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 */
14 #include <linux/hugetlb.h> /* hstate_index_to_shift */
15 #include <linux/prefetch.h> /* prefetchw */
16 #include <linux/context_tracking.h> /* exception_enter(), ... */
18 #include <asm/traps.h> /* dotraplinkage, ... */
19 #include <asm/pgalloc.h> /* pgd_*(), ... */
20 #include <asm/kmemcheck.h> /* kmemcheck_*(), ... */
21 #include <asm/fixmap.h> /* VSYSCALL_ADDR */
22 #include <asm/vsyscall.h> /* emulate_vsyscall */
24 #define CREATE_TRACE_POINTS
25 #include <asm/trace/exceptions.h>
28 * Page fault error code bits:
30 * bit 0 == 0: no page found 1: protection fault
31 * bit 1 == 0: read access 1: write access
32 * bit 2 == 0: kernel-mode access 1: user-mode access
33 * bit 3 == 1: use of reserved bit detected
34 * bit 4 == 1: fault was an instruction fetch
36 enum x86_pf_error_code
{
46 * Returns 0 if mmiotrace is disabled, or if the fault is not
47 * handled by mmiotrace:
49 static inline int __kprobes
50 kmmio_fault(struct pt_regs
*regs
, unsigned long addr
)
52 if (unlikely(is_kmmio_active()))
53 if (kmmio_handler(regs
, addr
) == 1)
58 static inline int __kprobes
kprobes_fault(struct pt_regs
*regs
)
62 /* kprobe_running() needs smp_processor_id() */
63 if (kprobes_built_in() && !user_mode_vm(regs
)) {
65 if (kprobe_running() && kprobe_fault_handler(regs
, 14))
78 * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch.
79 * Check that here and ignore it.
83 * Sometimes the CPU reports invalid exceptions on prefetch.
84 * Check that here and ignore it.
86 * Opcode checker based on code by Richard Brunner.
89 check_prefetch_opcode(struct pt_regs
*regs
, unsigned char *instr
,
90 unsigned char opcode
, int *prefetch
)
92 unsigned char instr_hi
= opcode
& 0xf0;
93 unsigned char instr_lo
= opcode
& 0x0f;
99 * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes.
100 * In X86_64 long mode, the CPU will signal invalid
101 * opcode if some of these prefixes are present so
102 * X86_64 will never get here anyway
104 return ((instr_lo
& 7) == 0x6);
108 * In AMD64 long mode 0x40..0x4F are valid REX prefixes
109 * Need to figure out under what instruction mode the
110 * instruction was issued. Could check the LDT for lm,
111 * but for now it's good enough to assume that long
112 * mode only uses well known segments or kernel.
114 return (!user_mode(regs
) || user_64bit_mode(regs
));
117 /* 0x64 thru 0x67 are valid prefixes in all modes. */
118 return (instr_lo
& 0xC) == 0x4;
120 /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */
121 return !instr_lo
|| (instr_lo
>>1) == 1;
123 /* Prefetch instruction is 0x0F0D or 0x0F18 */
124 if (probe_kernel_address(instr
, opcode
))
127 *prefetch
= (instr_lo
== 0xF) &&
128 (opcode
== 0x0D || opcode
== 0x18);
136 is_prefetch(struct pt_regs
*regs
, unsigned long error_code
, unsigned long addr
)
138 unsigned char *max_instr
;
139 unsigned char *instr
;
143 * If it was a exec (instruction fetch) fault on NX page, then
144 * do not ignore the fault:
146 if (error_code
& PF_INSTR
)
149 instr
= (void *)convert_ip_to_linear(current
, regs
);
150 max_instr
= instr
+ 15;
152 if (user_mode(regs
) && instr
>= (unsigned char *)TASK_SIZE
)
155 while (instr
< max_instr
) {
156 unsigned char opcode
;
158 if (probe_kernel_address(instr
, opcode
))
163 if (!check_prefetch_opcode(regs
, instr
, opcode
, &prefetch
))
170 force_sig_info_fault(int si_signo
, int si_code
, unsigned long address
,
171 struct task_struct
*tsk
, int fault
)
176 info
.si_signo
= si_signo
;
178 info
