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>
61 /* Avoid __ASSEMBLER__'ifying <linux/audit.h> just for this. */
62 #include <linux/elf-em.h>
63 #define AUDIT_ARCH_I386 (EM_386|__AUDIT_ARCH_LE)
64 #define __AUDIT_ARCH_LE 0x40000000
66 #ifndef CONFIG_AUDITSYSCALL
67 #define sysenter_audit syscall_trace_entry
68 #define sysexit_audit syscall_exit_work
71 .section .entry.text, "ax"
74 * We use macros for low-level operations which need to be overridden
75 * for paravirtualization. The following will never clobber any registers:
76 * INTERRUPT_RETURN (aka. "iret")
77 * GET_CR0_INTO_EAX (aka. "movl %cr0, %eax")
78 * ENABLE_INTERRUPTS_SYSEXIT (aka "sti; sysexit").
80 * For DISABLE_INTERRUPTS/ENABLE_INTERRUPTS (aka "cli"/"sti"), you must
81 * specify what registers can be overwritten (CLBR_NONE, CLBR_EAX/EDX/ECX/ANY).
82 * Allowing a register to be clobbered can shrink the paravirt replacement
83 * enough to patch inline, increasing performance.
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 * User gs save/restore
105 * %gs is used for userland TLS and kernel only uses it for stack
106 * canary which is required to be at %gs:20 by gcc. Read the comment
107 * at the top of stackprotector.h for more info.
109 * Local labels 98 and 99 are used.
111 #ifdef CONFIG_X86_32_LAZY_GS
113 /* unfortunately push/pop can't be no-op */
118 addl $(4 + \pop), %esp
119 CFI_ADJUST_CFA_OFFSET -(4 + \pop)
124 /* all the rest are no-op */
131 .macro REG_TO_PTGS reg
133 .macro SET_KERNEL_GS reg
136 #else /* CONFIG_X86_32_LAZY_GS */
140 /*CFI_REL_OFFSET gs, 0*/
148 CFI_ADJUST_CFA_OFFSET -\pop
152 .pushsection .fixup, "ax"
156 _ASM_EXTABLE(98b,99b)
160 98: mov PT_GS(%esp), %gs
163 .pushsection .fixup, "ax"
164 99: movl $0, PT_GS(%esp)
167 _ASM_EXTABLE(98b,99b)
172 /*CFI_REGISTER gs, \reg*/
174 .macro REG_TO_PTGS reg
175 movl \reg, PT_GS(%esp)
176 /*CFI_REL_OFFSET gs, PT_GS*/
178 .macro SET_KERNEL_GS reg
179 movl $(__KERNEL_STACK_CANARY), \reg
183 #endif /* CONFIG_X86_32_LAZY_GS */
189 /*CFI_REL_OFFSET fs, 0;*/
191 /*CFI_REL_OFFSET es, 0;*/
193 /*CFI_REL_OFFSET ds, 0;*/
195 CFI_REL_OFFSET eax, 0
197 CFI_REL_OFFSET ebp, 0
199 CFI_REL_OFFSET edi, 0
201 CFI_REL_OFFSET esi, 0
203 CFI_REL_OFFSET edx, 0
205 CFI_REL_OFFSET ecx, 0
207 CFI_REL_OFFSET ebx, 0
208 movl $(__USER_DS), %edx
211 movl $(__KERNEL_PERCPU), %edx
216 .macro RESTORE_INT_REGS
233 .macro RESTORE_REGS pop=0
242 .pushsection .fixup, "ax"
256 .macro RING0_INT_FRAME
260 /*CFI_OFFSET cs, -2*4;*/
264 .macro RING0_EC_FRAME
268 /*CFI_OFFSET cs, -2*4;*/
272 .macro RING0_PTREGS_FRAME
275 CFI_DEF_CFA esp, PT_OLDESP-PT_EBX
276 /*CFI_OFFSET cs, PT_CS-PT_OLDESP;*/
277 CFI_OFFSET eip, PT_EIP-PT_OLDESP
278 /*CFI_OFFSET es, PT_ES-PT_OLDESP;*/
279 /*CFI_OFFSET ds, PT_DS-PT_OLDESP;*/
