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/sched.h> /* test_thread_flag(), ... */
7 #include <linux/kdebug.h> /* oops_begin/end, ... */
8 #include <linux/module.h> /* search_exception_table */
9 #include <linux/bootmem.h> /* max_low_pfn */
10 #include <linux/kprobes.h> /* NOKPROBE_SYMBOL, ... */
11 #include <linux/mmiotrace.h> /* kmmio_handler, ... */
12 #include <linux/perf_event.h> /* perf_sw_event */
13 #include <linux/hugetlb.h> /* hstate_index_to_shift */
14 #include <linux/prefetch.h> /* prefetchw */
15 #include <linux/context_tracking.h> /* exception_enter(), ... */
16 #include <linux/uaccess.h> /* faulthandler_disabled() */
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 */
23 #include <asm/vm86.h> /* struct vm86 */
25 #define CREATE_TRACE_POINTS
26 #include <asm/trace/exceptions.h>
29 * Page fault error code bits:
31 * bit 0 == 0: no page found 1: protection fault
32 * bit 1 == 0: read access 1: write access
33 * bit 2 == 0: kernel-mode access 1: user-mode access
34 * bit 3 == 1: use of reserved bit detected
35 * bit 4 == 1: fault was an instruction fetch
36 * bit 5 == 1: protection keys block access
38 enum x86_pf_error_code
{
49 * Returns 0 if mmiotrace is disabled, or if the fault is not
50 * handled by mmiotrace:
52 static nokprobe_inline
int
53 kmmio_fault(struct pt_regs
*regs
, unsigned long addr
)
55 if (unlikely(is_kmmio_active()))
56 if (kmmio_handler(regs
, addr
) == 1)
61 static nokprobe_inline
int kprobes_fault(struct pt_regs
*regs
)
65 /* kprobe_running() needs smp_processor_id() */
66 if (kprobes_built_in() && !user_mode(regs
)) {
68 if (kprobe_running() && kprobe_fault_handler(regs
, 14))
81 * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch.
82 * Check that here and ignore it.
86 * Sometimes the CPU reports invalid exceptions on prefetch.
87 * Check that here and ignore it.
89 * Opcode checker based on code by Richard Brunner.
92 check_prefetch_opcode(struct pt_regs
*regs
, unsigned char *instr
,
93 unsigned char opcode
, int *prefetch
)
95 unsigned char instr_hi
= opcode
& 0xf0;
96 unsigned char instr_lo
= opcode
& 0x0f;
102 * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes.
103 * In X86_64 long mode, the CPU will signal invalid
104 * opcode if some of these prefixes are present so
105 * X86_64 will never get here anyway
107 return ((instr_lo
& 7) == 0x6);
111 * In AMD64 long mode 0x40..0x4F are valid REX prefixes
112 * Need to figure out under what instruction mode the
113 * instruction was issued. Could check the LDT for lm,
114 * but for now it's good enough to assume that long
115 * mode only uses well known segments or kernel.
117 return (!user_mode(regs
) || user_64bit_mode(regs
));
120 /* 0x64 thru 0x67 are valid prefixes in all modes. */
121 return (instr_lo
& 0xC) == 0x4;
123 /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */
124 return !instr_lo
|| (instr_lo
>>1) == 1;
126 /* Prefetch instruction is 0x0F0D or 0x0F18 */
127 if (probe_kernel_address(instr
, opcode
))
130 *prefetch
= (instr_lo
== 0xF) &&
131 (opcode
== 0x0D || opcode
== 0x18);
139 is_prefetch(struct pt_regs
*regs
, unsigned long error_code
, unsigned long addr
)
141 unsigned char *max_instr
;
142 unsigned char *instr
;
146 * If it was a exec (instruction fetch) fault on NX page, then
147 * do not ignore the fault:
149 if (error_code
& PF_INSTR
)
152 instr
= (void *)convert_ip_to_linear(current
, regs
);
153 max_instr
= instr
+ 15;
155 if (user_mode(regs
) && instr
>= (unsigned char *)TASK_SIZE_MAX
)
158 while (instr
< max_instr
) {
159 unsigned char opcode
;
161 if (probe_kernel_address(instr
, opcode
))
166 if (!check_prefetch_opcode(regs
, instr
, opcode
, &prefetch
))
173 force_sig_info_fault(int si_signo
