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 <linux/err.h>
46 #include <asm/thread_info.h>
47 #include <asm/irqflags.h>
48 #include <asm/errno.h>
49 #include <asm/segment.h>
51 #include <asm/page_types.h>
52 #include <asm/percpu.h>
53 #include <asm/dwarf2.h>
54 #include <asm/processor-flags.h>
55 #include <asm/ftrace.h>
56 #include <asm/irq_vectors.h>
57 #include <asm/cpufeature.h>
58 #include <asm/alternative-asm.h>
62 /* Avoid __ASSEMBLER__'ifying <linux/audit.h> just for this. */
63 #include <linux/elf-em.h>
64 #define AUDIT_ARCH_I386 (EM_386|__AUDIT_ARCH_LE)
65 #define __AUDIT_ARCH_LE 0x40000000
67 #ifndef CONFIG_AUDITSYSCALL
68 #define sysenter_audit syscall_trace_entry
69 #define sysexit_audit syscall_exit_work
72 .section .entry.text, "ax"
75 * We use macros for low-level operations which need to be overridden
76 * for paravirtualization. The following will never clobber any registers:
77 * INTERRUPT_RETURN (aka. "iret")
78 * GET_CR0_INTO_EAX (aka. "movl %cr0, %eax")
79 * ENABLE_INTERRUPTS_SYSEXIT (aka "sti; sysexit").
81 * For DISABLE_INTERRUPTS/ENABLE_INTERRUPTS (aka "cli"/"sti"), you must
82 * specify what registers can be overwritten (CLBR_NONE, CLBR_EAX/EDX/ECX/ANY).
83 * Allowing a register to be clobbered can shrink the paravirt replacement
84 * enough to patch inline, increasing performance.
88 #define preempt_stop(clobbers) DISABLE_INTERRUPTS(clobbers); TRACE_IRQS_OFF
90 #define preempt_stop(clobbers)
91 #define resume_kernel restore_all
94 .macro TRACE_IRQS_IRET
95 #ifdef CONFIG_TRACE_IRQFLAGS
96 testl $X86_EFLAGS_IF,PT_EFLAGS(%esp) # interrupts off?
104 * User gs save/restore
106 * %gs is used for userland TLS and kernel only uses it for stack
107 * canary which is required to be at %gs:20 by gcc. Read the comment
108 * at the top of stackprotector.h for more info.
110 * Local labels 98 and 99 are used.
112 #ifdef CONFIG_X86_32_LAZY_GS
114 /* unfortunately push/pop can't be no-op */
119 addl $(4 + \pop), %esp
120 CFI_ADJUST_CFA_OFFSET -(4 + \pop)
125 /* all the rest are no-op */
132 .macro REG_TO_PTGS reg
134 .macro SET_KERNEL_GS reg
137 #else /* CONFIG_X86_32_LAZY_GS */
141 /*CFI_REL_OFFSET gs, 0*/
149 CFI_ADJUST_CFA_OFFSET -\pop
153 .pushsection .fixup, "ax"
157 _ASM_EXTABLE(98b,99b)
161 98: mov PT_GS(%esp), %gs
164 .pushsection .fixup, "ax"
165 99: movl $0, PT_GS(%esp)
168 _ASM_EXTABLE(98b,99b)
173 /*CFI_REGISTER gs, \reg*/
175 .macro REG_TO_PTGS reg
176 movl \reg, PT_GS(%esp)
177 /*CFI_REL_OFFSET gs, PT_GS*/
179 .macro SET_KERNEL_GS reg
180 movl $(__KERNEL_STACK_CANARY), \reg
184 #endif /* CONFIG_X86_32_LAZY_GS */
190 /*CFI_REL_OFFSET fs, 0;*/
192 /*CFI_REL_OFFSET es, 0;*/
194 /*CFI_REL_OFFSET ds, 0;*/
196 CFI_REL_OFFSET eax, 0
198 CFI_REL_OFFSET ebp, 0
200 CFI_REL_OFFSET edi, 0
202 CFI_REL_OFFSET esi, 0
204 CFI_REL_OFFSET edx, 0
206 CFI_REL_OFFSET ecx, 0
208 CFI_REL_OFFSET ebx, 0
209 movl $(__USER_DS), %edx
212 movl $(__KERNEL_PERCPU), %edx
217 .macro RESTORE_INT_REGS
