1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 1995 Linus Torvalds
4 * Copyright (C) 2001, 2002 Andi Kleen, SuSE Labs.
5 * Copyright (C) 2008-2009, Red Hat Inc., Ingo Molnar
7 #include <linux/sched.h> /* test_thread_flag(), ... */
8 #include <linux/sched/task_stack.h> /* task_stack_*(), ... */
9 #include <linux/kdebug.h> /* oops_begin/end, ... */
10 #include <linux/extable.h> /* search_exception_tables */
11 #include <linux/bootmem.h> /* max_low_pfn */
12 #include <linux/kprobes.h> /* NOKPROBE_SYMBOL, ... */
13 #include <linux/mmiotrace.h> /* kmmio_handler, ... */
14 #include <linux/perf_event.h> /* perf_sw_event */
15 #include <linux/hugetlb.h> /* hstate_index_to_shift */
16 #include <linux/prefetch.h> /* prefetchw */
17 #include <linux/context_tracking.h> /* exception_enter(), ... */
18 #include <linux/uaccess.h> /* faulthandler_disabled() */
20 #include <asm/cpufeature.h> /* boot_cpu_has, ... */
21 #include <asm/traps.h> /* dotraplinkage, ... */
22 #include <asm/pgalloc.h> /* pgd_*(), ... */
23 #include <asm/kmemcheck.h> /* kmemcheck_*(), ... */
24 #include <asm/fixmap.h> /* VSYSCALL_ADDR */
25 #include <asm/vsyscall.h> /* emulate_vsyscall */
26 #include <asm/vm86.h> /* struct vm86 */
27 #include <asm/mmu_context.h> /* vma_pkey() */
29 #define CREATE_TRACE_POINTS
30 #include <asm/trace/exceptions.h>
33 * Page fault error code bits:
35 * bit 0 == 0: no page found 1: protection fault
36 * bit 1 == 0: read access 1: write access
37 * bit 2 == 0: kernel-mode access 1: user-mode access
38 * bit 3 == 1: use of reserved bit detected
39 * bit 4 == 1: fault was an instruction fetch
40 * bit 5 == 1: protection keys block access
42 enum x86_pf_error_code
{
53 * Returns 0 if mmiotrace is disabled, or if the fault is not
54 * handled by mmiotrace:
56 static nokprobe_inline
int
57 kmmio_fault(struct pt_regs
*regs
, unsigned long addr
)
59 if (unlikely(is_kmmio_active()))
60 if (kmmio_handler(regs
, addr
) == 1)
65 static nokprobe_inline
int kprobes_fault(struct pt_regs
*regs
)
69 /* kprobe_running() needs smp_processor_id() */
70 if (kprobes_built_in() && !user_mode(regs
)) {
72 if (kprobe_running() && kprobe_fault_handler(regs
, 14))
85 * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch.
86 * Check that here and ignore it.
90 * Sometimes the CPU reports invalid exceptions on prefetch.
91 * Check that here and ignore it.
93 * Opcode checker based on code by Richard Brunner.
96 check_prefetch_opcode(struct pt_regs
*regs
, unsigned char *instr
,
97 unsigned char opcode
, int *prefetch
)
99 unsigned char instr_hi
= opcode
& 0xf0;
100 unsigned char instr_lo
= opcode
& 0x0f;
106 * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes.
107 * In X86_64 long mode, the CPU will signal invalid
108 * opcode if some of these prefixes are present so
109 * X86_64 will never get here anyway
111 return ((instr_lo
& 7) == 0x6);
115 * In AMD64 long mode 0x40..0x4F are valid REX prefixes
116 * Need to figure out under what instruction mode the
117 * instruction was issued. Could check the LDT for lm,
118 * but for now it's good enough to assume that long
119 * mode only uses well known segments or kernel.
121 return (!user_mode(regs
) || user_64bit_mode(regs
));
124 /* 0x64 thru 0x67 are valid prefixes in all modes. */
125 return (instr_lo
& 0xC) == 0x4;
127 /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */
128 return !instr_lo
|| (instr_lo
>>1) == 1;
130 /* Prefetch instruction is 0x0F0D or 0x0F18 */
131 if (probe_kernel_address(instr
, opcode
))
134 *prefetch
= (instr_lo
== 0xF) &&
135 (opcode
== 0x0D || opcode
== 0x18);
143 is_prefetch(struct pt_regs
*regs
, unsigned long error_code
, unsigned long addr
)
145 unsigned char *max_instr
;
146 unsigned char *instr
;
150 * If it was a exec (instruction fetch) fault on NX page, then
151 * do not ignore the fault:
153 if (error_code
& PF_INSTR
)
156 instr
= (void *)convert_ip_to_linear(current
, regs
);
157 max_instr
= instr
+ 15;
159 if (user_mode(regs
) && instr
>= (unsigned char *)TASK_SIZE_MAX
)
162 while (instr
< max_instr
) {
163 unsigned char opcode
;
165 if (probe_kernel_address(instr
, opcode
))
170 if (!check_prefetch_opcode(regs
, instr
, opcode
, &prefetch
))
177 * A protection key fault means that the PKRU value did not allow
178 * access to some PTE. Userspace can figure out what PKRU was
179 * from the XSAVE state, and this function fills out a field in
180 * siginfo so userspace can discover which protection key was set
183 * If we get here, we know that the hardware signaled a PF_PK
184 * fault and that there was a VMA once we got in the fault
185 * handler. It does *not* guarantee that the VMA we find here
186 * was the one that we faulted on.