.si_code
= si_code
;
179 info
.si_addr
= (void __user
*)address
;
180 if (fault
& VM_FAULT_HWPOISON_LARGE
)
181 lsb
= hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault
));
182 if (fault
& VM_FAULT_HWPOISON
)
184 info
.si_addr_lsb
= lsb
;
186 force_sig_info(si_signo
, &info
, tsk
);
189 DEFINE_SPINLOCK(pgd_lock
);
193 static inline pmd_t
*vmalloc_sync_one(pgd_t
*pgd
, unsigned long address
)
195 unsigned index
= pgd_index(address
);
201 pgd_k
= init_mm
.pgd
+ index
;
203 if (!pgd_present(*pgd_k
))
207 * set_pgd(pgd, *pgd_k); here would be useless on PAE
208 * and redundant with the set_pmd() on non-PAE. As would
211 pud
= pud_offset(pgd
, address
);
212 pud_k
= pud_offset(pgd_k
, address
);
213 if (!pud_present(*pud_k
))
216 pmd
= pmd_offset(pud
, address
);
217 pmd_k
= pmd_offset(pud_k
, address
);
218 if (!pmd_present(*pmd_k
))
221 if (!pmd_present(*pmd
))
222 set_pmd(pmd
, *pmd_k
);
224 BUG_ON(pmd_page(*pmd
) != pmd_page(*pmd_k
));
229 void vmalloc_sync_all(void)
231 unsigned long address
;
233 if (SHARED_KERNEL_PMD
)
236 for (address
= VMALLOC_START
& PMD_MASK
;
237 address
>= TASK_SIZE
&& address
< FIXADDR_TOP
;
238 address
+= PMD_SIZE
) {
241 spin_lock(&pgd_lock
);
242 list_for_each_entry(page
, &pgd_list
, lru
) {
243 spinlock_t
*pgt_lock
;
246 /* the pgt_lock only for Xen */
247 pgt_lock
= &pgd_page_get_mm(page
)->page_table_lock
;
250 ret
= vmalloc_sync_one(page_address(page
), address
);
251 spin_unlock(pgt_lock
);
256 spin_unlock(&pgd_lock
);
263 * Handle a fault on the vmalloc or module mapping area
265 static noinline __kprobes
int vmalloc_fault(unsigned long address
)
267 unsigned long pgd_paddr
;
271 /* Make sure we are in vmalloc area: */
272 if (!(address
>= VMALLOC_START
&& address
< VMALLOC_END
))
275 WARN_ON_ONCE(in_nmi());
278 * Synchronize this task's top level page-table
279 * with the 'reference' page table.
281 * Do _not_ use "current" here. We might be inside
282 * an interrupt in the middle of a task switch..
284 pgd_paddr
= read_cr3();
285 pmd_k
= vmalloc_sync_one(__va(pgd_paddr
), address
);
289 pte_k
= pte_offset_kernel(pmd_k
, address
);
290 if (!pte_present(*pte_k
))
297 * Did it hit the DOS screen memory VA from vm86 mode?
300 check_v8086_mode(struct pt_regs
*regs
, unsigned long address
,
301 struct task_struct
*tsk
)
305 if (!v8086_mode(regs
))
308 bit
= (address
- 0xA0000) >> PAGE_SHIFT
;
310 tsk
->thread
.screen_bitmap
|= 1 << bit
;
313 static bool low_pfn(unsigned long pfn
)
315 return pfn
< max_low_pfn
;
318 static void dump_pagetable(unsigned long address
)
320 pgd_t
*base
= __va(read_cr3());
321 pgd_t
*pgd
= &base
[pgd_index(address
)];
325 #ifdef CONFIG_X86_PAE
326 printk("*pdpt = %016Lx ", pgd_val(*pgd
));
327 if (!low_pfn(pgd_val(*pgd
) >> PAGE_SHIFT
) || !pgd_present(*pgd
))
330 pmd
= pmd_offset(pud_offset(pgd
, address
), address
);
331 printk(KERN_CONT
"*pde = %0*Lx ", sizeof(*pmd
) * 2, (u64
)pmd_val(*pmd
));
334 * We must not directly access the pte in the highpte
335 * case if the page table is located in highmem.
336 * And let's rather not kmap-atomic the pte, just in case
337 * it's allocated already:
339 if (!low_pfn(pmd_pfn(*pmd
)) || !pmd_present(*pmd
) || pmd_large(*pmd
))
342 pte
= pte_offset_kernel(pmd
, address
);
343 printk("*pte = %0*Lx ", sizeof(*pte
) * 2, (u64
)pte_val(*pte
));
348 #else /* CONFIG_X86_64: */
350 void vmalloc_sync_all(void)
352 sync_global_pgds(VMALLOC_START
& PGDIR_MASK
, VMALLOC_END
);
358 * Handle a fault on the vmalloc area
360 * This assumes no large pages in there.