234 .macro RESTORE_REGS pop=0
243 .pushsection .fixup, "ax"
257 .macro RING0_INT_FRAME
261 /*CFI_OFFSET cs, -2*4;*/
265 .macro RING0_EC_FRAME
269 /*CFI_OFFSET cs, -2*4;*/
273 .macro RING0_PTREGS_FRAME
276 CFI_DEF_CFA esp, PT_OLDESP-PT_EBX
277 /*CFI_OFFSET cs, PT_CS-PT_OLDESP;*/
278 CFI_OFFSET eip, PT_EIP-PT_OLDESP
279 /*CFI_OFFSET es, PT_ES-PT_OLDESP;*/
280 /*CFI_OFFSET ds, PT_DS-PT_OLDESP;*/
281 CFI_OFFSET eax, PT_EAX-PT_OLDESP
282 CFI_OFFSET ebp, PT_EBP-PT_OLDESP
283 CFI_OFFSET edi, PT_EDI-PT_OLDESP
284 CFI_OFFSET esi, PT_ESI-PT_OLDESP
285 CFI_OFFSET edx, PT_EDX-PT_OLDESP
286 CFI_OFFSET ecx, PT_ECX-PT_OLDESP
287 CFI_OFFSET ebx, PT_EBX-PT_OLDESP
294 GET_THREAD_INFO(%ebp)
296 pushl_cfi $0x0202 # Reset kernel eflags
302 ENTRY(ret_from_kernel_thread)
306 GET_THREAD_INFO(%ebp)
308 pushl_cfi $0x0202 # Reset kernel eflags
310 movl PT_EBP(%esp),%eax
315 ENDPROC(ret_from_kernel_thread)
318 * Interrupt exit functions should be protected against kprobes
320 .pushsection .kprobes.text, "ax"
322 * Return to user mode is not as complex as all this looks,
323 * but we want the default path for a system call return to
324 * go as quickly as possible which is why some of this is
325 * less clear than it otherwise should be.
328 # userspace resumption stub bypassing syscall exit tracing
332 preempt_stop(CLBR_ANY)
334 GET_THREAD_INFO(%ebp)
336 movl PT_EFLAGS(%esp), %eax # mix EFLAGS and CS
337 movb PT_CS(%esp), %al
338 andl $(X86_EFLAGS_VM | SEGMENT_RPL_MASK), %eax
341 * We can be coming here from child spawned by kernel_thread().
343 movl PT_CS(%esp), %eax
344 andl $SEGMENT_RPL_MASK, %eax
347 jb resume_kernel # not returning to v8086 or userspace
349 ENTRY(resume_userspace)
351 DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt
352 # setting need_resched or sigpending
353 # between sampling and the iret
355 movl TI_flags(%ebp), %ecx
356 andl $_TIF_WORK_MASK, %ecx # is there any work to be done on
357 # int/exception return?
360 END(ret_from_exception)
362 #ifdef CONFIG_PREEMPT
364 DISABLE_INTERRUPTS(CLBR_ANY)
365 cmpl $0,TI_preempt_count(%ebp) # non-zero preempt_count ?
368 movl TI_flags(%ebp), %ecx # need_resched set ?
369 testb $_TIF_NEED_RESCHED, %cl
371 testl $X86_EFLAGS_IF,PT_EFLAGS(%esp) # interrupts off (exception path) ?
373 call preempt_schedule_irq
379 * End of kprobes section
383 /* SYSENTER_RETURN points to after the "sysenter" instruction in
384 the vsyscall page. See vsyscall-sysentry.S, which defines the symbol. */
386 # sysenter call handler stub
387 ENTRY(ia32_sysenter_target)
391 CFI_REGISTER esp, ebp
392 movl TSS_sysenter_sp0(%esp),%esp
395 * Interrupts are disabled here, but we can't trace it until
396 * enough kernel state to call TRACE_IRQS_OFF can be called - but
397 * we immediately enable interrupts at that point anyway.
400 /*CFI_REL_OFFSET ss, 0*/
402 CFI_REL_OFFSET esp, 0
404 orl $X86_EFLAGS_IF, (%esp)
406 /*CFI_REL_OFFSET cs, 0*/
408 * Push current_thread_info()->sysenter_return to the stack.
