3 * Copyright (C) 1991, 1992 Linus Torvalds
7 * entry.S contains the system-call and fault low-level handling routines.
8 * This also contains the timer-interrupt handler, as well as all interrupts
9 * and faults that can result in a task-switch.
11 * NOTE: This code handles signal-recognition, which happens every time
12 * after a timer-interrupt and after each system call.
14 * I changed all the .align's to 4 (16 byte alignment), as that's faster
17 * Stack layout in 'syscall_exit':
18 * ptrace needs to have all regs on the stack.
19 * if the order here is changed, it needs to be
20 * updated in fork.c:copy_process, signal.c:do_signal,
21 * ptrace.c and ptrace.h
33 * 28(%esp) - %gs saved iff !CONFIG_X86_32_LAZY_GS
41 * "current" is in register %ebx during any slow entries.
44 #include <linux/linkage.h>
45 #include <asm/thread_info.h>
46 #include <asm/irqflags.h>
47 #include <asm/errno.h>
48 #include <asm/segment.h>
50 #include <asm/page_types.h>
51 #include <asm/percpu.h>
52 #include <asm/dwarf2.h>
53 #include <asm/processor-flags.h>
54 #include <asm/ftrace.h>
55 #include <asm/irq_vectors.h>
56 #include <asm/cpufeature.h>
57 #include <asm/alternative-asm.h>
59 /* Avoid __ASSEMBLER__'ifying <linux/audit.h> just for this. */
60 #include <linux/elf-em.h>
61 #define AUDIT_ARCH_I386 (EM_386|__AUDIT_ARCH_LE)
62 #define __AUDIT_ARCH_LE 0x40000000
64 #ifndef CONFIG_AUDITSYSCALL
65 #define sysenter_audit syscall_trace_entry
66 #define sysexit_audit syscall_exit_work
69 .section .entry.text, "ax"
72 * We use macros for low-level operations which need to be overridden
73 * for paravirtualization. The following will never clobber any registers:
74 * INTERRUPT_RETURN (aka. "iret")
75 * GET_CR0_INTO_EAX (aka. "movl %cr0, %eax")
76 * ENABLE_INTERRUPTS_SYSEXIT (aka "sti; sysexit").
78 * For DISABLE_INTERRUPTS/ENABLE_INTERRUPTS (aka "cli"/"sti"), you must
79 * specify what registers can be overwritten (CLBR_NONE, CLBR_EAX/EDX/ECX/ANY).
80 * Allowing a register to be clobbered can shrink the paravirt replacement
81 * enough to patch inline, increasing performance.
84 #define nr_syscalls ((syscall_table_size)/4)
87 #define preempt_stop(clobbers) DISABLE_INTERRUPTS(clobbers); TRACE_IRQS_OFF
89 #define preempt_stop(clobbers)
90 #define resume_kernel restore_all
93 .macro TRACE_IRQS_IRET
94 #ifdef CONFIG_TRACE_IRQFLAGS
95 testl $X86_EFLAGS_IF,PT_EFLAGS(%esp) # interrupts off?
103 #define resume_userspace_sig check_userspace
105 #define resume_userspace_sig resume_userspace
109 * User gs save/restore
111 * %gs is used for userland TLS and kernel only uses it for stack
112 * canary which is required to be at %gs:20 by gcc. Read the comment
113 * at the top of stackprotector.h for more info.
115 * Local labels 98 and 99 are used.