362 static noinline __kprobes
int vmalloc_fault(unsigned long address
)
364 pgd_t
*pgd
, *pgd_ref
;
365 pud_t
*pud
, *pud_ref
;
366 pmd_t
*pmd
, *pmd_ref
;
367 pte_t
*pte
, *pte_ref
;
369 /* Make sure we are in vmalloc area: */
370 if (!(address
>= VMALLOC_START
&& address
< VMALLOC_END
))
373 WARN_ON_ONCE(in_nmi());
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
))
385 if (pgd_none(*pgd
)) {
386 set_pgd(pgd
, *pgd_ref
);
387 arch_flush_lazy_mmu_mode();
389 BUG_ON(pgd_page_vaddr(*pgd
) != pgd_page_vaddr(*pgd_ref
));
393 * Below here mismatches are bugs because these lower tables
397 pud
= pud_offset(pgd
, address
);
398 pud_ref
= pud_offset(pgd_ref
, address
);
399 if (pud_none(*pud_ref
))
402 if (pud_none(*pud
) || pud_page_vaddr(*pud
) != pud_page_vaddr(*pud_ref
))
405 pmd
= pmd_offset(pud
, address
);
406 pmd_ref
= pmd_offset(pud_ref
, address
);
407 if (pmd_none(*pmd_ref
))
410 if (pmd_none(*pmd
) || pmd_page(*pmd
) != pmd_page(*pmd_ref
))
413 pte_ref
= pte_offset_kernel(pmd_ref
, address
);
414 if (!pte_present(*pte_ref
))
417 pte
= pte_offset_kernel(pmd
, address
);
420 * Don't use pte_page here, because the mappings can point
421 * outside mem_map, and the NUMA hash lookup cannot handle
424 if (!pte_present(*pte
) || pte_pfn(*pte
) != pte_pfn(*pte_ref
))
430 #ifdef CONFIG_CPU_SUP_AMD
431 static const char errata93_warning
[] =
433 "******* Your BIOS seems to not contain a fix for K8 errata #93\n"
434 "******* Working around it, but it may cause SEGVs or burn power.\n"
435 "******* Please consider a BIOS update.\n"
436 "******* Disabling USB legacy in the BIOS may also help.\n";
440 * No vm86 mode in 64-bit mode:
443 check_v8086_mode(struct pt_regs
*regs
, unsigned long address
,
444 struct task_struct
*tsk
)
448 static int bad_address(void *p
)
452 return probe_kernel_address((unsigned long *)p
, dummy
);
455 static void dump_pagetable(unsigned long address
)
457 pgd_t
*base
= __va(read_cr3() & PHYSICAL_PAGE_MASK
);
458 pgd_t
*pgd
= base
+ pgd_index(address
);
463 if (bad_address(pgd
))
466 printk("PGD %lx ", pgd_val(*pgd
));
468 if (!pgd_present(*pgd
))
471 pud
= pud_offset(pgd
, address
);
472 if (bad_address(pud
))
475 printk("PUD %lx ", pud_val(*pud
));
476 if (!pud_present(*pud
) || pud_large(*pud
))
479 pmd
= pmd_offset(pud
, address
);
480 if (bad_address(pmd
))
483 printk("PMD %lx ", pmd_val(*pmd
));
484 if (!pmd_present(*pmd
) || pmd_large(*pmd
))
487 pte
= pte_offset_kernel(pmd
, address
);
488 if (bad_address(pte
))
491 printk("PTE %lx", pte_val(*pte
));
499 #endif /* CONFIG_X86_64 */
502 * Workaround for K8 erratum #93 & buggy BIOS.
504 * BIOS SMM functions are required to use a specific workaround
505 * to avoid corruption of the 64bit RIP register on C stepping K8.
507 * A lot of BIOS that didn't get tested properly miss this.
509 * The OS sees this as a page fault with the upper 32bits of RIP cleared.
510 * Try to work around it here.
512 * Note we only handle faults in kernel here.
513 * Does nothing on 32-bit.
