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/memblock.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() */
19 #include <linux/efi.h> /* efi_recover_from_page_fault()*/
20 #include <linux/mm_types.h>
22 #include <asm/cpufeature.h> /* boot_cpu_has, ... */
23 #include <asm/traps.h> /* dotraplinkage, ... */
24 #include <asm/pgalloc.h> /* pgd_*(), ... */
25 #include <asm/fixmap.h> /* VSYSCALL_ADDR */
26 #include <asm/vsyscall.h> /* emulate_vsyscall */
27 #include <asm/vm86.h> /* struct vm86 */
28 #include <asm/mmu_context.h> /* vma_pkey() */
29 #include <asm/efi.h> /* efi_recover_from_page_fault()*/
30 #include <asm/desc.h> /* store_idt(), ... */
31 #include <asm/cpu_entry_area.h> /* exception stack */
32 #include <asm/pgtable_areas.h> /* VMALLOC_START, ... */
34 #define CREATE_TRACE_POINTS
35 #include <asm/trace/exceptions.h>
38 * Returns 0 if mmiotrace is disabled, or if the fault is not
39 * handled by mmiotrace:
41 static nokprobe_inline
int
42 kmmio_fault(struct pt_regs
*regs
, unsigned long addr
)
44 if (unlikely(is_kmmio_active()))
45 if (kmmio_handler(regs
, addr
) == 1)
55 * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch.
56 * Check that here and ignore it.
60 * Sometimes the CPU reports invalid exceptions on prefetch.
61 * Check that here and ignore it.
63 * Opcode checker based on code by Richard Brunner.
66 check_prefetch_opcode(struct pt_regs
*regs
, unsigned char *instr
,
67 unsigned char opcode
, int *prefetch
)
69 unsigned char instr_hi
= opcode
& 0xf0;
70 unsigned char instr_lo
= opcode
& 0x0f;
76 * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes.
77 * In X86_64 long mode, the CPU will signal invalid
78 * opcode if some of these prefixes are present so
79 * X86_64 will never get here anyway
81 return ((instr_lo
& 7) == 0x6);
85 * In AMD64 long mode 0x40..0x4F are valid REX prefixes
86 * Need to figure out under what instruction mode the
87 * instruction was issued. Could check the LDT for lm,
88 * but for now it's good enough to assume that long
89 * mode only uses well known segments or kernel.
91 return (!user_mode(regs
) || user_64bit_mode(regs
));
94 /* 0x64 thru 0x67 are valid prefixes in all modes. */
95 return (instr_lo
& 0xC) == 0x4;
97 /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */
98 return !instr_lo
|| (instr_lo
>>1) == 1;
100 /* Prefetch instruction is 0x0F0D or 0x0F18 */
101 if (probe_kernel_address(instr
, opcode
))
104 *prefetch
= (instr_lo
== 0xF) &&
105 (opcode
== 0x0D || opcode
== 0x18);
113 is_prefetch(struct pt_regs
*regs
, unsigned long error_code
, unsigned long addr
)
115 unsigned char *max_instr
;
116 unsigned char *instr
;
120 * If it was a exec (instruction fetch) fault on NX page, then
121 * do not ignore the fault:
123 if (error_code
& X86_PF_INSTR
)
126 instr
= (void *)convert_ip_to_linear(current
, regs
);
127 max_instr
= instr
+ 15;
129 if (user_mode(regs
) && instr
>= (unsigned char *)TASK_SIZE_MAX
)
132 while (instr
< max_instr
) {
133 unsigned char opcode
;
135 if (probe_kernel_address(instr
, opcode
))
140 if (!check_prefetch_opcode(regs
, instr
, opcode
, &prefetch
))
146 DEFINE_SPINLOCK(pgd_lock
);
150 static inline pmd_t
*vmalloc_sync_one(pgd_t
*pgd
, unsigned long address
)
152 unsigned index
= pgd_index(address
);
159 pgd_k
= init_mm
.pgd
+ index
;
161 if (!pgd_present(*pgd_k
))
165 * set_pgd(pgd, *pgd_k); here would be useless on PAE
166 * and redundant with the set_pmd() on non-PAE. As would
169 p4d
= p4d_offset(pgd
, address
);
170 p4d_k
= p4d_offset(pgd_k
, address
);
171 if (!p4d_present(*p4d_k
))
174 pud
= pud_offset(p4d
, address
);
175 pud_k
= pud_offset(p4d_k
, address
);
176 if (!pud_present(*pud_k
))
179 pmd
= pmd_offset(pud
, address
);
180 pmd_k
= pmd_offset(pud_k
, address
);
182 if (pmd_present(*pmd
) != pmd_present(*pmd_k
))
183 set_pmd(pmd
, *pmd_k
);
185 if (!pmd_present(*pmd_k
))
188 BUG_ON(pmd_pfn(*pmd
) != pmd_pfn(*pmd_k
));
193 static void vmalloc_sync(void)
195 unsigned long address
;
197 if (SHARED_KERNEL_PMD
)
200 for (address
= VMALLOC_START
& PMD_MASK
;
201 address
>= TASK_SIZE_MAX
&& address
< VMALLOC_END
;
202 address
+= PMD_SIZE
) {
205 spin_lock(&pgd_lock
);
206 list_for_each_entry(page
, &pgd_list
, lru
) {
207 spinlock_t
*pgt_lock
;
209 /* the pgt_lock only for Xen */
210 pgt_lock
= &pgd_page_get_mm(page
)->page_table_lock
;
213 vmalloc_sync_one(page_address(page
), address
);
214 spin_unlock(pgt_lock
);
216 spin_unlock(&pgd_lock
);
220 void vmalloc_sync_mappings(void)
225 void vmalloc_sync_unmappings(void)
233 * Handle a fault on the vmalloc or module mapping area
235 static noinline
int vmalloc_fault(unsigned long address
)
237 unsigned long pgd_paddr
;
241 /* Make sure we are in vmalloc area: */
242 if (!(address
>= VMALLOC_START
&& address
< VMALLOC_END
))
246 * Synchronize this task's top level page-table
247 * with the 'reference' page table.
