsigned long error_code
,
750 unsigned long address
)
752 __bad_area(regs
, error_code
, address
, SEGV_ACCERR
);
755 /* TODO: fixup for "mm-invoke-oom-killer-from-page-fault.patch" */
757 out_of_memory(struct pt_regs
*regs
, unsigned long error_code
,
758 unsigned long address
)
761 * We ran out of memory, call the OOM killer, and return the userspace
762 * (which will retry the fault, or kill us if we got oom-killed):
764 up_read(¤t
->mm
->mmap_sem
);
766 pagefault_out_of_memory();
770 do_sigbus(struct pt_regs
*regs
, unsigned long error_code
, unsigned long address
,
773 struct task_struct
*tsk
= current
;
774 struct mm_struct
*mm
= tsk
->mm
;
775 int code
= BUS_ADRERR
;
777 up_read(&mm
->mmap_sem
);
779 /* Kernel mode? Handle exceptions or die: */
780 if (!(error_code
& PF_USER
)) {
781 no_context(regs
, error_code
, address
);
785 /* User-space => ok to do another page fault: */
786 if (is_prefetch(regs
, error_code
, address
))
789 tsk
->thread
.cr2
= address
;
790 tsk
->thread
.error_code
= error_code
;
791 tsk
->thread
.trap_no
= 14;
793 #ifdef CONFIG_MEMORY_FAILURE
794 if (fault
& VM_FAULT_HWPOISON
) {
796 "MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n",
797 tsk
->comm
, tsk
->pid
, address
);
798 code
= BUS_MCEERR_AR
;
801 force_sig_info_fault(SIGBUS
, code
, address
, tsk
);
805 mm_fault_error(struct pt_regs
*regs
, unsigned long error_code
,
806 unsigned long address
, unsigned int fault
)
808 if (fault
& VM_FAULT_OOM
) {
809 out_of_memory(regs
, error_code
, address
);
811 if (fault
& (VM_FAULT_SIGBUS
|VM_FAULT_HWPOISON
))
812 do_sigbus(regs
, error_code
, address
, fault
);
818 static int spurious_fault_check(unsigned long error_code
, pte_t
*pte
)
820 if ((error_code
& PF_WRITE
) && !pte_write(*pte
))
823 if ((error_code
& PF_INSTR
) && !pte_exec(*pte
))
830 * Handle a spurious fault caused by a stale TLB entry.
832 * This allows us to lazily refresh the TLB when increasing the
833 * permissions of a kernel page (RO -> RW or NX -> X). Doing it
834 * eagerly is very expensive since that implies doing a full
835 * cross-processor TLB flush, even if no stale TLB entries exist
836 * on other processors.
838 * There are no security implications to leaving a stale TLB when
839 * increasing the permissions on a page.
841 static noinline __kprobes
int
842 spurious_fault(unsigned long error_code
, unsigned long address
)
850 /* Reserved-bit violation or user access to kernel space? */
851 if (error_code
& (PF_USER
| PF_RSVD
))
854 pgd
= init_mm
.pgd
+ pgd_index(address
);
855 if (!pgd_present(*pgd
))
858 pud
= pud_offset(pgd
, address
);
859 if (!pud_present(*pud
))
863 return spurious_fault_check(error_code
, (pte_t
*) pud
);
865 pmd
= pmd_offset(pud
, address
);
866 if (!pmd_present(*pmd
))
870 return spurious_fault_check(error_code
, (pte_t
*) pmd
);
872 pte
= pte_offset_kernel(pmd
, address
);
873 if (!pte_present(*pte
))
876 ret
= spurious_fault_check(error_code
, pte
);
881 * Make sure we have permissions in PMD.
882 * If not, then there's a bug in the page tables:
884 ret
= spurious_fault_check(error_code
, (pte_t
*) pmd
);
885 WARN_ONCE(!ret
, "PMD has incorrect permission bits\n");
890 int show_unhandled_signals
= 1;
893 access_error(unsigned long error_code
, int write
, struct vm_area_struct
*vma
)
896 /* write, present and write, not present: */
897 if (unlikely(!(vma
->vm_flags
& VM_WRITE
)))
903 if (unlikely(error_code
& PF_PROT
))
906 /* read, not present: */
907 if (unlikely(!(vma
->vm_flags
& (VM_READ
| VM_EXEC
| VM_WRITE
))))
913 static int fault_in_kernel_space(unsigned long address
)
915 return address
>= TASK_SIZE_MAX
;
919 * This routine handles page faults. It determines the address,
920 * and the problem, and then passes it off to one of the appropriate
923 dotraplinkage
void __kprobes
924 do_page_fault(struct pt_regs
*regs
, unsigned long error_code
)
926 struct vm_area_struct
*vma
;
927 struct task_struct
*tsk
;
928 unsigned long address
;
929 struct mm_struct
*mm
;
936 /* Get the faulting address: */
937 address
= read_cr2();
940 * Detect and handle instructions that would cause a page fault for
941 * both a tracked kernel page and a userspace page.
