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>
57 /* Avoid __ASSEMBLER__'ifying <linux/audit.h> just for this. */
58 #include <linux/elf-em.h>
59 #define AUDIT_ARCH_I386 (EM_386|__AUDIT_ARCH_LE)
60 #define __AUDIT_ARCH_LE 0x40000000
62 #ifndef CONFIG_AUDITSYSCALL
63 #define sysenter_audit syscall_trace_entry
64 #define sysexit_audit syscall_exit_work
68 * We use macros for low-level operations which need to be overridden
69 * for paravirtualization. The following will never clobber any registers:
70 * INTERRUPT_RETURN (aka. "iret")
71 * GET_CR0_INTO_EAX (aka. "movl %cr0, %eax")
72 * ENABLE_INTERRUPTS_SYSEXIT (aka "sti; sysexit").
74 * For DISABLE_INTERRUPTS/ENABLE_INTERRUPTS (aka "cli"/"sti"), you must
75 * specify what registers can be overwritten (CLBR_NONE, CLBR_EAX/EDX/ECX/ANY).
76 * Allowing a register to be clobbered can shrink the paravirt replacement
77 * enough to patch inline, increasing performance.
80 #define nr_syscalls ((syscall_table_size)/4)
83 #define preempt_stop(clobbers) DISABLE_INTERRUPTS(clobbers); TRACE_IRQS_OFF
85 #define preempt_stop(clobbers)
86 #define resume_kernel restore_all
89 .macro TRACE_IRQS_IRET
90 #ifdef CONFIG_TRACE_IRQFLAGS
91 testl $X86_EFLAGS_IF,PT_EFLAGS(%esp) # interrupts off?
99 #define resume_userspace_sig check_userspace
101 #define resume_userspace_sig resume_userspace
105 * User gs save/restore
107 * %gs is used for userland TLS and kernel only uses it for stack
108 * canary which is required to be at %gs:20 by gcc. Read the comment
109 * at the top of stackprotector.h for more info.
111 * Local labels 98 and 99 are used.
113 #ifdef CONFIG_X86_32_LAZY_GS
115 /* unfortunately push/pop can't be no-op */
118 CFI_ADJUST_CFA_OFFSET 4
121 addl $(4 + \pop), %esp
122 CFI_ADJUST_CFA_OFFSET -(4 + \pop)
127 /* all the rest are no-op */
134 .macro REG_TO_PTGS reg
136 .macro SET_KERNEL_GS reg
139 #else /* CONFIG_X86_32_LAZY_GS */
143 CFI_ADJUST_CFA_OFFSET 4
144 /*CFI_REL_OFFSET gs, 0*/
149 CFI_ADJUST_CFA_OFFSET -4
153 CFI_ADJUST_CFA_OFFSET -\pop
157 .pushsection .fixup, "ax"
160 .section __ex_table, "a"
167 98: mov PT_GS(%esp), %gs
170 .pushsection .fixup, "ax"
171 99: movl $0, PT_GS(%esp)
173 .section __ex_table, "a"
181 /*CFI_REGISTER gs, \reg*/
183 .macro REG_TO_PTGS reg
184 movl \reg, PT_GS(%esp)
185 /*CFI_REL_OFFSET gs, PT_GS*/
187 .macro SET_KERNEL_GS reg
188 movl $(__KERNEL_STACK_CANARY), \reg
192 #endif /* CONFIG_X86_32_LAZY_GS */
198 CFI_ADJUST_CFA_OFFSET 4
199 /*CFI_REL_OFFSET fs, 0;*/
201 CFI_ADJUST_CFA_OFFSET 4
202 /*CFI_REL_OFFSET es, 0;*/
204 CFI_ADJUST_CFA_OFFSET 4
205 /*CFI_REL_OFFSET ds, 0;*/
207 CFI_ADJUST_CFA_OFFSET 4
208 CFI_REL_OFFSET eax, 0
210 CFI_ADJUST_CFA_OFFSET 4