249 * Do _not_ use "current" here. We might be inside
250 * an interrupt in the middle of a task switch..
252 pgd_paddr
= read_cr3_pa();
253 pmd_k
= vmalloc_sync_one(__va(pgd_paddr
), address
);
257 if (pmd_large(*pmd_k
))
260 pte_k
= pte_offset_kernel(pmd_k
, address
);
261 if (!pte_present(*pte_k
))
266 NOKPROBE_SYMBOL(vmalloc_fault
);
269 * Did it hit the DOS screen memory VA from vm86 mode?
272 check_v8086_mode(struct pt_regs
*regs
, unsigned long address
,
273 struct task_struct
*tsk
)
278 if (!v8086_mode(regs
) || !tsk
->thread
.vm86
)
281 bit
= (address
- 0xA0000) >> PAGE_SHIFT
;
283 tsk
->thread
.vm86
->screen_bitmap
|= 1 << bit
;
287 static bool low_pfn(unsigned long pfn
)
289 return pfn
< max_low_pfn
;
292 static void dump_pagetable(unsigned long address
)
294 pgd_t
*base
= __va(read_cr3_pa());
295 pgd_t
*pgd
= &base
[pgd_index(address
)];
301 #ifdef CONFIG_X86_PAE
302 pr_info("*pdpt = %016Lx ", pgd_val(*pgd
));
303 if (!low_pfn(pgd_val(*pgd
) >> PAGE_SHIFT
) || !pgd_present(*pgd
))
305 #define pr_pde pr_cont
307 #define pr_pde pr_info
309 p4d
= p4d_offset(pgd
, address
);
310 pud
= pud_offset(p4d
, address
);
311 pmd
= pmd_offset(pud
, address
);
312 pr_pde("*pde = %0*Lx ", sizeof(*pmd
) * 2, (u64
)pmd_val(*pmd
));
316 * We must not directly access the pte in the highpte
317 * case if the page table is located in highmem.
318 * And let's rather not kmap-atomic the pte, just in case
319 * it's allocated already:
321 if (!low_pfn(pmd_pfn(*pmd
)) || !pmd_present(*pmd
) || pmd_large(*pmd
))
324 pte
= pte_offset_kernel(pmd
, address
);
325 pr_cont("*pte = %0*Lx ", sizeof(*pte
) * 2, (u64
)pte_val(*pte
));
330 #else /* CONFIG_X86_64: */
332 void vmalloc_sync_mappings(void)
335 * 64-bit mappings might allocate new p4d/pud pages
336 * that need to be propagated to all tasks' PGDs.
338 sync_global_pgds(VMALLOC_START
& PGDIR_MASK
, VMALLOC_END
);
341 void vmalloc_sync_unmappings(void)
344 * Unmappings never allocate or free p4d/pud pages.
345 * No work is required here.
352 * Handle a fault on the vmalloc area
354 static noinline
int vmalloc_fault(unsigned long address
)
362 /* Make sure we are in vmalloc area: */
363 if (!(address
>= VMALLOC_START
&& address
< VMALLOC_END
))
367 * Copy kernel mappings over when needed. This can also
368 * happen within a race in page table update. In the later
371 pgd
= (pgd_t
*)__va(read_cr3_pa()) + pgd_index(address
);
372 pgd_k
= pgd_offset_k(address
);
373 if (pgd_none(*pgd_k
))
376 if (pgtable_l5_enabled()) {
377 if (pgd_none(*pgd
)) {
378 set_pgd(pgd
, *pgd_k
);
379 arch_flush_lazy_mmu_mode();
381 BUG_ON(pgd_page_vaddr(*pgd
) != pgd_page_vaddr(*pgd_k
));
385 /* With 4-level paging, copying happens on the p4d level. */
386 p4d
= p4d_offset(pgd
, address
);
387 p4d_k
= p4d_offset(pgd_k
, address
);
388 if (p4d_none(*p4d_k
))
391 if (p4d_none(*p4d
) && !pgtable_l5_enabled()) {
392 set_p4d(p4d
, *p4d_k
);
393 arch_flush_lazy_mmu_mode();
395 BUG_ON(p4d_pfn(*p4d
) != p4d_pfn(*p4d_k
));
398 BUILD_BUG_ON(CONFIG_PGTABLE_LEVELS
< 4);
400 pud
= pud_offset(p4d
, address
);
407 pmd
= pmd_offset(pud
, address
);
414 pte
= pte_offset_kernel(pmd
, address
);
415 if (!pte_present(*pte
))
420 NOKPROBE_SYMBOL(vmalloc_fault
);
422 #ifdef CONFIG_CPU_SUP_AMD
423 static const char errata93_warning
[] =
425 "******* Your BIOS seems to not contain a fix for K8 errata #93\n"
426 "******* Working around it, but it may cause SEGVs or burn power.\n"
427 "******* Please consider a BIOS update.\n"
428 "******* Disabling USB legacy in the BIOS may also help.\n";
432 * No vm86 mode in 64-bit mode:
435 check_v8086_mode(struct pt_regs
*regs
, unsigned long address
,
436 struct task_struct
*tsk
)
440 static int bad_address(void *p
)
444 return probe_kernel_address((unsigned long *)p
, dummy
);
447 static void dump_pagetable(unsigned long address
)
449 pgd_t
*base
= __va(read_cr3_pa());
450 pgd_t
*pgd
= base
+ pgd_index(address
);
456 if (bad_address(pgd
))
459 pr_info("PGD %lx ", pgd_val(*pgd
));
461 if (!pgd_present(*pgd
))
464 p4d
= p4d_offset(pgd
, address
);
465 if (bad_address(p4d
))
468 pr_cont("P4D %lx ", p4d_val(*p4d
));
469 if (!p4d_present(*p4d
) || p4d_large(*p4d
))
472 pud
= pud_offset(p4d
, address
);
473 if (bad_address(pud
))
476 pr_cont("PUD %lx ", pud_val(*pud
));
477 if (!pud_present(*pud
) || pud_large(*pud
))
480 pmd
= pmd_offset(pud
, address
);
481 if (bad_address(pmd
))
484 pr_cont("PMD %lx ", pmd_val(*pmd
));
485 if (!pmd_present(*pmd
) || pmd_large(*pmd
))
488 pte
= pte_offset_kernel(pmd
, address
);
489 if (bad_address(pte
))
492 pr_cont("PTE %lx", pte_val(*pte
));
500 #endif /* CONFIG_X86_64 */
503 * Workaround for K8 erratum #93 & buggy BIOS.
