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_counter.h> /* perf_swcounter_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 kmmio_fault(struct pt_regs
*regs
, unsigned long addr
)
43 if (unlikely(is_kmmio_active()))
44 if (kmmio_handler(regs
, addr
) == 1)
49 static inline int notify_page_fault(struct pt_regs
*regs
)
53 /* kprobe_running() needs smp_processor_id() */
54 if (kprobes_built_in() && !user_mode_vm(regs
)) {
56 if (kprobe_running() && kprobe_fault_handler(regs
, 14))
69 * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch.
70 * Check that here and ignore it.
74 * Sometimes the CPU reports invalid exceptions on prefetch.
75 * Check that here and ignore it.
77 * Opcode checker based on code by Richard Brunner.
80 check_prefetch_opcode(struct pt_regs
*regs
, unsigned char *instr
,
81 unsigned char opcode
, int *prefetch
)
83 unsigned char instr_hi
= opcode
& 0xf0;
84 unsigned char instr_lo
= opcode
& 0x0f;
90 * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes.
91 * In X86_64 long mode, the CPU will signal invalid
92 * opcode if some of these prefixes are present so
93 * X86_64 will never get here anyway
95 return ((instr_lo
& 7) == 0x6);
99 * In AMD64 long mode 0x40..0x4F are valid REX prefixes
100 * Need to figure out under what instruction mode the
101 * instruction was issued. Could check the LDT for lm,
102 * but for now it's good enough to assume that long
103 * mode only uses well known segments or kernel.
105 return (!user_mode(regs
)) || (regs
->cs
== __USER_CS
);
108 /* 0x64 thru 0x67 are valid prefixes in all modes. */
109 return (instr_lo
& 0xC) == 0x4;
111 /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */
112 return !instr_lo
|| (instr_lo
>>1) == 1;
114 /* Prefetch instruction is 0x0F0D or 0x0F18 */
115 if (probe_kernel_address(instr
, opcode
))
118 *prefetch
= (instr_lo
== 0xF) &&
119 (opcode
== 0x0D || opcode
== 0x18);
127 is_prefetch(struct pt_regs
*regs
, unsigned long error_code
, unsigned long addr
)
129 unsigned char *max_instr
;
130 unsigned char *instr
;
134 * If it was a exec (instruction fetch) fault on NX page, then
135 * do not ignore the fault:
137 if (error_code
& PF_INSTR
)
140 instr
= (void *)convert_ip_to_linear(current
, regs
);
141 max_instr
= instr
+ 15;
143 if (user_mode(regs
) && instr
>= (unsigned char *)TASK_SIZE
)
146 while (instr
< max_instr
) {
147 unsigned char opcode
;
149 if (probe_kernel_address(instr
, opcode
))
154 if (!check_prefetch_opcode(regs
, instr
, opcode
, &prefetch
))
161 force_sig_info_fault(int si_signo
, int si_code
, unsigned long address
,
162 struct task_struct
*tsk
)
166 info
.si_signo
= si_signo
;
168 info
.si_code
= si_code
;
169 info
.si_addr
= (void __user
*)address
;
171 force_sig_info(si_signo
, &info
, tsk
);
174 DEFINE_SPINLOCK(pgd_lock
);
178 static inline pmd_t
*vmalloc_sync_one(pgd_t
*pgd
, unsigned long address
)
180 unsigned index
= pgd_index(address
);
186 pgd_k
= init_mm
.pgd
+ index
;
188 if (!pgd_present(*pgd_k
))
192 * set_pgd(pgd, *pgd_k); here would be useless on PAE
193 * and redundant with the set_pmd() on non-PAE. As would
196 pud
= pud_offset(pgd
, address
);
197 pud_k
= pud_offset(pgd_k
, address
);
198 if (!pud_present(*pud_k
))
201 pmd
= pmd_offset(pud
, address
);
202 pmd_k
= pmd_offset(pud_k
, address
);
203 if (!pmd_present(*pmd_k
))
206 if (!pmd_present(*pmd
))
207 set_pmd(pmd
, *pmd_k
);
209 BUG_ON(pmd_page(*pmd
) != pmd_page(*pmd_k
));
214 void vmalloc_sync_all(void)
216 unsigned long address
;
218 if (SHARED_KERNEL_PMD
)
221 for (address
= VMALLOC_START
& PMD_MASK
;
222 address
>= TASK_SIZE
&& address
< FIXADDR_TOP
;
223 address
+= PMD_SIZE
) {
228 spin_lock_irqsave(&pgd_lock
, flags
);
229 list_for_each_entry(page
, &pgd_list
, lru
) {
230 if (!vmalloc_sync_one(page_address(page
), address
))
233 spin_unlock_irqrestore(&pgd_lock
, flags
);
240 * Handle a fault on the vmalloc or module mapping area
242 static noinline
int vmalloc_fault(unsigned long address
)
244 unsigned long pgd_paddr
;
248 /* Make sure we are in vmalloc area: */
249 if (!(address
>= VMALLOC_START
&& address
< VMALLOC_END
))
253 * Synchronize this task's top level page-table
254 * with the 'reference' page table.