280 CFI_OFFSET eax, PT_EAX-PT_OLDESP
281 CFI_OFFSET ebp, PT_EBP-PT_OLDESP
282 CFI_OFFSET edi, PT_EDI-PT_OLDESP
283 CFI_OFFSET esi, PT_ESI-PT_OLDESP
284 CFI_OFFSET edx, PT_EDX-PT_OLDESP
285 CFI_OFFSET ecx, PT_ECX-PT_OLDESP
286 CFI_OFFSET ebx, PT_EBX-PT_OLDESP
293 GET_THREAD_INFO(%ebp)
295 pushl_cfi $0x0202 # Reset kernel eflags
302 * Interrupt exit functions should be protected against kprobes
304 .pushsection .kprobes.text, "ax"
306 * Return to user mode is not as complex as all this looks,
307 * but we want the default path for a system call return to
308 * go as quickly as possible which is why some of this is
309 * less clear than it otherwise should be.
312 # userspace resumption stub bypassing syscall exit tracing
316 preempt_stop(CLBR_ANY)
318 GET_THREAD_INFO(%ebp)
319 resume_userspace_sig:
321 movl PT_EFLAGS(%esp), %eax # mix EFLAGS and CS
322 movb PT_CS(%esp), %al
323 andl $(X86_EFLAGS_VM | SEGMENT_RPL_MASK), %eax
326 * We can be coming here from a syscall done in the kernel space,
327 * e.g. a failed kernel_execve().
329 movl PT_CS(%esp), %eax
330 andl $SEGMENT_RPL_MASK, %eax
333 jb resume_kernel # not returning to v8086 or userspace
335 ENTRY(resume_userspace)
337 DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt
338 # setting need_resched or sigpending
339 # between sampling and the iret
341 movl TI_flags(%ebp), %ecx
342 andl $_TIF_WORK_MASK, %ecx # is there any work to be done on
343 # int/exception return?
346 END(ret_from_exception)
348 #ifdef CONFIG_PREEMPT
350 DISABLE_INTERRUPTS(CLBR_ANY)
351 cmpl $0,TI_preempt_count(%ebp) # non-zero preempt_count ?
354 movl TI_flags(%ebp), %ecx # need_resched set ?
355 testb $_TIF_NEED_RESCHED, %cl
357 testl $X86_EFLAGS_IF,PT_EFLAGS(%esp) # interrupts off (exception path) ?
359 call preempt_schedule_irq
365 * End of kprobes section
369 /* SYSENTER_RETURN points to after the "sysenter" instruction in
370 the vsyscall page. See vsyscall-sysentry.S, which defines the symbol. */
372 # sysenter call handler stub
373 ENTRY(ia32_sysenter_target)
377 CFI_REGISTER esp, ebp
378 movl TSS_sysenter_sp0(%esp),%esp
381 * Interrupts are disabled here, but we can't trace it until
382 * enough kernel state to call TRACE_IRQS_OFF can be called - but
383 * we immediately enable interrupts at that point anyway.
386 /*CFI_REL_OFFSET ss, 0*/
388 CFI_REL_OFFSET esp, 0
390 orl $X86_EFLAGS_IF, (%esp)
392 /*CFI_REL_OFFSET cs, 0*/
394 * Push current_thread_info()->sysenter_return to the stack.
395 * A tiny bit of offset fixup is necessary - 4*4 means the 4 words
396 * pushed above; +8 corresponds to copy_thread's esp0 setting.
398 pushl_cfi ((TI_sysenter_return)-THREAD_SIZE+8+4*4)(%esp)
399 CFI_REL_OFFSET eip, 0
403 ENABLE_INTERRUPTS(CLBR_NONE)
406 * Load the potential sixth argument from user stack.
