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>
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>
58 /* Avoid __ASSEMBLER__'ifying <linux/audit.h> just for this. */
59 #include <linux/elf-em.h>
60 #define AUDIT_ARCH_I386 (EM_386|__AUDIT_ARCH_LE)
61 #define __AUDIT_ARCH_LE 0x40000000
63 #ifndef CONFIG_AUDITSYSCALL
64 #define sysenter_audit syscall_trace_entry
65 #define sysexit_audit syscall_exit_work
69 * We use macros for low-level operations which need to be overridden
70 * for paravirtualization. The following will never clobber any registers:
71 * INTERRUPT_RETURN (aka. "iret")
72 * GET_CR0_INTO_EAX (aka. "movl %cr0, %eax")
73 * ENABLE_INTERRUPTS_SYSEXIT (aka "sti; sysexit").
75 * For DISABLE_INTERRUPTS/ENABLE_INTERRUPTS (aka "cli"/"sti"), you must
76 * specify what registers can be overwritten (CLBR_NONE, CLBR_EAX/EDX/ECX/ANY).
77 * Allowing a register to be clobbered can shrink the paravirt replacement
78 * enough to patch inline, increasing performance.
81 #define nr_syscalls ((syscall_table_size)/4)
84 #define preempt_stop(clobbers) DISABLE_INTERRUPTS(clobbers); TRACE_IRQS_OFF
86 #define preempt_stop(clobbers)
87 #define resume_kernel restore_nocheck
90 .macro TRACE_IRQS_IRET
91 #ifdef CONFIG_TRACE_IRQFLAGS
92 testl $X86_EFLAGS_IF,PT_EFLAGS(%esp) # interrupts off?
100 #define resume_userspace_sig check_userspace
102 #define resume_userspace_sig resume_userspace
106 * User gs save/restore
108 * %gs is used for userland TLS and kernel only uses it for stack
109 * canary which is required to be at %gs:20 by gcc. Read the comment
110 * at the top of stackprotector.h for more info.
112 * Local labels 98 and 99 are used.
114 #ifdef CONFIG_X86_32_LAZY_GS
116 /* unfortunately push/pop can't be no-op */
119 CFI_ADJUST_CFA_OFFSET 4
122 addl $(4 + \pop), %esp
123 CFI_ADJUST_CFA_OFFSET -(4 + \pop)
128 /* all the rest are no-op */
135 .macro REG_TO_PTGS reg
137 .macro SET_KERNEL_GS reg
140 #else /* CONFIG_X86_32_LAZY_GS */
144 CFI_ADJUST_CFA_OFFSET 4
145 /*CFI_REL_OFFSET gs, 0*/
150 CFI_ADJUST_CFA_OFFSET -4
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_ADJUST_CFA_OFFSET 4
200 /*CFI_REL_OFFSET fs, 0;*/
202 CFI_ADJUST_CFA_OFFSET 4
203 /*CFI_REL_OFFSET es, 0;*/
205 CFI_ADJUST_CFA_OFFSET 4
206 /*CFI_REL_OFFSET ds, 0;*/
208 CFI_ADJUST_CFA_OFFSET 4
209 CFI_REL_OFFSET eax, 0
211 CFI_ADJUST_CFA_OFFSET 4
212 CFI_REL_OFFSET ebp, 0
214 CFI_ADJUST_CFA_OFFSET 4
215 CFI_REL_OFFSET edi, 0
217 CFI_ADJUST_CFA_OFFSET 4
218 CFI_REL_OFFSET esi, 0
220 CFI_ADJUST_CFA_OFFSET 4
221 CFI_REL_OFFSET edx, 0
223 CFI_ADJUST_CFA_OFFSET 4
224 CFI_REL_OFFSET ecx, 0
226 CFI_ADJUST_CFA_OFFSET 4
227 CFI_REL_OFFSET ebx, 0
228 movl $(__USER_DS), %edx
231 movl $(__KERNEL_PERCPU), %edx
236 .macro RESTORE_INT_REGS
238 CFI_ADJUST_CFA_OFFSET -4