256 * Do _not_ use "current" here. We might be inside
257 * an interrupt in the middle of a task switch..
259 pgd_paddr
= read_cr3();
260 pmd_k
= vmalloc_sync_one(__va(pgd_paddr
), address
);
264 pte_k
= pte_offset_kernel(pmd_k
, address
);
265 if (!pte_present(*pte_k
))
272 * Did it hit the DOS screen memory VA from vm86 mode?
275 check_v8086_mode(struct pt_regs
*regs
, unsigned long address
,
276 struct task_struct
*tsk
)
280 if (!v8086_mode(regs
))
283 bit
= (address
- 0xA0000) >> PAGE_SHIFT
;
285 tsk
->thread
.screen_bitmap
|= 1 << bit
;
288 static void dump_pagetable(unsigned long address
)
290 __typeof__(pte_val(__pte(0))) page
;
293 page
= ((__typeof__(page
) *) __va(page
))[address
>> PGDIR_SHIFT
];
295 #ifdef CONFIG_X86_PAE
296 printk("*pdpt = %016Lx ", page
);
297 if ((page
>> PAGE_SHIFT
) < max_low_pfn
298 && page
& _PAGE_PRESENT
) {
300 page
= ((__typeof__(page
) *) __va(page
))[(address
>> PMD_SHIFT
)
301 & (PTRS_PER_PMD
- 1)];
302 printk(KERN_CONT
"*pde = %016Lx ", page
);
306 printk("*pde = %08lx ", page
);
310 * We must not directly access the pte in the highpte
311 * case if the page table is located in highmem.
312 * And let's rather not kmap-atomic the pte, just in case
313 * it's allocated already:
315 if ((page
>> PAGE_SHIFT
) < max_low_pfn
316 && (page
& _PAGE_PRESENT
)
317 && !(page
& _PAGE_PSE
)) {
320 page
= ((__typeof__(page
) *) __va(page
))[(address
>> PAGE_SHIFT
)
321 & (PTRS_PER_PTE
- 1)];
322 printk("*pte = %0*Lx ", sizeof(page
)*2, (u64
)page
);
328 #else /* CONFIG_X86_64: */
330 void vmalloc_sync_all(void)
332 unsigned long address
;
334 for (address
= VMALLOC_START
& PGDIR_MASK
; address
<= VMALLOC_END
;
335 address
+= PGDIR_SIZE
) {
337 const pgd_t
*pgd_ref
= pgd_offset_k(address
);
341 if (pgd_none(*pgd_ref
))
344 spin_lock_irqsave(&pgd_lock
, flags
);
345 list_for_each_entry(page
, &pgd_list
, lru
) {
347 pgd
= (pgd_t
*)page_address(page
) + pgd_index(address
);
349 set_pgd(pgd
, *pgd_ref
);
351 BUG_ON(pgd_page_vaddr(*pgd
) != pgd_page_vaddr(*pgd_ref
));
353 spin_unlock_irqrestore(&pgd_lock
, flags
);
360 * Handle a fault on the vmalloc area
362 * This assumes no large pages in there.