, int si_code
, unsigned long address
,
174 struct task_struct
*tsk
, int fault
)
179 info
.si_signo
= si_signo
;
181 info
.si_code
= si_code
;
182 info
.si_addr
= (void __user
*)address
;
183 if (fault
& VM_FAULT_HWPOISON_LARGE
)
184 lsb
= hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault
));
185 if (fault
& VM_FAULT_HWPOISON
)
187 info
.si_addr_lsb
= lsb
;
189 force_sig_info(si_signo
, &info
, tsk
);
192 DEFINE_SPINLOCK(pgd_lock
);
196 static inline pmd_t
*vmalloc_sync_one(pgd_t
*pgd
, unsigned long address
)
198 unsigned index
= pgd_index(address
);
204 pgd_k
= init_mm
.pgd
+ index
;
206 if (!pgd_present(*pgd_k
))
210 * set_pgd(pgd, *pgd_k); here would be useless on PAE
211 * and redundant with the set_pmd() on non-PAE. As would
214 pud
= pud_offset(pgd
, address
);
215 pud_k
= pud_offset(pgd_k
, address
);
216 if (!pud_present(*pud_k
))
219 pmd
= pmd_offset(pud
, address
);
220 pmd_k
= pmd_offset(pud_k
, address
);
221 if (!pmd_present(*pmd_k
))
224 if (!pmd_present(*pmd
))
225 set_pmd(pmd
, *pmd_k
);
227 BUG_ON(pmd_page(*pmd
) != pmd_page(*pmd_k
));
232 void vmalloc_sync_all(void)
234 unsigned long address
;
236 if (SHARED_KERNEL_PMD
)
239 for (address
= VMALLOC_START
& PMD_MASK
;
240 address
>= TASK_SIZE
&& address
< FIXADDR_TOP
;
241 address
+= PMD_SIZE
) {
244 spin_lock(&pgd_lock
);
245 list_for_each_entry(page
, &pgd_list
, lru
) {
246 spinlock_t
*pgt_lock
;
249 /* the pgt_lock only for Xen */
250 pgt_lock
= &pgd_page_get_mm(page
)->page_table_lock
;
253 ret
= vmalloc_sync_one(page_address(page
), address
);
254 spin_unlock(pgt_lock
);
259 spin_unlock(&pgd_lock
);
266 * Handle a fault on the vmalloc or module mapping area
268 static noinline
int vmalloc_fault(unsigned long address
)
270 unsigned long pgd_paddr
;
274 /* Make sure we are in vmalloc area: */
275 if (!(address
>= VMALLOC_START
&& address
< VMALLOC_END
))
278 WARN_ON_ONCE(in_nmi());
281 * Synchronize this task's top level page-table
282 * with the 'reference' page table.
284 * Do _not_ use "current" here. We might be inside
285 * an interrupt in the middle of a task switch..
287 pgd_paddr
= read_cr3();
288 pmd_k
= vmalloc_sync_one(__va(pgd_paddr
), address
);
292 pte_k
= pte_offset_kernel(pmd_k
, address
);
293 if (!pte_present(*pte_k
))
298 NOKPROBE_SYMBOL(vmalloc_fault
);
301 * Did it hit the DOS screen memory VA from vm86 mode?
304 check_v8086_mode(struct pt_regs
*regs
, unsigned long address
,
305 struct task_struct
*tsk
)
310 if (!v8086_mode(regs
) || !tsk
->thread
.vm86
)
313 bit
= (address
- 0xA0000) >> PAGE_SHIFT
;
315 tsk
->thread
.vm86
->screen_bitmap
|= 1 << bit
;
319 static bool low_pfn(unsigned long pfn
)
321 return pfn
< max_low_pfn
;
324 static void dump_pagetable(unsigned long address
)
326 pgd_t
*base
= __va(read_cr3());
327 pgd_t
*pgd
= &base
[pgd_index(address
)];
331 #ifdef CONFIG_X86_PAE
332 printk("*pdpt = %016Lx ", pgd_val(*pgd
));
333 if (!low_pfn(pgd_val(*pgd
) >> PAGE_SHIFT
) || !pgd_present(*pgd
))
336 pmd
= pmd_offset(pud_offset(pgd
, address
), address
);
337 printk(KERN_CONT
"*pde = %0*Lx ", sizeof(*pmd
) * 2, (u64
)pmd_val(*pmd
));
340 * We must not directly access the pte in the highpte
341 * case if the page table is located in highmem.
342 * And let's rather not kmap-atomic the pte, just in case
343 * it's allocated already:
345 if (!low_pfn(pmd_pfn(*pmd
)) || !pmd_present(*pmd
) || pmd_large(*pmd
))
348 pte
= pte_offset_kernel(pmd
, address
);
349 printk("*pte = %0*Lx ", sizeof(*pte
) * 2, (u64
)pte_val(*pte
));
354 #else /* CONFIG_X86_64: */
356 void vmalloc_sync_all(void)
358 sync_global_pgds(VMALLOC_START
& PGDIR_MASK
, VMALLOC_END
, 0);
364 * Handle a fault on the vmalloc area
366 * This assumes no large pages in there.