241 CFI_ADJUST_CFA_OFFSET -4
244 CFI_ADJUST_CFA_OFFSET -4
247 CFI_ADJUST_CFA_OFFSET -4
250 CFI_ADJUST_CFA_OFFSET -4
253 CFI_ADJUST_CFA_OFFSET -4
256 CFI_ADJUST_CFA_OFFSET -4
260 .macro RESTORE_REGS pop=0
263 CFI_ADJUST_CFA_OFFSET -4
266 CFI_ADJUST_CFA_OFFSET -4
269 CFI_ADJUST_CFA_OFFSET -4
272 .pushsection .fixup, "ax"
279 .section __ex_table, "a"
288 .macro RING0_INT_FRAME
292 /*CFI_OFFSET cs, -2*4;*/
296 .macro RING0_EC_FRAME
300 /*CFI_OFFSET cs, -2*4;*/
304 .macro RING0_PTREGS_FRAME
307 CFI_DEF_CFA esp, PT_OLDESP-PT_EBX
308 /*CFI_OFFSET cs, PT_CS-PT_OLDESP;*/
309 CFI_OFFSET eip, PT_EIP-PT_OLDESP
310 /*CFI_OFFSET es, PT_ES-PT_OLDESP;*/
311 /*CFI_OFFSET ds, PT_DS-PT_OLDESP;*/
312 CFI_OFFSET eax, PT_EAX-PT_OLDESP
313 CFI_OFFSET ebp, PT_EBP-PT_OLDESP
314 CFI_OFFSET edi, PT_EDI-PT_OLDESP
315 CFI_OFFSET esi, PT_ESI-PT_OLDESP
316 CFI_OFFSET edx, PT_EDX-PT_OLDESP
317 CFI_OFFSET ecx, PT_ECX-PT_OLDESP
318 CFI_OFFSET ebx, PT_EBX-PT_OLDESP
324 CFI_ADJUST_CFA_OFFSET 4
326 GET_THREAD_INFO(%ebp)
328 CFI_ADJUST_CFA_OFFSET -4
329 pushl $0x0202 # Reset kernel eflags
330 CFI_ADJUST_CFA_OFFSET 4
332 CFI_ADJUST_CFA_OFFSET -4
338 * Return to user mode is not as complex as all this looks,
339 * but we want the default path for a system call return to
340 * go as quickly as possible which is why some of this is
341 * less clear than it otherwise should be.
344 # userspace resumption stub bypassing syscall exit tracing
348 preempt_stop(CLBR_ANY)
350 GET_THREAD_INFO(%ebp)
352 movl PT_EFLAGS(%esp), %eax # mix EFLAGS and CS
353 movb PT_CS(%esp), %al
354 andl $(X86_EFLAGS_VM | SEGMENT_RPL_MASK), %eax
356 jb resume_kernel # not returning to v8086 or userspace
358 ENTRY(resume_userspace)
360 DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt
361 # setting need_resched or sigpending
362 # between sampling and the iret
364 movl TI_flags(%ebp), %ecx
365 andl $_TIF_WORK_MASK, %ecx # is there any work to be done on
366 # int/exception return?
369 END(ret_from_exception)
371 #ifdef CONFIG_PREEMPT
373 DISABLE_INTERRUPTS(CLBR_ANY)
374 cmpl $0,TI_preempt_count(%ebp) # non-zero preempt_count ?
377 movl TI_flags(%ebp), %ecx # need_resched set ?
378 testb $_TIF_NEED_RESCHED, %cl
380 testl $X86_EFLAGS_IF,PT_EFLAGS(%esp) # interrupts off (exception path) ?
382 call preempt_schedule_irq
388 /* SYSENTER_RETURN points to after the "sysenter" instruction in
389 the vsyscall page. See vsyscall-sysentry.S, which defines the symbol. */
391 # sysenter call handler stub
392 ENTRY(ia32_sysenter_target)
396 CFI_REGISTER esp, ebp
397 movl TSS_sysenter_sp0(%esp),%esp
400 * Interrupts are disabled here, but we can't trace it until
401 * enough kernel state to call TRACE_IRQS_OFF can be called - but
402 * we immediately enable interrupts at that point anyway.