364 static noinline
int vmalloc_fault(unsigned long address
)
366 pgd_t
*pgd
, *pgd_ref
;
367 pud_t
*pud
, *pud_ref
;
368 pmd_t
*pmd
, *pmd_ref
;
369 pte_t
*pte
, *pte_ref
;
371 /* Make sure we are in vmalloc area: */
372 if (!(address
>= VMALLOC_START
&& address
< VMALLOC_END
))
376 * Copy kernel mappings over when needed. This can also
377 * happen within a race in page table update. In the later
380 pgd
= pgd_offset(current
->active_mm
, address
);
381 pgd_ref
= pgd_offset_k(address
);
382 if (pgd_none(*pgd_ref
))
386 set_pgd(pgd
, *pgd_ref
);
388 BUG_ON(pgd_page_vaddr(*pgd
) != pgd_page_vaddr(*pgd_ref
));
391 * Below here mismatches are bugs because these lower tables
395 pud
= pud_offset(pgd
, address
);
396 pud_ref
= pud_offset(pgd_ref
, address
);
397 if (pud_none(*pud_ref
))
400 if (pud_none(*pud
) || pud_page_vaddr(*pud
) != pud_page_vaddr(*pud_ref
))
403 pmd
= pmd_offset(pud
, address
);
404 pmd_ref
= pmd_offset(pud_ref
, address
);
405 if (pmd_none(*pmd_ref
))
408 if (pmd_none(*pmd
) || pmd_page(*pmd
) != pmd_page(*pmd_ref
))
411 pte_ref
= pte_offset_kernel(pmd_ref
, address
);
412 if (!pte_present(*pte_ref
))
415 pte
= pte_offset_kernel(pmd
, address
);
418 * Don't use pte_page here, because the mappings can point
419 * outside mem_map, and the NUMA hash lookup cannot handle
422 if (!pte_present(*pte
) || pte_pfn(*pte
) != pte_pfn(*pte_ref
))
428 static const char errata93_warning
[] =
429 KERN_ERR
"******* Your BIOS seems to not contain a fix for K8 errata #93\n"
430 KERN_ERR
"******* Working around it, but it may cause SEGVs or burn power.\n"
431 KERN_ERR
"******* Please consider a BIOS update.\n"
432 KERN_ERR
"******* Disabling USB legacy in the BIOS may also help.\n";
435 * No vm86 mode in 64-bit mode:
438 check_v8086_mode(struct pt_regs
*regs
, unsigned long address
,
439 struct task_struct
*tsk
)
443 static int bad_address(void *p
)
447 return probe_kernel_address((unsigned long *)p
, dummy
);
450 static void dump_pagetable(unsigned long address
)
457 pgd
= (pgd_t
*)read_cr3();
459 pgd
= __va((unsigned long)pgd
& PHYSICAL_PAGE_MASK
);
461 pgd
+= pgd_index(address
);
462 if (bad_address(pgd
))
465 printk("PGD %lx ", pgd_val(*pgd
));
467 if (!pgd_present(*pgd
))
470 pud
= pud_offset(pgd
, address
);
471 if (bad_address(pud
))
474 printk("PUD %lx ", pud_val(*pud
));
475 if (!pud_present(*pud
) || pud_large(*pud
))
478 pmd
= pmd_offset(pud
, address
);
479 if (bad_address(pmd
))
482 printk("PMD %lx ", pmd_val(*pmd
));
483 if (!pmd_present(*pmd
) || pmd_large(*pmd
))
486 pte
= pte_offset_kernel(pmd
, address
);
487 if (bad_address(pte
))
490 printk("PTE %lx", pte_val(*pte
));
498 #endif /* CONFIG_X86_64 */
501 * Workaround for K8 erratum #93 & buggy BIOS.
503 * BIOS SMM functions are required to use a specific workaround
504 * to avoid corruption of the 64bit RIP register on C stepping K8.
506 * A lot of BIOS that didn't get tested properly miss this.
508 * The OS sees this as a page fault with the upper 32bits of RIP cleared.
509 * Try to work around it here.
511 * Note we only handle faults in kernel here.
512 * Does nothing on 32-bit.