505 * BIOS SMM functions are required to use a specific workaround
506 * to avoid corruption of the 64bit RIP register on C stepping K8.
508 * A lot of BIOS that didn't get tested properly miss this.
510 * The OS sees this as a page fault with the upper 32bits of RIP cleared.
511 * Try to work around it here.
513 * Note we only handle faults in kernel here.
514 * Does nothing on 32-bit.
516 static int is_errata93(struct pt_regs
*regs
, unsigned long address
)
518 #if defined(CONFIG_X86_64) && defined(CONFIG_CPU_SUP_AMD)
519 if (boot_cpu_data
.x86_vendor
!= X86_VENDOR_AMD
520 || boot_cpu_data
.x86
!= 0xf)
523 if (address
!= regs
->ip
)
526 if ((address
>> 32) != 0)
529 address
|= 0xffffffffUL
<< 32;
530 if ((address
>= (u64
)_stext
&& address
<= (u64
)_etext
) ||
531 (address
>= MODULES_VADDR
&& address
<= MODULES_END
)) {
532 printk_once(errata93_warning
);
541 * Work around K8 erratum #100 K8 in compat mode occasionally jumps
542 * to illegal addresses >4GB.
544 * We catch this in the page fault handler because these addresses
545 * are not reachable. Just detect this case and return. Any code
546 * segment in LDT is compatibility mode.
548 static int is_errata100(struct pt_regs
*regs
, unsigned long address
)
551 if ((regs
->cs
== __USER32_CS
|| (regs
->cs
& (1<<2))) && (address
>> 32))
557 static int is_f00f_bug(struct pt_regs
*regs
, unsigned long address
)
559 #ifdef CONFIG_X86_F00F_BUG
563 * Pentium F0 0F C7 C8 bug workaround:
565 if (boot_cpu_has_bug(X86_BUG_F00F
)) {
566 nr
= (address
- idt_descr
.address
) >> 3;
569 do_invalid_op(regs
, 0);
577 static void show_ldttss(const struct desc_ptr
*gdt
, const char *name
, u16 index
)
579 u32 offset
= (index
>> 3) * sizeof(struct desc_struct
);
581 struct ldttss_desc desc
;
584 pr_alert("%s: NULL\n", name
);
588 if (offset
+ sizeof(struct ldttss_desc
) >= gdt
->size
) {
589 pr_alert("%s: 0x%hx -- out of bounds\n", name
, index
);
593 if (probe_kernel_read(&desc
, (void *)(gdt
->address
+ offset
),
594 sizeof(struct ldttss_desc
))) {
595 pr_alert("%s: 0x%hx -- GDT entry is not readable\n",
600 addr
= desc
.base0
| (desc
.base1
<< 16) | ((unsigned long)desc
.base2
<< 24);
602 addr
|= ((u64
)desc
.base3
<< 32);
604 pr_alert("%s: 0x%hx -- base=0x%lx limit=0x%x\n",
605 name
, index
, addr
, (desc
.limit0
| (desc
.limit1
<< 16)));
609 show_fault_oops(struct pt_regs
*regs
, unsigned long error_code
, unsigned long address
)
611 if (!oops_may_print())
614 if (error_code
& X86_PF_INSTR
) {
619 pgd
= __va(read_cr3_pa());
620 pgd
+= pgd_index(address
);
622 pte
= lookup_address_in_pgd(pgd
, address
, &level
);
624 if (pte
&& pte_present(*pte
) && !pte_exec(*pte
))
625 pr_crit("kernel tried to execute NX-protected page - exploit attempt? (uid: %d)\n",
626 from_kuid(&init_user_ns
, current_uid()));
627 if (pte
&& pte_present(*pte
) && pte_exec(*pte
) &&
628 (pgd_flags(*pgd
) & _PAGE_USER
) &&
629 (__read_cr4() & X86_CR4_SMEP
))
630 pr_crit("unable to execute userspace code (SMEP?) (uid: %d)\n",
631 from_kuid(&init_user_ns
, current_uid()));
634 if (address
< PAGE_SIZE
&& !user_mode(regs
))
635 pr_alert("BUG: kernel NULL pointer dereference, address: %px\n",
638 pr_alert("BUG: unable to handle page fault for address: %px\n",
641 pr_alert("#PF: %s %s in %s mode\n",
642 (error_code
& X86_PF_USER
) ? "user" : "supervisor",
643 (error_code
& X86_PF_INSTR
) ? "instruction fetch" :
644 (error_code
& X86_PF_WRITE
) ? "write access" :
646 user_mode(regs
) ? "user" : "kernel");
647 pr_alert("#PF: error_code(0x%04lx) - %s\n", error_code
,
648 !(error_code
& X86_PF_PROT
) ? "not-present page" :
649 (error_code
& X86_PF_RSVD
) ? "reserved bit violation" :
650 (error_code
& X86_PF_PK
) ? "protection keys violation" :
651 "permissions violation");
653 if (!(error_code
& X86_PF_USER
) && user_mode(regs
)) {
654 struct desc_ptr idt
, gdt
;
658 * This can happen for quite a few reasons. The more obvious
659 * ones are faults accessing the GDT, or LDT. Perhaps
660 * surprisingly, if the CPU tries to deliver a benign or
661 * contributory exception from user code and gets a page fault
662 * during delivery, the page fault can be delivered as though
663 * it originated directly from user code. This could happen
664 * due to wrong permissions on the IDT, GDT, LDT, TSS, or
665 * kernel or IST stack.