117 #ifdef CONFIG_X86_32_LAZY_GS
119 /* unfortunately push/pop can't be no-op */
124 addl $(4 + \pop), %esp
125 CFI_ADJUST_CFA_OFFSET -(4 + \pop)
130 /* all the rest are no-op */
137 .macro REG_TO_PTGS reg
139 .macro SET_KERNEL_GS reg
142 #else /* CONFIG_X86_32_LAZY_GS */
146 /*CFI_REL_OFFSET gs, 0*/
154 CFI_ADJUST_CFA_OFFSET -\pop
158 .pushsection .fixup, "ax"
161 .section __ex_table, "a"
168 98: mov PT_GS(%esp), %gs
171 .pushsection .fixup, "ax"
172 99: movl $0, PT_GS(%esp)
174 .section __ex_table, "a"
182 /*CFI_REGISTER gs, \reg*/
184 .macro REG_TO_PTGS reg
185 movl \reg, PT_GS(%esp)
186 /*CFI_REL_OFFSET gs, PT_GS*/
188 .macro SET_KERNEL_GS reg
189 movl $(__KERNEL_STACK_CANARY), \reg
193 #endif /* CONFIG_X86_32_LAZY_GS */
199 /*CFI_REL_OFFSET fs, 0;*/
201 /*CFI_REL_OFFSET es, 0;*/
203 /*CFI_REL_OFFSET ds, 0;*/
205 CFI_REL_OFFSET eax, 0
207 CFI_REL_OFFSET ebp, 0
209 CFI_REL_OFFSET edi, 0
211 CFI_REL_OFFSET esi, 0
213 CFI_REL_OFFSET edx, 0
215 CFI_REL_OFFSET ecx, 0
217 CFI_REL_OFFSET ebx, 0
218 movl $(__USER_DS), %edx
221 movl $(__KERNEL_PERCPU), %edx
226 .macro RESTORE_INT_REGS
243 .macro RESTORE_REGS pop=0
252 .pushsection .fixup, "ax"
259 .section __ex_table, "a"
268 .macro RING0_INT_FRAME
272 /*CFI_OFFSET cs, -2*4;*/
276 .macro RING0_EC_FRAME
280 /*CFI_OFFSET cs, -2*4;*/
284 .macro RING0_PTREGS_FRAME
287 CFI_DEF_CFA esp, PT_OLDESP-PT_EBX
288 /*CFI_OFFSET cs, PT_CS-PT_OLDESP;*/
289 CFI_OFFSET eip, PT_EIP-PT_OLDESP
290 /*CFI_OFFSET es, PT_ES-PT_OLDESP;*/
291 /*CFI_OFFSET ds, PT_DS-PT_OLDESP;*/
292 CFI_OFFSET eax, PT_EAX-PT_OLDESP
293 CFI_OFFSET ebp, PT_EBP-PT_OLDESP
294 CFI_OFFSET edi, PT_EDI-PT_OLDESP
295 CFI_OFFSET esi, PT_ESI-PT_OLDESP
296 CFI_OFFSET edx, PT_EDX-PT_OLDESP
297 CFI_OFFSET ecx, PT_ECX-PT_OLDESP
298 CFI_OFFSET ebx, PT_EBX-PT_OLDESP
305 GET_THREAD_INFO(%ebp)
307 pushl_cfi $0x0202 # Reset kernel eflags
314 * Interrupt exit functions should be protected against kprobes
316 .pushsection .kprobes.text, "ax"
318 * Return to user mode is not as complex as all this looks,
319 * but we want the default path for a system call return to
320 * go as quickly as possible which is why some of this is
321 * less clear than it otherwise should be.
324 # userspace resumption stub bypassing syscall exit tracing
328 preempt_stop(CLBR_ANY)
330 GET_THREAD_INFO(%ebp)
332 movl PT_EFLAGS(%esp), %eax # mix EFLAGS and CS
333 movb PT_CS(%esp), %al
334 andl $(X86_EFLAGS_VM | SEGMENT_RPL_MASK), %eax
336 jb resume_kernel # not returning to v8086 or userspace
338 ENTRY(resume_userspace)
340 DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt
341 # setting need_resched or sigpending
342 # between sampling and the iret
344 movl TI_flags(%ebp), %ecx
345 andl $_TIF_WORK_MASK, %ecx # is there any work to be done on
346 # int/exception return?
349 END(ret_from_exception)
351 #ifdef CONFIG_PREEMPT
353 DISABLE_INTERRUPTS(CLBR_ANY)
354 cmpl $0,TI_preempt_count(%ebp) # non-zero preempt_count ?
357 movl TI_flags(%ebp), %ecx # need_resched set ?
358 testb $_TIF_NEED_RESCHED, %cl
360 testl $X86_EFLAGS_IF,PT_EFLAGS(%esp) # interrupts off (exception path) ?