405 CFI_ADJUST_CFA_OFFSET 4
406 /*CFI_REL_OFFSET ss, 0*/
408 CFI_ADJUST_CFA_OFFSET 4
409 CFI_REL_OFFSET esp, 0
411 orl $X86_EFLAGS_IF, (%esp)
412 CFI_ADJUST_CFA_OFFSET 4
414 CFI_ADJUST_CFA_OFFSET 4
415 /*CFI_REL_OFFSET cs, 0*/
417 * Push current_thread_info()->sysenter_return to the stack.
418 * A tiny bit of offset fixup is necessary - 4*4 means the 4 words
419 * pushed above; +8 corresponds to copy_thread's esp0 setting.
421 pushl (TI_sysenter_return-THREAD_SIZE+8+4*4)(%esp)
422 CFI_ADJUST_CFA_OFFSET 4
423 CFI_REL_OFFSET eip, 0
426 CFI_ADJUST_CFA_OFFSET 4
428 ENABLE_INTERRUPTS(CLBR_NONE)
431 * Load the potential sixth argument from user stack.
432 * Careful about security.
434 cmpl $__PAGE_OFFSET-3,%ebp
437 movl %ebp,PT_EBP(%esp)
438 .section __ex_table,"a"
440 .long 1b,syscall_fault
443 GET_THREAD_INFO(%ebp)
445 /* Note, _TIF_SECCOMP is bit number 8, and so it needs testw and not testb */
446 testw $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%ebp)
449 cmpl $(nr_syscalls), %eax
451 call *sys_call_table(,%eax,4)
452 movl %eax,PT_EAX(%esp)
454 DISABLE_INTERRUPTS(CLBR_ANY)
456 movl TI_flags(%ebp), %ecx
457 testw $_TIF_ALLWORK_MASK, %cx
460 /* if something modifies registers it must also disable sysexit */
461 movl PT_EIP(%esp), %edx
462 movl PT_OLDESP(%esp), %ecx
465 1: mov PT_FS(%esp), %fs
467 ENABLE_INTERRUPTS_SYSEXIT
469 #ifdef CONFIG_AUDITSYSCALL
471 testw $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags(%ebp)
472 jnz syscall_trace_entry
474 CFI_ADJUST_CFA_OFFSET -4
475 /* %esi already in 8(%esp) 6th arg: 4th syscall arg */
476 /* %edx already in 4(%esp) 5th arg: 3rd syscall arg */
477 /* %ecx already in 0(%esp) 4th arg: 2nd syscall arg */
478 movl %ebx,%ecx /* 3rd arg: 1st syscall arg */
479 movl %eax,%edx /* 2nd arg: syscall number */
480 movl $AUDIT_ARCH_I386,%eax /* 1st arg: audit arch */
481 call audit_syscall_entry
483 CFI_ADJUST_CFA_OFFSET 4
484 movl PT_EAX(%esp),%eax /* reload syscall number */
488 testw $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT), %cx
489 jne syscall_exit_work
491 ENABLE_INTERRUPTS(CLBR_ANY)
492 movl %eax,%edx /* second arg, syscall return value */
493 cmpl $0,%eax /* is it < 0? */
494 setl %al /* 1 if so, 0 if not */
495 movzbl %al,%eax /* zero-extend that */
496 inc %eax /* first arg, 0->1(AUDITSC_SUCCESS), 1->2(AUDITSC_FAILURE) */
497 call audit_syscall_exit
498 DISABLE_INTERRUPTS(CLBR_ANY)
500 movl TI_flags(%ebp), %ecx
501 testw $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT), %cx
502 jne syscall_exit_work
503 movl PT_EAX(%esp),%eax /* reload syscall return value */
508 .pushsection .fixup,"ax"
509 2: movl $0,PT_FS(%esp)
511 .section __ex_table,"a"
516 ENDPROC(ia32_sysenter_target)
518 # system call handler stub
520 RING0_INT_FRAME # can't unwind into user space anyway
521 pushl %eax # save orig_eax
522 CFI_ADJUST_CFA_OFFSET 4
524 GET_THREAD_INFO(%ebp)
525 # system call tracing in operation / emulation
526 /* Note, _TIF_SECCOMP is bit number 8, and so it needs testw and not testb */
527 testw $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%ebp)
528 jnz syscall_trace_entry
529 cmpl $(nr_syscalls), %eax
532 call *sys_call_table(,%eax,4)
533 movl %eax,PT_EAX(%esp) # store the return value
536 DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt
537 # setting need_resched or sigpending
538 # between sampling and the iret
540 movl TI_flags(%ebp), %ecx
541 testw $_TIF_ALLWORK_MASK, %cx # current->work
542 jne syscall_exit_work
545 movl PT_EFLAGS(%esp), %eax # mix EFLAGS, SS and CS
546 # Warning: PT_OLDSS(%esp) contains the wrong/random values if we
547 # are returning to the kernel.