514 static int is_errata93(struct pt_regs
*regs
, unsigned long address
)
517 if (address
!= regs
->ip
)
520 if ((address
>> 32) != 0)
523 address
|= 0xffffffffUL
<< 32;
524 if ((address
>= (u64
)_stext
&& address
<= (u64
)_etext
) ||
525 (address
>= MODULES_VADDR
&& address
<= MODULES_END
)) {
526 printk_once(errata93_warning
);
535 * Work around K8 erratum #100 K8 in compat mode occasionally jumps
536 * to illegal addresses >4GB.
538 * We catch this in the page fault handler because these addresses
539 * are not reachable. Just detect this case and return. Any code
540 * segment in LDT is compatibility mode.
542 static int is_errata100(struct pt_regs
*regs
, unsigned long address
)
545 if ((regs
->cs
== __USER32_CS
|| (regs
->cs
& (1<<2))) && (address
>> 32))
551 static int is_f00f_bug(struct pt_regs
*regs
, unsigned long address
)
553 #ifdef CONFIG_X86_F00F_BUG
557 * Pentium F0 0F C7 C8 bug workaround:
559 if (boot_cpu_data
.f00f_bug
) {
560 nr
= (address
- idt_descr
.address
) >> 3;
563 do_invalid_op(regs
, 0);
571 static const char nx_warning
[] = KERN_CRIT
572 "kernel tried to execute NX-protected page - exploit attempt? (uid: %d)\n";
575 show_fault_oops(struct pt_regs
*regs
, unsigned long error_code
,
576 unsigned long address
)
578 if (!oops_may_print())
581 if (error_code
& PF_INSTR
) {
584 pte_t
*pte
= lookup_address(address
, &level
);
586 if (pte
&& pte_present(*pte
) && !pte_exec(*pte
))
587 printk(nx_warning
, current_uid());
590 printk(KERN_ALERT
"BUG: unable to handle kernel ");
591 if (address
< PAGE_SIZE
)
592 printk(KERN_CONT
"NULL pointer dereference");
594 printk(KERN_CONT
"paging request");
596 printk(KERN_CONT
" at %p\n", (void *) address
);
597 printk(KERN_ALERT
"IP:");
598 printk_address(regs
->ip
, 1);
600 dump_pagetable(address
);
604 pgtable_bad(struct pt_regs
*regs
, unsigned long error_code
,
605 unsigned long address
)
607 struct task_struct
*tsk
;
611 flags
= oops_begin();
615 printk(KERN_ALERT
"%s: Corrupted page table at address %lx\n",
617 dump_pagetable(address
);
619 tsk
->thread
.cr2
= address
;
620 tsk
->thread
.trap_no
= 14;
621 tsk
->thread
.error_code
= error_code
;
623 if (__die("Bad pagetable", regs
, error_code
))
626 oops_end(flags
, regs
, sig
);
630 no_context(struct pt_regs
*regs
, unsigned long error_code
,
631 unsigned long address
)
633 struct task_struct
*tsk
= current
;
634 unsigned long *stackend
;
638 /* Are we prepared to handle this kernel fault? */
639 if (fixup_exception(regs
))
645 * Valid to do another page fault here, because if this fault
646 * had been triggered by is_prefetch fixup_exception would have
651 * Hall of shame of CPU/BIOS bugs.