407 * Careful about security.
409 cmpl $__PAGE_OFFSET-3,%ebp
412 movl %ebp,PT_EBP(%esp)
413 _ASM_EXTABLE(1b,syscall_fault)
415 GET_THREAD_INFO(%ebp)
417 testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%ebp)
420 cmpl $(NR_syscalls), %eax
422 call *sys_call_table(,%eax,4)
423 movl %eax,PT_EAX(%esp)
425 DISABLE_INTERRUPTS(CLBR_ANY)
427 movl TI_flags(%ebp), %ecx
428 testl $_TIF_ALLWORK_MASK, %ecx
431 /* if something modifies registers it must also disable sysexit */
432 movl PT_EIP(%esp), %edx
433 movl PT_OLDESP(%esp), %ecx
436 1: mov PT_FS(%esp), %fs
438 ENABLE_INTERRUPTS_SYSEXIT
440 #ifdef CONFIG_AUDITSYSCALL
442 testl $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags(%ebp)
443 jnz syscall_trace_entry
445 CFI_ADJUST_CFA_OFFSET -4
446 /* %esi already in 8(%esp) 6th arg: 4th syscall arg */
447 /* %edx already in 4(%esp) 5th arg: 3rd syscall arg */
448 /* %ecx already in 0(%esp) 4th arg: 2nd syscall arg */
449 movl %ebx,%ecx /* 3rd arg: 1st syscall arg */
450 movl %eax,%edx /* 2nd arg: syscall number */
451 movl $AUDIT_ARCH_I386,%eax /* 1st arg: audit arch */
452 call __audit_syscall_entry
454 movl PT_EAX(%esp),%eax /* reload syscall number */
458 testl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT), %ecx
459 jne syscall_exit_work
461 ENABLE_INTERRUPTS(CLBR_ANY)
462 movl %eax,%edx /* second arg, syscall return value */
463 cmpl $-MAX_ERRNO,%eax /* is it an error ? */
464 setbe %al /* 1 if so, 0 if not */
465 movzbl %al,%eax /* zero-extend that */
466 call __audit_syscall_exit
467 DISABLE_INTERRUPTS(CLBR_ANY)
469 movl TI_flags(%ebp), %ecx
470 testl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT), %ecx
471 jne syscall_exit_work
472 movl PT_EAX(%esp),%eax /* reload syscall return value */
477 .pushsection .fixup,"ax"
478 2: movl $0,PT_FS(%esp)
483 ENDPROC(ia32_sysenter_target)
486 * syscall stub including irq exit should be protected against kprobes
488 .pushsection .kprobes.text, "ax"
489 # system call handler stub
491 RING0_INT_FRAME # can't unwind into user space anyway
492 pushl_cfi %eax # save orig_eax
494 GET_THREAD_INFO(%ebp)
495 # system call tracing in operation / emulation
496 testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%ebp)
497 jnz syscall_trace_entry
498 cmpl $(NR_syscalls), %eax
501 call *sys_call_table(,%eax,4)
502 movl %eax,PT_EAX(%esp) # store the return value
505 DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt
506 # setting need_resched or sigpending
507 # between sampling and the iret
509 movl TI_flags(%ebp), %ecx
510 testl $_TIF_ALLWORK_MASK, %ecx # current->work
511 jne syscall_exit_work
516 movl PT_EFLAGS(%esp), %eax # mix EFLAGS, SS and CS
517 # Warning: PT_OLDSS(%esp) contains the wrong/random values if we
518 # are returning to the kernel.
519 # See comments in process.c:copy_thread() for details.