211 CFI_REL_OFFSET ebp, 0
213 CFI_ADJUST_CFA_OFFSET 4
214 CFI_REL_OFFSET edi, 0
216 CFI_ADJUST_CFA_OFFSET 4
217 CFI_REL_OFFSET esi, 0
219 CFI_ADJUST_CFA_OFFSET 4
220 CFI_REL_OFFSET edx, 0
222 CFI_ADJUST_CFA_OFFSET 4
223 CFI_REL_OFFSET ecx, 0
225 CFI_ADJUST_CFA_OFFSET 4
226 CFI_REL_OFFSET ebx, 0
227 movl $(__USER_DS), %edx
230 movl $(__KERNEL_PERCPU), %edx
235 .macro RESTORE_INT_REGS
237 CFI_ADJUST_CFA_OFFSET -4
240 CFI_ADJUST_CFA_OFFSET -4
243 CFI_ADJUST_CFA_OFFSET -4
246 CFI_ADJUST_CFA_OFFSET -4
249 CFI_ADJUST_CFA_OFFSET -4
252 CFI_ADJUST_CFA_OFFSET -4
255 CFI_ADJUST_CFA_OFFSET -4
259 .macro RESTORE_REGS pop=0
262 CFI_ADJUST_CFA_OFFSET -4
265 CFI_ADJUST_CFA_OFFSET -4
268 CFI_ADJUST_CFA_OFFSET -4
271 .pushsection .fixup, "ax"
278 .section __ex_table, "a"
287 .macro RING0_INT_FRAME
291 /*CFI_OFFSET cs, -2*4;*/
295 .macro RING0_EC_FRAME
299 /*CFI_OFFSET cs, -2*4;*/
303 .macro RING0_PTREGS_FRAME
306 CFI_DEF_CFA esp, PT_OLDESP-PT_EBX
307 /*CFI_OFFSET cs, PT_CS-PT_OLDESP;*/
308 CFI_OFFSET eip, PT_EIP-PT_OLDESP
309 /*CFI_OFFSET es, PT_ES-PT_OLDESP;*/
310 /*CFI_OFFSET ds, PT_DS-PT_OLDESP;*/
311 CFI_OFFSET eax, PT_EAX-PT_OLDESP
312 CFI_OFFSET ebp, PT_EBP-PT_OLDESP
313 CFI_OFFSET edi, PT_EDI-PT_OLDESP
314 CFI_OFFSET esi, PT_ESI-PT_OLDESP
315 CFI_OFFSET edx, PT_EDX-PT_OLDESP
316 CFI_OFFSET ecx, PT_ECX-PT_OLDESP
317 CFI_OFFSET ebx, PT_EBX-PT_OLDESP
323 CFI_ADJUST_CFA_OFFSET 4
325 GET_THREAD_INFO(%ebp)
327 CFI_ADJUST_CFA_OFFSET -4
328 pushl $0x0202 # Reset kernel eflags
329 CFI_ADJUST_CFA_OFFSET 4
331 CFI_ADJUST_CFA_OFFSET -4
337 * Return to user mode is not as complex as all this looks,
338 * but we want the default path for a system call return to
339 * go as quickly as possible which is why some of this is
340 * less clear than it otherwise should be.
343 # userspace resumption stub bypassing syscall exit tracing
347 preempt_stop(CLBR_ANY)
349 GET_THREAD_INFO(%ebp)
351 movl PT_EFLAGS(%esp), %eax # mix EFLAGS and CS
352 movb PT_CS(%esp), %al
353 andl $(X86_EFLAGS_VM | SEGMENT_RPL_MASK), %eax
355 jb resume_kernel # not returning to v8086 or userspace
357 ENTRY(resume_userspace)
359 DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt
360 # setting need_resched or sigpending
361 # between sampling and the iret
363 movl TI_flags(%ebp), %ecx
364 andl $_TIF_WORK_MASK, %ecx # is there any work to be done on
365 # int/exception return?
368 END(ret_from_exception)
370 #ifdef CONFIG_PREEMPT
372 DISABLE_INTERRUPTS(CLBR_ANY)
373 cmpl $0,TI_preempt_count(%ebp) # non-zero preempt_count ?
376 movl TI_flags(%ebp), %ecx # need_resched set ?
377 testb $_TIF_NEED_RESCHED, %cl
379 testl $X86_EFLAGS_IF,PT_EFLAGS(%esp) # interrupts off (exception path) ?