188 * 1. T1 : mprotect_key(foo, PAGE_SIZE, pkey=4);
189 * 2. T1 : set PKRU to deny access to pkey=4, touches page
191 * 4. T2: mprotect_key(foo, PAGE_SIZE, pkey=5);
192 * 5. T1 : enters fault handler, takes mmap_sem, etc...
193 * 6. T1 : reaches here, sees vma_pkey(vma)=5, when we really
194 * faulted on a pte with its pkey=4.
196 static void fill_sig_info_pkey(int si_code
, siginfo_t
*info
, u32
*pkey
)
198 /* This is effectively an #ifdef */
199 if (!boot_cpu_has(X86_FEATURE_OSPKE
))
202 /* Fault not from Protection Keys: nothing to do */
203 if (si_code
!= SEGV_PKUERR
)
206 * force_sig_info_fault() is called from a number of
207 * contexts, some of which have a VMA and some of which
208 * do not. The PF_PK handing happens after we have a
209 * valid VMA, so we should never reach this without a
213 WARN_ONCE(1, "PKU fault with no VMA passed in");
218 * si_pkey should be thought of as a strong hint, but not
219 * absolutely guranteed to be 100% accurate because of
220 * the race explained above.
222 info
->si_pkey
= *pkey
;
226 force_sig_info_fault(int si_signo
, int si_code
, unsigned long address
,
227 struct task_struct
*tsk
, u32
*pkey
, int fault
)
232 info
.si_signo
= si_signo
;
234 info
.si_code
= si_code
;
235 info
.si_addr
= (void __user
*)address
;
236 if (fault
& VM_FAULT_HWPOISON_LARGE
)
237 lsb
= hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault
));
238 if (fault
& VM_FAULT_HWPOISON
)
240 info
.si_addr_lsb
= lsb
;
242 fill_sig_info_pkey(si_code
, &info
, pkey
);
244 force_sig_info(si_signo
, &info
, tsk
);
247 DEFINE_SPINLOCK(pgd_lock
);
251 static inline pmd_t
*vmalloc_sync_one(pgd_t
*pgd
, unsigned long address
)
253 unsigned index
= pgd_index(address
);
260 pgd_k
= init_mm
.pgd
+ index
;
262 if (!pgd_present(*pgd_k
))
266 * set_pgd(pgd, *pgd_k); here would be useless on PAE
267 * and redundant with the set_pmd() on non-PAE. As would
270 p4d
= p4d_offset(pgd
, address
);
271 p4d_k
= p4d_offset(pgd_k
, address
);
272 if (!p4d_present(*p4d_k
))
275 pud
= pud_offset(p4d
, address
);
276 pud_k
= pud_offset(p4d_k
, address
);
277 if (!pud_present(*pud_k
))
280 pmd
= pmd_offset(pud
, address
);
281 pmd_k
= pmd_offset(pud_k
, address
);
282 if (!pmd_present(*pmd_k
))
285 if (!pmd_present(*pmd
))
286 set_pmd(pmd
, *pmd_k
);
288 BUG_ON(pmd_page(*pmd
) != pmd_page(*pmd_k
));
293 void vmalloc_sync_all(void)
295 unsigned long address
;
297 if (SHARED_KERNEL_PMD
)
300 for (address
= VMALLOC_START
& PMD_MASK
;
301 address
>= TASK_SIZE_MAX
&& address
< FIXADDR_TOP
;
302 address
+= PMD_SIZE
) {
305 spin_lock(&pgd_lock
);
306 list_for_each_entry(page
, &pgd_list
, lru
) {
307 spinlock_t
*pgt_lock
;
310 /* the pgt_lock only for Xen */
311 pgt_lock
= &pgd_page_get_mm(page
)->page_table_lock
;
314 ret
= vmalloc_sync_one(page_address(page
), address
);
315 spin_unlock(pgt_lock
);
320 spin_unlock(&pgd_lock
);
327 * Handle a fault on the vmalloc or module mapping area
329 static noinline
int vmalloc_fault(unsigned long address
)
331 unsigned long pgd_paddr
;
335 /* Make sure we are in vmalloc area: */
336 if (!(address
>= VMALLOC_START
&& address
< VMALLOC_END
))
339 WARN_ON_ONCE(in_nmi());
342 * Synchronize this task's top level page-table
343 * with the 'reference' page table.
345 * Do _not_ use "current" here. We might be inside
346 * an interrupt in the middle of a task switch..
348 pgd_paddr
= read_cr3_pa();
349 pmd_k
= vmalloc_sync_one(__va(pgd_paddr
), address
);
353 if (pmd_huge(*pmd_k
))
356 pte_k
= pte_offset_kernel(pmd_k
, address
);
357 if (!pte_present(*pte_k
))
362 NOKPROBE_SYMBOL(vmalloc_fault
);
365 * Did it hit the DOS screen memory VA from vm86 mode?
368 check_v8086_mode(struct pt_regs
*regs
, unsigned long address
,
369 struct task_struct
*tsk
)
374 if (!v8086_mode(regs
) || !tsk
->thread
.vm86
)
377 bit
= (address
- 0xA0000) >> PAGE_SHIFT
;
379 tsk
->thread
.vm86
->screen_bitmap
|= 1 << bit
;
383 static bool low_pfn(unsigned long pfn
)
385 return pfn
< max_low_pfn
;
388 static void dump_pagetable(unsigned long address
)
390 pgd_t
*base
= __va(read_cr3_pa());
391 pgd_t
*pgd
= &base
[pgd_index(address
)];
397 #ifdef CONFIG_X86_PAE
398 pr_info("*pdpt = %016Lx ", pgd_val(*pgd
));
399 if (!low_pfn(pgd_val(*pgd
) >> PAGE_SHIFT
) || !pgd_present(*pgd
))
401 #define pr_pde pr_cont
403 #define pr_pde pr_info
405 p4d
= p4d_offset(pgd
, address
);
406 pud
= pud_offset(p4d
, address
);
407 pmd
= pmd_offset(pud
, address
);
408 pr_pde("*pde = %0*Lx ", sizeof(*pmd
) * 2, (u64
)pmd_val(*pmd
));
412 * We must not directly access the pte in the highpte
413 * case if the page table is located in highmem.