515 static int is_errata93(struct pt_regs
*regs
, unsigned long address
)
517 #if defined(CONFIG_X86_64) && defined(CONFIG_CPU_SUP_AMD)
518 if (boot_cpu_data
.x86_vendor
!= X86_VENDOR_AMD
519 || boot_cpu_data
.x86
!= 0xf)
522 if (address
!= regs
->ip
)
525 if ((address
>> 32) != 0)
528 address
|= 0xffffffffUL
<< 32;
529 if ((address
>= (u64
)_stext
&& address
<= (u64
)_etext
) ||
530 (address
>= MODULES_VADDR
&& address
<= MODULES_END
)) {
531 printk_once(errata93_warning
);
540 * Work around K8 erratum #100 K8 in compat mode occasionally jumps
541 * to illegal addresses >4GB.
543 * We catch this in the page fault handler because these addresses
544 * are not reachable. Just detect this case and return. Any code
545 * segment in LDT is compatibility mode.
547 static int is_errata100(struct pt_regs
*regs
, unsigned long address
)
550 if ((regs
->cs
== __USER32_CS
|| (regs
->cs
& (1<<2))) && (address
>> 32))
556 static int is_f00f_bug(struct pt_regs
*regs
, unsigned long address
)
558 #ifdef CONFIG_X86_F00F_BUG
562 * Pentium F0 0F C7 C8 bug workaround:
564 if (boot_cpu_has_bug(X86_BUG_F00F
)) {
565 nr
= (address
- idt_descr
.address
) >> 3;
568 do_invalid_op(regs
, 0);
576 static const char nx_warning
[] = KERN_CRIT
577 "kernel tried to execute NX-protected page - exploit attempt? (uid: %d)\n";
580 show_fault_oops(struct pt_regs
*regs
, unsigned long error_code
,
581 unsigned long address
)
583 if (!oops_may_print())
586 if (error_code
& PF_INSTR
) {
591 pgd
= __va(read_cr3() & PHYSICAL_PAGE_MASK
);
592 pgd
+= pgd_index(address
);
594 pte
= lookup_address_in_pgd(pgd
, address
, &level
);
596 if (pte
&& pte_present(*pte
) && !pte_exec(*pte
))
597 printk(nx_warning
, from_kuid(&init_user_ns
, current_uid()));
600 printk(KERN_ALERT
"BUG: unable to handle kernel ");
601 if (address
< PAGE_SIZE
)
602 printk(KERN_CONT
"NULL pointer dereference");
604 printk(KERN_CONT
"paging request");
606 printk(KERN_CONT
" at %p\n", (void *) address
);
607 printk(KERN_ALERT
"IP:");
608 printk_address(regs
->ip
);
610 dump_pagetable(address
);
614 pgtable_bad(struct pt_regs
*regs
, unsigned long error_code
,
615 unsigned long address
)
617 struct task_struct
*tsk
;
621 flags
= oops_begin();
625 printk(KERN_ALERT
"%s: Corrupted page table at address %lx\n",
627 dump_pagetable(address
);
629 tsk
->thread
.cr2
= address
;
630 tsk
->thread
.trap_nr
= X86_TRAP_PF
;
631 tsk
->thread
.error_code
= error_code
;
633 if (__die("Bad pagetable", regs
, error_code
))
636 oops_end(flags
, regs
, sig
);
640 no_context(struct pt_regs
*regs
, unsigned long error_code
,
641 unsigned long address
, int signal
, int si_code
)
643 struct task_struct
*tsk
= current
;
644 unsigned long *stackend
;
648 /* Are we prepared to handle this kernel fault? */
649 if (fixup_exception(regs
)) {
651 * Any interrupt that takes a fault gets the fixup. This makes
652 * the below recursive fault logic only apply to a faults from
659 * Per the above we're !in_interrupt(), aka. task context.
661 * In this case we need to make sure we're not recursively
662 * faulting through the emulate_vsyscall() logic.
664 if (current_thread_info()->sig_on_uaccess_error
&& signal
) {
665 tsk
->thread
.trap_nr
= X86_TRAP_PF
;
666 tsk
->thread
.error_code
= error_code
| PF_USER
;
667 tsk
->thread
.cr2
= address
;
669 /* XXX: hwpoison faults will set the wrong code. */
670 force_sig_info_fault(signal
, si_code
, address
, tsk
, 0);
674 * Barring that, we can do the fixup and be happy.
682 * Valid to do another page fault here, because if this fault
683 * had been triggered by is_prefetch fixup_exception would have
688 * Hall of shame of CPU/BIOS bugs.