362 call preempt_schedule_irq
368 * End of kprobes section
372 /* SYSENTER_RETURN points to after the "sysenter" instruction in
373 the vsyscall page. See vsyscall-sysentry.S, which defines the symbol. */
375 # sysenter call handler stub
376 ENTRY(ia32_sysenter_target)
380 CFI_REGISTER esp, ebp
381 movl TSS_sysenter_sp0(%esp),%esp
384 * Interrupts are disabled here, but we can't trace it until
385 * enough kernel state to call TRACE_IRQS_OFF can be called - but
386 * we immediately enable interrupts at that point anyway.
389 /*CFI_REL_OFFSET ss, 0*/
391 CFI_REL_OFFSET esp, 0
393 orl $X86_EFLAGS_IF, (%esp)
395 /*CFI_REL_OFFSET cs, 0*/
397 * Push current_thread_info()->sysenter_return to the stack.
398 * A tiny bit of offset fixup is necessary - 4*4 means the 4 words
399 * pushed above; +8 corresponds to copy_thread's esp0 setting.
401 pushl_cfi ((TI_sysenter_return)-THREAD_SIZE+8+4*4)(%esp)
402 CFI_REL_OFFSET eip, 0
406 ENABLE_INTERRUPTS(CLBR_NONE)
409 * Load the potential sixth argument from user stack.
410 * Careful about security.
412 cmpl $__PAGE_OFFSET-3,%ebp
415 movl %ebp,PT_EBP(%esp)
416 .section __ex_table,"a"
418 .long 1b,syscall_fault
421 GET_THREAD_INFO(%ebp)
423 testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%ebp)
426 cmpl $(nr_syscalls), %eax
428 call *sys_call_table(,%eax,4)
429 movl %eax,PT_EAX(%esp)
431 DISABLE_INTERRUPTS(CLBR_ANY)
433 movl TI_flags(%ebp), %ecx
434 testl $_TIF_ALLWORK_MASK, %ecx
437 /* if something modifies registers it must also disable sysexit */
438 movl PT_EIP(%esp), %edx
439 movl PT_OLDESP(%esp), %ecx
442 1: mov PT_FS(%esp), %fs
444 ENABLE_INTERRUPTS_SYSEXIT
446 #ifdef CONFIG_AUDITSYSCALL
448 testl $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags(%ebp)
449 jnz syscall_trace_entry
451 CFI_ADJUST_CFA_OFFSET -4
452 /* %esi already in 8(%esp) 6th arg: 4th syscall arg */
453 /* %edx already in 4(%esp) 5th arg: 3rd syscall arg */
454 /* %ecx already in 0(%esp) 4th arg: 2nd syscall arg */
455 movl %ebx,%ecx /* 3rd arg: 1st syscall arg */
456 movl %eax,%edx /* 2nd arg: syscall number */
457 movl $AUDIT_ARCH_I386,%eax /* 1st arg: audit arch */
458 call audit_syscall_entry
460 movl PT_EAX(%esp),%eax /* reload syscall number */
464 testl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT), %ecx
465 jne syscall_exit_work
467 ENABLE_INTERRUPTS(CLBR_ANY)
468 movl %eax,%edx /* second arg, syscall return value */
469 cmpl $0,%eax /* is it < 0? */
470 setl %al /* 1 if so, 0 if not */
471 movzbl %al,%eax /* zero-extend that */
472 inc %eax /* first arg, 0->1(AUDITSC_SUCCESS), 1->2(AUDITSC_FAILURE) */
473 call audit_syscall_exit
474 DISABLE_INTERRUPTS(CLBR_ANY)
476 movl TI_flags(%ebp), %ecx
477 testl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT), %ecx
478 jne syscall_exit_work
479 movl PT_EAX(%esp),%eax /* reload syscall return value */
484 .pushsection .fixup,"ax"
485 2: movl $0,PT_FS(%esp)
487 .section __ex_table,"a"
492 ENDPROC(ia32_sysenter_target)
495 * syscall stub including irq exit should be protected against kprobes
497 .pushsection .kprobes.text, "ax"
498 # system call handler stub
500 RING0_INT_FRAME # can't unwind into user space anyway
501 pushl_cfi %eax # save orig_eax
503 GET_THREAD_INFO(%ebp)
504 # system call tracing in operation / emulation
505 testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%ebp)
506 jnz syscall_trace_entry
507 cmpl $(nr_syscalls), %eax
510 call *sys_call_table(,%eax,4)
511 movl %eax,PT_EAX(%esp) # store the return value
514 DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt
515 # setting need_resched or sigpending
516 # between sampling and the iret
518 movl TI_flags(%ebp), %ecx
519 testl $_TIF_ALLWORK_MASK, %ecx # current->work
520 jne syscall_exit_work
525 movl PT_EFLAGS(%esp), %eax # mix EFLAGS, SS and CS
526 # Warning: PT_OLDSS(%esp) contains the wrong/random values if we
527 # are returning to the kernel.