409 * A tiny bit of offset fixup is necessary - 4*4 means the 4 words
410 * pushed above; +8 corresponds to copy_thread's esp0 setting.
412 pushl_cfi ((TI_sysenter_return)-THREAD_SIZE+8+4*4)(%esp)
413 CFI_REL_OFFSET eip, 0
417 ENABLE_INTERRUPTS(CLBR_NONE)
420 * Load the potential sixth argument from user stack.
421 * Careful about security.
423 cmpl $__PAGE_OFFSET-3,%ebp
428 movl %ebp,PT_EBP(%esp)
429 _ASM_EXTABLE(1b,syscall_fault)
431 GET_THREAD_INFO(%ebp)
433 testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%ebp)
436 cmpl $(NR_syscalls), %eax
438 call *sys_call_table(,%eax,4)
440 movl %eax,PT_EAX(%esp)
442 DISABLE_INTERRUPTS(CLBR_ANY)
444 movl TI_flags(%ebp), %ecx
445 testl $_TIF_ALLWORK_MASK, %ecx
448 /* if something modifies registers it must also disable sysexit */
449 movl PT_EIP(%esp), %edx
450 movl PT_OLDESP(%esp), %ecx
453 1: mov PT_FS(%esp), %fs
455 ENABLE_INTERRUPTS_SYSEXIT
457 #ifdef CONFIG_AUDITSYSCALL
459 testl $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags(%ebp)
460 jnz syscall_trace_entry
462 CFI_ADJUST_CFA_OFFSET -4
463 /* %esi already in 8(%esp) 6th arg: 4th syscall arg */
464 /* %edx already in 4(%esp) 5th arg: 3rd syscall arg */
465 /* %ecx already in 0(%esp) 4th arg: 2nd syscall arg */
466 movl %ebx,%ecx /* 3rd arg: 1st syscall arg */
467 movl %eax,%edx /* 2nd arg: syscall number */
468 movl $AUDIT_ARCH_I386,%eax /* 1st arg: audit arch */
469 call __audit_syscall_entry
471 movl PT_EAX(%esp),%eax /* reload syscall number */
475 testl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT), %ecx
476 jne syscall_exit_work
478 ENABLE_INTERRUPTS(CLBR_ANY)
479 movl %eax,%edx /* second arg, syscall return value */
480 cmpl $-MAX_ERRNO,%eax /* is it an error ? */
481 setbe %al /* 1 if so, 0 if not */
482 movzbl %al,%eax /* zero-extend that */
483 call __audit_syscall_exit
484 DISABLE_INTERRUPTS(CLBR_ANY)
486 movl TI_flags(%ebp), %ecx
487 testl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT), %ecx
488 jne syscall_exit_work
489 movl PT_EAX(%esp),%eax /* reload syscall return value */
494 .pushsection .fixup,"ax"
495 2: movl $0,PT_FS(%esp)
500 ENDPROC(ia32_sysenter_target)
503 * syscall stub including irq exit should be protected against kprobes
505 .pushsection .kprobes.text, "ax"
506 # system call handler stub
508 RING0_INT_FRAME # can't unwind into user space anyway
510 pushl_cfi %eax # save orig_eax
512 GET_THREAD_INFO(%ebp)
513 # system call tracing in operation / emulation
514 testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%ebp)
515 jnz syscall_trace_entry
516 cmpl $(NR_syscalls), %eax
519 call *sys_call_table(,%eax,4)
521 movl %eax,PT_EAX(%esp) # store the return value
524 DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt
525 # setting need_resched or sigpending
526 # between sampling and the iret
528 movl TI_flags(%ebp), %ecx
529 testl $_TIF_ALLWORK_MASK, %ecx # current->work
530 jne syscall_exit_work
535 #ifdef CONFIG_X86_ESPFIX32
536 movl PT_EFLAGS(%esp), %eax # mix EFLAGS, SS and CS
537 # Warning: PT_OLDSS(%esp) contains the wrong/random values if we
538 # are returning to the kernel.
539 # See comments in process.c:copy_thread() for details.