548 # See comments in process.c:copy_thread() for details.
549 movb PT_OLDSS(%esp), %ah
550 movb PT_CS(%esp), %al
551 andl $(X86_EFLAGS_VM | (SEGMENT_TI_MASK << 8) | SEGMENT_RPL_MASK), %eax
552 cmpl $((SEGMENT_LDT << 8) | USER_RPL), %eax
554 je ldt_ss # returning to user-space with LDT SS
557 restore_nocheck_notrace:
558 RESTORE_REGS 4 # skip orig_eax/error_code
559 CFI_ADJUST_CFA_OFFSET -4
564 pushl $0 # no error code
568 .section __ex_table,"a"
570 .long irq_return,iret_exc
575 larl PT_OLDSS(%esp), %eax
577 testl $0x00400000, %eax # returning to 32bit stack?
578 jnz restore_nocheck # allright, normal return
580 #ifdef CONFIG_PARAVIRT
582 * The kernel can't run on a non-flat stack if paravirt mode
583 * is active. Rather than try to fixup the high bits of
584 * ESP, bypass this code entirely. This may break DOSemu
585 * and/or Wine support in a paravirt VM, although the option
586 * is still available to implement the setting of the high
587 * 16-bits in the INTERRUPT_RETURN paravirt-op.
589 cmpl $0, pv_info+PARAVIRT_enabled
593 /* If returning to userspace with 16bit stack,
594 * try to fix the higher word of ESP, as the CPU
596 * This is an "official" bug of all the x86-compatible
597 * CPUs, which we can try to work around to make
598 * dosemu and wine happy. */
599 movl PT_OLDESP(%esp), %eax
601 call patch_espfix_desc
603 CFI_ADJUST_CFA_OFFSET 4
605 CFI_ADJUST_CFA_OFFSET 4
606 DISABLE_INTERRUPTS(CLBR_EAX)
609 CFI_ADJUST_CFA_OFFSET -8
614 # perform work that needs to be done immediately before resumption
616 RING0_PTREGS_FRAME # can't unwind into user space anyway
618 testb $_TIF_NEED_RESCHED, %cl
623 DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt
624 # setting need_resched or sigpending
625 # between sampling and the iret
627 movl TI_flags(%ebp), %ecx
628 andl $_TIF_WORK_MASK, %ecx # is there any work to be done other
629 # than syscall tracing?
631 testb $_TIF_NEED_RESCHED, %cl
634 work_notifysig: # deal with pending signals and
635 # notify-resume requests
637 testl $X86_EFLAGS_VM, PT_EFLAGS(%esp)
639 jne work_notifysig_v86 # returning to kernel-space or
642 call do_notify_resume
643 jmp resume_userspace_sig
647 pushl %ecx # save ti_flags for do_notify_resume
648 CFI_ADJUST_CFA_OFFSET 4
649 call save_v86_state # %eax contains pt_regs pointer
651 CFI_ADJUST_CFA_OFFSET -4
657 call do_notify_resume
658 jmp resume_userspace_sig
661 # perform syscall exit tracing
664 movl $-ENOSYS,PT_EAX(%esp)
666 call syscall_trace_enter
667 /* What it returned is what we'll actually use. */
668 cmpl $(nr_syscalls), %eax
671 END(syscall_trace_entry)
673 # perform syscall exit tracing
676 testb $_TIF_WORK_SYSCALL_EXIT, %cl
679 ENABLE_INTERRUPTS(CLBR_ANY) # could let syscall_trace_leave() call
682 call syscall_trace_leave
684 END(syscall_exit_work)
687 RING0_INT_FRAME # can't unwind into user space anyway
689 GET_THREAD_INFO(%ebp)
690 movl $-EFAULT,PT_EAX(%esp)
695 movl $-ENOSYS,PT_EAX(%esp)
701 * System calls that need a pt_regs pointer.