414 * And let's rather not kmap-atomic the pte, just in case
415 * it's allocated already:
417 if (!low_pfn(pmd_pfn(*pmd
)) || !pmd_present(*pmd
) || pmd_large(*pmd
))
420 pte
= pte_offset_kernel(pmd
, address
);
421 pr_cont("*pte = %0*Lx ", sizeof(*pte
) * 2, (u64
)pte_val(*pte
));
426 #else /* CONFIG_X86_64: */
428 void vmalloc_sync_all(void)
430 sync_global_pgds(VMALLOC_START
& PGDIR_MASK
, VMALLOC_END
);
436 * Handle a fault on the vmalloc area
438 static noinline
int vmalloc_fault(unsigned long address
)
440 pgd_t
*pgd
, *pgd_ref
;
441 p4d_t
*p4d
, *p4d_ref
;
442 pud_t
*pud
, *pud_ref
;
443 pmd_t
*pmd
, *pmd_ref
;
444 pte_t
*pte
, *pte_ref
;
446 /* Make sure we are in vmalloc area: */
447 if (!(address
>= VMALLOC_START
&& address
< VMALLOC_END
))
450 WARN_ON_ONCE(in_nmi());
453 * Copy kernel mappings over when needed. This can also
454 * happen within a race in page table update. In the later
457 pgd
= (pgd_t
*)__va(read_cr3_pa()) + pgd_index(address
);
458 pgd_ref
= pgd_offset_k(address
);
459 if (pgd_none(*pgd_ref
))
462 if (pgd_none(*pgd
)) {
463 set_pgd(pgd
, *pgd_ref
);
464 arch_flush_lazy_mmu_mode();
465 } else if (CONFIG_PGTABLE_LEVELS
> 4) {
467 * With folded p4d, pgd_none() is always false, so the pgd may
468 * point to an empty page table entry and pgd_page_vaddr()
469 * will return garbage.
471 * We will do the correct sanity check on the p4d level.
473 BUG_ON(pgd_page_vaddr(*pgd
) != pgd_page_vaddr(*pgd_ref
));
476 /* With 4-level paging, copying happens on the p4d level. */
477 p4d
= p4d_offset(pgd
, address
);
478 p4d_ref
= p4d_offset(pgd_ref
, address
);
479 if (p4d_none(*p4d_ref
))
482 if (p4d_none(*p4d
)) {
483 set_p4d(p4d
, *p4d_ref
);
484 arch_flush_lazy_mmu_mode();
486 BUG_ON(p4d_pfn(*p4d
) != p4d_pfn(*p4d_ref
));
490 * Below here mismatches are bugs because these lower tables
494 pud
= pud_offset(p4d
, address
);
495 pud_ref
= pud_offset(p4d_ref
, address
);
496 if (pud_none(*pud_ref
))
499 if (pud_none(*pud
) || pud_pfn(*pud
) != pud_pfn(*pud_ref
))
505 pmd
= pmd_offset(pud
, address
);
506 pmd_ref
= pmd_offset(pud_ref
, address
);
507 if (pmd_none(*pmd_ref
))
510 if (pmd_none(*pmd
) || pmd_pfn(*pmd
) != pmd_pfn(*pmd_ref
))
516 pte_ref
= pte_offset_kernel(pmd_ref
, address
);
517 if (!pte_present(*pte_ref
))
520 pte
= pte_offset_kernel(pmd
, address
);
523 * Don't use pte_page here, because the mappings can point
524 * outside mem_map, and the NUMA hash lookup cannot handle
527 if (!pte_present(*pte
) || pte_pfn(*pte
) != pte_pfn(*pte_ref
))
532 NOKPROBE_SYMBOL(vmalloc_fault
);
534 #ifdef CONFIG_CPU_SUP_AMD
535 static const char errata93_warning
[] =
537 "******* Your BIOS seems to not contain a fix for K8 errata #93\n"
538 "******* Working around it, but it may cause SEGVs or burn power.\n"
539 "******* Please consider a BIOS update.\n"
540 "******* Disabling USB legacy in the BIOS may also help.\n";
544 * No vm86 mode in 64-bit mode:
547 check_v8086_mode(struct pt_regs
*regs
, unsigned long address
,
548 struct task_struct
*tsk
)
552 static int bad_address(void *p
)
556 return probe_kernel_address((unsigned long *)p
, dummy
);
559 static void dump_pagetable(unsigned long address
)
561 pgd_t
*base
= __va(read_cr3_pa());
562 pgd_t
*pgd
= base
+ pgd_index(address
);
568 if (bad_address(pgd
))
571 pr_info("PGD %lx ", pgd_val(*pgd
));
573 if (!pgd_present(*pgd
))
576 p4d
= p4d_offset(pgd
, address
);
577 if (bad_address(p4d
))
580 pr_cont("P4D %lx ", p4d_val(*p4d
));
581 if (!p4d_present(*p4d
) || p4d_large(*p4d
))
584 pud
= pud_offset(p4d
, address
);
585 if (bad_address(pud
))
588 pr_cont("PUD %lx ", pud_val(*pud
));
589 if (!pud_present(*pud
) || pud_large(*pud
))
592 pmd
= pmd_offset(pud
, address
);
593 if (bad_address(pmd
))
596 pr_cont("PMD %lx ", pmd_val(*pmd
));
597 if (!pmd_present(*pmd
) || pmd_large(*pmd
))
600 pte
= pte_offset_kernel(pmd
, address
);
601 if (bad_address(pte
))
604 pr_cont("PTE %lx", pte_val(*pte
));
612 #endif /* CONFIG_X86_64 */
615 * Workaround for K8 erratum #93 & buggy BIOS.