540 movb PT_OLDSS(%esp), %ah
541 movb PT_CS(%esp), %al
542 andl $(X86_EFLAGS_VM | (SEGMENT_TI_MASK << 8) | SEGMENT_RPL_MASK), %eax
543 cmpl $((SEGMENT_LDT << 8) | USER_RPL), %eax
545 je ldt_ss # returning to user-space with LDT SS
548 RESTORE_REGS 4 # skip orig_eax/error_code
553 pushl $0 # no error code
557 _ASM_EXTABLE(irq_return,iret_exc)
559 #ifdef CONFIG_X86_ESPFIX32
562 #ifdef CONFIG_PARAVIRT
564 * The kernel can't run on a non-flat stack if paravirt mode
565 * is active. Rather than try to fixup the high bits of
566 * ESP, bypass this code entirely. This may break DOSemu
567 * and/or Wine support in a paravirt VM, although the option
568 * is still available to implement the setting of the high
569 * 16-bits in the INTERRUPT_RETURN paravirt-op.
571 cmpl $0, pv_info+PARAVIRT_enabled
576 * Setup and switch to ESPFIX stack
578 * We're returning to userspace with a 16 bit stack. The CPU will not
579 * restore the high word of ESP for us on executing iret... This is an
580 * "official" bug of all the x86-compatible CPUs, which we can work
581 * around to make dosemu and wine happy. We do this by preloading the
582 * high word of ESP with the high word of the userspace ESP while
583 * compensating for the offset by changing to the ESPFIX segment with
584 * a base address that matches for the difference.
586 #define GDT_ESPFIX_SS PER_CPU_VAR(gdt_page) + (GDT_ENTRY_ESPFIX_SS * 8)
587 mov %esp, %edx /* load kernel esp */
588 mov PT_OLDESP(%esp), %eax /* load userspace esp */
589 mov %dx, %ax /* eax: new kernel esp */
590 sub %eax, %edx /* offset (low word is 0) */
592 mov %dl, GDT_ESPFIX_SS + 4 /* bits 16..23 */
593 mov %dh, GDT_ESPFIX_SS + 7 /* bits 24..31 */
594 pushl_cfi $__ESPFIX_SS
595 pushl_cfi %eax /* new kernel esp */
596 /* Disable interrupts, but do not irqtrace this section: we
597 * will soon execute iret and the tracer was already set to
598 * the irqstate after the iret */
599 DISABLE_INTERRUPTS(CLBR_EAX)
600 lss (%esp), %esp /* switch to espfix segment */
601 CFI_ADJUST_CFA_OFFSET -8
607 # perform work that needs to be done immediately before resumption
609 RING0_PTREGS_FRAME # can't unwind into user space anyway
611 testb $_TIF_NEED_RESCHED, %cl
616 DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt
617 # setting need_resched or sigpending
618 # between sampling and the iret
620 movl TI_flags(%ebp), %ecx
621 andl $_TIF_WORK_MASK, %ecx # is there any work to be done other
622 # than syscall tracing?
624 testb $_TIF_NEED_RESCHED, %cl
627 work_notifysig: # deal with pending signals and
628 # notify-resume requests
630 testl $X86_EFLAGS_VM, PT_EFLAGS(%esp)
632 jne work_notifysig_v86 # returning to kernel-space or
639 ENABLE_INTERRUPTS(CLBR_NONE)
640 movb PT_CS(%esp), %bl
641 andb $SEGMENT_RPL_MASK, %bl
645 call do_notify_resume
651 pushl_cfi %ecx # save ti_flags for do_notify_resume
652 call save_v86_state # %eax contains pt_regs pointer
659 # perform syscall exit tracing
662 movl $-ENOSYS,PT_EAX(%esp)
664 call syscall_trace_enter
665 /* What it returned is what we'll actually use. */
666 cmpl $(NR_syscalls), %eax
669 END(syscall_trace_entry)
671 # perform syscall exit tracing
674 testl $_TIF_WORK_SYSCALL_EXIT, %ecx
677 ENABLE_INTERRUPTS(CLBR_ANY) # could let syscall_trace_leave() call
680 call syscall_trace_leave
682 END(syscall_exit_work)
685 RING0_INT_FRAME # can't unwind into user space anyway
688 GET_THREAD_INFO(%ebp)
689 movl $-EFAULT,PT_EAX(%esp)
695 jmp syscall_after_call
700 jmp sysenter_after_call
704 * End of kprobes section
708 .macro FIXUP_ESPFIX_STACK
710 * Switch back for ESPFIX stack to the normal zerobased stack
712 * We can't call C functions using the ESPFIX stack. This code reads
713 * the high word of the segment base from the GDT and swiches to the
714 * normal stack and adjusts ESP with the matching offset.