520 movb PT_OLDSS(%esp), %ah
521 movb PT_CS(%esp), %al
522 andl $(X86_EFLAGS_VM | (SEGMENT_TI_MASK << 8) | SEGMENT_RPL_MASK), %eax
523 cmpl $((SEGMENT_LDT << 8) | USER_RPL), %eax
525 je ldt_ss # returning to user-space with LDT SS
527 RESTORE_REGS 4 # skip orig_eax/error_code
532 pushl $0 # no error code
536 _ASM_EXTABLE(irq_return,iret_exc)
540 larl PT_OLDSS(%esp), %eax
542 testl $0x00400000, %eax # returning to 32bit stack?
543 jnz restore_nocheck # allright, normal return
545 #ifdef CONFIG_PARAVIRT
547 * The kernel can't run on a non-flat stack if paravirt mode
548 * is active. Rather than try to fixup the high bits of
549 * ESP, bypass this code entirely. This may break DOSemu
550 * and/or Wine support in a paravirt VM, although the option
551 * is still available to implement the setting of the high
552 * 16-bits in the INTERRUPT_RETURN paravirt-op.
554 cmpl $0, pv_info+PARAVIRT_enabled
559 * Setup and switch to ESPFIX stack
561 * We're returning to userspace with a 16 bit stack. The CPU will not
562 * restore the high word of ESP for us on executing iret... This is an
563 * "official" bug of all the x86-compatible CPUs, which we can work
564 * around to make dosemu and wine happy. We do this by preloading the
565 * high word of ESP with the high word of the userspace ESP while
566 * compensating for the offset by changing to the ESPFIX segment with
567 * a base address that matches for the difference.
569 #define GDT_ESPFIX_SS PER_CPU_VAR(gdt_page) + (GDT_ENTRY_ESPFIX_SS * 8)
570 mov %esp, %edx /* load kernel esp */
571 mov PT_OLDESP(%esp), %eax /* load userspace esp */
572 mov %dx, %ax /* eax: new kernel esp */
573 sub %eax, %edx /* offset (low word is 0) */
575 mov %dl, GDT_ESPFIX_SS + 4 /* bits 16..23 */
576 mov %dh, GDT_ESPFIX_SS + 7 /* bits 24..31 */
577 pushl_cfi $__ESPFIX_SS
578 pushl_cfi %eax /* new kernel esp */
579 /* Disable interrupts, but do not irqtrace this section: we
580 * will soon execute iret and the tracer was already set to
581 * the irqstate after the iret */
582 DISABLE_INTERRUPTS(CLBR_EAX)
583 lss (%esp), %esp /* switch to espfix segment */
584 CFI_ADJUST_CFA_OFFSET -8
589 # perform work that needs to be done immediately before resumption
591 RING0_PTREGS_FRAME # can't unwind into user space anyway
593 testb $_TIF_NEED_RESCHED, %cl
598 DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt
599 # setting need_resched or sigpending
600 # between sampling and the iret
602 movl TI_flags(%ebp), %ecx
603 andl $_TIF_WORK_MASK, %ecx # is there any work to be done other
604 # than syscall tracing?
606 testb $_TIF_NEED_RESCHED, %cl
609 work_notifysig: # deal with pending signals and
610 # notify-resume requests
612 testl $X86_EFLAGS_VM, PT_EFLAGS(%esp)
614 jne work_notifysig_v86 # returning to kernel-space or
617 ENABLE_INTERRUPTS(CLBR_NONE)
619 call do_notify_resume
620 jmp resume_userspace_sig
624 pushl_cfi %ecx # save ti_flags for do_notify_resume
625 call save_v86_state # %eax contains pt_regs pointer
632 ENABLE_INTERRUPTS(CLBR_NONE)
634 call do_notify_resume
635 jmp resume_userspace_sig
638 # perform syscall exit tracing
641 movl $-ENOSYS,PT_EAX(%esp)
643 call syscall_trace_enter
644 /* What it returned is what we'll actually use. */
645 cmpl $(NR_syscalls), %eax
648 END(syscall_trace_entry)
650 # perform syscall exit tracing
653 testl $_TIF_WORK_SYSCALL_EXIT, %ecx
656 ENABLE_INTERRUPTS(CLBR_ANY) # could let syscall_trace_leave() call
659 call syscall_trace_leave
661 END(syscall_exit_work)
664 RING0_INT_FRAME # can't unwind into user space anyway
666 GET_THREAD_INFO(%ebp)
667 movl $-EFAULT,PT_EAX(%esp)
672 movl $-ENOSYS,PT_EAX(%esp)
677 * End of kprobes section
682 * System calls that need a pt_regs pointer.