368 static noinline
int vmalloc_fault(unsigned long address
)
370 pgd_t
*pgd
, *pgd_ref
;
371 pud_t
*pud
, *pud_ref
;
372 pmd_t
*pmd
, *pmd_ref
;
373 pte_t
*pte
, *pte_ref
;
375 /* Make sure we are in vmalloc area: */
376 if (!(address
>= VMALLOC_START
&& address
< VMALLOC_END
))
379 WARN_ON_ONCE(in_nmi());
382 * Copy kernel mappings over when needed. This can also
383 * happen within a race in page table update. In the later
386 pgd
= pgd_offset(current
->active_mm
, address
);
387 pgd_ref
= pgd_offset_k(address
);
388 if (pgd_none(*pgd_ref
))
391 if (pgd_none(*pgd
)) {
392 set_pgd(pgd
, *pgd_ref
);
393 arch_flush_lazy_mmu_mode();
395 BUG_ON(pgd_page_vaddr(*pgd
) != pgd_page_vaddr(*pgd_ref
));
399 * Below here mismatches are bugs because these lower tables
403 pud
= pud_offset(pgd
, address
);
404 pud_ref
= pud_offset(pgd_ref
, address
);
405 if (pud_none(*pud_ref
))
408 if (pud_none(*pud
) || pud_page_vaddr(*pud
) != pud_page_vaddr(*pud_ref
))
411 pmd
= pmd_offset(pud
, address
);
412 pmd_ref
= pmd_offset(pud_ref
, address
);
413 if (pmd_none(*pmd_ref
))
416 if (pmd_none(*pmd
) || pmd_page(*pmd
) != pmd_page(*pmd_ref
))
419 pte_ref
= pte_offset_kernel(pmd_ref
, address
);
420 if (!pte_present(*pte_ref
))
423 pte
= pte_offset_kernel(pmd
, address
);
426 * Don't use pte_page here, because the mappings can point
427 * outside mem_map, and the NUMA hash lookup cannot handle
430 if (!pte_present(*pte
) || pte_pfn(*pte
) != pte_pfn(*pte_ref
))
435 NOKPROBE_SYMBOL(vmalloc_fault
);
437 #ifdef CONFIG_CPU_SUP_AMD
438 static const char errata93_warning
[] =
440 "******* Your BIOS seems to not contain a fix for K8 errata #93\n"
441 "******* Working around it, but it may cause SEGVs or burn power.\n"
442 "******* Please consider a BIOS update.\n"
443 "******* Disabling USB legacy in the BIOS may also help.\n";
447 * No vm86 mode in 64-bit mode:
450 check_v8086_mode(struct pt_regs
*regs
, unsigned long address
,
451 struct task_struct
*tsk
)
455 static int bad_address(void *p
)
459 return probe_kernel_address((unsigned long *)p
, dummy
);
462 static void dump_pagetable(unsigned long address
)
464 pgd_t
*base
= __va(read_cr3() & PHYSICAL_PAGE_MASK
);
465 pgd_t
*pgd
= base
+ pgd_index(address
);
470 if (bad_address(pgd
))
473 printk("PGD %lx ", pgd_val(*pgd
));
475 if (!pgd_present(*pgd
))
478 pud
= pud_offset(pgd
, address
);
479 if (bad_address(pud
))
482 printk("PUD %lx ", pud_val(*pud
));
483 if (!pud_present(*pud
) || pud_large(*pud
))
486 pmd
= pmd_offset(pud
, address
);
487 if (bad_address(pmd
))
490 printk("PMD %lx ", pmd_val(*pmd
));
491 if (!pmd_present(*pmd
) || pmd_large(*pmd
))
494 pte
= pte_offset_kernel(pmd
, address
);
495 if (bad_address(pte
))
498 printk("PTE %lx", pte_val(*pte
));
506 #endif /* CONFIG_X86_64 */
509 * Workaround for K8 erratum #93 & buggy BIOS.
511 * BIOS SMM functions are required to use a specific workaround
512 * to avoid corruption of the 64bit RIP register on C stepping K8.
514 * A lot of BIOS that didn't get tested properly miss this.
516 * The OS sees this as a page fault with the upper 32bits of RIP cleared.
517 * Try to work around it here.
519 * Note we only handle faults in kernel here.
520 * Does nothing on 32-bit.
522 static int is_errata93(struct pt_regs
*regs
, unsigned long address
)
524 #if defined(CONFIG_X86_64) && defined(CONFIG_CPU_SUP_AMD)
525 if (boot_cpu_data
.x86_vendor
!= X86_VENDOR_AMD
526 || boot_cpu_data
.x86
!= 0xf)
529 if (address
!= regs
->ip
)
532 if ((address
>> 32) != 0)
535 address
|= 0xffffffffUL
<< 32;
536 if ((address
>= (u64
)_stext
&& address
<= (u64
)_etext
) ||
537 (address
>= MODULES_VADDR
&& address
<= MODULES_END
)) {
538 printk_once(errata93_warning
);
547 * Work around K8 erratum #100 K8 in compat mode occasionally jumps
548 * to illegal addresses >4GB.