617 * BIOS SMM functions are required to use a specific workaround
618 * to avoid corruption of the 64bit RIP register on C stepping K8.
620 * A lot of BIOS that didn't get tested properly miss this.
622 * The OS sees this as a page fault with the upper 32bits of RIP cleared.
623 * Try to work around it here.
625 * Note we only handle faults in kernel here.
626 * Does nothing on 32-bit.
628 static int is_errata93(struct pt_regs
*regs
, unsigned long address
)
630 #if defined(CONFIG_X86_64) && defined(CONFIG_CPU_SUP_AMD)
631 if (boot_cpu_data
.x86_vendor
!= X86_VENDOR_AMD
632 || boot_cpu_data
.x86
!= 0xf)
635 if (address
!= regs
->ip
)
638 if ((address
>> 32) != 0)
641 address
|= 0xffffffffUL
<< 32;
642 if ((address
>= (u64
)_stext
&& address
<= (u64
)_etext
) ||
643 (address
>= MODULES_VADDR
&& address
<= MODULES_END
)) {
644 printk_once(errata93_warning
);
653 * Work around K8 erratum #100 K8 in compat mode occasionally jumps
654 * to illegal addresses >4GB.
656 * We catch this in the page fault handler because these addresses
657 * are not reachable. Just detect this case and return. Any code
658 * segment in LDT is compatibility mode.
660 static int is_errata100(struct pt_regs
*regs
, unsigned long address
)
663 if ((regs
->cs
== __USER32_CS
|| (regs
->cs
& (1<<2))) && (address
>> 32))
669 static int is_f00f_bug(struct pt_regs
*regs
, unsigned long address
)
671 #ifdef CONFIG_X86_F00F_BUG
675 * Pentium F0 0F C7 C8 bug workaround:
677 if (boot_cpu_has_bug(X86_BUG_F00F
)) {
678 nr
= (address
- idt_descr
.address
) >> 3;
681 do_invalid_op(regs
, 0);
689 static const char nx_warning
[] = KERN_CRIT
690 "kernel tried to execute NX-protected page - exploit attempt? (uid: %d)\n";
691 static const char smep_warning
[] = KERN_CRIT
692 "unable to execute userspace code (SMEP?) (uid: %d)\n";
695 show_fault_oops(struct pt_regs
*regs
, unsigned long error_code
,
696 unsigned long address
)
698 if (!oops_may_print())
701 if (error_code
& PF_INSTR
) {
706 pgd
= __va(read_cr3_pa());
707 pgd
+= pgd_index(address
);
709 pte
= lookup_address_in_pgd(pgd
, address
, &level
);
711 if (pte
&& pte_present(*pte
) && !pte_exec(*pte
))
712 printk(nx_warning
, from_kuid(&init_user_ns
, current_uid()));
713 if (pte
&& pte_present(*pte
) && pte_exec(*pte
) &&
714 (pgd_flags(*pgd
) & _PAGE_USER
) &&
715 (__read_cr4() & X86_CR4_SMEP
))
716 printk(smep_warning
, from_kuid(&init_user_ns
, current_uid()));
719 printk(KERN_ALERT
"BUG: unable to handle kernel ");
720 if (address
< PAGE_SIZE
)
721 printk(KERN_CONT
"NULL pointer dereference");
723 printk(KERN_CONT
"paging request");
725 printk(KERN_CONT
" at %p\n", (void *) address
);
726 printk(KERN_ALERT
"IP: %pS\n", (void *)regs
->ip
);
728 dump_pagetable(address
);
732 pgtable_bad(struct pt_regs
*regs
, unsigned long error_code
,
733 unsigned long address
)
735 struct task_struct
*tsk
;
739 flags
= oops_begin();
743 printk(KERN_ALERT
"%s: Corrupted page table at address %lx\n",
745 dump_pagetable(address
);
747 tsk
->thread
.cr2
= address
;
748 tsk
->thread
.trap_nr
= X86_TRAP_PF
;
749 tsk
->thread
.error_code
= error_code
;
751 if (__die("Bad pagetable", regs
, error_code
))
754 oops_end(flags
, regs
, sig
);
758 no_context(struct pt_regs
*regs
, unsigned long error_code
,
759 unsigned long address
, int signal
, int si_code
)
761 struct task_struct
*tsk
= current
;
765 /* Are we prepared to handle this kernel fault? */
766 if (fixup_exception(regs
, X86_TRAP_PF
)) {
768 * Any interrupt that takes a fault gets the fixup. This makes
769 * the below recursive fault logic only apply to a faults from
776 * Per the above we're !in_interrupt(), aka. task context.
778 * In this case we need to make sure we're not recursively
779 * faulting through the emulate_vsyscall() logic.
781 if (current
->thread
.sig_on_uaccess_err
&& signal
) {
782 tsk
->thread
.trap_nr
= X86_TRAP_PF
;
783 tsk
->thread
.error_code
= error_code
| PF_USER
;
784 tsk
->thread
.cr2
= address
;
786 /* XXX: hwpoison faults will set the wrong code. */
787 force_sig_info_fault(signal
, si_code
, address
,
792 * Barring that, we can do the fixup and be happy.