528 # See comments in process.c:copy_thread() for details.
529 movb PT_OLDSS(%esp), %ah
530 movb PT_CS(%esp), %al
531 andl $(X86_EFLAGS_VM | (SEGMENT_TI_MASK << 8) | SEGMENT_RPL_MASK), %eax
532 cmpl $((SEGMENT_LDT << 8) | USER_RPL), %eax
534 je ldt_ss # returning to user-space with LDT SS
536 RESTORE_REGS 4 # skip orig_eax/error_code
541 pushl $0 # no error code
545 .section __ex_table,"a"
547 .long irq_return,iret_exc
552 larl PT_OLDSS(%esp), %eax
554 testl $0x00400000, %eax # returning to 32bit stack?
555 jnz restore_nocheck # allright, normal return
557 #ifdef CONFIG_PARAVIRT
559 * The kernel can't run on a non-flat stack if paravirt mode
560 * is active. Rather than try to fixup the high bits of
561 * ESP, bypass this code entirely. This may break DOSemu
562 * and/or Wine support in a paravirt VM, although the option
563 * is still available to implement the setting of the high
564 * 16-bits in the INTERRUPT_RETURN paravirt-op.
566 cmpl $0, pv_info+PARAVIRT_enabled
571 * Setup and switch to ESPFIX stack
573 * We're returning to userspace with a 16 bit stack. The CPU will not
574 * restore the high word of ESP for us on executing iret... This is an
575 * "official" bug of all the x86-compatible CPUs, which we can work
576 * around to make dosemu and wine happy. We do this by preloading the
577 * high word of ESP with the high word of the userspace ESP while
578 * compensating for the offset by changing to the ESPFIX segment with
579 * a base address that matches for the difference.
581 #define GDT_ESPFIX_SS PER_CPU_VAR(gdt_page) + (GDT_ENTRY_ESPFIX_SS * 8)
582 mov %esp, %edx /* load kernel esp */
583 mov PT_OLDESP(%esp), %eax /* load userspace esp */
584 mov %dx, %ax /* eax: new kernel esp */
585 sub %eax, %edx /* offset (low word is 0) */
587 mov %dl, GDT_ESPFIX_SS + 4 /* bits 16..23 */
588 mov %dh, GDT_ESPFIX_SS + 7 /* bits 24..31 */
589 pushl_cfi $__ESPFIX_SS
590 pushl_cfi %eax /* new kernel esp */
591 /* Disable interrupts, but do not irqtrace this section: we
592 * will soon execute iret and the tracer was already set to
593 * the irqstate after the iret */
594 DISABLE_INTERRUPTS(CLBR_EAX)
595 lss (%esp), %esp /* switch to espfix segment */
596 CFI_ADJUST_CFA_OFFSET -8
601 # perform work that needs to be done immediately before resumption
603 RING0_PTREGS_FRAME # can't unwind into user space anyway
605 testb $_TIF_NEED_RESCHED, %cl
610 DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt
611 # setting need_resched or sigpending
612 # between sampling and the iret
614 movl TI_flags(%ebp), %ecx
615 andl $_TIF_WORK_MASK, %ecx # is there any work to be done other
616 # than syscall tracing?