653 if (is_prefetch(regs
, error_code
, address
))
656 if (is_errata93(regs
, address
))
660 * Oops. The kernel tried to access some bad page. We'll have to
661 * terminate things with extreme prejudice:
663 flags
= oops_begin();
665 show_fault_oops(regs
, error_code
, address
);
667 stackend
= end_of_stack(tsk
);
668 if (*stackend
!= STACK_END_MAGIC
)
669 printk(KERN_ALERT
"Thread overran stack, or stack corrupted\n");
671 tsk
->thread
.cr2
= address
;
672 tsk
->thread
.trap_no
= 14;
673 tsk
->thread
.error_code
= error_code
;
676 if (__die("Oops", regs
, error_code
))
679 /* Executive summary in case the body of the oops scrolled away */
680 printk(KERN_EMERG
"CR2: %016lx\n", address
);
682 oops_end(flags
, regs
, sig
);
686 * Print out info about fatal segfaults, if the show_unhandled_signals
690 show_signal_msg(struct pt_regs
*regs
, unsigned long error_code
,
691 unsigned long address
, struct task_struct
*tsk
)
693 if (!unhandled_signal(tsk
, SIGSEGV
))
696 if (!printk_ratelimit())
699 printk(KERN_CONT
"%s%s[%d]: segfault at %lx ip %p sp %p error %lx",
700 task_pid_nr(tsk
) > 1 ? KERN_INFO
: KERN_EMERG
,
701 tsk
->comm
, task_pid_nr(tsk
), address
,
702 (void *)regs
->ip
, (void *)regs
->sp
, error_code
);
704 print_vma_addr(KERN_CONT
" in ", regs
->ip
);
706 printk(KERN_CONT
"\n");
710 __bad_area_nosemaphore(struct pt_regs
*regs
, unsigned long error_code
,
711 unsigned long address
, int si_code
)
713 struct task_struct
*tsk
= current
;
715 /* User mode accesses just cause a SIGSEGV */
716 if (error_code
& PF_USER
) {
718 * It's possible to have interrupts off here:
723 * Valid to do another page fault here because this one came
726 if (is_prefetch(regs
, error_code
, address
))
729 if (is_errata100(regs
, address
))
732 if (unlikely(show_unhandled_signals
))
733 show_signal_msg(regs
, error_code
, address
, tsk
);
735 /* Kernel addresses are always protection faults: */
736 tsk
->thread
.cr2
= address
;
737 tsk
->thread
.error_code
= error_code
| (address
>= TASK_SIZE
);
738 tsk
->thread
.trap_no
= 14;
740 force_sig_info_fault(SIGSEGV
, si_code
, address
, tsk
);
745 if (is_f00f_bug(regs
, address
))
748 no_context(regs
, error_code
, address
);
752 bad_area_nosemaphore(struct pt_regs
*regs
, unsigned long error_code
,
753 unsigned long address
)
755 __bad_area_nosemaphore(regs
, error_code
, address
, SEGV_MAPERR
);
759 __bad_area(struct pt_regs
*regs
, unsigned long error_code
,
760 unsigned long address
, int si_code
)
762 struct mm_struct
*mm
= current
->mm
;
765 * Something tried to access memory that isn't in our memory map..
766 * Fix it, but check if it's kernel or user first..
768 up_read(&mm
->mmap_sem
);
770 __bad_area_nosemaphore(regs
, error_code
, address
, si_code
);
774 bad_area(struct pt_regs
*regs
, unsigned long error_code
, unsigned long address
)
776 __bad_area(regs
, error_code
, address
, SEGV_MAPERR
);
780 bad_area_access_error(struct pt_regs
*regs
, unsigned long error_code
,
781 unsigned long address
)
783 __bad_area(regs
, error_code
, address
, SEGV_ACCERR
);
786 /* TODO: fixup for "mm-invoke-oom-killer-from-page-fault.patch" */
788 out_of_memory(struct pt_regs
*regs
, unsigned long error_code
,
789 unsigned long address
)
792 * We ran out of memory, call the OOM killer, and return the userspace
793 * (which will retry the fault, or kill us if we got oom-killed):
795 up_read(¤t
->mm
->mmap_sem
);
797 pagefault_out_of_memory();
801 do_sigbus(struct pt_regs
*regs
, unsigned long error_code
, unsigned long address
)
803 struct task_struct
*tsk
= current
;
804 struct mm_struct
*mm
= tsk
->mm
;
806 up_read(&mm
->mmap_sem
);
808 /* Kernel mode? Handle exceptions or die: */
809 if (!(error_code
& PF_USER
))
810 no_context(regs
, error_code
, address
);
812 /* User-space => ok to do another page fault: */
813 if (is_prefetch(regs
, error_code
, address
))
816 tsk
->thread
.cr2
= address
;
817 tsk
->thread
.error_code
= error_code
;
818 tsk
->thread
.trap_no
= 14;
820 force_sig_info_fault(SIGBUS
, BUS_ADRERR
, address
, tsk
);
824 mm_fault_error(struct pt_regs
*regs
, unsigned long error_code
,
825 unsigned long address
, unsigned int fault
)
827 if (fault
& VM_FAULT_OOM
) {
828 out_of_memory(regs
, error_code
, address
);
830 if (fault
& VM_FAULT_SIGBUS
)
831 do_sigbus(regs
, error_code
, address
);
837 static int spurious_fault_check(unsigned long error_code
, pte_t
*pte
)
839 if ((error_code
& PF_WRITE
) && !pte_write(*pte
))
842 if ((error_code
& PF_INSTR
) && !pte_exec(*pte
))
849 * Handle a spurious fault caused by a stale TLB entry.