716 #ifdef CONFIG_X86_ESPFIX32
717 /* fixup the stack */
718 mov GDT_ESPFIX_SS + 4, %al /* bits 16..23 */
719 mov GDT_ESPFIX_SS + 7, %ah /* bits 24..31 */
721 addl %esp, %eax /* the adjusted stack pointer */
722 pushl_cfi $__KERNEL_DS
724 lss (%esp), %esp /* switch to the normal stack segment */
725 CFI_ADJUST_CFA_OFFSET -8
728 .macro UNWIND_ESPFIX_STACK
729 #ifdef CONFIG_X86_ESPFIX32
731 /* see if on espfix stack */
732 cmpw $__ESPFIX_SS, %ax
734 movl $__KERNEL_DS, %eax
737 /* switch to normal stack */
744 * Build the entry stubs and pointer table with some assembler magic.
745 * We pack 7 stubs into a single 32-byte chunk, which will fit in a
746 * single cache line on all modern x86 implementations.
748 .section .init.rodata,"a"
750 .section .entry.text, "ax"
752 .p2align CONFIG_X86_L1_CACHE_SHIFT
753 ENTRY(irq_entries_start)
755 vector=FIRST_EXTERNAL_VECTOR
756 .rept (NR_VECTORS-FIRST_EXTERNAL_VECTOR+6)/7
759 .if vector < NR_VECTORS
760 .if vector <> FIRST_EXTERNAL_VECTOR
761 CFI_ADJUST_CFA_OFFSET -4
763 1: pushl_cfi $(~vector+0x80) /* Note: always in signed byte range */
764 .if ((vector-FIRST_EXTERNAL_VECTOR)%7) <> 6
769 .section .entry.text, "ax"
773 2: jmp common_interrupt
775 END(irq_entries_start)
782 * the CPU automatically disables interrupts when executing an IRQ vector,
783 * so IRQ-flags tracing has to follow that:
785 .p2align CONFIG_X86_L1_CACHE_SHIFT
788 addl $-0x80,(%esp) /* Adjust vector into the [-256,-1] range */
794 ENDPROC(common_interrupt)
798 * Irq entries should be protected against kprobes
800 .pushsection .kprobes.text, "ax"
801 #define BUILD_INTERRUPT3(name, nr, fn) \
815 #ifdef CONFIG_TRACING
816 #define TRACE_BUILD_INTERRUPT(name, nr) \
817 BUILD_INTERRUPT3(trace_##name, nr, smp_trace_##name)
819 #define TRACE_BUILD_INTERRUPT(name, nr)
822 #define BUILD_INTERRUPT(name, nr) \
823 BUILD_INTERRUPT3(name, nr, smp_##name); \
824 TRACE_BUILD_INTERRUPT(name, nr)
826 /* The include is where all of the SMP etc. interrupts come from */
827 #include <asm/entry_arch.h>
829 ENTRY(coprocessor_error)
833 pushl_cfi $do_coprocessor_error
836 END(coprocessor_error)
838 ENTRY(simd_coprocessor_error)
842 #ifdef CONFIG_X86_INVD_BUG
843 /* AMD 486 bug: invd from userspace calls exception 19 instead of #GP */
844 661: pushl_cfi $do_general_protection
846 .section .altinstructions,"a"
847 altinstruction_entry 661b, 663f, X86_FEATURE_XMM, 662b-661b, 664f-663f
849 .section .altinstr_replacement,"ax"
850 663: pushl $do_simd_coprocessor_error
854 pushl_cfi $do_simd_coprocessor_error
858 END(simd_coprocessor_error)
860 ENTRY(device_not_available)
863 pushl_cfi $-1 # mark this as an int
864 pushl_cfi $do_device_not_available
867 END(device_not_available)
869 #ifdef CONFIG_PARAVIRT
872 _ASM_EXTABLE(native_iret, iret_exc)
875 ENTRY(native_irq_enable_sysexit)
878 END(native_irq_enable_sysexit)
885 pushl_cfi $do_overflow
903 pushl_cfi $do_invalid_op
908 ENTRY(coprocessor_segment_overrun)
912 pushl_cfi $do_coprocessor_segment_overrun
915 END(coprocessor_segment_overrun)
920 pushl_cfi $do_invalid_TSS
925 ENTRY(segment_not_present)
928 pushl_cfi $do_segment_not_present