381 call preempt_schedule_irq
387 /* SYSENTER_RETURN points to after the "sysenter" instruction in
388 the vsyscall page. See vsyscall-sysentry.S, which defines the symbol. */
390 # sysenter call handler stub
391 ENTRY(ia32_sysenter_target)
395 CFI_REGISTER esp, ebp
396 movl TSS_sysenter_sp0(%esp),%esp
399 * Interrupts are disabled here, but we can't trace it until
400 * enough kernel state to call TRACE_IRQS_OFF can be called - but
401 * we immediately enable interrupts at that point anyway.
404 CFI_ADJUST_CFA_OFFSET 4
405 /*CFI_REL_OFFSET ss, 0*/
407 CFI_ADJUST_CFA_OFFSET 4
408 CFI_REL_OFFSET esp, 0
410 orl $X86_EFLAGS_IF, (%esp)
411 CFI_ADJUST_CFA_OFFSET 4
413 CFI_ADJUST_CFA_OFFSET 4
414 /*CFI_REL_OFFSET cs, 0*/
416 * Push current_thread_info()->sysenter_return to the stack.
417 * A tiny bit of offset fixup is necessary - 4*4 means the 4 words
418 * pushed above; +8 corresponds to copy_thread's esp0 setting.
420 pushl (TI_sysenter_return-THREAD_SIZE+8+4*4)(%esp)
421 CFI_ADJUST_CFA_OFFSET 4
422 CFI_REL_OFFSET eip, 0
425 CFI_ADJUST_CFA_OFFSET 4
427 ENABLE_INTERRUPTS(CLBR_NONE)
430 * Load the potential sixth argument from user stack.
431 * Careful about security.
433 cmpl $__PAGE_OFFSET-3,%ebp
436 movl %ebp,PT_EBP(%esp)
437 .section __ex_table,"a"
439 .long 1b,syscall_fault
442 GET_THREAD_INFO(%ebp)
444 testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%ebp)
447 cmpl $(nr_syscalls), %eax
449 call *sys_call_table(,%eax,4)
450 movl %eax,PT_EAX(%esp)
452 DISABLE_INTERRUPTS(CLBR_ANY)
454 movl TI_flags(%ebp), %ecx
455 testl $_TIF_ALLWORK_MASK, %ecx
458 /* if something modifies registers it must also disable sysexit */
459 movl PT_EIP(%esp), %edx
460 movl PT_OLDESP(%esp), %ecx
463 1: mov PT_FS(%esp), %fs
465 ENABLE_INTERRUPTS_SYSEXIT
467 #ifdef CONFIG_AUDITSYSCALL
469 testl $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags(%ebp)
470 jnz syscall_trace_entry
472 CFI_ADJUST_CFA_OFFSET -4
473 /* %esi already in 8(%esp) 6th arg: 4th syscall arg */
474 /* %edx already in 4(%esp) 5th arg: 3rd syscall arg */
475 /* %ecx already in 0(%esp) 4th arg: 2nd syscall arg */
476 movl %ebx,%ecx /* 3rd arg: 1st syscall arg */
477 movl %eax,%edx /* 2nd arg: syscall number */
478 movl $AUDIT_ARCH_I386,%eax /* 1st arg: audit arch */
479 call audit_syscall_entry
481 CFI_ADJUST_CFA_OFFSET 4
482 movl PT_EAX(%esp),%eax /* reload syscall number */
486 testl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT), %ecx
487 jne syscall_exit_work
489 ENABLE_INTERRUPTS(CLBR_ANY)
490 movl %eax,%edx /* second arg, syscall return value */
491 cmpl $0,%eax /* is it < 0? */
492 setl %al /* 1 if so, 0 if not */
493 movzbl %al,%eax /* zero-extend that */
494 inc %eax /* first arg, 0->1(AUDITSC_SUCCESS), 1->2(AUDITSC_FAILURE) */
495 call audit_syscall_exit
496 DISABLE_INTERRUPTS(CLBR_ANY)
498 movl TI_flags(%ebp), %ecx
499 testl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT), %ecx
500 jne syscall_exit_work
501 movl PT_EAX(%esp),%eax /* reload syscall return value */
506 .pushsection .fixup,"ax"
507 2: movl $0,PT_FS(%esp)
509 .section __ex_table,"a"
514 ENDPROC(ia32_sysenter_target)
516 # system call handler stub
518 RING0_INT_FRAME # can't unwind into user space anyway
519 pushl %eax # save orig_eax
520 CFI_ADJUST_CFA_OFFSET 4
522 GET_THREAD_INFO(%ebp)
523 # system call tracing in operation / emulation
524 testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%ebp)
525 jnz syscall_trace_entry
526 cmpl $(nr_syscalls), %eax
529 call *sys_call_table(,%eax,4)
530 movl %eax,PT_EAX(%esp) # store the return value
533 DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt
534 # setting need_resched or sigpending
535 # between sampling and the iret
537 movl TI_flags(%ebp), %ecx
538 testl $_TIF_ALLWORK_MASK, %ecx # current->work
539 jne syscall_exit_work
544 movl PT_EFLAGS(%esp), %eax # mix EFLAGS, SS and CS
545 # Warning: PT_OLDSS(%esp) contains the wrong/random values if we
546 # are returning to the kernel.