618 testb $_TIF_NEED_RESCHED, %cl
621 work_notifysig: # deal with pending signals and
622 # notify-resume requests
624 testl $X86_EFLAGS_VM, PT_EFLAGS(%esp)
626 jne work_notifysig_v86 # returning to kernel-space or
629 call do_notify_resume
630 jmp resume_userspace_sig
634 pushl_cfi %ecx # save ti_flags for do_notify_resume
635 call save_v86_state # %eax contains pt_regs pointer
642 call do_notify_resume
643 jmp resume_userspace_sig
646 # perform syscall exit tracing
649 movl $-ENOSYS,PT_EAX(%esp)
651 call syscall_trace_enter
652 /* What it returned is what we'll actually use. */
653 cmpl $(nr_syscalls), %eax
656 END(syscall_trace_entry)
658 # perform syscall exit tracing
661 testl $_TIF_WORK_SYSCALL_EXIT, %ecx
664 ENABLE_INTERRUPTS(CLBR_ANY) # could let syscall_trace_leave() call
667 call syscall_trace_leave
669 END(syscall_exit_work)
672 RING0_INT_FRAME # can't unwind into user space anyway
674 GET_THREAD_INFO(%ebp)
675 movl $-EFAULT,PT_EAX(%esp)
680 movl $-ENOSYS,PT_EAX(%esp)
685 * End of kprobes section
690 * System calls that need a pt_regs pointer.
692 #define PTREGSCALL0(name) \
698 #define PTREGSCALL1(name) \
702 movl (PT_EBX+4)(%esp),%eax; \
705 #define PTREGSCALL2(name) \
709 movl (PT_ECX+4)(%esp),%edx; \
710 movl (PT_EBX+4)(%esp),%eax; \
713 #define PTREGSCALL3(name) \
719 movl PT_EDX(%eax),%ecx; \
720 movl PT_ECX(%eax),%edx; \
721 movl PT_EBX(%eax),%eax; \
724 CFI_ADJUST_CFA_OFFSET -4; \
727 ENDPROC(ptregs_##name)
733 PTREGSCALL2(sigaltstack)
734 PTREGSCALL0(sigreturn)
735 PTREGSCALL0(rt_sigreturn)
739 /* Clone is an oddball. The 4th arg is in %edi */
745 pushl_cfi PT_EDI(%eax)
746 movl PT_EDX(%eax),%ecx
747 movl PT_ECX(%eax),%edx
748 movl PT_EBX(%eax),%eax
751 CFI_ADJUST_CFA_OFFSET -8
754 ENDPROC(ptregs_clone)
756 .macro FIXUP_ESPFIX_STACK
758 * Switch back for ESPFIX stack to the normal zerobased stack
760 * We can't call C functions using the ESPFIX stack. This code reads
761 * the high word of the segment base from the GDT and swiches to the
762 * normal stack and adjusts ESP with the matching offset.
764 /* fixup the stack */
765 mov GDT_ESPFIX_SS + 4, %al /* bits 16..23 */
766 mov GDT_ESPFIX_SS + 7, %ah /* bits 24..31 */
768 addl %esp, %eax /* the adjusted stack pointer */
769 pushl_cfi $__KERNEL_DS
771 lss (%esp), %esp /* switch to the normal stack segment */
772 CFI_ADJUST_CFA_OFFSET -8
774 .macro UNWIND_ESPFIX_STACK
776 /* see if on espfix stack */
777 cmpw $__ESPFIX_SS, %ax
779 movl $__KERNEL_DS, %eax
782 /* switch to normal stack */
788 * Build the entry stubs and pointer table with some assembler magic.
789 * We pack 7 stubs into a single 32-byte chunk, which will fit in a
790 * single cache line on all modern x86 implementations.