703 #define PTREGSCALL(name) \
714 PTREGSCALL(sigaltstack)
715 PTREGSCALL(sigreturn)
716 PTREGSCALL(rt_sigreturn)
720 .macro FIXUP_ESPFIX_STACK
721 /* since we are on a wrong stack, we cant make it a C code :( */
722 PER_CPU(gdt_page, %ebx)
723 GET_DESC_BASE(GDT_ENTRY_ESPFIX_SS, %ebx, %eax, %ax, %al, %ah)
726 CFI_ADJUST_CFA_OFFSET 4
728 CFI_ADJUST_CFA_OFFSET 4
730 CFI_ADJUST_CFA_OFFSET -8
732 .macro UNWIND_ESPFIX_STACK
734 /* see if on espfix stack */
735 cmpw $__ESPFIX_SS, %ax
737 movl $__KERNEL_DS, %eax
740 /* switch to normal stack */
746 * Build the entry stubs and pointer table with some assembler magic.
747 * We pack 7 stubs into a single 32-byte chunk, which will fit in a
748 * single cache line on all modern x86 implementations.
750 .section .init.rodata,"a"
754 .p2align CONFIG_X86_L1_CACHE_SHIFT
755 ENTRY(irq_entries_start)
757 vector=FIRST_EXTERNAL_VECTOR
758 .rept (NR_VECTORS-FIRST_EXTERNAL_VECTOR+6)/7
761 .if vector < NR_VECTORS
762 .if vector <> FIRST_EXTERNAL_VECTOR
763 CFI_ADJUST_CFA_OFFSET -4
765 1: pushl $(~vector+0x80) /* Note: always in signed byte range */
766 CFI_ADJUST_CFA_OFFSET 4
767 .if ((vector-FIRST_EXTERNAL_VECTOR)%7) <> 6
776 2: jmp common_interrupt
778 END(irq_entries_start)
785 * the CPU automatically disables interrupts when executing an IRQ vector,
786 * so IRQ-flags tracing has to follow that:
788 .p2align CONFIG_X86_L1_CACHE_SHIFT
790 addl $-0x80,(%esp) /* Adjust vector into the [-256,-1] range */
796 ENDPROC(common_interrupt)
799 #define BUILD_INTERRUPT3(name, nr, fn) \
803 CFI_ADJUST_CFA_OFFSET 4; \
812 #define BUILD_INTERRUPT(name, nr) BUILD_INTERRUPT3(name, nr, smp_##name)
814 /* The include is where all of the SMP etc. interrupts come from */
815 #include <asm/entry_arch.h>
817 ENTRY(coprocessor_error)
820 CFI_ADJUST_CFA_OFFSET 4
821 pushl $do_coprocessor_error
822 CFI_ADJUST_CFA_OFFSET 4
825 END(coprocessor_error)
827 ENTRY(simd_coprocessor_error)
830 CFI_ADJUST_CFA_OFFSET 4
831 pushl $do_simd_coprocessor_error
832 CFI_ADJUST_CFA_OFFSET 4
835 END(simd_coprocessor_error)
837 ENTRY(device_not_available)
839 pushl $-1 # mark this as an int
840 CFI_ADJUST_CFA_OFFSET 4
841 pushl $do_device_not_available
842 CFI_ADJUST_CFA_OFFSET 4
845 END(device_not_available)
847 #ifdef CONFIG_PARAVIRT
850 .section __ex_table,"a"
852 .long native_iret, iret_exc
856 ENTRY(native_irq_enable_sysexit)
859 END(native_irq_enable_sysexit)
865 CFI_ADJUST_CFA_OFFSET 4
867 CFI_ADJUST_CFA_OFFSET 4
875 CFI_ADJUST_CFA_OFFSET 4
877 CFI_ADJUST_CFA_OFFSET 4
885 CFI_ADJUST_CFA_OFFSET 4
887 CFI_ADJUST_CFA_OFFSET 4