547 # See comments in process.c:copy_thread() for details.
548 movb PT_OLDSS(%esp), %ah
549 movb PT_CS(%esp), %al
550 andl $(X86_EFLAGS_VM | (SEGMENT_TI_MASK << 8) | SEGMENT_RPL_MASK), %eax
551 cmpl $((SEGMENT_LDT << 8) | USER_RPL), %eax
553 je ldt_ss # returning to user-space with LDT SS
555 RESTORE_REGS 4 # skip orig_eax/error_code
556 CFI_ADJUST_CFA_OFFSET -4
561 pushl $0 # no error code
565 .section __ex_table,"a"
567 .long irq_return,iret_exc
572 larl PT_OLDSS(%esp), %eax
574 testl $0x00400000, %eax # returning to 32bit stack?
575 jnz restore_nocheck # allright, normal return
577 #ifdef CONFIG_PARAVIRT
579 * The kernel can't run on a non-flat stack if paravirt mode
580 * is active. Rather than try to fixup the high bits of
581 * ESP, bypass this code entirely. This may break DOSemu
582 * and/or Wine support in a paravirt VM, although the option
583 * is still available to implement the setting of the high
584 * 16-bits in the INTERRUPT_RETURN paravirt-op.
586 cmpl $0, pv_info+PARAVIRT_enabled
591 * Setup and switch to ESPFIX stack
593 * We're returning to userspace with a 16 bit stack. The CPU will not
594 * restore the high word of ESP for us on executing iret... This is an
595 * "official" bug of all the x86-compatible CPUs, which we can work
596 * around to make dosemu and wine happy. We do this by preloading the
597 * high word of ESP with the high word of the userspace ESP while
598 * compensating for the offset by changing to the ESPFIX segment with
599 * a base address that matches for the difference.
601 mov %esp, %edx /* load kernel esp */
602 mov PT_OLDESP(%esp), %eax /* load userspace esp */
603 mov %dx, %ax /* eax: new kernel esp */
604 sub %eax, %edx /* offset (low word is 0) */
605 PER_CPU(gdt_page, %ebx)
607 mov %dl, GDT_ENTRY_ESPFIX_SS * 8 + 4(%ebx) /* bits 16..23 */
608 mov %dh, GDT_ENTRY_ESPFIX_SS * 8 + 7(%ebx) /* bits 24..31 */
610 CFI_ADJUST_CFA_OFFSET 4
611 push %eax /* new kernel esp */
612 CFI_ADJUST_CFA_OFFSET 4
613 /* Disable interrupts, but do not irqtrace this section: we
614 * will soon execute iret and the tracer was already set to
615 * the irqstate after the iret */
616 DISABLE_INTERRUPTS(CLBR_EAX)
617 lss (%esp), %esp /* switch to espfix segment */
618 CFI_ADJUST_CFA_OFFSET -8
623 # perform work that needs to be done immediately before resumption
625 RING0_PTREGS_FRAME # can't unwind into user space anyway
627 testb $_TIF_NEED_RESCHED, %cl
632 DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt
633 # setting need_resched or sigpending
634 # between sampling and the iret
636 movl TI_flags(%ebp), %ecx
637 andl $_TIF_WORK_MASK, %ecx # is there any work to be done other
638 # than syscall tracing?