690 if (is_prefetch(regs
, error_code
, address
))
693 if (is_errata93(regs
, address
))
697 * Oops. The kernel tried to access some bad page. We'll have to
698 * terminate things with extreme prejudice:
700 flags
= oops_begin();
702 show_fault_oops(regs
, error_code
, address
);
704 stackend
= end_of_stack(tsk
);
705 if (tsk
!= &init_task
&& *stackend
!= STACK_END_MAGIC
)
706 printk(KERN_EMERG
"Thread overran stack, or stack corrupted\n");
708 tsk
->thread
.cr2
= address
;
709 tsk
->thread
.trap_nr
= X86_TRAP_PF
;
710 tsk
->thread
.error_code
= error_code
;
713 if (__die("Oops", regs
, error_code
))
716 /* Executive summary in case the body of the oops scrolled away */
717 printk(KERN_DEFAULT
"CR2: %016lx\n", address
);
719 oops_end(flags
, regs
, sig
);
723 * Print out info about fatal segfaults, if the show_unhandled_signals
727 show_signal_msg(struct pt_regs
*regs
, unsigned long error_code
,
728 unsigned long address
, struct task_struct
*tsk
)
730 if (!unhandled_signal(tsk
, SIGSEGV
))
733 if (!printk_ratelimit())
736 printk("%s%s[%d]: segfault at %lx ip %p sp %p error %lx",
737 task_pid_nr(tsk
) > 1 ? KERN_INFO
: KERN_EMERG
,
738 tsk
->comm
, task_pid_nr(tsk
), address
,
739 (void *)regs
->ip
, (void *)regs
->sp
, error_code
);
741 print_vma_addr(KERN_CONT
" in ", regs
->ip
);
743 printk(KERN_CONT
"\n");
747 __bad_area_nosemaphore(struct pt_regs
*regs
, unsigned long error_code
,
748 unsigned long address
, int si_code
)
750 struct task_struct
*tsk
= current
;
752 /* User mode accesses just cause a SIGSEGV */
753 if (error_code
& PF_USER
) {
755 * It's possible to have interrupts off here:
760 * Valid to do another page fault here because this one came
763 if (is_prefetch(regs
, error_code
, address
))
766 if (is_errata100(regs
, address
))
771 * Instruction fetch faults in the vsyscall page might need
774 if (unlikely((error_code
& PF_INSTR
) &&
775 ((address
& ~0xfff) == VSYSCALL_ADDR
))) {
776 if (emulate_vsyscall(regs
, address
))
780 /* Kernel addresses are always protection faults: */
781 if (address
>= TASK_SIZE
)
782 error_code
|= PF_PROT
;
784 if (likely(show_unhandled_signals
))
785 show_signal_msg(regs
, error_code
, address
, tsk
);
787 tsk
->thread
.cr2
= address
;
788 tsk
->thread
.error_code
= error_code
;
789 tsk
->thread
.trap_nr
= X86_TRAP_PF
;
791 force_sig_info_fault(SIGSEGV
, si_code
, address
, tsk
, 0);
796 if (is_f00f_bug(regs
, address
))
799 no_context(regs
, error_code
, address
, SIGSEGV
, si_code
);
803 bad_area_nosemaphore(struct pt_regs
*regs
, unsigned long error_code
,
804 unsigned long address
)
806 __bad_area_nosemaphore(regs
, error_code
, address
, SEGV_MAPERR
);
810 __bad_area(struct pt_regs
*regs
, unsigned long error_code
,
811 unsigned long address
, int si_code
)
813 struct mm_struct
*mm
= current
->mm
;
816 * Something tried to access memory that isn't in our memory map..
817 * Fix it, but check if it's kernel or user first..