792 .section .init.rodata,"a"
794 .section .entry.text, "ax"
796 .p2align CONFIG_X86_L1_CACHE_SHIFT
797 ENTRY(irq_entries_start)
799 vector=FIRST_EXTERNAL_VECTOR
800 .rept (NR_VECTORS-FIRST_EXTERNAL_VECTOR+6)/7
803 .if vector < NR_VECTORS
804 .if vector <> FIRST_EXTERNAL_VECTOR
805 CFI_ADJUST_CFA_OFFSET -4
807 1: pushl_cfi $(~vector+0x80) /* Note: always in signed byte range */
808 .if ((vector-FIRST_EXTERNAL_VECTOR)%7) <> 6
813 .section .entry.text, "ax"
817 2: jmp common_interrupt
819 END(irq_entries_start)
826 * the CPU automatically disables interrupts when executing an IRQ vector,
827 * so IRQ-flags tracing has to follow that:
829 .p2align CONFIG_X86_L1_CACHE_SHIFT
831 addl $-0x80,(%esp) /* Adjust vector into the [-256,-1] range */
837 ENDPROC(common_interrupt)
841 * Irq entries should be protected against kprobes
843 .pushsection .kprobes.text, "ax"
844 #define BUILD_INTERRUPT3(name, nr, fn) \
856 #define BUILD_INTERRUPT(name, nr) BUILD_INTERRUPT3(name, nr, smp_##name)
858 /* The include is where all of the SMP etc. interrupts come from */
859 #include <asm/entry_arch.h>
861 ENTRY(coprocessor_error)
864 pushl_cfi $do_coprocessor_error
867 END(coprocessor_error)
869 ENTRY(simd_coprocessor_error)
872 #ifdef CONFIG_X86_INVD_BUG
873 /* AMD 486 bug: invd from userspace calls exception 19 instead of #GP */
874 661: pushl_cfi $do_general_protection
876 .section .altinstructions,"a"
877 altinstruction_entry 661b, 663f, X86_FEATURE_XMM, 662b-661b, 664f-663f
879 .section .altinstr_replacement,"ax"
880 663: pushl $do_simd_coprocessor_error
884 pushl_cfi $do_simd_coprocessor_error
888 END(simd_coprocessor_error)
890 ENTRY(device_not_available)
892 pushl_cfi $-1 # mark this as an int
893 pushl_cfi $do_device_not_available
896 END(device_not_available)
898 #ifdef CONFIG_PARAVIRT
901 .section __ex_table,"a"
903 .long native_iret, iret_exc
907 ENTRY(native_irq_enable_sysexit)
910 END(native_irq_enable_sysexit)
916 pushl_cfi $do_overflow
932 pushl_cfi $do_invalid_op
937 ENTRY(coprocessor_segment_overrun)
940 pushl_cfi $do_coprocessor_segment_overrun
943 END(coprocessor_segment_overrun)
947 pushl_cfi $do_invalid_TSS
952 ENTRY(segment_not_present)
954 pushl_cfi $do_segment_not_present
957 END(segment_not_present)
961 pushl_cfi $do_stack_segment
966 ENTRY(alignment_check)
968 pushl_cfi $do_alignment_check
975 pushl_cfi $0 # no error code
976 pushl_cfi $do_divide_error
981 #ifdef CONFIG_X86_MCE
985 pushl_cfi machine_check_vector
991 ENTRY(spurious_interrupt_bug)
994 pushl_cfi $do_spurious_interrupt_bug
997 END(spurious_interrupt_bug)
999 * End of kprobes section
1003 ENTRY(kernel_thread_helper)
1004 pushl $0 # fake return address for unwinder
1009 ud2 # padding for call trace
1011 ENDPROC(kernel_thread_helper)
1014 /* Xen doesn't set %esp to be precisely what the normal sysenter
1015 entrypoint expects, so fix it up before using the normal path. */
1016 ENTRY(xen_sysenter_target)
1018 addl $5*4, %esp /* remove xen-provided frame */
1019 CFI_ADJUST_CFA_OFFSET -5*4
1020 jmp sysenter_past_esp
1023 ENTRY(xen_hypervisor_callback)
1029 /* Check to see if we got the event in the critical
1030 region in xen_iret_direct, after we've reenabled
1031 events and checked for pending events. This simulates
1032 iret instruction's behaviour where it delivers a
1033 pending interrupt when enabling interrupts. */
1034 movl PT_EIP(%esp),%eax
1035 cmpl $xen_iret_start_crit,%eax
1037 cmpl $xen_iret_end_crit,%eax
1040 jmp xen_iret_crit_fixup
1042 ENTRY(xen_do_upcall)
1044 call xen_evtchn_do_upcall
1047 ENDPROC(xen_hypervisor_callback)
1049 # Hypervisor uses this for application faults while it executes.