669 /* Usable even on Xen PV -- it's just slow. */
670 native_store_gdt(&gdt
);
672 pr_alert("IDT: 0x%lx (limit=0x%hx) GDT: 0x%lx (limit=0x%hx)\n",
673 idt
.address
, idt
.size
, gdt
.address
, gdt
.size
);
676 show_ldttss(&gdt
, "LDTR", ldtr
);
679 show_ldttss(&gdt
, "TR", tr
);
682 dump_pagetable(address
);
686 pgtable_bad(struct pt_regs
*regs
, unsigned long error_code
,
687 unsigned long address
)
689 struct task_struct
*tsk
;
693 flags
= oops_begin();
697 printk(KERN_ALERT
"%s: Corrupted page table at address %lx\n",
699 dump_pagetable(address
);
701 if (__die("Bad pagetable", regs
, error_code
))
704 oops_end(flags
, regs
, sig
);
707 static void set_signal_archinfo(unsigned long address
,
708 unsigned long error_code
)
710 struct task_struct
*tsk
= current
;
713 * To avoid leaking information about the kernel page
714 * table layout, pretend that user-mode accesses to
715 * kernel addresses are always protection faults.
717 * NB: This means that failed vsyscalls with vsyscall=none
718 * will have the PROT bit. This doesn't leak any
719 * information and does not appear to cause any problems.
721 if (address
>= TASK_SIZE_MAX
)
722 error_code
|= X86_PF_PROT
;
724 tsk
->thread
.trap_nr
= X86_TRAP_PF
;
725 tsk
->thread
.error_code
= error_code
| X86_PF_USER
;
726 tsk
->thread
.cr2
= address
;
730 no_context(struct pt_regs
*regs
, unsigned long error_code
,
731 unsigned long address
, int signal
, int si_code
)
733 struct task_struct
*tsk
= current
;
737 if (user_mode(regs
)) {
739 * This is an implicit supervisor-mode access from user
740 * mode. Bypass all the kernel-mode recovery code and just
746 /* Are we prepared to handle this kernel fault? */
747 if (fixup_exception(regs
, X86_TRAP_PF
, error_code
, address
)) {
749 * Any interrupt that takes a fault gets the fixup. This makes
750 * the below recursive fault logic only apply to a faults from
757 * Per the above we're !in_interrupt(), aka. task context.
759 * In this case we need to make sure we're not recursively
760 * faulting through the emulate_vsyscall() logic.
762 if (current
->thread
.sig_on_uaccess_err
&& signal
) {
763 set_signal_archinfo(address
, error_code
);
765 /* XXX: hwpoison faults will set the wrong code. */
766 force_sig_fault(signal
, si_code
, (void __user
*)address
);
770 * Barring that, we can do the fixup and be happy.
775 #ifdef CONFIG_VMAP_STACK
777 * Stack overflow? During boot, we can fault near the initial
778 * stack in the direct map, but that's not an overflow -- check
779 * that we're in vmalloc space to avoid this.
781 if (is_vmalloc_addr((void *)address
) &&
782 (((unsigned long)tsk
->stack
- 1 - address
< PAGE_SIZE
) ||
783 address
- ((unsigned long)tsk
->stack
+ THREAD_SIZE
) < PAGE_SIZE
)) {
784 unsigned long stack
= __this_cpu_ist_top_va(DF
) - sizeof(void *);
786 * We're likely to be running with very little stack space
787 * left. It's plausible that we'd hit this condition but
788 * double-fault even before we get this far, in which case
789 * we're fine: the double-fault handler will deal with it.
791 * We don't want to make it all the way into the oops code
792 * and then double-fault, though, because we're likely to
793 * break the console driver and lose most of the stack dump.
795 asm volatile ("movq %[stack], %%rsp\n\t"
796 "call handle_stack_overflow\n\t"
798 : ASM_CALL_CONSTRAINT
799 : "D" ("kernel stack overflow (page fault)"),
800 "S" (regs
), "d" (address
),
801 [stack
] "rm" (stack
));
809 * Valid to do another page fault here, because if this fault
810 * had been triggered by is_prefetch fixup_exception would have
815 * Hall of shame of CPU/BIOS bugs.
817 if (is_prefetch(regs
, error_code
, address
))
820 if (is_errata93(regs
, address
))
824 * Buggy firmware could access regions which might page fault, try to
825 * recover from such faults.