684 #define PTREGSCALL0(name) \
685 ENTRY(ptregs_##name) ; \
688 ENDPROC(ptregs_##name)
690 #define PTREGSCALL1(name) \
691 ENTRY(ptregs_##name) ; \
693 movl (PT_EBX+4)(%esp),%eax; \
695 ENDPROC(ptregs_##name)
697 #define PTREGSCALL2(name) \
698 ENTRY(ptregs_##name) ; \
700 movl (PT_ECX+4)(%esp),%edx; \
701 movl (PT_EBX+4)(%esp),%eax; \
703 ENDPROC(ptregs_##name)
705 #define PTREGSCALL3(name) \
706 ENTRY(ptregs_##name) ; \
710 movl PT_EDX(%eax),%ecx; \
711 movl PT_ECX(%eax),%edx; \
712 movl PT_EBX(%eax),%eax; \
715 CFI_ADJUST_CFA_OFFSET -4; \
718 ENDPROC(ptregs_##name)
724 PTREGSCALL2(sigaltstack)
725 PTREGSCALL0(sigreturn)
726 PTREGSCALL0(rt_sigreturn)
730 /* Clone is an oddball. The 4th arg is in %edi */
735 pushl_cfi PT_EDI(%eax)
736 movl PT_EDX(%eax),%ecx
737 movl PT_ECX(%eax),%edx
738 movl PT_EBX(%eax),%eax
741 CFI_ADJUST_CFA_OFFSET -8
744 ENDPROC(ptregs_clone)
746 .macro FIXUP_ESPFIX_STACK
748 * Switch back for ESPFIX stack to the normal zerobased stack
750 * We can't call C functions using the ESPFIX stack. This code reads
751 * the high word of the segment base from the GDT and swiches to the
752 * normal stack and adjusts ESP with the matching offset.
754 /* fixup the stack */
755 mov GDT_ESPFIX_SS + 4, %al /* bits 16..23 */
756 mov GDT_ESPFIX_SS + 7, %ah /* bits 24..31 */
758 addl %esp, %eax /* the adjusted stack pointer */
759 pushl_cfi $__KERNEL_DS
761 lss (%esp), %esp /* switch to the normal stack segment */
762 CFI_ADJUST_CFA_OFFSET -8
764 .macro UNWIND_ESPFIX_STACK
766 /* see if on espfix stack */
767 cmpw $__ESPFIX_SS, %ax
769 movl $__KERNEL_DS, %eax
772 /* switch to normal stack */
778 * Build the entry stubs and pointer table with some assembler magic.
779 * We pack 7 stubs into a single 32-byte chunk, which will fit in a
780 * single cache line on all modern x86 implementations.