1050 # We get here for two reasons:
1051 # 1. Fault while reloading DS, ES, FS or GS
1052 # 2. Fault while executing IRET
1053 # Category 1 we fix up by reattempting the load, and zeroing the segment
1054 # register if the load fails.
1055 # Category 2 we fix up by jumping to do_iret_error. We cannot use the
1056 # normal Linux return path in this case because if we use the IRET hypercall
1057 # to pop the stack frame we end up in an infinite loop of failsafe callbacks.
1058 # We distinguish between categories by maintaining a status value in EAX.
1059 ENTRY(xen_failsafe_callback)
1070 CFI_ADJUST_CFA_OFFSET -16
1073 jmp iret_exc # EAX != 0 => Category 2 (Bad IRET)
1074 5: pushl_cfi $0 # EAX == 0 => Category 1 (Bad segment)
1076 jmp ret_from_exception
1079 .section .fixup,"ax"
1093 .section __ex_table,"a"
1100 ENDPROC(xen_failsafe_callback)
1102 BUILD_INTERRUPT3(xen_hvm_callback_vector, XEN_HVM_EVTCHN_CALLBACK,
1103 xen_evtchn_do_upcall)
1105 #endif /* CONFIG_XEN */
1107 #ifdef CONFIG_FUNCTION_TRACER
1108 #ifdef CONFIG_DYNAMIC_FTRACE
1114 ENTRY(ftrace_caller)
1115 cmpl $0, function_trace_stop
1121 movl 0xc(%esp), %eax
1122 movl 0x4(%ebp), %edx
1123 subl $MCOUNT_INSN_SIZE, %eax
1132 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1133 .globl ftrace_graph_call
1143 #else /* ! CONFIG_DYNAMIC_FTRACE */
1146 cmpl $0, function_trace_stop
1149 cmpl $ftrace_stub, ftrace_trace_function
1151 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1152 cmpl $ftrace_stub, ftrace_graph_return
1153 jnz ftrace_graph_caller
1155 cmpl $ftrace_graph_entry_stub, ftrace_graph_entry
1156 jnz ftrace_graph_caller
1162 /* taken from glibc */
1167 movl 0xc(%esp), %eax
1168 movl 0x4(%ebp), %edx
1169 subl $MCOUNT_INSN_SIZE, %eax
1171 call *ftrace_trace_function
1178 #endif /* CONFIG_DYNAMIC_FTRACE */
1179 #endif /* CONFIG_FUNCTION_TRACER */
1181 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1182 ENTRY(ftrace_graph_caller)
1183 cmpl $0, function_trace_stop
1189 movl 0xc(%esp), %edx
1192 subl $MCOUNT_INSN_SIZE, %edx
1193 call prepare_ftrace_return
1198 END(ftrace_graph_caller)
1200 .globl return_to_handler
1205 call ftrace_return_to_handler
1212 .section .rodata,"a"
1213 #include "syscall_table_32.S"
1215 syscall_table_size=(.-sys_call_table)
1218 * Some functions should be protected against kprobes
1220 .pushsection .kprobes.text, "ax"
1224 pushl_cfi $do_page_fault
1227 /* the function address is in %gs's slot on the stack */
1229 /*CFI_REL_OFFSET fs, 0*/
1231 /*CFI_REL_OFFSET es, 0*/
1233 /*CFI_REL_OFFSET ds, 0*/
1235 CFI_REL_OFFSET eax, 0
1237 CFI_REL_OFFSET ebp, 0
1239 CFI_REL_OFFSET edi, 0
1241 CFI_REL_OFFSET esi, 0
1243 CFI_REL_OFFSET edx, 0
1245 CFI_REL_OFFSET ecx, 0
1247 CFI_REL_OFFSET ebx, 0
1249 movl $(__KERNEL_PERCPU), %ecx
1253 movl PT_GS(%esp), %edi # get the function address
1254 movl PT_ORIG_EAX(%esp), %edx # get the error code
1255 movl $-1, PT_ORIG_EAX(%esp) # no syscall to restart
1258 movl $(__USER_DS), %ecx
1262 movl %esp,%eax # pt_regs pointer
1264 jmp ret_from_exception
1269 * Debug traps and NMI can happen at the one SYSENTER instruction
1270 * that sets up the real kernel stack. Check here, since we can't
1271 * allow the wrong stack to be used.