797 #ifdef CONFIG_VMAP_STACK
799 * Stack overflow? During boot, we can fault near the initial
800 * stack in the direct map, but that's not an overflow -- check
801 * that we're in vmalloc space to avoid this.
803 if (is_vmalloc_addr((void *)address
) &&
804 (((unsigned long)tsk
->stack
- 1 - address
< PAGE_SIZE
) ||
805 address
- ((unsigned long)tsk
->stack
+ THREAD_SIZE
) < PAGE_SIZE
)) {
806 unsigned long stack
= this_cpu_read(orig_ist
.ist
[DOUBLEFAULT_STACK
]) - sizeof(void *);
808 * We're likely to be running with very little stack space
809 * left. It's plausible that we'd hit this condition but
810 * double-fault even before we get this far, in which case
811 * we're fine: the double-fault handler will deal with it.
813 * We don't want to make it all the way into the oops code
814 * and then double-fault, though, because we're likely to
815 * break the console driver and lose most of the stack dump.
817 asm volatile ("movq %[stack], %%rsp\n\t"
818 "call handle_stack_overflow\n\t"
820 : ASM_CALL_CONSTRAINT
821 : "D" ("kernel stack overflow (page fault)"),
822 "S" (regs
), "d" (address
),
823 [stack
] "rm" (stack
));
831 * Valid to do another page fault here, because if this fault
832 * had been triggered by is_prefetch fixup_exception would have
837 * Hall of shame of CPU/BIOS bugs.
839 if (is_prefetch(regs
, error_code
, address
))
842 if (is_errata93(regs
, address
))
846 * Oops. The kernel tried to access some bad page. We'll have to
847 * terminate things with extreme prejudice:
849 flags
= oops_begin();
851 show_fault_oops(regs
, error_code
, address
);
853 if (task_stack_end_corrupted(tsk
))
854 printk(KERN_EMERG
"Thread overran stack, or stack corrupted\n");
856 tsk
->thread
.cr2
= address
;
857 tsk
->thread
.trap_nr
= X86_TRAP_PF
;
858 tsk
->thread
.error_code
= error_code
;
861 if (__die("Oops", regs
, error_code
))
864 /* Executive summary in case the body of the oops scrolled away */
865 printk(KERN_DEFAULT
"CR2: %016lx\n", address
);
867 oops_end(flags
, regs
, sig
);
871 * Print out info about fatal segfaults, if the show_unhandled_signals
875 show_signal_msg(struct pt_regs
*regs
, unsigned long error_code
,
876 unsigned long address
, struct task_struct
*tsk
)
878 if (!unhandled_signal(tsk
, SIGSEGV
))
881 if (!printk_ratelimit())
884 printk("%s%s[%d]: segfault at %lx ip %p sp %p error %lx",
885 task_pid_nr(tsk
) > 1 ? KERN_INFO
: KERN_EMERG
,
886 tsk
->comm
, task_pid_nr(tsk
), address
,
887 (void *)regs
->ip
, (void *)regs
->sp
, error_code
);
889 print_vma_addr(KERN_CONT
" in ", regs
->ip
);
891 printk(KERN_CONT
"\n");
895 __bad_area_nosemaphore(struct pt_regs
*regs
, unsigned long error_code
,
896 unsigned long address
, u32
*pkey
, int si_code
)
898 struct task_struct
*tsk
= current
;
900 /* User mode accesses just cause a SIGSEGV */
901 if (error_code
& PF_USER
) {
903 * It's possible to have interrupts off here:
908 * Valid to do another page fault here because this one came
911 if (is_prefetch(regs
, error_code
, address
))
914 if (is_errata100(regs
, address
))
919 * Instruction fetch faults in the vsyscall page might need
922 if (unlikely((error_code
& PF_INSTR
) &&
923 ((address
& ~0xfff) == VSYSCALL_ADDR
))) {
924 if (emulate_vsyscall(regs
, address
))
930 * To avoid leaking information about the kernel page table
931 * layout, pretend that user-mode accesses to kernel addresses
932 * are always protection faults.
934 if (address
>= TASK_SIZE_MAX
)
935 error_code
|= PF_PROT
;
937 if (likely(show_unhandled_signals
))
938 show_signal_msg(regs
, error_code
, address
, tsk
);
940 tsk
->thread
.cr2
= address
;
941 tsk
->thread
.error_code
= error_code
;
942 tsk
->thread
.trap_nr
= X86_TRAP_PF
;
944 force_sig_info_fault(SIGSEGV
, si_code
, address
, tsk
, pkey
, 0);
949 if (is_f00f_bug(regs
, address
))
952 no_context(regs
, error_code
, address
, SIGSEGV
, si_code
);
956 bad_area_nosemaphore(struct pt_regs
*regs
, unsigned long error_code
,
957 unsigned long address
, u32
*pkey
)
959 __bad_area_nosemaphore(regs
, error_code
, address
, pkey
, SEGV_MAPERR
);
963 __bad_area(struct pt_regs
*regs
, unsigned long error_code
,
964 unsigned long address
, struct vm_area_struct
*vma
, int si_code
)
966 struct mm_struct
*mm
= current
->mm
;
970 pkey
= vma_pkey(vma
);
973 * Something tried to access memory that isn't in our memory map..
974 * Fix it, but check if it's kernel or user first..