827 if (IS_ENABLED(CONFIG_EFI
))
828 efi_recover_from_page_fault(address
);
832 * Oops. The kernel tried to access some bad page. We'll have to
833 * terminate things with extreme prejudice:
835 flags
= oops_begin();
837 show_fault_oops(regs
, error_code
, address
);
839 if (task_stack_end_corrupted(tsk
))
840 printk(KERN_EMERG
"Thread overran stack, or stack corrupted\n");
843 if (__die("Oops", regs
, error_code
))
846 /* Executive summary in case the body of the oops scrolled away */
847 printk(KERN_DEFAULT
"CR2: %016lx\n", address
);
849 oops_end(flags
, regs
, sig
);
853 * Print out info about fatal segfaults, if the show_unhandled_signals
857 show_signal_msg(struct pt_regs
*regs
, unsigned long error_code
,
858 unsigned long address
, struct task_struct
*tsk
)
860 const char *loglvl
= task_pid_nr(tsk
) > 1 ? KERN_INFO
: KERN_EMERG
;
862 if (!unhandled_signal(tsk
, SIGSEGV
))
865 if (!printk_ratelimit())
868 printk("%s%s[%d]: segfault at %lx ip %px sp %px error %lx",
869 loglvl
, tsk
->comm
, task_pid_nr(tsk
), address
,
870 (void *)regs
->ip
, (void *)regs
->sp
, error_code
);
872 print_vma_addr(KERN_CONT
" in ", regs
->ip
);
874 printk(KERN_CONT
"\n");
876 show_opcodes(regs
, loglvl
);
880 * The (legacy) vsyscall page is the long page in the kernel portion
881 * of the address space that has user-accessible permissions.
883 static bool is_vsyscall_vaddr(unsigned long vaddr
)
885 return unlikely((vaddr
& PAGE_MASK
) == VSYSCALL_ADDR
);
889 __bad_area_nosemaphore(struct pt_regs
*regs
, unsigned long error_code
,
890 unsigned long address
, u32 pkey
, int si_code
)
892 struct task_struct
*tsk
= current
;
894 /* User mode accesses just cause a SIGSEGV */
895 if (user_mode(regs
) && (error_code
& X86_PF_USER
)) {
897 * It's possible to have interrupts off here:
902 * Valid to do another page fault here because this one came
905 if (is_prefetch(regs
, error_code
, address
))
908 if (is_errata100(regs
, address
))
912 * To avoid leaking information about the kernel page table
913 * layout, pretend that user-mode accesses to kernel addresses
914 * are always protection faults.
916 if (address
>= TASK_SIZE_MAX
)
917 error_code
|= X86_PF_PROT
;
919 if (likely(show_unhandled_signals
))
920 show_signal_msg(regs
, error_code
, address
, tsk
);
922 set_signal_archinfo(address
, error_code
);
924 if (si_code
== SEGV_PKUERR
)
925 force_sig_pkuerr((void __user
*)address
, pkey
);
927 force_sig_fault(SIGSEGV
, si_code
, (void __user
*)address
);
932 if (is_f00f_bug(regs
, address
))
935 no_context(regs
, error_code
, address
, SIGSEGV
, si_code
);
939 bad_area_nosemaphore(struct pt_regs
*regs
, unsigned long error_code
,
940 unsigned long address
)
942 __bad_area_nosemaphore(regs
, error_code
, address
, 0, SEGV_MAPERR
);
946 __bad_area(struct pt_regs
*regs
, unsigned long error_code
,
947 unsigned long address
, u32 pkey
, int si_code
)
949 struct mm_struct
*mm
= current
->mm
;
951 * Something tried to access memory that isn't in our memory map..
952 * Fix it, but check if it's kernel or user first..
954 up_read(&mm
->mmap_sem
);
956 __bad_area_nosemaphore(regs
, error_code
, address
, pkey
, si_code
);
960 bad_area(struct pt_regs
*regs
, unsigned long error_code
, unsigned long address
)
962 __bad_area(regs
, error_code
, address
, 0, SEGV_MAPERR
);
965 static inline bool bad_area_access_from_pkeys(unsigned long error_code
,
966 struct vm_area_struct
*vma
)
968 /* This code is always called on the current mm */
969 bool foreign
= false;
971 if (!boot_cpu_has(X86_FEATURE_OSPKE
))
973 if (error_code
& X86_PF_PK
)
975 /* this checks permission keys on the VMA: */
976 if (!arch_vma_access_permitted(vma
, (error_code
& X86_PF_WRITE
),
977 (error_code
& X86_PF_INSTR
), foreign
))
983 bad_area_access_error(struct pt_regs
*regs
, unsigned long error_code
,
984 unsigned long address
, struct vm_area_struct
*vma
)
987 * This OSPKE check is not strictly necessary at runtime.
988 * But, doing it this way allows compiler optimizations
989 * if pkeys are compiled out.
991 if (bad_area_access_from_pkeys(error_code
, vma
)) {
993 * A protection key fault means that the PKRU value did not allow
994 * access to some PTE. Userspace can figure out what PKRU was
995 * from the XSAVE state. This function captures the pkey from
996 * the vma and passes it to userspace so userspace can discover
997 * which protection key was set on the PTE.
999 * If we get here, we know that the hardware signaled a X86_PF_PK
1000 * fault and that there was a VMA once we got in the fault
1001 * handler. It does *not* guarantee that the VMA we find here
1002 * was the one that we faulted on.
1004 * 1. T1 : mprotect_key(foo, PAGE_SIZE, pkey=4);
1005 * 2. T1 : set PKRU to deny access to pkey=4, touches page
1007 * 4. T2: mprotect_key(foo, PAGE_SIZE, pkey=5);
1008 * 5. T1 : enters fault handler, takes mmap_sem, etc...
1009 * 6. T1 : reaches here, sees vma_pkey(vma)=5, when we really
1010 * faulted on a pte with its pkey=4.