851 * This allows us to lazily refresh the TLB when increasing the
852 * permissions of a kernel page (RO -> RW or NX -> X). Doing it
853 * eagerly is very expensive since that implies doing a full
854 * cross-processor TLB flush, even if no stale TLB entries exist
855 * on other processors.
857 * There are no security implications to leaving a stale TLB when
858 * increasing the permissions on a page.
861 spurious_fault(unsigned long error_code
, unsigned long address
)
869 /* Reserved-bit violation or user access to kernel space? */
870 if (error_code
& (PF_USER
| PF_RSVD
))
873 pgd
= init_mm
.pgd
+ pgd_index(address
);
874 if (!pgd_present(*pgd
))
877 pud
= pud_offset(pgd
, address
);
878 if (!pud_present(*pud
))
882 return spurious_fault_check(error_code
, (pte_t
*) pud
);
884 pmd
= pmd_offset(pud
, address
);
885 if (!pmd_present(*pmd
))
889 return spurious_fault_check(error_code
, (pte_t
*) pmd
);
891 pte
= pte_offset_kernel(pmd
, address
);
892 if (!pte_present(*pte
))
895 ret
= spurious_fault_check(error_code
, pte
);
900 * Make sure we have permissions in PMD.
901 * If not, then there's a bug in the page tables:
903 ret
= spurious_fault_check(error_code
, (pte_t
*) pmd
);
904 WARN_ONCE(!ret
, "PMD has incorrect permission bits\n");
909 int show_unhandled_signals
= 1;
912 access_error(unsigned long error_code
, int write
, struct vm_area_struct
*vma
)
915 /* write, present and write, not present: */
916 if (unlikely(!(vma
->vm_flags
& VM_WRITE
)))
922 if (unlikely(error_code
& PF_PROT
))
925 /* read, not present: */
926 if (unlikely(!(vma
->vm_flags
& (VM_READ
| VM_EXEC
| VM_WRITE
))))
932 static int fault_in_kernel_space(unsigned long address
)
934 return address
>= TASK_SIZE_MAX
;
938 * This routine handles page faults. It determines the address,
939 * and the problem, and then passes it off to one of the appropriate
942 dotraplinkage
void __kprobes
943 do_page_fault(struct pt_regs
*regs
, unsigned long error_code
)
945 struct vm_area_struct
*vma
;
946 struct task_struct
*tsk
;
947 unsigned long address
;
948 struct mm_struct
*mm
;
955 prefetchw(&mm
->mmap_sem
);
957 /* Get the faulting address: */
958 address
= read_cr2();
961 * Detect and handle instructions that would cause a page fault for
962 * both a tracked kernel page and a userspace page.
964 if (kmemcheck_active(regs
))
965 kmemcheck_hide(regs
);
967 if (unlikely(kmmio_fault(regs
, address
)))
971 * We fault-in kernel-space virtual memory on-demand. The
972 * 'reference' page table is init_mm.pgd.
974 * NOTE! We MUST NOT take any locks for this case. We may
975 * be in an interrupt or a critical region, and should
976 * only copy the information from the master page table,
979 * This verifies that the fault happens in kernel space
980 * (error_code & 4) == 0, and that the fault was not a
981 * protection error (error_code & 9) == 0.
983 if (unlikely(fault_in_kernel_space(address
))) {
984 if (!(error_code
& (PF_RSVD
| PF_USER
| PF_PROT
))) {
985 if (vmalloc_fault(address
) >= 0)
988 if (kmemcheck_fault(regs
, address
, error_code
))
992 /* Can handle a stale RO->RW TLB: */
993 if (spurious_fault(error_code
, address
))
996 /* kprobes don't want to hook the spurious faults: */
997 if (notify_page_fault(regs
))
1000 * Don't take the mm semaphore here. If we fixup a prefetch
1001 * fault we could otherwise deadlock:
1003 bad_area_nosemaphore(regs
, error_code
, address
);
1008 /* kprobes don't want to hook the spurious faults: */
1009 if (unlikely(notify_page_fault(regs
)))
1012 * It's safe to allow irq's after cr2 has been saved and the
1013 * vmalloc fault has been handled.