550 * We catch this in the page fault handler because these addresses
551 * are not reachable. Just detect this case and return. Any code
552 * segment in LDT is compatibility mode.
554 static int is_errata100(struct pt_regs
*regs
, unsigned long address
)
557 if ((regs
->cs
== __USER32_CS
|| (regs
->cs
& (1<<2))) && (address
>> 32))
563 static int is_f00f_bug(struct pt_regs
*regs
, unsigned long address
)
565 #ifdef CONFIG_X86_F00F_BUG
569 * Pentium F0 0F C7 C8 bug workaround:
571 if (boot_cpu_has_bug(X86_BUG_F00F
)) {
572 nr
= (address
- idt_descr
.address
) >> 3;
575 do_invalid_op(regs
, 0);
583 static const char nx_warning
[] = KERN_CRIT
584 "kernel tried to execute NX-protected page - exploit attempt? (uid: %d)\n";
585 static const char smep_warning
[] = KERN_CRIT
586 "unable to execute userspace code (SMEP?) (uid: %d)\n";
589 show_fault_oops(struct pt_regs
*regs
, unsigned long error_code
,
590 unsigned long address
)
592 if (!oops_may_print())
595 if (error_code
& PF_INSTR
) {
600 pgd
= __va(read_cr3() & PHYSICAL_PAGE_MASK
);
601 pgd
+= pgd_index(address
);
603 pte
= lookup_address_in_pgd(pgd
, address
, &level
);
605 if (pte
&& pte_present(*pte
) && !pte_exec(*pte
))
606 printk(nx_warning
, from_kuid(&init_user_ns
, current_uid()));
607 if (pte
&& pte_present(*pte
) && pte_exec(*pte
) &&
608 (pgd_flags(*pgd
) & _PAGE_USER
) &&
609 (__read_cr4() & X86_CR4_SMEP
))
610 printk(smep_warning
, from_kuid(&init_user_ns
, current_uid()));
613 printk(KERN_ALERT
"BUG: unable to handle kernel ");
614 if (address
< PAGE_SIZE
)
615 printk(KERN_CONT
"NULL pointer dereference");
617 printk(KERN_CONT
"paging request");
619 printk(KERN_CONT
" at %p\n", (void *) address
);
620 printk(KERN_ALERT
"IP:");
621 printk_address(regs
->ip
);
623 dump_pagetable(address
);
627 pgtable_bad(struct pt_regs
*regs
, unsigned long error_code
,
628 unsigned long address
)
630 struct task_struct
*tsk
;
634 flags
= oops_begin();
638 printk(KERN_ALERT
"%s: Corrupted page table at address %lx\n",
640 dump_pagetable(address
);
642 tsk
->thread
.cr2
= address
;
643 tsk
->thread
.trap_nr
= X86_TRAP_PF
;
644 tsk
->thread
.error_code
= error_code
;
646 if (__die("Bad pagetable", regs
, error_code
))
649 oops_end(flags
, regs
, sig
);
653 no_context(struct pt_regs
*regs
, unsigned long error_code
,
654 unsigned long address
, int signal
, int si_code
)
656 struct task_struct
*tsk
= current
;
660 /* Are we prepared to handle this kernel fault? */
661 if (fixup_exception(regs
)) {
663 * Any interrupt that takes a fault gets the fixup. This makes
664 * the below recursive fault logic only apply to a faults from
671 * Per the above we're !in_interrupt(), aka. task context.
673 * In this case we need to make sure we're not recursively
674 * faulting through the emulate_vsyscall() logic.
676 if (current_thread_info()->sig_on_uaccess_error
&& signal
) {
677 tsk
->thread
.trap_nr
= X86_TRAP_PF
;
678 tsk
->thread
.error_code
= error_code
| PF_USER
;
679 tsk
->thread
.cr2
= address
;
681 /* XXX: hwpoison faults will set the wrong code. */
682 force_sig_info_fault(signal
, si_code
, address
, tsk
, 0);
686 * Barring that, we can do the fixup and be happy.
694 * Valid to do another page fault here, because if this fault
695 * had been triggered by is_prefetch fixup_exception would have
700 * Hall of shame of CPU/BIOS bugs.