931 END(segment_not_present)
936 pushl_cfi $do_stack_segment
941 ENTRY(alignment_check)
944 pushl_cfi $do_alignment_check
952 pushl_cfi $0 # no error code
953 pushl_cfi $do_divide_error
958 #ifdef CONFIG_X86_MCE
963 pushl_cfi machine_check_vector
969 ENTRY(spurious_interrupt_bug)
973 pushl_cfi $do_spurious_interrupt_bug
976 END(spurious_interrupt_bug)
978 * End of kprobes section
983 /* Xen doesn't set %esp to be precisely what the normal sysenter
984 entrypoint expects, so fix it up before using the normal path. */
985 ENTRY(xen_sysenter_target)
987 addl $5*4, %esp /* remove xen-provided frame */
988 CFI_ADJUST_CFA_OFFSET -5*4
989 jmp sysenter_past_esp
992 ENTRY(xen_hypervisor_callback)
994 pushl_cfi $-1 /* orig_ax = -1 => not a system call */
998 /* Check to see if we got the event in the critical
999 region in xen_iret_direct, after we've reenabled
1000 events and checked for pending events. This simulates
1001 iret instruction's behaviour where it delivers a
1002 pending interrupt when enabling interrupts. */
1003 movl PT_EIP(%esp),%eax
1004 cmpl $xen_iret_start_crit,%eax
1006 cmpl $xen_iret_end_crit,%eax
1009 jmp xen_iret_crit_fixup
1011 ENTRY(xen_do_upcall)
1013 call xen_evtchn_do_upcall
1016 ENDPROC(xen_hypervisor_callback)
1018 # Hypervisor uses this for application faults while it executes.
1019 # We get here for two reasons:
1020 # 1. Fault while reloading DS, ES, FS or GS
1021 # 2. Fault while executing IRET
1022 # Category 1 we fix up by reattempting the load, and zeroing the segment
1023 # register if the load fails.
1024 # Category 2 we fix up by jumping to do_iret_error. We cannot use the
1025 # normal Linux return path in this case because if we use the IRET hypercall
1026 # to pop the stack frame we end up in an infinite loop of failsafe callbacks.
1027 # We distinguish between categories by maintaining a status value in EAX.
1028 ENTRY(xen_failsafe_callback)
1036 /* EAX == 0 => Category 1 (Bad segment)
1037 EAX != 0 => Category 2 (Bad IRET) */
1041 CFI_ADJUST_CFA_OFFSET -16
1044 5: pushl_cfi $-1 /* orig_ax = -1 => not a system call */
1046 jmp ret_from_exception
1049 .section .fixup,"ax"
1067 ENDPROC(xen_failsafe_callback)
1069 BUILD_INTERRUPT3(xen_hvm_callback_vector, HYPERVISOR_CALLBACK_VECTOR,
1070 xen_evtchn_do_upcall)
1072 #endif /* CONFIG_XEN */
1074 #if IS_ENABLED(CONFIG_HYPERV)
1076 BUILD_INTERRUPT3(hyperv_callback_vector, HYPERVISOR_CALLBACK_VECTOR,
1077 hyperv_vector_handler)
1079 #endif /* CONFIG_HYPERV */
1081 #ifdef CONFIG_FUNCTION_TRACER
1082 #ifdef CONFIG_DYNAMIC_FTRACE
1088 ENTRY(ftrace_caller)
1089 cmpl $0, function_trace_stop
1095 pushl $0 /* Pass NULL as regs pointer */
1096 movl 4*4(%esp), %eax
1097 movl 0x4(%ebp), %edx
1098 movl function_trace_op, %ecx
1099 subl $MCOUNT_INSN_SIZE, %eax
1105 addl $4,%esp /* skip NULL pointer */
1110 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1111 .globl ftrace_graph_call
1121 ENTRY(ftrace_regs_caller)
1122 pushf /* push flags before compare (in cs location) */
1123 cmpl $0, function_trace_stop
1124 jne ftrace_restore_flags
1127 * i386 does not save SS and ESP when coming from kernel.