976 up_read(&mm
->mmap_sem
);
978 __bad_area_nosemaphore(regs
, error_code
, address
,
979 (vma
) ? &pkey
: NULL
, si_code
);
983 bad_area(struct pt_regs
*regs
, unsigned long error_code
, unsigned long address
)
985 __bad_area(regs
, error_code
, address
, NULL
, SEGV_MAPERR
);
988 static inline bool bad_area_access_from_pkeys(unsigned long error_code
,
989 struct vm_area_struct
*vma
)
991 /* This code is always called on the current mm */
992 bool foreign
= false;
994 if (!boot_cpu_has(X86_FEATURE_OSPKE
))
996 if (error_code
& PF_PK
)
998 /* this checks permission keys on the VMA: */
999 if (!arch_vma_access_permitted(vma
, (error_code
& PF_WRITE
),
1000 (error_code
& PF_INSTR
), foreign
))
1005 static noinline
void
1006 bad_area_access_error(struct pt_regs
*regs
, unsigned long error_code
,
1007 unsigned long address
, struct vm_area_struct
*vma
)
1010 * This OSPKE check is not strictly necessary at runtime.
1011 * But, doing it this way allows compiler optimizations
1012 * if pkeys are compiled out.
1014 if (bad_area_access_from_pkeys(error_code
, vma
))
1015 __bad_area(regs
, error_code
, address
, vma
, SEGV_PKUERR
);
1017 __bad_area(regs
, error_code
, address
, vma
, SEGV_ACCERR
);
1021 do_sigbus(struct pt_regs
*regs
, unsigned long error_code
, unsigned long address
,
1022 u32
*pkey
, unsigned int fault
)
1024 struct task_struct
*tsk
= current
;
1025 int code
= BUS_ADRERR
;
1027 /* Kernel mode? Handle exceptions or die: */
1028 if (!(error_code
& PF_USER
)) {
1029 no_context(regs
, error_code
, address
, SIGBUS
, BUS_ADRERR
);
1033 /* User-space => ok to do another page fault: */
1034 if (is_prefetch(regs
, error_code
, address
))
1037 tsk
->thread
.cr2
= address
;
1038 tsk
->thread
.error_code
= error_code
;
1039 tsk
->thread
.trap_nr
= X86_TRAP_PF
;
1041 #ifdef CONFIG_MEMORY_FAILURE
1042 if (fault
& (VM_FAULT_HWPOISON
|VM_FAULT_HWPOISON_LARGE
)) {
1044 "MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n",
1045 tsk
->comm
, tsk
->pid
, address
);
1046 code
= BUS_MCEERR_AR
;
1049 force_sig_info_fault(SIGBUS
, code
, address
, tsk
, pkey
, fault
);
1052 static noinline
void
1053 mm_fault_error(struct pt_regs
*regs
, unsigned long error_code
,
1054 unsigned long address
, u32
*pkey
, unsigned int fault
)
1056 if (fatal_signal_pending(current
) && !(error_code
& PF_USER
)) {
1057 no_context(regs
, error_code
, address
, 0, 0);
1061 if (fault
& VM_FAULT_OOM
) {
1062 /* Kernel mode? Handle exceptions or die: */
1063 if (!(error_code
& PF_USER
)) {
1064 no_context(regs
, error_code
, address
,
1065 SIGSEGV
, SEGV_MAPERR
);
1070 * We ran out of memory, call the OOM killer, and return the
1071 * userspace (which will retry the fault, or kill us if we got
1074 pagefault_out_of_memory();
1076 if (fault
& (VM_FAULT_SIGBUS
|VM_FAULT_HWPOISON
|
1077 VM_FAULT_HWPOISON_LARGE
))
1078 do_sigbus(regs
, error_code
, address
, pkey
, fault
);
1079 else if (fault
& VM_FAULT_SIGSEGV
)
1080 bad_area_nosemaphore(regs
, error_code
, address
, pkey
);
1086 static int spurious_fault_check(unsigned long error_code
, pte_t
*pte
)
1088 if ((error_code
& PF_WRITE
) && !pte_write(*pte
))
1091 if ((error_code
& PF_INSTR
) && !pte_exec(*pte
))
1094 * Note: We do not do lazy flushing on protection key
1095 * changes, so no spurious fault will ever set PF_PK.
1097 if ((error_code
& PF_PK
))
1104 * Handle a spurious fault caused by a stale TLB entry.
1106 * This allows us to lazily refresh the TLB when increasing the
1107 * permissions of a kernel page (RO -> RW or NX -> X). Doing it
1108 * eagerly is very expensive since that implies doing a full
1109 * cross-processor TLB flush, even if no stale TLB entries exist
1110 * on other processors.
1112 * Spurious faults may only occur if the TLB contains an entry with
1113 * fewer permission than the page table entry. Non-present (P = 0)
1114 * and reserved bit (R = 1) faults are never spurious.
1116 * There are no security implications to leaving a stale TLB when
1117 * increasing the permissions on a page.
1119 * Returns non-zero if a spurious fault was handled, zero otherwise.
1121 * See Intel Developer's Manual Vol 3 Section 4.10.4.3, bullet 3
1122 * (Optional Invalidation).
1125 spurious_fault(unsigned long error_code
, unsigned long address
)
1135 * Only writes to RO or instruction fetches from NX may cause
1138 * These could be from user or supervisor accesses but the TLB
1139 * is only lazily flushed after a kernel mapping protection
1140 * change, so user accesses are not expected to cause spurious
1143 if (error_code
!= (PF_WRITE
| PF_PROT
)
1144 && error_code
!= (PF_INSTR
| PF_PROT
))
1147 pgd
= init_mm
.pgd
+ pgd_index(address
);
1148 if (!pgd_present(*pgd
))
1151 p4d
= p4d_offset(pgd
, address
);
1152 if (!p4d_present(*p4d
))
1155 if (p4d_large(*p4d
))
1156 return spurious_fault_check(error_code
, (pte_t
*) p4d
);
1158 pud
= pud_offset(p4d
, address
);
1159 if (!pud_present(*pud
))
1162 if (pud_large(*pud
))
1163 return spurious_fault_check(error_code
, (pte_t
*) pud
);
1165 pmd
= pmd_offset(pud
, address
);
1166 if (!pmd_present(*pmd
))
1169 if (pmd_large(*pmd
))
1170 return spurious_fault_check(error_code
, (pte_t
*) pmd
);
1172 pte
= pte_offset_kernel(pmd
, address
);
1173 if (!pte_present(*pte
))
1176 ret
= spurious_fault_check(error_code
, pte
);
1181 * Make sure we have permissions in PMD.