702 if (is_prefetch(regs
, error_code
, address
))
705 if (is_errata93(regs
, address
))
709 * Oops. The kernel tried to access some bad page. We'll have to
710 * terminate things with extreme prejudice:
712 flags
= oops_begin();
714 show_fault_oops(regs
, error_code
, address
);
716 if (task_stack_end_corrupted(tsk
))
717 printk(KERN_EMERG
"Thread overran stack, or stack corrupted\n");
719 tsk
->thread
.cr2
= address
;
720 tsk
->thread
.trap_nr
= X86_TRAP_PF
;
721 tsk
->thread
.error_code
= error_code
;
724 if (__die("Oops", regs
, error_code
))
727 /* Executive summary in case the body of the oops scrolled away */
728 printk(KERN_DEFAULT
"CR2: %016lx\n", address
);
730 oops_end(flags
, regs
, sig
);
734 * Print out info about fatal segfaults, if the show_unhandled_signals
738 show_signal_msg(struct pt_regs
*regs
, unsigned long error_code
,
739 unsigned long address
, struct task_struct
*tsk
)
741 if (!unhandled_signal(tsk
, SIGSEGV
))
744 if (!printk_ratelimit())
747 printk("%s%s[%d]: segfault at %lx ip %p sp %p error %lx",
748 task_pid_nr(tsk
) > 1 ? KERN_INFO
: KERN_EMERG
,
749 tsk
->comm
, task_pid_nr(tsk
), address
,
750 (void *)regs
->ip
, (void *)regs
->sp
, error_code
);
752 print_vma_addr(KERN_CONT
" in ", regs
->ip
);
754 printk(KERN_CONT
"\n");
758 __bad_area_nosemaphore(struct pt_regs
*regs
, unsigned long error_code
,
759 unsigned long address
, int si_code
)
761 struct task_struct
*tsk
= current
;
763 /* User mode accesses just cause a SIGSEGV */
764 if (error_code
& PF_USER
) {
766 * It's possible to have interrupts off here:
771 * Valid to do another page fault here because this one came
774 if (is_prefetch(regs
, error_code
, address
))
777 if (is_errata100(regs
, address
))
782 * Instruction fetch faults in the vsyscall page might need
785 if (unlikely((error_code
& PF_INSTR
) &&
786 ((address
& ~0xfff) == VSYSCALL_ADDR
))) {
787 if (emulate_vsyscall(regs
, address
))
791 /* Kernel addresses are always protection faults: */
792 if (address
>= TASK_SIZE
)
793 error_code
|= PF_PROT
;
795 if (likely(show_unhandled_signals
))
796 show_signal_msg(regs
, error_code
, address
, tsk
);
798 tsk
->thread
.cr2
= address
;
799 tsk
->thread
.error_code
= error_code
;
800 tsk
->thread
.trap_nr
= X86_TRAP_PF
;
802 force_sig_info_fault(SIGSEGV
, si_code
, address
, tsk
, 0);
807 if (is_f00f_bug(regs
, address
))
810 no_context(regs
, error_code
, address
, SIGSEGV
, si_code
);
814 bad_area_nosemaphore(struct pt_regs
*regs
, unsigned long error_code
,
815 unsigned long address
)
817 __bad_area_nosemaphore(regs
, error_code
, address
, SEGV_MAPERR
);
821 __bad_area(struct pt_regs
*regs
, unsigned long error_code
,
822 unsigned long address
, int si_code
)
824 struct mm_struct
*mm
= current
->mm
;
827 * Something tried to access memory that isn't in our memory map..
828 * Fix it, but check if it's kernel or user first..
830 up_read(&mm
->mmap_sem
);
832 __bad_area_nosemaphore(regs
, error_code
, address
, si_code
);
836 bad_area(struct pt_regs
*regs
, unsigned long error_code
, unsigned long address
)
838 __bad_area(regs
, error_code
, address
, SEGV_MAPERR
);
842 bad_area_access_error(struct pt_regs
*regs
, unsigned long error_code
,
843 unsigned long address
)
845 __bad_area(regs
, error_code
, address
, SEGV_ACCERR
);
849 do_sigbus(struct pt_regs
*regs
, unsigned long error_code
, unsigned long address
,
852 struct task_struct
*tsk
= current
;
853 int code
= BUS_ADRERR
;
855 /* Kernel mode? Handle exceptions or die: */
856 if (!(error_code
& PF_USER
)) {
857 no_context(regs
, error_code
, address
, SIGBUS
, BUS_ADRERR
);
861 /* User-space => ok to do another page fault: */
862 if (is_prefetch(regs
, error_code
, address
))
865 tsk
->thread
.cr2
= address
;
866 tsk
->thread
.error_code
= error_code
;
867 tsk
->thread
.trap_nr
= X86_TRAP_PF
;
869 #ifdef CONFIG_MEMORY_FAILURE
870 if (fault
& (VM_FAULT_HWPOISON
|VM_FAULT_HWPOISON_LARGE
)) {
872 "MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n",
873 tsk
->comm
, tsk
->pid
, address
);
874 code
= BUS_MCEERR_AR
;
877 force_sig_info_fault(SIGBUS
, code
, address
, tsk
, fault
);
881 mm_fault_error(struct pt_regs
*regs
, unsigned long error_code