943 if (kmemcheck_active(regs
))
944 kmemcheck_hide(regs
);
945 prefetchw(&mm
->mmap_sem
);
947 if (unlikely(kmmio_fault(regs
, address
)))
951 * We fault-in kernel-space virtual memory on-demand. The
952 * 'reference' page table is init_mm.pgd.
954 * NOTE! We MUST NOT take any locks for this case. We may
955 * be in an interrupt or a critical region, and should
956 * only copy the information from the master page table,
959 * This verifies that the fault happens in kernel space
960 * (error_code & 4) == 0, and that the fault was not a
961 * protection error (error_code & 9) == 0.
963 if (unlikely(fault_in_kernel_space(address
))) {
964 if (!(error_code
& (PF_RSVD
| PF_USER
| PF_PROT
))) {
965 if (vmalloc_fault(address
) >= 0)
968 if (kmemcheck_fault(regs
, address
, error_code
))
972 /* Can handle a stale RO->RW TLB: */
973 if (spurious_fault(error_code
, address
))
976 /* kprobes don't want to hook the spurious faults: */
977 if (notify_page_fault(regs
))
980 * Don't take the mm semaphore here. If we fixup a prefetch
981 * fault we could otherwise deadlock:
983 bad_area_nosemaphore(regs
, error_code
, address
);
988 /* kprobes don't want to hook the spurious faults: */
989 if (unlikely(notify_page_fault(regs
)))
992 * It's safe to allow irq's after cr2 has been saved and the
993 * vmalloc fault has been handled.
995 * User-mode registers count as a user access even for any
996 * potential system fault or CPU buglet:
998 if (user_mode_vm(regs
)) {
1000 error_code
|= PF_USER
;
1002 if (regs
->flags
& X86_EFLAGS_IF
)
1006 if (unlikely(error_code
& PF_RSVD
))
1007 pgtable_bad(regs
, error_code
, address
);
1009 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS
, 1, 0, regs
, address
);
1012 * If we're in an interrupt, have no user context or are running
1013 * in an atomic region then we must not take the fault:
1015 if (unlikely(in_atomic() || !mm
)) {
1016 bad_area_nosemaphore(regs
, error_code
, address
);
1021 * When running in the kernel we expect faults to occur only to
1022 * addresses in user space. All other faults represent errors in
1023 * the kernel and should generate an OOPS. Unfortunately, in the
1024 * case of an erroneous fault occurring in a code path which already
1025 * holds mmap_sem we will deadlock attempting to validate the fault
1026 * against the address space. Luckily the kernel only validly
1027 * references user space from well defined areas of code, which are
1028 * listed in the exceptions table.
1030 * As the vast majority of faults will be valid we will only perform
1031 * the source reference check when there is a possibility of a
1032 * deadlock. Attempt to lock the address space, if we cannot we then
1033 * validate the source. If this is invalid we can skip the address
1034 * space check, thus avoiding the deadlock:
1036 if (unlikely(!down_read_trylock(&mm
->mmap_sem
))) {
1037 if ((error_code
& PF_USER
) == 0 &&
1038 !search_exception_tables(regs
->ip
)) {
1039 bad_area_nosemaphore(regs
, error_code
, address
);
1042 down_read(&mm
->mmap_sem
);
1045 * The above down_read_trylock() might have succeeded in
1046 * which case we'll have missed the might_sleep() from
1052 vma
= find_vma(mm
, address
);
1053 if (unlikely(!vma
)) {
1054 bad_area(regs
, error_code
, address
);
1057 if (likely(vma
->vm_start
<= address
))
1059 if (unlikely(!(vma
->vm_flags
& VM_GROWSDOWN
))) {
1060 bad_area(regs
, error_code
, address
);
1063 if (error_code
& PF_USER
) {
1065 * Accessing the stack below %sp is always a bug.
1066 * The large cushion allows instructions like enter
1067 * and pusha to work. ("enter $65535, $31" pushes
1068 * 32 pointers and then decrements %sp by 65535.)
1070 if (unlikely(address
+ 65536 + 32 * sizeof(unsigned long) < regs
->sp
)) {
1071 bad_area(regs
, error_code
, address
);
1075 if (unlikely(expand_stack(vma
, address
))) {
1076 bad_area(regs
, error_code
, address
);
1081 * Ok, we have a good vm_area for this memory access, so
1082 * we can handle it..
1085 write
= error_code
& PF_WRITE
;
1087 if (unlikely(access_error(error_code
, write
, vma
))) {
1088 bad_area_access_error(regs
, error_code
, address
);
1093 * If for any reason at all we couldn't handle the fault,
1094 * make sure we exit gracefully rather than endlessly redo
1097 fault
= handle_mm_fault(mm
, vma
, address
, write
? FAULT_FLAG_WRITE
: 0);
1099 if (unlikely(fault
& VM_FAULT_ERROR
)) {
1100 mm_fault_error(regs
, error_code
, address
, fault
);
1104 if (fault
& VM_FAULT_MAJOR
) {
1106 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ
, 1, 0,
1110 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN
, 1, 0,
1114 check_v8086_mode(regs
, address
, tsk
);
1116 up_read(&mm
->mmap_sem
);