1128 * Instead, to get sp, ®s->sp is used (see ptrace.h).
1129 * Unfortunately, that means eflags must be at the same location
1130 * as the current return ip is. We move the return ip into the
1131 * ip location, and move flags into the return ip location.
1133 pushl 4(%esp) /* save return ip into ip slot */
1135 pushl $0 /* Load 0 into orig_ax */
1148 movl 13*4(%esp), %eax /* Get the saved flags */
1149 movl %eax, 14*4(%esp) /* Move saved flags into regs->flags location */
1150 /* clobbering return ip */
1151 movl $__KERNEL_CS,13*4(%esp)
1153 movl 12*4(%esp), %eax /* Load ip (1st parameter) */
1154 subl $MCOUNT_INSN_SIZE, %eax /* Adjust ip */
1155 movl 0x4(%ebp), %edx /* Load parent ip (2nd parameter) */
1156 movl function_trace_op, %ecx /* Save ftrace_pos in 3rd parameter */
1157 pushl %esp /* Save pt_regs as 4th parameter */
1159 GLOBAL(ftrace_regs_call)
1162 addl $4, %esp /* Skip pt_regs */
1163 movl 14*4(%esp), %eax /* Move flags back into cs */
1164 movl %eax, 13*4(%esp) /* Needed to keep addl from modifying flags */
1165 movl 12*4(%esp), %eax /* Get return ip from regs->ip */
1166 movl %eax, 14*4(%esp) /* Put return ip back for ret */
1179 addl $8, %esp /* Skip orig_ax and ip */
1180 popf /* Pop flags at end (no addl to corrupt flags) */
1183 ftrace_restore_flags:
1186 #else /* ! CONFIG_DYNAMIC_FTRACE */
1189 cmpl $__PAGE_OFFSET, %esp
1190 jb ftrace_stub /* Paging not enabled yet? */
1192 cmpl $0, function_trace_stop
1195 cmpl $ftrace_stub, ftrace_trace_function
1197 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1198 cmpl $ftrace_stub, ftrace_graph_return
1199 jnz ftrace_graph_caller
1201 cmpl $ftrace_graph_entry_stub, ftrace_graph_entry
1202 jnz ftrace_graph_caller
1208 /* taken from glibc */
1213 movl 0xc(%esp), %eax
1214 movl 0x4(%ebp), %edx
1215 subl $MCOUNT_INSN_SIZE, %eax
1217 call *ftrace_trace_function
1224 #endif /* CONFIG_DYNAMIC_FTRACE */
1225 #endif /* CONFIG_FUNCTION_TRACER */
1227 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1228 ENTRY(ftrace_graph_caller)
1232 movl 0xc(%esp), %edx
1235 subl $MCOUNT_INSN_SIZE, %edx
1236 call prepare_ftrace_return
1241 END(ftrace_graph_caller)
1243 .globl return_to_handler
1248 call ftrace_return_to_handler
1256 * Some functions should be protected against kprobes
1258 .pushsection .kprobes.text, "ax"
1263 pushl_cfi $do_page_fault
1266 /* the function address is in %gs's slot on the stack */
1268 /*CFI_REL_OFFSET fs, 0*/
1270 /*CFI_REL_OFFSET es, 0*/
1272 /*CFI_REL_OFFSET ds, 0*/
1274 CFI_REL_OFFSET eax, 0
1276 CFI_REL_OFFSET ebp, 0
1278 CFI_REL_OFFSET edi, 0
1280 CFI_REL_OFFSET esi, 0
1282 CFI_REL_OFFSET edx, 0
1284 CFI_REL_OFFSET ecx, 0
1286 CFI_REL_OFFSET ebx, 0
1288 movl $(__KERNEL_PERCPU), %ecx
1292 movl PT_GS(%esp), %edi # get the function address
1293 movl PT_ORIG_EAX(%esp), %edx # get the error code
1294 movl $-1, PT_ORIG_EAX(%esp) # no syscall to restart
1297 movl $(__USER_DS), %ecx
1301 movl %esp,%eax # pt_regs pointer
1303 jmp ret_from_exception
1308 * Debug traps and NMI can happen at the one SYSENTER instruction
1309 * that sets up the real kernel stack. Check here, since we can't
1310 * allow the wrong stack to be used.