640 testb $_TIF_NEED_RESCHED, %cl
643 work_notifysig: # deal with pending signals and
644 # notify-resume requests
646 testl $X86_EFLAGS_VM, PT_EFLAGS(%esp)
648 jne work_notifysig_v86 # returning to kernel-space or
651 call do_notify_resume
652 jmp resume_userspace_sig
656 pushl %ecx # save ti_flags for do_notify_resume
657 CFI_ADJUST_CFA_OFFSET 4
658 call save_v86_state # %eax contains pt_regs pointer
660 CFI_ADJUST_CFA_OFFSET -4
666 call do_notify_resume
667 jmp resume_userspace_sig
670 # perform syscall exit tracing
673 movl $-ENOSYS,PT_EAX(%esp)
675 call syscall_trace_enter
676 /* What it returned is what we'll actually use. */
677 cmpl $(nr_syscalls), %eax
680 END(syscall_trace_entry)
682 # perform syscall exit tracing
685 testl $_TIF_WORK_SYSCALL_EXIT, %ecx
688 ENABLE_INTERRUPTS(CLBR_ANY) # could let syscall_trace_leave() call
691 call syscall_trace_leave
693 END(syscall_exit_work)
696 RING0_INT_FRAME # can't unwind into user space anyway
698 GET_THREAD_INFO(%ebp)
699 movl $-EFAULT,PT_EAX(%esp)
704 movl $-ENOSYS,PT_EAX(%esp)
710 * System calls that need a pt_regs pointer.
712 #define PTREGSCALL(name) \
723 PTREGSCALL(sigaltstack)
724 PTREGSCALL(sigreturn)
725 PTREGSCALL(rt_sigreturn)
729 .macro FIXUP_ESPFIX_STACK
731 * Switch back for ESPFIX stack to the normal zerobased stack
733 * We can't call C functions using the ESPFIX stack. This code reads
734 * the high word of the segment base from the GDT and swiches to the
735 * normal stack and adjusts ESP with the matching offset.
737 /* fixup the stack */
738 PER_CPU(gdt_page, %ebx)
739 mov GDT_ENTRY_ESPFIX_SS * 8 + 4(%ebx), %al /* bits 16..23 */
740 mov GDT_ENTRY_ESPFIX_SS * 8 + 7(%ebx), %ah /* bits 24..31 */
742 addl %esp, %eax /* the adjusted stack pointer */
744 CFI_ADJUST_CFA_OFFSET 4
746 CFI_ADJUST_CFA_OFFSET 4
747 lss (%esp), %esp /* switch to the normal stack segment */
748 CFI_ADJUST_CFA_OFFSET -8
750 .macro UNWIND_ESPFIX_STACK
752 /* see if on espfix stack */
753 cmpw $__ESPFIX_SS, %ax
755 movl $__KERNEL_DS, %eax
758 /* switch to normal stack */
764 * Build the entry stubs and pointer table with some assembler magic.
765 * We pack 7 stubs into a single 32-byte chunk, which will fit in a
766 * single cache line on all modern x86 implementations.