819 up_read(&mm
->mmap_sem
);
821 __bad_area_nosemaphore(regs
, error_code
, address
, si_code
);
825 bad_area(struct pt_regs
*regs
, unsigned long error_code
, unsigned long address
)
827 __bad_area(regs
, error_code
, address
, SEGV_MAPERR
);
831 bad_area_access_error(struct pt_regs
*regs
, unsigned long error_code
,
832 unsigned long address
)
834 __bad_area(regs
, error_code
, address
, SEGV_ACCERR
);
838 do_sigbus(struct pt_regs
*regs
, unsigned long error_code
, unsigned long address
,
841 struct task_struct
*tsk
= current
;
842 struct mm_struct
*mm
= tsk
->mm
;
843 int code
= BUS_ADRERR
;
845 up_read(&mm
->mmap_sem
);
847 /* Kernel mode? Handle exceptions or die: */
848 if (!(error_code
& PF_USER
)) {
849 no_context(regs
, error_code
, address
, SIGBUS
, BUS_ADRERR
);
853 /* User-space => ok to do another page fault: */
854 if (is_prefetch(regs
, error_code
, address
))
857 tsk
->thread
.cr2
= address
;
858 tsk
->thread
.error_code
= error_code
;
859 tsk
->thread
.trap_nr
= X86_TRAP_PF
;
861 #ifdef CONFIG_MEMORY_FAILURE
862 if (fault
& (VM_FAULT_HWPOISON
|VM_FAULT_HWPOISON_LARGE
)) {
864 "MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n",
865 tsk
->comm
, tsk
->pid
, address
);
866 code
= BUS_MCEERR_AR
;
869 force_sig_info_fault(SIGBUS
, code
, address
, tsk
, fault
);
873 mm_fault_error(struct pt_regs
*regs
, unsigned long error_code
,
874 unsigned long address
, unsigned int fault
)
876 if (fatal_signal_pending(current
) && !(error_code
& PF_USER
)) {
877 up_read(¤t
->mm
->mmap_sem
);
878 no_context(regs
, error_code
, address
, 0, 0);
882 if (fault
& VM_FAULT_OOM
) {
883 /* Kernel mode? Handle exceptions or die: */
884 if (!(error_code
& PF_USER
)) {
885 up_read(¤t
->mm
->mmap_sem
);
886 no_context(regs
, error_code
, address
,
887 SIGSEGV
, SEGV_MAPERR
);
891 up_read(¤t
->mm
->mmap_sem
);
894 * We ran out of memory, call the OOM killer, and return the
895 * userspace (which will retry the fault, or kill us if we got
898 pagefault_out_of_memory();
900 if (fault
& (VM_FAULT_SIGBUS
|VM_FAULT_HWPOISON
|
901 VM_FAULT_HWPOISON_LARGE
))
902 do_sigbus(regs
, error_code
, address
, fault
);
908 static int spurious_fault_check(unsigned long error_code
, pte_t
*pte
)
910 if ((error_code
& PF_WRITE
) && !pte_write(*pte
))
913 if ((error_code
& PF_INSTR
) && !pte_exec(*pte
))
920 * Handle a spurious fault caused by a stale TLB entry.
922 * This allows us to lazily refresh the TLB when increasing the
923 * permissions of a kernel page (RO -> RW or NX -> X). Doing it
924 * eagerly is very expensive since that implies doing a full
925 * cross-processor TLB flush, even if no stale TLB entries exist
926 * on other processors.
928 * There are no security implications to leaving a stale TLB when
929 * increasing the permissions on a page.
931 static noinline __kprobes
int
932 spurious_fault(unsigned long error_code
, unsigned long address
)
940 /* Reserved-bit violation or user access to kernel space? */
941 if (error_code
& (PF_USER
| PF_RSVD
))
944 pgd
= init_mm
.pgd
+ pgd_index(address
);
945 if (!pgd_present(*pgd
))
948 pud
= pud_offset(pgd
, address
);
949 if (!pud_present(*pud
))
953 return spurious_fault_check(error_code
, (pte_t
*) pud
);
955 pmd
= pmd_offset(pud
, address
);
956 if (!pmd_present(*pmd
))
960 return spurious_fault_check(error_code
, (pte_t
*) pmd
);
962 pte
= pte_offset_kernel(pmd
, address
);
963 if (!pte_present(*pte
))
966 ret
= spurious_fault_check(error_code
, pte
);
971 * Make sure we have permissions in PMD.
972 * If not, then there's a bug in the page tables:
974 ret
= spurious_fault_check(error_code
, (pte_t
*) pmd
);
975 WARN_ONCE(!ret
, "PMD has incorrect permission bits\n");
980 int show_unhandled_signals
= 1;
983 access_error(unsigned long error_code
, struct vm_area_struct
*vma
)
985 if (error_code
& PF_WRITE
) {
986 /* write, present and write, not present: */
987 if (unlikely(!(vma
->vm_flags
& VM_WRITE
)))
993 if (unlikely(error_code
& PF_PROT
))
996 /* read, not present: */
997 if (unlikely(!(vma
->vm_flags
& (VM_READ
| VM_EXEC
| VM_WRITE
))))
1003 static int fault_in_kernel_space(unsigned long address
)
1005 return address
>= TASK_SIZE_MAX
;
1008 static inline bool smap_violation(int error_code
, struct pt_regs
*regs
)
1010 if (!IS_ENABLED(CONFIG_X86_SMAP
))
1013 if (!static_cpu_has(X86_FEATURE_SMAP
))
1016 if (error_code
& PF_USER
)
1019 if (!user_mode_vm(regs
) && (regs
->flags
& X86_EFLAGS_AC
))
1026 * This routine handles page faults. It determines the address,
1027 * and the problem, and then passes it off to one of the appropriate
1030 * This function must have noinline because both callers
1031 * {,trace_}do_page_fault() have notrace on. Having this an actual function
1032 * guarantees there's a function trace entry.