1012 u32 pkey
= vma_pkey(vma
);
1014 __bad_area(regs
, error_code
, address
, pkey
, SEGV_PKUERR
);
1016 __bad_area(regs
, error_code
, address
, 0, SEGV_ACCERR
);
1021 do_sigbus(struct pt_regs
*regs
, unsigned long error_code
, unsigned long address
,
1024 /* Kernel mode? Handle exceptions or die: */
1025 if (!(error_code
& X86_PF_USER
)) {
1026 no_context(regs
, error_code
, address
, SIGBUS
, BUS_ADRERR
);
1030 /* User-space => ok to do another page fault: */
1031 if (is_prefetch(regs
, error_code
, address
))
1034 set_signal_archinfo(address
, error_code
);
1036 #ifdef CONFIG_MEMORY_FAILURE
1037 if (fault
& (VM_FAULT_HWPOISON
|VM_FAULT_HWPOISON_LARGE
)) {
1038 struct task_struct
*tsk
= current
;
1042 "MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n",
1043 tsk
->comm
, tsk
->pid
, address
);
1044 if (fault
& VM_FAULT_HWPOISON_LARGE
)
1045 lsb
= hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault
));
1046 if (fault
& VM_FAULT_HWPOISON
)
1048 force_sig_mceerr(BUS_MCEERR_AR
, (void __user
*)address
, lsb
);
1052 force_sig_fault(SIGBUS
, BUS_ADRERR
, (void __user
*)address
);
1055 static noinline
void
1056 mm_fault_error(struct pt_regs
*regs
, unsigned long error_code
,
1057 unsigned long address
, vm_fault_t fault
)
1059 if (fatal_signal_pending(current
) && !(error_code
& X86_PF_USER
)) {
1060 no_context(regs
, error_code
, address
, 0, 0);
1064 if (fault
& VM_FAULT_OOM
) {
1065 /* Kernel mode? Handle exceptions or die: */
1066 if (!(error_code
& X86_PF_USER
)) {
1067 no_context(regs
, error_code
, address
,
1068 SIGSEGV
, SEGV_MAPERR
);
1073 * We ran out of memory, call the OOM killer, and return the
1074 * userspace (which will retry the fault, or kill us if we got
1077 pagefault_out_of_memory();
1079 if (fault
& (VM_FAULT_SIGBUS
|VM_FAULT_HWPOISON
|
1080 VM_FAULT_HWPOISON_LARGE
))
1081 do_sigbus(regs
, error_code
, address
, fault
);
1082 else if (fault
& VM_FAULT_SIGSEGV
)
1083 bad_area_nosemaphore(regs
, error_code
, address
);
1089 static int spurious_kernel_fault_check(unsigned long error_code
, pte_t
*pte
)
1091 if ((error_code
& X86_PF_WRITE
) && !pte_write(*pte
))
1094 if ((error_code
& X86_PF_INSTR
) && !pte_exec(*pte
))
1101 * Handle a spurious fault caused by a stale TLB entry.
1103 * This allows us to lazily refresh the TLB when increasing the
1104 * permissions of a kernel page (RO -> RW or NX -> X). Doing it
1105 * eagerly is very expensive since that implies doing a full
1106 * cross-processor TLB flush, even if no stale TLB entries exist
1107 * on other processors.
1109 * Spurious faults may only occur if the TLB contains an entry with
1110 * fewer permission than the page table entry. Non-present (P = 0)
1111 * and reserved bit (R = 1) faults are never spurious.
1113 * There are no security implications to leaving a stale TLB when
1114 * increasing the permissions on a page.
1116 * Returns non-zero if a spurious fault was handled, zero otherwise.
1118 * See Intel Developer's Manual Vol 3 Section 4.10.4.3, bullet 3
1119 * (Optional Invalidation).
1122 spurious_kernel_fault(unsigned long error_code
, unsigned long address
)
1132 * Only writes to RO or instruction fetches from NX may cause
1135 * These could be from user or supervisor accesses but the TLB
1136 * is only lazily flushed after a kernel mapping protection
1137 * change, so user accesses are not expected to cause spurious
1140 if (error_code
!= (X86_PF_WRITE
| X86_PF_PROT
) &&
1141 error_code
!= (X86_PF_INSTR
| X86_PF_PROT
))
1144 pgd
= init_mm
.pgd
+ pgd_index(address
);
1145 if (!pgd_present(*pgd
))
1148 p4d
= p4d_offset(pgd
, address
);
1149 if (!p4d_present(*p4d
))
1152 if (p4d_large(*p4d
))
1153 return spurious_kernel_fault_check(error_code
, (pte_t
*) p4d
);
1155 pud
= pud_offset(p4d
, address
);
1156 if (!pud_present(*pud
))
1159 if (pud_large(*pud
))
1160 return spurious_kernel_fault_check(error_code
, (pte_t
*) pud
);
1162 pmd
= pmd_offset(pud
, address
);
1163 if (!pmd_present(*pmd
))
1166 if (pmd_large(*pmd
))
1167 return spurious_kernel_fault_check(error_code
, (pte_t
*) pmd
);
1169 pte
= pte_offset_kernel(pmd
, address
);
1170 if (!pte_present(*pte
))
1173 ret
= spurious_kernel_fault_check(error_code
, pte
);
1178 * Make sure we have permissions in PMD.
1179 * If not, then there's a bug in the page tables:
1181 ret
= spurious_kernel_fault_check(error_code
, (pte_t
*) pmd
);
1182 WARN_ONCE(!ret
, "PMD has incorrect permission bits\n");
1186 NOKPROBE_SYMBOL(spurious_kernel_fault
);
1188 int show_unhandled_signals
= 1;
1191 access_error(unsigned long error_code
, struct vm_area_struct
*vma
)
1193 /* This is only called for the current mm, so: */
1194 bool foreign
= false;
1197 * Read or write was blocked by protection keys. This is
1198 * always an unconditional error and can never result in
1199 * a follow-up action to resolve the fault, like a COW.