768 .section .init.rodata,"a"
772 .p2align CONFIG_X86_L1_CACHE_SHIFT
773 ENTRY(irq_entries_start)
775 vector=FIRST_EXTERNAL_VECTOR
776 .rept (NR_VECTORS-FIRST_EXTERNAL_VECTOR+6)/7
779 .if vector < NR_VECTORS
780 .if vector <> FIRST_EXTERNAL_VECTOR
781 CFI_ADJUST_CFA_OFFSET -4
783 1: pushl $(~vector+0x80) /* Note: always in signed byte range */
784 CFI_ADJUST_CFA_OFFSET 4
785 .if ((vector-FIRST_EXTERNAL_VECTOR)%7) <> 6
794 2: jmp common_interrupt
796 END(irq_entries_start)
803 * the CPU automatically disables interrupts when executing an IRQ vector,
804 * so IRQ-flags tracing has to follow that:
806 .p2align CONFIG_X86_L1_CACHE_SHIFT
808 addl $-0x80,(%esp) /* Adjust vector into the [-256,-1] range */
814 ENDPROC(common_interrupt)
817 #define BUILD_INTERRUPT3(name, nr, fn) \
821 CFI_ADJUST_CFA_OFFSET 4; \
830 #define BUILD_INTERRUPT(name, nr) BUILD_INTERRUPT3(name, nr, smp_##name)
832 /* The include is where all of the SMP etc. interrupts come from */
833 #include <asm/entry_arch.h>
835 ENTRY(coprocessor_error)
838 CFI_ADJUST_CFA_OFFSET 4
839 pushl $do_coprocessor_error
840 CFI_ADJUST_CFA_OFFSET 4
843 END(coprocessor_error)
845 ENTRY(simd_coprocessor_error)
848 CFI_ADJUST_CFA_OFFSET 4
849 pushl $do_simd_coprocessor_error
850 CFI_ADJUST_CFA_OFFSET 4
853 END(simd_coprocessor_error)
855 ENTRY(device_not_available)
857 pushl $-1 # mark this as an int
858 CFI_ADJUST_CFA_OFFSET 4
859 pushl $do_device_not_available
860 CFI_ADJUST_CFA_OFFSET 4
863 END(device_not_available)
865 #ifdef CONFIG_PARAVIRT
868 .section __ex_table,"a"
870 .long native_iret, iret_exc
874 ENTRY(native_irq_enable_sysexit)
877 END(native_irq_enable_sysexit)
883 CFI_ADJUST_CFA_OFFSET 4
885 CFI_ADJUST_CFA_OFFSET 4
893 CFI_ADJUST_CFA_OFFSET 4
895 CFI_ADJUST_CFA_OFFSET 4
903 CFI_ADJUST_CFA_OFFSET 4
905 CFI_ADJUST_CFA_OFFSET 4
910 ENTRY(coprocessor_segment_overrun)
913 CFI_ADJUST_CFA_OFFSET 4
914 pushl $do_coprocessor_segment_overrun
915 CFI_ADJUST_CFA_OFFSET 4
918 END(coprocessor_segment_overrun)
922 pushl $do_invalid_TSS
923 CFI_ADJUST_CFA_OFFSET 4
928 ENTRY(segment_not_present)
930 pushl $do_segment_not_present
931 CFI_ADJUST_CFA_OFFSET 4
934 END(segment_not_present)
938 pushl $do_stack_segment
939 CFI_ADJUST_CFA_OFFSET 4
944 ENTRY(alignment_check)
946 pushl $do_alignment_check
947 CFI_ADJUST_CFA_OFFSET 4
954 pushl $0 # no error code
955 CFI_ADJUST_CFA_OFFSET 4
956 pushl $do_divide_error
957 CFI_ADJUST_CFA_OFFSET 4
962 #ifdef CONFIG_X86_MCE
966 CFI_ADJUST_CFA_OFFSET 4
967 pushl machine_check_vector
968 CFI_ADJUST_CFA_OFFSET 4
974 ENTRY(spurious_interrupt_bug)
977 CFI_ADJUST_CFA_OFFSET 4
978 pushl $do_spurious_interrupt_bug
979 CFI_ADJUST_CFA_OFFSET 4
982 END(spurious_interrupt_bug)
984 ENTRY(kernel_thread_helper)
985 pushl $0 # fake return address for unwinder
989 CFI_ADJUST_CFA_OFFSET 4
992 CFI_ADJUST_CFA_OFFSET 4
994 ud2 # padding for call trace
996 ENDPROC(kernel_thread_helper)
999 /* Xen doesn't set %esp to be precisely what the normal sysenter
1000 entrypoint expects, so fix it up before using the normal path. */
1001 ENTRY(xen_sysenter_target)
1003 addl $5*4, %esp /* remove xen-provided frame */
1004 CFI_ADJUST_CFA_OFFSET -5*4
1005 jmp sysenter_past_esp
1008 ENTRY(xen_hypervisor_callback)
1011 CFI_ADJUST_CFA_OFFSET 4
1015 /* Check to see if we got the event in the critical
1016 region in xen_iret_direct, after we've reenabled
1017 events and checked for pending events. This simulates
1018 iret instruction's behaviour where it delivers a
1019 pending interrupt when enabling interrupts. */
1020 movl PT_EIP(%esp),%eax
1021 cmpl $xen_iret_start_crit,%eax
1023 cmpl $xen_iret_end_crit,%eax
1026 jmp xen_iret_crit_fixup
1028 ENTRY(xen_do_upcall)
1030 call xen_evtchn_do_upcall
1033 ENDPROC(xen_hypervisor_callback)
1035 # Hypervisor uses this for application faults while it executes.