1034 static void __kprobes noinline
1035 __do_page_fault(struct pt_regs
*regs
, unsigned long error_code
,
1036 unsigned long address
)
1038 struct vm_area_struct
*vma
;
1039 struct task_struct
*tsk
;
1040 struct mm_struct
*mm
;
1042 unsigned int flags
= FAULT_FLAG_ALLOW_RETRY
| FAULT_FLAG_KILLABLE
;
1048 * Detect and handle instructions that would cause a page fault for
1049 * both a tracked kernel page and a userspace page.
1051 if (kmemcheck_active(regs
))
1052 kmemcheck_hide(regs
);
1053 prefetchw(&mm
->mmap_sem
);
1055 if (unlikely(kmmio_fault(regs
, address
)))
1059 * We fault-in kernel-space virtual memory on-demand. The
1060 * 'reference' page table is init_mm.pgd.
1062 * NOTE! We MUST NOT take any locks for this case. We may
1063 * be in an interrupt or a critical region, and should
1064 * only copy the information from the master page table,
1067 * This verifies that the fault happens in kernel space
1068 * (error_code & 4) == 0, and that the fault was not a
1069 * protection error (error_code & 9) == 0.
1071 if (unlikely(fault_in_kernel_space(address
))) {
1072 if (!(error_code
& (PF_RSVD
| PF_USER
| PF_PROT
))) {
1073 if (vmalloc_fault(address
) >= 0)
1076 if (kmemcheck_fault(regs
, address
, error_code
))
1080 /* Can handle a stale RO->RW TLB: */
1081 if (spurious_fault(error_code
, address
))
1084 /* kprobes don't want to hook the spurious faults: */
1085 if (kprobes_fault(regs
))
1088 * Don't take the mm semaphore here. If we fixup a prefetch
1089 * fault we could otherwise deadlock:
1091 bad_area_nosemaphore(regs
, error_code
, address
);
1096 /* kprobes don't want to hook the spurious faults: */
1097 if (unlikely(kprobes_fault(regs
)))
1100 if (unlikely(error_code
& PF_RSVD
))
1101 pgtable_bad(regs
, error_code
, address
);
1103 if (unlikely(smap_violation(error_code
, regs
))) {
1104 bad_area_nosemaphore(regs
, error_code
, address
);
1109 * If we're in an interrupt, have no user context or are running
1110 * in an atomic region then we must not take the fault:
1112 if (unlikely(in_atomic() || !mm
)) {
1113 bad_area_nosemaphore(regs
, error_code
, address
);
1118 * It's safe to allow irq's after cr2 has been saved and the
1119 * vmalloc fault has been handled.
1121 * User-mode registers count as a user access even for any
1122 * potential system fault or CPU buglet:
1124 if (user_mode_vm(regs
)) {
1126 error_code
|= PF_USER
;
1127 flags
|= FAULT_FLAG_USER
;
1129 if (regs
->flags
& X86_EFLAGS_IF
)
1133 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS
, 1, regs
, address
);
1135 if (error_code
& PF_WRITE
)
1136 flags
|= FAULT_FLAG_WRITE
;
1139 * When running in the kernel we expect faults to occur only to
1140 * addresses in user space. All other faults represent errors in
1141 * the kernel and should generate an OOPS. Unfortunately, in the
1142 * case of an erroneous fault occurring in a code path which already
1143 * holds mmap_sem we will deadlock attempting to validate the fault
1144 * against the address space. Luckily the kernel only validly
1145 * references user space from well defined areas of code, which are
1146 * listed in the exceptions table.