1201 if (error_code
& X86_PF_PK
)
1205 * Make sure to check the VMA so that we do not perform
1206 * faults just to hit a X86_PF_PK as soon as we fill in a
1209 if (!arch_vma_access_permitted(vma
, (error_code
& X86_PF_WRITE
),
1210 (error_code
& X86_PF_INSTR
), foreign
))
1213 if (error_code
& X86_PF_WRITE
) {
1214 /* write, present and write, not present: */
1215 if (unlikely(!(vma
->vm_flags
& VM_WRITE
)))
1220 /* read, present: */
1221 if (unlikely(error_code
& X86_PF_PROT
))
1224 /* read, not present: */
1225 if (unlikely(!vma_is_accessible(vma
)))
1231 static int fault_in_kernel_space(unsigned long address
)
1234 * On 64-bit systems, the vsyscall page is at an address above
1235 * TASK_SIZE_MAX, but is not considered part of the kernel
1238 if (IS_ENABLED(CONFIG_X86_64
) && is_vsyscall_vaddr(address
))
1241 return address
>= TASK_SIZE_MAX
;
1245 * Called for all faults where 'address' is part of the kernel address
1246 * space. Might get called for faults that originate from *code* that
1247 * ran in userspace or the kernel.
1250 do_kern_addr_fault(struct pt_regs
*regs
, unsigned long hw_error_code
,
1251 unsigned long address
)
1254 * Protection keys exceptions only happen on user pages. We
1255 * have no user pages in the kernel portion of the address
1256 * space, so do not expect them here.
1258 WARN_ON_ONCE(hw_error_code
& X86_PF_PK
);
1261 * We can fault-in kernel-space virtual memory on-demand. The
1262 * 'reference' page table is init_mm.pgd.
1264 * NOTE! We MUST NOT take any locks for this case. We may
1265 * be in an interrupt or a critical region, and should
1266 * only copy the information from the master page table,
1269 * Before doing this on-demand faulting, ensure that the
1270 * fault is not any of the following:
1271 * 1. A fault on a PTE with a reserved bit set.
1272 * 2. A fault caused by a user-mode access. (Do not demand-
1273 * fault kernel memory due to user-mode accesses).
1274 * 3. A fault caused by a page-level protection violation.
1275 * (A demand fault would be on a non-present page which
1276 * would have X86_PF_PROT==0).
1278 if (!(hw_error_code
& (X86_PF_RSVD
| X86_PF_USER
| X86_PF_PROT
))) {
1279 if (vmalloc_fault(address
) >= 0)
1283 /* Was the fault spurious, caused by lazy TLB invalidation? */
1284 if (spurious_kernel_fault(hw_error_code
, address
))
1287 /* kprobes don't want to hook the spurious faults: */
1288 if (kprobe_page_fault(regs
, X86_TRAP_PF
))
1292 * Note, despite being a "bad area", there are quite a few
1293 * acceptable reasons to get here, such as erratum fixups
1294 * and handling kernel code that can fault, like get_user().
1296 * Don't take the mm semaphore here. If we fixup a prefetch
1297 * fault we could otherwise deadlock:
1299 bad_area_nosemaphore(regs
, hw_error_code
, address
);
1301 NOKPROBE_SYMBOL(do_kern_addr_fault
);
1303 /* Handle faults in the user portion of the address space */
1305 void do_user_addr_fault(struct pt_regs
*regs
,
1306 unsigned long hw_error_code
,
1307 unsigned long address
)
1309 struct vm_area_struct
*vma
;
1310 struct task_struct
*tsk
;
1311 struct mm_struct
*mm
;
1312 vm_fault_t fault
, major
= 0;
1313 unsigned int flags
= FAULT_FLAG_DEFAULT
;
1318 /* kprobes don't want to hook the spurious faults: */
1319 if (unlikely(kprobe_page_fault(regs
, X86_TRAP_PF
)))
1323 * Reserved bits are never expected to be set on
1324 * entries in the user portion of the page tables.
1326 if (unlikely(hw_error_code
& X86_PF_RSVD
))
1327 pgtable_bad(regs
, hw_error_code
, address
);
1330 * If SMAP is on, check for invalid kernel (supervisor) access to user
1331 * pages in the user address space. The odd case here is WRUSS,
1332 * which, according to the preliminary documentation, does not respect
1333 * SMAP and will have the USER bit set so, in all cases, SMAP
1334 * enforcement appears to be consistent with the USER bit.
1336 if (unlikely(cpu_feature_enabled(X86_FEATURE_SMAP
) &&
1337 !(hw_error_code
& X86_PF_USER
) &&
1338 !(regs
->flags
& X86_EFLAGS_AC
)))
1340 bad_area_nosemaphore(regs
, hw_error_code
, address
);
1345 * If we're in an interrupt, have no user context or are running
1346 * in a region with pagefaults disabled then we must not take the fault
1348 if (unlikely(faulthandler_disabled() || !mm
)) {
1349 bad_area_nosemaphore(regs
, hw_error_code
, address
);
1354 * It's safe to allow irq's after cr2 has been saved and the
1355 * vmalloc fault has been handled.