1036 # We get here for two reasons:
1037 # 1. Fault while reloading DS, ES, FS or GS
1038 # 2. Fault while executing IRET
1039 # Category 1 we fix up by reattempting the load, and zeroing the segment
1040 # register if the load fails.
1041 # Category 2 we fix up by jumping to do_iret_error. We cannot use the
1042 # normal Linux return path in this case because if we use the IRET hypercall
1043 # to pop the stack frame we end up in an infinite loop of failsafe callbacks.
1044 # We distinguish between categories by maintaining a status value in EAX.
1045 ENTRY(xen_failsafe_callback)
1048 CFI_ADJUST_CFA_OFFSET 4
1056 CFI_ADJUST_CFA_OFFSET -4
1058 CFI_ADJUST_CFA_OFFSET -16
1061 jmp iret_exc # EAX != 0 => Category 2 (Bad IRET)
1062 5: pushl $0 # EAX == 0 => Category 1 (Bad segment)
1063 CFI_ADJUST_CFA_OFFSET 4
1065 jmp ret_from_exception
1068 .section .fixup,"ax"
1082 .section __ex_table,"a"
1089 ENDPROC(xen_failsafe_callback)
1091 #endif /* CONFIG_XEN */
1093 #ifdef CONFIG_FUNCTION_TRACER
1094 #ifdef CONFIG_DYNAMIC_FTRACE
1100 ENTRY(ftrace_caller)
1101 cmpl $0, function_trace_stop
1107 movl 0xc(%esp), %eax
1108 movl 0x4(%ebp), %edx
1109 subl $MCOUNT_INSN_SIZE, %eax
1118 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1119 .globl ftrace_graph_call
1129 #else /* ! CONFIG_DYNAMIC_FTRACE */
1132 cmpl $0, function_trace_stop
1135 cmpl $ftrace_stub, ftrace_trace_function
1137 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1138 cmpl $ftrace_stub, ftrace_graph_return
1139 jnz ftrace_graph_caller
1141 cmpl $ftrace_graph_entry_stub, ftrace_graph_entry
1142 jnz ftrace_graph_caller
1148 /* taken from glibc */
1153 movl 0xc(%esp), %eax
1154 movl 0x4(%ebp), %edx
1155 subl $MCOUNT_INSN_SIZE, %eax
1157 call *ftrace_trace_function
1164 #endif /* CONFIG_DYNAMIC_FTRACE */
1165 #endif /* CONFIG_FUNCTION_TRACER */
1167 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1168 ENTRY(ftrace_graph_caller)
1169 cmpl $0, function_trace_stop
1175 movl 0xc(%esp), %edx
1178 subl $MCOUNT_INSN_SIZE, %edx
1179 call prepare_ftrace_return
1184 END(ftrace_graph_caller)
1186 .globl return_to_handler
1193 call ftrace_return_to_handler
1194 movl %eax, 0xc(%esp)
1201 .section .rodata,"a"
1202 #include "syscall_table_32.S"
1204 syscall_table_size=(.-sys_call_table)
1207 * Some functions should be protected against kprobes
1209 .pushsection .kprobes.text, "ax"
1213 pushl $do_page_fault
1214 CFI_ADJUST_CFA_OFFSET 4
1217 /* the function address is in %gs's slot on the stack */
1219 CFI_ADJUST_CFA_OFFSET 4
1220 /*CFI_REL_OFFSET fs, 0*/
1222 CFI_ADJUST_CFA_OFFSET 4
1223 /*CFI_REL_OFFSET es, 0*/
1225 CFI_ADJUST_CFA_OFFSET 4
1226 /*CFI_REL_OFFSET ds, 0*/
1228 CFI_ADJUST_CFA_OFFSET 4
1229 CFI_REL_OFFSET eax, 0
1231 CFI_ADJUST_CFA_OFFSET 4
1232 CFI_REL_OFFSET ebp, 0
1234 CFI_ADJUST_CFA_OFFSET 4