1273 * "TSS_sysenter_sp0+12" is because the NMI/debug handler will have
1274 * already pushed 3 words if it hits on the sysenter instruction:
1275 * eflags, cs and eip.
1277 * We just load the right stack, and push the three (known) values
1278 * by hand onto the new stack - while updating the return eip past
1279 * the instruction that would have done it for sysenter.
1281 .macro FIX_STACK offset ok label
1282 cmpw $__KERNEL_CS, 4(%esp)
1285 movl TSS_sysenter_sp0 + \offset(%esp), %esp
1289 pushl_cfi $__KERNEL_CS
1290 pushl_cfi $sysenter_past_esp
1291 CFI_REL_OFFSET eip, 0
1296 cmpl $ia32_sysenter_target,(%esp)
1297 jne debug_stack_correct
1298 FIX_STACK 12, debug_stack_correct, debug_esp_fix_insn
1299 debug_stack_correct:
1300 pushl_cfi $-1 # mark this as an int
1303 xorl %edx,%edx # error code 0
1304 movl %esp,%eax # pt_regs pointer
1306 jmp ret_from_exception
1311 * NMI is doubly nasty. It can happen _while_ we're handling
1312 * a debug fault, and the debug fault hasn't yet been able to
1313 * clear up the stack. So we first check whether we got an
1314 * NMI on the sysenter entry path, but after that we need to
1315 * check whether we got an NMI on the debug path where the debug
1316 * fault happened on the sysenter path.
1322 cmpw $__ESPFIX_SS, %ax
1325 cmpl $ia32_sysenter_target,(%esp)
1329 /* Do not access memory above the end of our stack page,
1330 * it might not exist.
1332 andl $(THREAD_SIZE-1),%eax
1333 cmpl $(THREAD_SIZE-20),%eax
1335 jae nmi_stack_correct
1336 cmpl $ia32_sysenter_target,12(%esp)
1337 je nmi_debug_stack_check
1339 /* We have a RING0_INT_FRAME here */
1342 xorl %edx,%edx # zero error code
1343 movl %esp,%eax # pt_regs pointer
1345 jmp restore_all_notrace
1350 FIX_STACK 12, nmi_stack_correct, 1
1351 jmp nmi_stack_correct
1353 nmi_debug_stack_check:
1354 /* We have a RING0_INT_FRAME here */
1355 cmpw $__KERNEL_CS,16(%esp)
1356 jne nmi_stack_correct
1358 jb nmi_stack_correct
1359 cmpl $debug_esp_fix_insn,(%esp)
1360 ja nmi_stack_correct
1361 FIX_STACK 24, nmi_stack_correct, 1
1362 jmp nmi_stack_correct
1365 /* We have a RING0_INT_FRAME here.
1367 * create the pointer to lss back
1372 /* copy the iret frame of 12 bytes */
1378 FIXUP_ESPFIX_STACK # %eax == %esp
1379 xorl %edx,%edx # zero error code
1382 lss 12+4(%esp), %esp # back to espfix stack
1383 CFI_ADJUST_CFA_OFFSET -24
1390 pushl_cfi $-1 # mark this as an int
1393 xorl %edx,%edx # zero error code
1394 movl %esp,%eax # pt_regs pointer
1396 jmp ret_from_exception
1400 ENTRY(general_protection)
1402 pushl_cfi $do_general_protection
1405 END(general_protection)
1407 #ifdef CONFIG_KVM_GUEST
1408 ENTRY(async_page_fault)
1410 pushl_cfi $do_async_page_fault
1413 END(async_page_fault)
1417 * End of kprobes section