1015 * User-mode registers count as a user access even for any
1016 * potential system fault or CPU buglet:
1018 if (user_mode_vm(regs
)) {
1020 error_code
|= PF_USER
;
1022 if (regs
->flags
& X86_EFLAGS_IF
)
1026 if (unlikely(error_code
& PF_RSVD
))
1027 pgtable_bad(regs
, error_code
, address
);
1029 perf_swcounter_event(PERF_COUNT_SW_PAGE_FAULTS
, 1, 0, regs
, address
);
1032 * If we're in an interrupt, have no user context or are running
1033 * in an atomic region then we must not take the fault:
1035 if (unlikely(in_atomic() || !mm
)) {
1036 bad_area_nosemaphore(regs
, error_code
, address
);
1041 * When running in the kernel we expect faults to occur only to
1042 * addresses in user space. All other faults represent errors in
1043 * the kernel and should generate an OOPS. Unfortunately, in the
1044 * case of an erroneous fault occurring in a code path which already
1045 * holds mmap_sem we will deadlock attempting to validate the fault
1046 * against the address space. Luckily the kernel only validly
1047 * references user space from well defined areas of code, which are
1048 * listed in the exceptions table.
1050 * As the vast majority of faults will be valid we will only perform
1051 * the source reference check when there is a possibility of a
1052 * deadlock. Attempt to lock the address space, if we cannot we then
1053 * validate the source. If this is invalid we can skip the address
1054 * space check, thus avoiding the deadlock:
1056 if (unlikely(!down_read_trylock(&mm
->mmap_sem
))) {
1057 if ((error_code
& PF_USER
) == 0 &&
1058 !search_exception_tables(regs
->ip
)) {
1059 bad_area_nosemaphore(regs
, error_code
, address
);
1062 down_read(&mm
->mmap_sem
);
1065 * The above down_read_trylock() might have succeeded in
1066 * which case we'll have missed the might_sleep() from
1072 vma
= find_vma(mm
, address
);
1073 if (unlikely(!vma
)) {
1074 bad_area(regs
, error_code
, address
);
1077 if (likely(vma
->vm_start
<= address
))
1079 if (unlikely(!(vma
->vm_flags
& VM_GROWSDOWN
))) {
1080 bad_area(regs
, error_code
, address
);
1083 if (error_code
& PF_USER
) {
1085 * Accessing the stack below %sp is always a bug.
1086 * The large cushion allows instructions like enter
1087 * and pusha to work. ("enter $65535, $31" pushes
1088 * 32 pointers and then decrements %sp by 65535.)
1090 if (unlikely(address
+ 65536 + 32 * sizeof(unsigned long) < regs
->sp
)) {
1091 bad_area(regs
, error_code
, address
);
1095 if (unlikely(expand_stack(vma
, address
))) {
1096 bad_area(regs
, error_code
, address
);
1101 * Ok, we have a good vm_area for this memory access, so
1102 * we can handle it..
1105 write
= error_code
& PF_WRITE
;
1107 if (unlikely(access_error(error_code
, write
, vma
))) {
1108 bad_area_access_error(regs
, error_code
, address
);
1113 * If for any reason at all we couldn't handle the fault,
1114 * make sure we exit gracefully rather than endlessly redo
1117 fault
= handle_mm_fault(mm
, vma
, address
, write
);
1119 if (unlikely(fault
& VM_FAULT_ERROR
)) {
1120 mm_fault_error(regs
, error_code
, address
, fault
);
1124 if (fault
& VM_FAULT_MAJOR
) {
1126 perf_swcounter_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ
, 1, 0,
1130 perf_swcounter_event(PERF_COUNT_SW_PAGE_FAULTS_MIN
, 1, 0,
1134 check_v8086_mode(regs
, address
, tsk
);
1136 up_read(&mm
->mmap_sem
);