,
882 unsigned long address
, unsigned int fault
)
884 if (fatal_signal_pending(current
) && !(error_code
& PF_USER
)) {
885 no_context(regs
, error_code
, address
, 0, 0);
889 if (fault
& VM_FAULT_OOM
) {
890 /* Kernel mode? Handle exceptions or die: */
891 if (!(error_code
& PF_USER
)) {
892 no_context(regs
, error_code
, address
,
893 SIGSEGV
, SEGV_MAPERR
);
898 * We ran out of memory, call the OOM killer, and return the
899 * userspace (which will retry the fault, or kill us if we got
902 pagefault_out_of_memory();
904 if (fault
& (VM_FAULT_SIGBUS
|VM_FAULT_HWPOISON
|
905 VM_FAULT_HWPOISON_LARGE
))
906 do_sigbus(regs
, error_code
, address
, fault
);
907 else if (fault
& VM_FAULT_SIGSEGV
)
908 bad_area_nosemaphore(regs
, error_code
, address
);
914 static int spurious_fault_check(unsigned long error_code
, pte_t
*pte
)
916 if ((error_code
& PF_WRITE
) && !pte_write(*pte
))
919 if ((error_code
& PF_INSTR
) && !pte_exec(*pte
))
922 * Note: We do not do lazy flushing on protection key
923 * changes, so no spurious fault will ever set PF_PK.
925 if ((error_code
& PF_PK
))
932 * Handle a spurious fault caused by a stale TLB entry.
934 * This allows us to lazily refresh the TLB when increasing the
935 * permissions of a kernel page (RO -> RW or NX -> X). Doing it
936 * eagerly is very expensive since that implies doing a full
937 * cross-processor TLB flush, even if no stale TLB entries exist
938 * on other processors.
940 * Spurious faults may only occur if the TLB contains an entry with
941 * fewer permission than the page table entry. Non-present (P = 0)
942 * and reserved bit (R = 1) faults are never spurious.
944 * There are no security implications to leaving a stale TLB when
945 * increasing the permissions on a page.
947 * Returns non-zero if a spurious fault was handled, zero otherwise.
949 * See Intel Developer's Manual Vol 3 Section 4.10.4.3, bullet 3
950 * (Optional Invalidation).
953 spurious_fault(unsigned long error_code
, unsigned long address
)
962 * Only writes to RO or instruction fetches from NX may cause
965 * These could be from user or supervisor accesses but the TLB
966 * is only lazily flushed after a kernel mapping protection
967 * change, so user accesses are not expected to cause spurious
970 if (error_code
!= (PF_WRITE
| PF_PROT
)
971 && error_code
!= (PF_INSTR
| PF_PROT
))
974 pgd
= init_mm
.pgd
+ pgd_index(address
);
975 if (!pgd_present(*pgd
))
978 pud
= pud_offset(pgd
, address
);
979 if (!pud_present(*pud
))
983 return spurious_fault_check(error_code
, (pte_t
*) pud
);
985 pmd
= pmd_offset(pud
, address
);
986 if (!pmd_present(*pmd
))
990 return spurious_fault_check(error_code
, (pte_t
*) pmd
);
992 pte
= pte_offset_kernel(pmd
, address
);
993 if (!pte_present(*pte
))
996 ret
= spurious_fault_check(error_code
, pte
);
1001 * Make sure we have permissions in PMD.
1002 * If not, then there's a bug in the page tables:
1004 ret
= spurious_fault_check(error_code
, (pte_t
*) pmd
);
1005 WARN_ONCE(!ret
, "PMD has incorrect permission bits\n");
1009 NOKPROBE_SYMBOL(spurious_fault
);
1011 int show_unhandled_signals
= 1;
1014 access_error(unsigned long error_code
, struct vm_area_struct
*vma
)
1016 if (error_code
& PF_WRITE
) {
1017 /* write, present and write, not present: */
1018 if (unlikely(!(vma
->vm_flags
& VM_WRITE
)))
1023 /* read, present: */
1024 if (unlikely(error_code
& PF_PROT
))
1027 /* read, not present: */
1028 if (unlikely(!(vma
->vm_flags
& (VM_READ
| VM_EXEC
| VM_WRITE
))))
1034 static int fault_in_kernel_space(unsigned long address
)
1036 return address
>= TASK_SIZE_MAX
;
1039 static inline bool smap_violation(int error_code
, struct pt_regs
*regs
)
1041 if (!IS_ENABLED(CONFIG_X86_SMAP
))
1044 if (!static_cpu_has(X86_FEATURE_SMAP
))
1047 if (error_code
& PF_USER
)
1050 if (!user_mode(regs
) && (regs
->flags
& X86_EFLAGS_AC
))
1057 * This routine handles page faults. It determines the address,
1058 * and the problem, and then passes it off to one of the appropriate
1061 * This function must have noinline because both callers
1062 * {,trace_}do_page_fault() have notrace on. Having this an actual function
1063 * guarantees there's a function trace entry.