782 .section .init.rodata,"a"
784 .section .entry.text, "ax"
786 .p2align CONFIG_X86_L1_CACHE_SHIFT
787 ENTRY(irq_entries_start)
789 vector=FIRST_EXTERNAL_VECTOR
790 .rept (NR_VECTORS-FIRST_EXTERNAL_VECTOR+6)/7
793 .if vector < NR_VECTORS
794 .if vector <> FIRST_EXTERNAL_VECTOR
795 CFI_ADJUST_CFA_OFFSET -4
797 1: pushl_cfi $(~vector+0x80) /* Note: always in signed byte range */
798 .if ((vector-FIRST_EXTERNAL_VECTOR)%7) <> 6
803 .section .entry.text, "ax"
807 2: jmp common_interrupt
809 END(irq_entries_start)
816 * the CPU automatically disables interrupts when executing an IRQ vector,
817 * so IRQ-flags tracing has to follow that:
819 .p2align CONFIG_X86_L1_CACHE_SHIFT
821 addl $-0x80,(%esp) /* Adjust vector into the [-256,-1] range */
827 ENDPROC(common_interrupt)
831 * Irq entries should be protected against kprobes
833 .pushsection .kprobes.text, "ax"
834 #define BUILD_INTERRUPT3(name, nr, fn) \
846 #define BUILD_INTERRUPT(name, nr) BUILD_INTERRUPT3(name, nr, smp_##name)
848 /* The include is where all of the SMP etc. interrupts come from */
849 #include <asm/entry_arch.h>
851 ENTRY(coprocessor_error)
854 pushl_cfi $do_coprocessor_error
857 END(coprocessor_error)
859 ENTRY(simd_coprocessor_error)
862 #ifdef CONFIG_X86_INVD_BUG
863 /* AMD 486 bug: invd from userspace calls exception 19 instead of #GP */
864 661: pushl_cfi $do_general_protection
866 .section .altinstructions,"a"
867 altinstruction_entry 661b, 663f, X86_FEATURE_XMM, 662b-661b, 664f-663f
869 .section .altinstr_replacement,"ax"
870 663: pushl $do_simd_coprocessor_error
874 pushl_cfi $do_simd_coprocessor_error
878 END(simd_coprocessor_error)
880 ENTRY(device_not_available)
882 pushl_cfi $-1 # mark this as an int
883 pushl_cfi $do_device_not_available
886 END(device_not_available)
888 #ifdef CONFIG_PARAVIRT
891 _ASM_EXTABLE(native_iret, iret_exc)
894 ENTRY(native_irq_enable_sysexit)
897 END(native_irq_enable_sysexit)
903 pushl_cfi $do_overflow
919 pushl_cfi $do_invalid_op
924 ENTRY(coprocessor_segment_overrun)
927 pushl_cfi $do_coprocessor_segment_overrun
930 END(coprocessor_segment_overrun)
934 pushl_cfi $do_invalid_TSS
939 ENTRY(segment_not_present)
941 pushl_cfi $do_segment_not_present
944 END(segment_not_present)
948 pushl_cfi $do_stack_segment
953 ENTRY(alignment_check)
955 pushl_cfi $do_alignment_check
962 pushl_cfi $0 # no error code
963 pushl_cfi $do_divide_error
968 #ifdef CONFIG_X86_MCE
972 pushl_cfi machine_check_vector
978 ENTRY(spurious_interrupt_bug)
981 pushl_cfi $do_spurious_interrupt_bug
984 END(spurious_interrupt_bug)
986 * End of kprobes section
990 ENTRY(kernel_thread_helper)
991 pushl $0 # fake return address for unwinder
996 ud2 # padding for call trace
998 ENDPROC(kernel_thread_helper)
1001 /* Xen doesn't set %esp to be precisely what the normal sysenter
1002 entrypoint expects, so fix it up before using the normal path. */
1003 ENTRY(xen_sysenter_target)
1005 addl $5*4, %esp /* remove xen-provided frame */
1006 CFI_ADJUST_CFA_OFFSET -5*4
1007 jmp sysenter_past_esp
1010 ENTRY(xen_hypervisor_callback)
1016 /* Check to see if we got the event in the critical
1017 region in xen_iret_direct, after we've reenabled
1018 events and checked for pending events. This simulates
1019 iret instruction's behaviour where it delivers a
1020 pending interrupt when enabling interrupts. */
1021 movl PT_EIP(%esp),%eax
1022 cmpl $xen_iret_start_crit,%eax
1024 cmpl $xen_iret_end_crit,%eax
1027 jmp xen_iret_crit_fixup
1029 ENTRY(xen_do_upcall)
1031 call xen_evtchn_do_upcall
1034 ENDPROC(xen_hypervisor_callback)
1036 # Hypervisor uses this for application faults while it executes.