1182 * If not, then there's a bug in the page tables:
1184 ret
= spurious_fault_check(error_code
, (pte_t
*) pmd
);
1185 WARN_ONCE(!ret
, "PMD has incorrect permission bits\n");
1189 NOKPROBE_SYMBOL(spurious_fault
);
1191 int show_unhandled_signals
= 1;
1194 access_error(unsigned long error_code
, struct vm_area_struct
*vma
)
1196 /* This is only called for the current mm, so: */
1197 bool foreign
= false;
1200 * Read or write was blocked by protection keys. This is
1201 * always an unconditional error and can never result in
1202 * a follow-up action to resolve the fault, like a COW.
1204 if (error_code
& PF_PK
)
1208 * Make sure to check the VMA so that we do not perform
1209 * faults just to hit a PF_PK as soon as we fill in a
1212 if (!arch_vma_access_permitted(vma
, (error_code
& PF_WRITE
),
1213 (error_code
& PF_INSTR
), foreign
))
1216 if (error_code
& PF_WRITE
) {
1217 /* write, present and write, not present: */
1218 if (unlikely(!(vma
->vm_flags
& VM_WRITE
)))
1223 /* read, present: */
1224 if (unlikely(error_code
& PF_PROT
))
1227 /* read, not present: */
1228 if (unlikely(!(vma
->vm_flags
& (VM_READ
| VM_EXEC
| VM_WRITE
))))
1234 static int fault_in_kernel_space(unsigned long address
)
1236 return address
>= TASK_SIZE_MAX
;
1239 static inline bool smap_violation(int error_code
, struct pt_regs
*regs
)
1241 if (!IS_ENABLED(CONFIG_X86_SMAP
))
1244 if (!static_cpu_has(X86_FEATURE_SMAP
))
1247 if (error_code
& PF_USER
)
1250 if (!user_mode(regs
) && (regs
->flags
& X86_EFLAGS_AC
))
1257 * This routine handles page faults. It determines the address,
1258 * and the problem, and then passes it off to one of the appropriate
1261 static noinline
void
1262 __do_page_fault(struct pt_regs
*regs
, unsigned long error_code
,
1263 unsigned long address
)
1265 struct vm_area_struct
*vma
;
1266 struct task_struct
*tsk
;
1267 struct mm_struct
*mm
;
1268 int fault
, major
= 0;
1269 unsigned int flags
= FAULT_FLAG_ALLOW_RETRY
| FAULT_FLAG_KILLABLE
;
1276 * Detect and handle instructions that would cause a page fault for
1277 * both a tracked kernel page and a userspace page.
1279 if (kmemcheck_active(regs
))
1280 kmemcheck_hide(regs
);
1281 prefetchw(&mm
->mmap_sem
);
1283 if (unlikely(kmmio_fault(regs
, address
)))
1287 * We fault-in kernel-space virtual memory on-demand. The
1288 * 'reference' page table is init_mm.pgd.
1290 * NOTE! We MUST NOT take any locks for this case. We may
1291 * be in an interrupt or a critical region, and should
1292 * only copy the information from the master page table,
1295 * This verifies that the fault happens in kernel space
1296 * (error_code & 4) == 0, and that the fault was not a
1297 * protection error (error_code & 9) == 0.
1299 if (unlikely(fault_in_kernel_space(address
))) {
1300 if (!(error_code
& (PF_RSVD
| PF_USER
| PF_PROT
))) {
1301 if (vmalloc_fault(address
) >= 0)
1304 if (kmemcheck_fault(regs
, address
, error_code
))
1308 /* Can handle a stale RO->RW TLB: */
1309 if (spurious_fault(error_code
, address
))
1312 /* kprobes don't want to hook the spurious faults: */
1313 if (kprobes_fault(regs
))
1316 * Don't take the mm semaphore here. If we fixup a prefetch
1317 * fault we could otherwise deadlock:
1319 bad_area_nosemaphore(regs
, error_code
, address
, NULL
);
1324 /* kprobes don't want to hook the spurious faults: */
1325 if (unlikely(kprobes_fault(regs
)))
1328 if (unlikely(error_code
& PF_RSVD
))
1329 pgtable_bad(regs
, error_code
, address
);
1331 if (unlikely(smap_violation(error_code
, regs
))) {
1332 bad_area_nosemaphore(regs
, error_code
, address
, NULL
);
1337 * If we're in an interrupt, have no user context or are running
1338 * in a region with pagefaults disabled then we must not take the fault
1340 if (unlikely(faulthandler_disabled() || !mm
)) {
1341 bad_area_nosemaphore(regs
, error_code
, address
, NULL
);
1346 * It's safe to allow irq's after cr2 has been saved and the
1347 * vmalloc fault has been handled.