1065 static noinline
void
1066 __do_page_fault(struct pt_regs
*regs
, unsigned long error_code
,
1067 unsigned long address
)
1069 struct vm_area_struct
*vma
;
1070 struct task_struct
*tsk
;
1071 struct mm_struct
*mm
;
1072 int fault
, major
= 0;
1073 unsigned int flags
= FAULT_FLAG_ALLOW_RETRY
| FAULT_FLAG_KILLABLE
;
1079 * Detect and handle instructions that would cause a page fault for
1080 * both a tracked kernel page and a userspace page.
1082 if (kmemcheck_active(regs
))
1083 kmemcheck_hide(regs
);
1084 prefetchw(&mm
->mmap_sem
);
1086 if (unlikely(kmmio_fault(regs
, address
)))
1090 * We fault-in kernel-space virtual memory on-demand. The
1091 * 'reference' page table is init_mm.pgd.
1093 * NOTE! We MUST NOT take any locks for this case. We may
1094 * be in an interrupt or a critical region, and should
1095 * only copy the information from the master page table,
1098 * This verifies that the fault happens in kernel space
1099 * (error_code & 4) == 0, and that the fault was not a
1100 * protection error (error_code & 9) == 0.
1102 if (unlikely(fault_in_kernel_space(address
))) {
1103 if (!(error_code
& (PF_RSVD
| PF_USER
| PF_PROT
))) {
1104 if (vmalloc_fault(address
) >= 0)
1107 if (kmemcheck_fault(regs
, address
, error_code
))
1111 /* Can handle a stale RO->RW TLB: */
1112 if (spurious_fault(error_code
, address
))
1115 /* kprobes don't want to hook the spurious faults: */
1116 if (kprobes_fault(regs
))
1119 * Don't take the mm semaphore here. If we fixup a prefetch
1120 * fault we could otherwise deadlock:
1122 bad_area_nosemaphore(regs
, error_code
, address
);
1127 /* kprobes don't want to hook the spurious faults: */
1128 if (unlikely(kprobes_fault(regs
)))
1131 if (unlikely(error_code
& PF_RSVD
))
1132 pgtable_bad(regs
, error_code
, address
);
1134 if (unlikely(smap_violation(error_code
, regs
))) {
1135 bad_area_nosemaphore(regs
, error_code
, address
);
1140 * If we're in an interrupt, have no user context or are running
1141 * in a region with pagefaults disabled then we must not take the fault
1143 if (unlikely(faulthandler_disabled() || !mm
)) {
1144 bad_area_nosemaphore(regs
, error_code
, address
);
1149 * It's safe to allow irq's after cr2 has been saved and the
1150 * vmalloc fault has been handled.
1152 * User-mode registers count as a user access even for any
1153 * potential system fault or CPU buglet:
1155 if (user_mode(regs
)) {
1157 error_code
|= PF_USER
;
1158 flags
|= FAULT_FLAG_USER
;
1160 if (regs
->flags
& X86_EFLAGS_IF
)
1164 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS
, 1, regs
, address
);
1166 if (error_code
& PF_WRITE
)
1167 flags
|= FAULT_FLAG_WRITE
;
1170 * When running in the kernel we expect faults to occur only to
1171 * addresses in user space. All other faults represent errors in
1172 * the kernel and should generate an OOPS. Unfortunately, in the
1173 * case of an erroneous fault occurring in a code path which already
1174 * holds mmap_sem we will deadlock attempting to validate the fault
1175 * against the address space. Luckily the kernel only validly
1176 * references user space from well defined areas of code, which are
1177 * listed in the exceptions table.
1179 * As the vast majority of faults will be valid we will only perform
1180 * the source reference check when there is a possibility of a
1181 * deadlock. Attempt to lock the address space, if we cannot we then
1182 * validate the source. If this is invalid we can skip the address
1183 * space check, thus avoiding the deadlock:
1185 if (unlikely(!down_read_trylock(&mm
->mmap_sem
))) {
1186 if ((error_code
& PF_USER
) == 0 &&
1187 !search_exception_tables(regs
->ip
)) {
1188 bad_area_nosemaphore(regs
, error_code
, address
);
1192 down_read(&mm
->mmap_sem
);
1195 * The above down_read_trylock() might have succeeded in
1196 * which case we'll have missed the might_sleep() from
1202 vma
= find_vma(mm
, address
);
1203 if (unlikely(!vma
)) {
1204 bad_area(regs
, error_code
, address
);
1207 if (likely(vma
->vm_start
<= address
))
1209 if (unlikely(!(vma
->vm_flags
& VM_GROWSDOWN
))) {
1210 bad_area(regs
, error_code
, address
);
1213 if (error_code
& PF_USER
) {
1215 * Accessing the stack below %sp is always a bug.