1037 # We get here for two reasons:
1038 # 1. Fault while reloading DS, ES, FS or GS
1039 # 2. Fault while executing IRET
1040 # Category 1 we fix up by reattempting the load, and zeroing the segment
1041 # register if the load fails.
1042 # Category 2 we fix up by jumping to do_iret_error. We cannot use the
1043 # normal Linux return path in this case because if we use the IRET hypercall
1044 # to pop the stack frame we end up in an infinite loop of failsafe callbacks.
1045 # We distinguish between categories by maintaining a status value in EAX.
1046 ENTRY(xen_failsafe_callback)
1057 CFI_ADJUST_CFA_OFFSET -16
1060 jmp iret_exc # EAX != 0 => Category 2 (Bad IRET)
1061 5: pushl_cfi $0 # EAX == 0 => Category 1 (Bad segment)
1063 jmp ret_from_exception
1066 .section .fixup,"ax"
1084 ENDPROC(xen_failsafe_callback)
1086 BUILD_INTERRUPT3(xen_hvm_callback_vector, XEN_HVM_EVTCHN_CALLBACK,
1087 xen_evtchn_do_upcall)
1089 #endif /* CONFIG_XEN */
1091 #ifdef CONFIG_FUNCTION_TRACER
1092 #ifdef CONFIG_DYNAMIC_FTRACE
1098 ENTRY(ftrace_caller)
1099 cmpl $0, function_trace_stop
1105 movl 0xc(%esp), %eax
1106 movl 0x4(%ebp), %edx
1107 subl $MCOUNT_INSN_SIZE, %eax
1116 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1117 .globl ftrace_graph_call
1127 #else /* ! CONFIG_DYNAMIC_FTRACE */
1130 cmpl $0, function_trace_stop
1133 cmpl $ftrace_stub, ftrace_trace_function
1135 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1136 cmpl $ftrace_stub, ftrace_graph_return
1137 jnz ftrace_graph_caller
1139 cmpl $ftrace_graph_entry_stub, ftrace_graph_entry
1140 jnz ftrace_graph_caller
1146 /* taken from glibc */
1151 movl 0xc(%esp), %eax
1152 movl 0x4(%ebp), %edx
1153 subl $MCOUNT_INSN_SIZE, %eax
1155 call *ftrace_trace_function
1162 #endif /* CONFIG_DYNAMIC_FTRACE */
1163 #endif /* CONFIG_FUNCTION_TRACER */
1165 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1166 ENTRY(ftrace_graph_caller)
1167 cmpl $0, function_trace_stop
1173 movl 0xc(%esp), %edx
1176 subl $MCOUNT_INSN_SIZE, %edx
1177 call prepare_ftrace_return
1182 END(ftrace_graph_caller)
1184 .globl return_to_handler
1189 call ftrace_return_to_handler
1197 * Some functions should be protected against kprobes
1199 .pushsection .kprobes.text, "ax"
1203 pushl_cfi $do_page_fault
1206 /* the function address is in %gs's slot on the stack */
1208 /*CFI_REL_OFFSET fs, 0*/
1210 /*CFI_REL_OFFSET es, 0*/
1212 /*CFI_REL_OFFSET ds, 0*/
1214 CFI_REL_OFFSET eax, 0
1216 CFI_REL_OFFSET ebp, 0
1218 CFI_REL_OFFSET edi, 0
1220 CFI_REL_OFFSET esi, 0
1222 CFI_REL_OFFSET edx, 0
1224 CFI_REL_OFFSET ecx, 0
1226 CFI_REL_OFFSET ebx, 0
1228 movl $(__KERNEL_PERCPU), %ecx
1232 movl PT_GS(%esp), %edi # get the function address
1233 movl PT_ORIG_EAX(%esp), %edx # get the error code
1234 movl $-1, PT_ORIG_EAX(%esp) # no syscall to restart
1237 movl $(__USER_DS), %ecx
1241 movl %esp,%eax # pt_regs pointer
1243 jmp ret_from_exception
1248 * Debug traps and NMI can happen at the one SYSENTER instruction
1249 * that sets up the real kernel stack. Check here, since we can't
1250 * allow the wrong stack to be used.