892 ENTRY(coprocessor_segment_overrun)
895 CFI_ADJUST_CFA_OFFSET 4
896 pushl $do_coprocessor_segment_overrun
897 CFI_ADJUST_CFA_OFFSET 4
900 END(coprocessor_segment_overrun)
904 pushl $do_invalid_TSS
905 CFI_ADJUST_CFA_OFFSET 4
910 ENTRY(segment_not_present)
912 pushl $do_segment_not_present
913 CFI_ADJUST_CFA_OFFSET 4
916 END(segment_not_present)
920 pushl $do_stack_segment
921 CFI_ADJUST_CFA_OFFSET 4
926 ENTRY(alignment_check)
928 pushl $do_alignment_check
929 CFI_ADJUST_CFA_OFFSET 4
936 pushl $0 # no error code
937 CFI_ADJUST_CFA_OFFSET 4
938 pushl $do_divide_error
939 CFI_ADJUST_CFA_OFFSET 4
944 #ifdef CONFIG_X86_MCE
948 CFI_ADJUST_CFA_OFFSET 4
949 pushl machine_check_vector
950 CFI_ADJUST_CFA_OFFSET 4
956 ENTRY(spurious_interrupt_bug)
959 CFI_ADJUST_CFA_OFFSET 4
960 pushl $do_spurious_interrupt_bug
961 CFI_ADJUST_CFA_OFFSET 4
964 END(spurious_interrupt_bug)
966 ENTRY(kernel_thread_helper)
967 pushl $0 # fake return address for unwinder
971 CFI_ADJUST_CFA_OFFSET 4
974 CFI_ADJUST_CFA_OFFSET 4
976 ud2 # padding for call trace
978 ENDPROC(kernel_thread_helper)
981 /* Xen doesn't set %esp to be precisely what the normal sysenter
982 entrypoint expects, so fix it up before using the normal path. */
983 ENTRY(xen_sysenter_target)
985 addl $5*4, %esp /* remove xen-provided frame */
986 CFI_ADJUST_CFA_OFFSET -5*4
987 jmp sysenter_past_esp
990 ENTRY(xen_hypervisor_callback)
993 CFI_ADJUST_CFA_OFFSET 4
997 /* Check to see if we got the event in the critical
998 region in xen_iret_direct, after we've reenabled
999 events and checked for pending events. This simulates
1000 iret instruction's behaviour where it delivers a
1001 pending interrupt when enabling interrupts. */
1002 movl PT_EIP(%esp),%eax
1003 cmpl $xen_iret_start_crit,%eax
1005 cmpl $xen_iret_end_crit,%eax
1008 jmp xen_iret_crit_fixup
1010 ENTRY(xen_do_upcall)
1012 call xen_evtchn_do_upcall
1015 ENDPROC(xen_hypervisor_callback)
1017 # Hypervisor uses this for application faults while it executes.
1018 # We get here for two reasons:
1019 # 1. Fault while reloading DS, ES, FS or GS
1020 # 2. Fault while executing IRET
1021 # Category 1 we fix up by reattempting the load, and zeroing the segment
1022 # register if the load fails.
1023 # Category 2 we fix up by jumping to do_iret_error. We cannot use the
1024 # normal Linux return path in this case because if we use the IRET hypercall
1025 # to pop the stack frame we end up in an infinite loop of failsafe callbacks.
1026 # We distinguish between categories by maintaining a status value in EAX.