1349 * User-mode registers count as a user access even for any
1350 * potential system fault or CPU buglet:
1352 if (user_mode(regs
)) {
1354 error_code
|= PF_USER
;
1355 flags
|= FAULT_FLAG_USER
;
1357 if (regs
->flags
& X86_EFLAGS_IF
)
1361 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS
, 1, regs
, address
);
1363 if (error_code
& PF_WRITE
)
1364 flags
|= FAULT_FLAG_WRITE
;
1365 if (error_code
& PF_INSTR
)
1366 flags
|= FAULT_FLAG_INSTRUCTION
;
1369 * When running in the kernel we expect faults to occur only to
1370 * addresses in user space. All other faults represent errors in
1371 * the kernel and should generate an OOPS. Unfortunately, in the
1372 * case of an erroneous fault occurring in a code path which already
1373 * holds mmap_sem we will deadlock attempting to validate the fault
1374 * against the address space. Luckily the kernel only validly
1375 * references user space from well defined areas of code, which are
1376 * listed in the exceptions table.
1378 * As the vast majority of faults will be valid we will only perform
1379 * the source reference check when there is a possibility of a
1380 * deadlock. Attempt to lock the address space, if we cannot we then
1381 * validate the source. If this is invalid we can skip the address
1382 * space check, thus avoiding the deadlock:
1384 if (unlikely(!down_read_trylock(&mm
->mmap_sem
))) {
1385 if ((error_code
& PF_USER
) == 0 &&
1386 !search_exception_tables(regs
->ip
)) {
1387 bad_area_nosemaphore(regs
, error_code
, address
, NULL
);
1391 down_read(&mm
->mmap_sem
);
1394 * The above down_read_trylock() might have succeeded in
1395 * which case we'll have missed the might_sleep() from
1401 vma
= find_vma(mm
, address
);
1402 if (unlikely(!vma
)) {
1403 bad_area(regs
, error_code
, address
);
1406 if (likely(vma
->vm_start
<= address
))
1408 if (unlikely(!(vma
->vm_flags
& VM_GROWSDOWN
))) {
1409 bad_area(regs
, error_code
, address
);
1412 if (error_code
& PF_USER
) {
1414 * Accessing the stack below %sp is always a bug.
1415 * The large cushion allows instructions like enter
1416 * and pusha to work. ("enter $65535, $31" pushes
1417 * 32 pointers and then decrements %sp by 65535.)
1419 if (unlikely(address
+ 65536 + 32 * sizeof(unsigned long) < regs
->sp
)) {
1420 bad_area(regs
, error_code
, address
);
1424 if (unlikely(expand_stack(vma
, address
))) {
1425 bad_area(regs
, error_code
, address
);
1430 * Ok, we have a good vm_area for this memory access, so
1431 * we can handle it..
1434 if (unlikely(access_error(error_code
, vma
))) {
1435 bad_area_access_error(regs
, error_code
, address
, vma
);
1440 * If for any reason at all we couldn't handle the fault,
1441 * make sure we exit gracefully rather than endlessly redo
1442 * the fault. Since we never set FAULT_FLAG_RETRY_NOWAIT, if
1443 * we get VM_FAULT_RETRY back, the mmap_sem has been unlocked.
1445 fault
= handle_mm_fault(vma
, address
, flags
);
1446 major
|= fault
& VM_FAULT_MAJOR
;
1449 * If we need to retry the mmap_sem has already been released,
1450 * and if there is a fatal signal pending there is no guarantee
1451 * that we made any progress. Handle this case first.
1453 if (unlikely(fault
& VM_FAULT_RETRY
)) {
1454 /* Retry at most once */
1455 if (flags
& FAULT_FLAG_ALLOW_RETRY
) {
1456 flags
&= ~FAULT_FLAG_ALLOW_RETRY
;
1457 flags
|= FAULT_FLAG_TRIED
;
1458 if (!fatal_signal_pending(tsk
))
1462 /* User mode? Just return to handle the fatal exception */
1463 if (flags
& FAULT_FLAG_USER
)
1466 /* Not returning to user mode? Handle exceptions or die: */
1467 no_context(regs
, error_code
, address
, SIGBUS
, BUS_ADRERR
);
1471 pkey
= vma_pkey(vma
);
1472 up_read(&mm
->mmap_sem
);
1473 if (unlikely(fault
& VM_FAULT_ERROR
)) {
1474 mm_fault_error(regs
, error_code
, address
, &pkey
, fault
);
1479 * Major/minor page fault accounting. If any of the events
1480 * returned VM_FAULT_MAJOR, we account it as a major fault.
1484 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ
, 1, regs
, address
);
1487 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN
, 1, regs
, address
);
1490 check_v8086_mode(regs
, address
, tsk
);
1492 NOKPROBE_SYMBOL(__do_page_fault
);
1494 static nokprobe_inline
void
1495 trace_page_fault_entries(unsigned long address
, struct pt_regs
*regs
,
1496 unsigned long error_code
)
1498 if (user_mode(regs
))
1499 trace_page_fault_user(address
, regs
, error_code
);
1501 trace_page_fault_kernel(address
, regs
, error_code
);
1505 * We must have this function blacklisted from kprobes, tagged with notrace
1506 * and call read_cr2() before calling anything else. To avoid calling any
1507 * kind of tracing machinery before we've observed the CR2 value.
1509 * exception_{enter,exit}() contains all sorts of tracepoints.
1511 dotraplinkage
void notrace
1512 do_page_fault(struct pt_regs
*regs
, unsigned long error_code
)
1514 unsigned long address
= read_cr2(); /* Get the faulting address */
1515 enum ctx_state prev_state
;
1517 prev_state
= exception_enter();
1518 if (trace_pagefault_enabled())
1519 trace_page_fault_entries(address
, regs
, error_code
);
1521 __do_page_fault(regs
, error_code
, address
);
1522 exception_exit(prev_state
);
1524 NOKPROBE_SYMBOL(do_page_fault
);