1357 * User-mode registers count as a user access even for any
1358 * potential system fault or CPU buglet:
1360 if (user_mode(regs
)) {
1362 flags
|= FAULT_FLAG_USER
;
1364 if (regs
->flags
& X86_EFLAGS_IF
)
1368 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS
, 1, regs
, address
);
1370 if (hw_error_code
& X86_PF_WRITE
)
1371 flags
|= FAULT_FLAG_WRITE
;
1372 if (hw_error_code
& X86_PF_INSTR
)
1373 flags
|= FAULT_FLAG_INSTRUCTION
;
1375 #ifdef CONFIG_X86_64
1377 * Faults in the vsyscall page might need emulation. The
1378 * vsyscall page is at a high address (>PAGE_OFFSET), but is
1379 * considered to be part of the user address space.
1381 * The vsyscall page does not have a "real" VMA, so do this
1382 * emulation before we go searching for VMAs.
1384 * PKRU never rejects instruction fetches, so we don't need
1385 * to consider the PF_PK bit.
1387 if (is_vsyscall_vaddr(address
)) {
1388 if (emulate_vsyscall(hw_error_code
, regs
, address
))
1394 * Kernel-mode access to the user address space should only occur
1395 * on well-defined single instructions listed in the exception
1396 * tables. But, an erroneous kernel fault occurring outside one of
1397 * those areas which also holds mmap_sem might deadlock attempting
1398 * to validate the fault against the address space.
1400 * Only do the expensive exception table search when we might be at
1401 * risk of a deadlock. This happens if we
1402 * 1. Failed to acquire mmap_sem, and
1403 * 2. The access did not originate in userspace.
1405 if (unlikely(!down_read_trylock(&mm
->mmap_sem
))) {
1406 if (!user_mode(regs
) && !search_exception_tables(regs
->ip
)) {
1408 * Fault from code in kernel from
1409 * which we do not expect faults.
1411 bad_area_nosemaphore(regs
, hw_error_code
, address
);
1415 down_read(&mm
->mmap_sem
);
1418 * The above down_read_trylock() might have succeeded in
1419 * which case we'll have missed the might_sleep() from
1425 vma
= find_vma(mm
, address
);
1426 if (unlikely(!vma
)) {
1427 bad_area(regs
, hw_error_code
, address
);
1430 if (likely(vma
->vm_start
<= address
))
1432 if (unlikely(!(vma
->vm_flags
& VM_GROWSDOWN
))) {
1433 bad_area(regs
, hw_error_code
, address
);
1436 if (unlikely(expand_stack(vma
, address
))) {
1437 bad_area(regs
, hw_error_code
, address
);
1442 * Ok, we have a good vm_area for this memory access, so
1443 * we can handle it..
1446 if (unlikely(access_error(hw_error_code
, vma
))) {
1447 bad_area_access_error(regs
, hw_error_code
, address
, vma
);
1452 * If for any reason at all we couldn't handle the fault,
1453 * make sure we exit gracefully rather than endlessly redo
1454 * the fault. Since we never set FAULT_FLAG_RETRY_NOWAIT, if
1455 * we get VM_FAULT_RETRY back, the mmap_sem has been unlocked.
1457 * Note that handle_userfault() may also release and reacquire mmap_sem
1458 * (and not return with VM_FAULT_RETRY), when returning to userland to
1459 * repeat the page fault later with a VM_FAULT_NOPAGE retval
1460 * (potentially after handling any pending signal during the return to
1461 * userland). The return to userland is identified whenever
1462 * FAULT_FLAG_USER|FAULT_FLAG_KILLABLE are both set in flags.
1464 fault
= handle_mm_fault(vma
, address
, flags
);
1465 major
|= fault
& VM_FAULT_MAJOR
;
1467 /* Quick path to respond to signals */
1468 if (fault_signal_pending(fault
, regs
)) {
1469 if (!user_mode(regs
))
1470 no_context(regs
, hw_error_code
, address
, SIGBUS
,
1476 * If we need to retry the mmap_sem has already been released,
1477 * and if there is a fatal signal pending there is no guarantee
1478 * that we made any progress. Handle this case first.
1480 if (unlikely((fault
& VM_FAULT_RETRY
) &&
1481 (flags
& FAULT_FLAG_ALLOW_RETRY
))) {
1482 flags
|= FAULT_FLAG_TRIED
;
1486 up_read(&mm
->mmap_sem
);
1487 if (unlikely(fault
& VM_FAULT_ERROR
)) {
1488 mm_fault_error(regs
, hw_error_code
, address
, fault
);
1493 * Major/minor page fault accounting. If any of the events
1494 * returned VM_FAULT_MAJOR, we account it as a major fault.
1498 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ
, 1, regs
, address
);
1501 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN
, 1, regs
, address
);
1504 check_v8086_mode(regs
, address
, tsk
);
1506 NOKPROBE_SYMBOL(do_user_addr_fault
);
1508 static __always_inline
void
1509 trace_page_fault_entries(struct pt_regs
*regs
, unsigned long error_code
,
1510 unsigned long address
)
1512 if (!trace_pagefault_enabled())
1515 if (user_mode(regs
))
1516 trace_page_fault_user(address
, regs
, error_code
);
1518 trace_page_fault_kernel(address
, regs
, error_code
);
1522 do_page_fault(struct pt_regs
*regs
, unsigned long hw_error_code
,
1523 unsigned long address
)
1525 prefetchw(¤t
->mm
->mmap_sem
);
1526 trace_page_fault_entries(regs
, hw_error_code
, address
);
1528 if (unlikely(kmmio_fault(regs
, address
)))
1531 /* Was the fault on kernel-controlled part of the address space? */
1532 if (unlikely(fault_in_kernel_space(address
)))
1533 do_kern_addr_fault(regs
, hw_error_code
, address
);
1535 do_user_addr_fault(regs
, hw_error_code
, address
);
1537 NOKPROBE_SYMBOL(do_page_fault
);