1235 CFI_REL_OFFSET edi, 0
1237 CFI_ADJUST_CFA_OFFSET 4
1238 CFI_REL_OFFSET esi, 0
1240 CFI_ADJUST_CFA_OFFSET 4
1241 CFI_REL_OFFSET edx, 0
1243 CFI_ADJUST_CFA_OFFSET 4
1244 CFI_REL_OFFSET ecx, 0
1246 CFI_ADJUST_CFA_OFFSET 4
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 CFI_ADJUST_CFA_OFFSET 4
1291 CFI_ADJUST_CFA_OFFSET 4
1292 pushl $sysenter_past_esp
1293 CFI_ADJUST_CFA_OFFSET 4
1294 CFI_REL_OFFSET eip, 0
1299 cmpl $ia32_sysenter_target,(%esp)
1300 jne debug_stack_correct
1301 FIX_STACK 12, debug_stack_correct, debug_esp_fix_insn
1302 debug_stack_correct:
1303 pushl $-1 # mark this as an int
1304 CFI_ADJUST_CFA_OFFSET 4
1307 xorl %edx,%edx # error code 0
1308 movl %esp,%eax # pt_regs pointer
1310 jmp ret_from_exception
1315 * NMI is doubly nasty. It can happen _while_ we're handling
1316 * a debug fault, and the debug fault hasn't yet been able to
1317 * clear up the stack. So we first check whether we got an
1318 * NMI on the sysenter entry path, but after that we need to
1319 * check whether we got an NMI on the debug path where the debug
1320 * fault happened on the sysenter path.
1325 CFI_ADJUST_CFA_OFFSET 4
1327 cmpw $__ESPFIX_SS, %ax
1329 CFI_ADJUST_CFA_OFFSET -4
1331 cmpl $ia32_sysenter_target,(%esp)
1334 CFI_ADJUST_CFA_OFFSET 4
1336 /* Do not access memory above the end of our stack page,
1337 * it might not exist.
1339 andl $(THREAD_SIZE-1),%eax
1340 cmpl $(THREAD_SIZE-20),%eax
1342 CFI_ADJUST_CFA_OFFSET -4
1343 jae nmi_stack_correct
1344 cmpl $ia32_sysenter_target,12(%esp)
1345 je nmi_debug_stack_check
1347 /* We have a RING0_INT_FRAME here */
1349 CFI_ADJUST_CFA_OFFSET 4
1351 xorl %edx,%edx # zero error code
1352 movl %esp,%eax # pt_regs pointer
1354 jmp restore_all_notrace
1359 FIX_STACK 12, nmi_stack_correct, 1
1360 jmp nmi_stack_correct
1362 nmi_debug_stack_check:
1363 /* We have a RING0_INT_FRAME here */
1364 cmpw $__KERNEL_CS,16(%esp)
1365 jne nmi_stack_correct
1367 jb nmi_stack_correct
1368 cmpl $debug_esp_fix_insn,(%esp)
1369 ja nmi_stack_correct
1370 FIX_STACK 24, nmi_stack_correct, 1
1371 jmp nmi_stack_correct
1374 /* We have a RING0_INT_FRAME here.
1376 * create the pointer to lss back
1379 CFI_ADJUST_CFA_OFFSET 4
1381 CFI_ADJUST_CFA_OFFSET 4
1383 /* copy the iret frame of 12 bytes */
1386 CFI_ADJUST_CFA_OFFSET 4
1389 CFI_ADJUST_CFA_OFFSET 4
1391 FIXUP_ESPFIX_STACK # %eax == %esp
1392 xorl %edx,%edx # zero error code
1395 lss 12+4(%esp), %esp # back to espfix stack
1396 CFI_ADJUST_CFA_OFFSET -24
1403 pushl $-1 # mark this as an int
1404 CFI_ADJUST_CFA_OFFSET 4
1407 xorl %edx,%edx # zero error code
1408 movl %esp,%eax # pt_regs pointer
1410 jmp ret_from_exception
1414 ENTRY(general_protection)
1416 pushl $do_general_protection
1417 CFI_ADJUST_CFA_OFFSET 4
1420 END(general_protection)
1423 * End of kprobes section