1252 * "TSS_sysenter_sp0+12" is because the NMI/debug handler will have
1253 * already pushed 3 words if it hits on the sysenter instruction:
1254 * eflags, cs and eip.
1256 * We just load the right stack, and push the three (known) values
1257 * by hand onto the new stack - while updating the return eip past
1258 * the instruction that would have done it for sysenter.
1260 .macro FIX_STACK offset ok label
1261 cmpw $__KERNEL_CS, 4(%esp)
1264 movl TSS_sysenter_sp0 + \offset(%esp), %esp
1268 pushl_cfi $__KERNEL_CS
1269 pushl_cfi $sysenter_past_esp
1270 CFI_REL_OFFSET eip, 0
1275 cmpl $ia32_sysenter_target,(%esp)
1276 jne debug_stack_correct
1277 FIX_STACK 12, debug_stack_correct, debug_esp_fix_insn
1278 debug_stack_correct:
1279 pushl_cfi $-1 # mark this as an int
1282 xorl %edx,%edx # error code 0
1283 movl %esp,%eax # pt_regs pointer
1285 jmp ret_from_exception
1290 * NMI is doubly nasty. It can happen _while_ we're handling
1291 * a debug fault, and the debug fault hasn't yet been able to
1292 * clear up the stack. So we first check whether we got an
1293 * NMI on the sysenter entry path, but after that we need to
1294 * check whether we got an NMI on the debug path where the debug
1295 * fault happened on the sysenter path.
1301 cmpw $__ESPFIX_SS, %ax
1304 cmpl $ia32_sysenter_target,(%esp)
1308 /* Do not access memory above the end of our stack page,
1309 * it might not exist.
1311 andl $(THREAD_SIZE-1),%eax
1312 cmpl $(THREAD_SIZE-20),%eax
1314 jae nmi_stack_correct
1315 cmpl $ia32_sysenter_target,12(%esp)
1316 je nmi_debug_stack_check
1318 /* We have a RING0_INT_FRAME here */
1321 xorl %edx,%edx # zero error code
1322 movl %esp,%eax # pt_regs pointer
1324 jmp restore_all_notrace
1329 FIX_STACK 12, nmi_stack_correct, 1
1330 jmp nmi_stack_correct
1332 nmi_debug_stack_check:
1333 /* We have a RING0_INT_FRAME here */
1334 cmpw $__KERNEL_CS,16(%esp)
1335 jne nmi_stack_correct
1337 jb nmi_stack_correct
1338 cmpl $debug_esp_fix_insn,(%esp)
1339 ja nmi_stack_correct
1340 FIX_STACK 24, nmi_stack_correct, 1
1341 jmp nmi_stack_correct
1344 /* We have a RING0_INT_FRAME here.
1346 * create the pointer to lss back
1351 /* copy the iret frame of 12 bytes */
1357 FIXUP_ESPFIX_STACK # %eax == %esp
1358 xorl %edx,%edx # zero error code
1361 lss 12+4(%esp), %esp # back to espfix stack
1362 CFI_ADJUST_CFA_OFFSET -24
1369 pushl_cfi $-1 # mark this as an int
1372 xorl %edx,%edx # zero error code
1373 movl %esp,%eax # pt_regs pointer
1375 jmp ret_from_exception
1379 ENTRY(general_protection)
1381 pushl_cfi $do_general_protection
1384 END(general_protection)
1386 #ifdef CONFIG_KVM_GUEST
1387 ENTRY(async_page_fault)
1389 pushl_cfi $do_async_page_fault
1392 END(async_page_fault)
1396 * End of kprobes section