1148 * As the vast majority of faults will be valid we will only perform
1149 * the source reference check when there is a possibility of a
1150 * deadlock. Attempt to lock the address space, if we cannot we then
1151 * validate the source. If this is invalid we can skip the address
1152 * space check, thus avoiding the deadlock:
1154 if (unlikely(!down_read_trylock(&mm
->mmap_sem
))) {
1155 if ((error_code
& PF_USER
) == 0 &&
1156 !search_exception_tables(regs
->ip
)) {
1157 bad_area_nosemaphore(regs
, error_code
, address
);
1161 down_read(&mm
->mmap_sem
);
1164 * The above down_read_trylock() might have succeeded in
1165 * which case we'll have missed the might_sleep() from
1171 vma
= find_vma(mm
, address
);
1172 if (unlikely(!vma
)) {
1173 bad_area(regs
, error_code
, address
);
1176 if (likely(vma
->vm_start
<= address
))
1178 if (unlikely(!(vma
->vm_flags
& VM_GROWSDOWN
))) {
1179 bad_area(regs
, error_code
, address
);
1182 if (error_code
& PF_USER
) {
1184 * Accessing the stack below %sp is always a bug.
1185 * The large cushion allows instructions like enter
1186 * and pusha to work. ("enter $65535, $31" pushes
1187 * 32 pointers and then decrements %sp by 65535.)
1189 if (unlikely(address
+ 65536 + 32 * sizeof(unsigned long) < regs
->sp
)) {
1190 bad_area(regs
, error_code
, address
);
1194 if (unlikely(expand_stack(vma
, address
))) {
1195 bad_area(regs
, error_code
, address
);
1200 * Ok, we have a good vm_area for this memory access, so
1201 * we can handle it..
1204 if (unlikely(access_error(error_code
, vma
))) {
1205 bad_area_access_error(regs
, error_code
, address
);
1210 * If for any reason at all we couldn't handle the fault,
1211 * make sure we exit gracefully rather than endlessly redo
1214 fault
= handle_mm_fault(mm
, vma
, address
, flags
);
1217 * If we need to retry but a fatal signal is pending, handle the
1218 * signal first. We do not need to release the mmap_sem because it
1219 * would already be released in __lock_page_or_retry in mm/filemap.c.
1221 if (unlikely((fault
& VM_FAULT_RETRY
) && fatal_signal_pending(current
)))
1224 if (unlikely(fault
& VM_FAULT_ERROR
)) {
1225 mm_fault_error(regs
, error_code
, address
, fault
);
1230 * Major/minor page fault accounting is only done on the
1231 * initial attempt. If we go through a retry, it is extremely
1232 * likely that the page will be found in page cache at that point.
1234 if (flags
& FAULT_FLAG_ALLOW_RETRY
) {
1235 if (fault
& VM_FAULT_MAJOR
) {
1237 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ
, 1,
1241 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN
, 1,
1244 if (fault
& VM_FAULT_RETRY
) {
1245 /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
1247 flags
&= ~FAULT_FLAG_ALLOW_RETRY
;
1248 flags
|= FAULT_FLAG_TRIED
;
1253 check_v8086_mode(regs
, address
, tsk
);
1255 up_read(&mm
->mmap_sem
);
1258 dotraplinkage
void __kprobes notrace
1259 do_page_fault(struct pt_regs
*regs
, unsigned long error_code
)
1261 unsigned long address
= read_cr2(); /* Get the faulting address */
1262 enum ctx_state prev_state
;
1265 * We must have this function tagged with __kprobes, notrace and call
1266 * read_cr2() before calling anything else. To avoid calling any kind
1267 * of tracing machinery before we've observed the CR2 value.
1269 * exception_{enter,exit}() contain all sorts of tracepoints.
1272 prev_state
= exception_enter();
1273 __do_page_fault(regs
, error_code
, address
);
1274 exception_exit(prev_state
);
1277 #ifdef CONFIG_TRACING
1278 static void trace_page_fault_entries(unsigned long address
, struct pt_regs
*regs
,
1279 unsigned long error_code
)
1281 if (user_mode(regs
))
1282 trace_page_fault_user(address
, regs
, error_code
);
1284 trace_page_fault_kernel(address
, regs
, error_code
);
1287 dotraplinkage
void __kprobes notrace
1288 trace_do_page_fault(struct pt_regs
*regs
, unsigned long error_code
)
1291 * The exception_enter and tracepoint processing could
1292 * trigger another page faults (user space callchain
1293 * reading) and destroy the original cr2 value, so read
1294 * the faulting address now.
1296 unsigned long address
= read_cr2();
1297 enum ctx_state prev_state
;
1299 prev_state
= exception_enter();
1300 trace_page_fault_entries(address
, regs
, error_code
);
1301 __do_page_fault(regs
, error_code
, address
);
1302 exception_exit(prev_state
);
1304 #endif /* CONFIG_TRACING */