1216 * The large cushion allows instructions like enter
1217 * and pusha to work. ("enter $65535, $31" pushes
1218 * 32 pointers and then decrements %sp by 65535.)
1220 if (unlikely(address
+ 65536 + 32 * sizeof(unsigned long) < regs
->sp
)) {
1221 bad_area(regs
, error_code
, address
);
1225 if (unlikely(expand_stack(vma
, address
))) {
1226 bad_area(regs
, error_code
, address
);
1231 * Ok, we have a good vm_area for this memory access, so
1232 * we can handle it..
1235 if (unlikely(access_error(error_code
, vma
))) {
1236 bad_area_access_error(regs
, error_code
, address
);
1241 * If for any reason at all we couldn't handle the fault,
1242 * make sure we exit gracefully rather than endlessly redo
1243 * the fault. Since we never set FAULT_FLAG_RETRY_NOWAIT, if
1244 * we get VM_FAULT_RETRY back, the mmap_sem has been unlocked.
1246 fault
= handle_mm_fault(mm
, vma
, address
, flags
);
1247 major
|= fault
& VM_FAULT_MAJOR
;
1250 * If we need to retry the mmap_sem has already been released,
1251 * and if there is a fatal signal pending there is no guarantee
1252 * that we made any progress. Handle this case first.
1254 if (unlikely(fault
& VM_FAULT_RETRY
)) {
1255 /* Retry at most once */
1256 if (flags
& FAULT_FLAG_ALLOW_RETRY
) {
1257 flags
&= ~FAULT_FLAG_ALLOW_RETRY
;
1258 flags
|= FAULT_FLAG_TRIED
;
1259 if (!fatal_signal_pending(tsk
))
1263 /* User mode? Just return to handle the fatal exception */
1264 if (flags
& FAULT_FLAG_USER
)
1267 /* Not returning to user mode? Handle exceptions or die: */
1268 no_context(regs
, error_code
, address
, SIGBUS
, BUS_ADRERR
);
1272 up_read(&mm
->mmap_sem
);
1273 if (unlikely(fault
& VM_FAULT_ERROR
)) {
1274 mm_fault_error(regs
, error_code
, address
, fault
);
1279 * Major/minor page fault accounting. If any of the events
1280 * returned VM_FAULT_MAJOR, we account it as a major fault.
1284 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ
, 1, regs
, address
);
1287 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN
, 1, regs
, address
);
1290 check_v8086_mode(regs
, address
, tsk
);
1292 NOKPROBE_SYMBOL(__do_page_fault
);
1294 dotraplinkage
void notrace
1295 do_page_fault(struct pt_regs
*regs
, unsigned long error_code
)
1297 unsigned long address
= read_cr2(); /* Get the faulting address */
1298 enum ctx_state prev_state
;
1301 * We must have this function tagged with __kprobes, notrace and call
1302 * read_cr2() before calling anything else. To avoid calling any kind
1303 * of tracing machinery before we've observed the CR2 value.
1305 * exception_{enter,exit}() contain all sorts of tracepoints.
1308 prev_state
= exception_enter();
1309 __do_page_fault(regs
, error_code
, address
);
1310 exception_exit(prev_state
);
1312 NOKPROBE_SYMBOL(do_page_fault
);
1314 #ifdef CONFIG_TRACING
1315 static nokprobe_inline
void
1316 trace_page_fault_entries(unsigned long address
, struct pt_regs
*regs
,
1317 unsigned long error_code
)
1319 if (user_mode(regs
))
1320 trace_page_fault_user(address
, regs
, error_code
);
1322 trace_page_fault_kernel(address
, regs
, error_code
);
1325 dotraplinkage
void notrace
1326 trace_do_page_fault(struct pt_regs
*regs
, unsigned long error_code
)
1329 * The exception_enter and tracepoint processing could
1330 * trigger another page faults (user space callchain
1331 * reading) and destroy the original cr2 value, so read
1332 * the faulting address now.
1334 unsigned long address
= read_cr2();
1335 enum ctx_state prev_state
;
1337 prev_state
= exception_enter();
1338 trace_page_fault_entries(address
, regs
, error_code
);
1339 __do_page_fault(regs
, error_code
, address
);
1340 exception_exit(prev_state
);
1342 NOKPROBE_SYMBOL(trace_do_page_fault
);
1343 #endif /* CONFIG_TRACING */