1312 * "TSS_sysenter_sp0+12" is because the NMI/debug handler will have
1313 * already pushed 3 words if it hits on the sysenter instruction:
1314 * eflags, cs and eip.
1316 * We just load the right stack, and push the three (known) values
1317 * by hand onto the new stack - while updating the return eip past
1318 * the instruction that would have done it for sysenter.
1320 .macro FIX_STACK offset ok label
1321 cmpw $__KERNEL_CS, 4(%esp)
1324 movl TSS_sysenter_sp0 + \offset(%esp), %esp
1328 pushl_cfi $__KERNEL_CS
1329 pushl_cfi $sysenter_past_esp
1330 CFI_REL_OFFSET eip, 0
1336 cmpl $ia32_sysenter_target,(%esp)
1337 jne debug_stack_correct
1338 FIX_STACK 12, debug_stack_correct, debug_esp_fix_insn
1339 debug_stack_correct:
1340 pushl_cfi $-1 # mark this as an int
1343 xorl %edx,%edx # error code 0
1344 movl %esp,%eax # pt_regs pointer
1346 jmp ret_from_exception
1351 * NMI is doubly nasty. It can happen _while_ we're handling
1352 * a debug fault, and the debug fault hasn't yet been able to
1353 * clear up the stack. So we first check whether we got an
1354 * NMI on the sysenter entry path, but after that we need to
1355 * check whether we got an NMI on the debug path where the debug
1356 * fault happened on the sysenter path.
1361 #ifdef CONFIG_X86_ESPFIX32
1364 cmpw $__ESPFIX_SS, %ax
1368 cmpl $ia32_sysenter_target,(%esp)
1372 /* Do not access memory above the end of our stack page,
1373 * it might not exist.
1375 andl $(THREAD_SIZE-1),%eax
1376 cmpl $(THREAD_SIZE-20),%eax
1378 jae nmi_stack_correct
1379 cmpl $ia32_sysenter_target,12(%esp)
1380 je nmi_debug_stack_check
1382 /* We have a RING0_INT_FRAME here */
1385 xorl %edx,%edx # zero error code
1386 movl %esp,%eax # pt_regs pointer
1388 jmp restore_all_notrace
1393 FIX_STACK 12, nmi_stack_correct, 1
1394 jmp nmi_stack_correct
1396 nmi_debug_stack_check:
1397 /* We have a RING0_INT_FRAME here */
1398 cmpw $__KERNEL_CS,16(%esp)
1399 jne nmi_stack_correct
1401 jb nmi_stack_correct
1402 cmpl $debug_esp_fix_insn,(%esp)
1403 ja nmi_stack_correct
1404 FIX_STACK 24, nmi_stack_correct, 1
1405 jmp nmi_stack_correct
1407 #ifdef CONFIG_X86_ESPFIX32
1409 /* We have a RING0_INT_FRAME here.
1411 * create the pointer to lss back
1416 /* copy the iret frame of 12 bytes */
1422 FIXUP_ESPFIX_STACK # %eax == %esp
1423 xorl %edx,%edx # zero error code
1426 lss 12+4(%esp), %esp # back to espfix stack
1427 CFI_ADJUST_CFA_OFFSET -24
1436 pushl_cfi $-1 # mark this as an int
1439 xorl %edx,%edx # zero error code
1440 movl %esp,%eax # pt_regs pointer
1442 jmp ret_from_exception
1446 ENTRY(general_protection)
1448 pushl_cfi $do_general_protection
1451 END(general_protection)
1453 #ifdef CONFIG_KVM_GUEST
1454 ENTRY(async_page_fault)
1457 pushl_cfi $do_async_page_fault
1460 END(async_page_fault)
1464 * End of kprobes section