1027 ENTRY(xen_failsafe_callback)
1030 CFI_ADJUST_CFA_OFFSET 4
1038 CFI_ADJUST_CFA_OFFSET -4
1040 CFI_ADJUST_CFA_OFFSET -16
1043 jmp iret_exc # EAX != 0 => Category 2 (Bad IRET)
1044 5: pushl $0 # EAX == 0 => Category 1 (Bad segment)
1045 CFI_ADJUST_CFA_OFFSET 4
1047 jmp ret_from_exception
1050 .section .fixup,"ax"
1064 .section __ex_table,"a"
1071 ENDPROC(xen_failsafe_callback)
1073 #endif /* CONFIG_XEN */
1075 #ifdef CONFIG_FUNCTION_TRACER
1076 #ifdef CONFIG_DYNAMIC_FTRACE
1082 ENTRY(ftrace_caller)
1083 cmpl $0, function_trace_stop
1089 movl 0xc(%esp), %eax
1090 movl 0x4(%ebp), %edx
1091 subl $MCOUNT_INSN_SIZE, %eax
1100 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1101 .globl ftrace_graph_call
1111 #else /* ! CONFIG_DYNAMIC_FTRACE */
1114 cmpl $0, function_trace_stop
1117 cmpl $ftrace_stub, ftrace_trace_function
1119 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1120 cmpl $ftrace_stub, ftrace_graph_return
1121 jnz ftrace_graph_caller
1123 cmpl $ftrace_graph_entry_stub, ftrace_graph_entry
1124 jnz ftrace_graph_caller
1130 /* taken from glibc */
1135 movl 0xc(%esp), %eax
1136 movl 0x4(%ebp), %edx
1137 subl $MCOUNT_INSN_SIZE, %eax
1139 call *ftrace_trace_function
1146 #endif /* CONFIG_DYNAMIC_FTRACE */
1147 #endif /* CONFIG_FUNCTION_TRACER */
1149 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1150 ENTRY(ftrace_graph_caller)
1151 cmpl $0, function_trace_stop
1157 movl 0xc(%esp), %edx
1159 subl $MCOUNT_INSN_SIZE, %edx
1160 call prepare_ftrace_return
1165 END(ftrace_graph_caller)
1167 .globl return_to_handler
1173 call ftrace_return_to_handler
1174 movl %eax, 0xc(%esp)
1181 .section .rodata,"a"
1182 #include "syscall_table_32.S"
1184 syscall_table_size=(.-sys_call_table)
1187 * Some functions should be protected against kprobes
1189 .pushsection .kprobes.text, "ax"
1193 pushl $do_page_fault
1194 CFI_ADJUST_CFA_OFFSET 4
1197 /* the function address is in %gs's slot on the stack */
1199 CFI_ADJUST_CFA_OFFSET 4
1200 /*CFI_REL_OFFSET fs, 0*/
1202 CFI_ADJUST_CFA_OFFSET 4
1203 /*CFI_REL_OFFSET es, 0*/
1205 CFI_ADJUST_CFA_OFFSET 4
1206 /*CFI_REL_OFFSET ds, 0*/
1208 CFI_ADJUST_CFA_OFFSET 4
1209 CFI_REL_OFFSET eax, 0
1211 CFI_ADJUST_CFA_OFFSET 4
1212 CFI_REL_OFFSET ebp, 0
1214 CFI_ADJUST_CFA_OFFSET 4
1215 CFI_REL_OFFSET edi, 0
1217 CFI_ADJUST_CFA_OFFSET 4
1218 CFI_REL_OFFSET esi, 0
1220 CFI_ADJUST_CFA_OFFSET 4
1221 CFI_REL_OFFSET edx, 0
1223 CFI_ADJUST_CFA_OFFSET 4
1224 CFI_REL_OFFSET ecx, 0
1226 CFI_ADJUST_CFA_OFFSET 4
1227 CFI_REL_OFFSET ebx, 0
1229 movl $(__KERNEL_PERCPU), %ecx
1233 movl PT_GS(%esp), %edi # get the function address
1234 movl PT_ORIG_EAX(%esp), %edx # get the error code
1235 movl $-1, PT_ORIG_EAX(%esp) # no syscall to restart
1238 movl $(__USER_DS), %ecx
1242 movl %esp,%eax # pt_regs pointer
1244 jmp ret_from_exception
1249 * Debug traps and NMI can happen at the one SYSENTER instruction
1250 * that sets up the real kernel stack. Check here, since we can't
1251 * allow the wrong stack to be used.
1253 * "TSS_sysenter_sp0+12" is because the NMI/debug handler will have
1254 * already pushed 3 words if it hits on the sysenter instruction:
1255 * eflags, cs and eip.
1257 * We just load the right stack, and push the three (known) values
1258 * by hand onto the new stack - while updating the return eip past
1259 * the instruction that would have done it for sysenter.
1261 .macro FIX_STACK offset ok label
1262 cmpw $__KERNEL_CS, 4(%esp)
1265 movl TSS_sysenter_sp0 + \offset(%esp), %esp
1269 CFI_ADJUST_CFA_OFFSET 4
1271 CFI_ADJUST_CFA_OFFSET 4
1272 pushl $sysenter_past_esp
1273 CFI_ADJUST_CFA_OFFSET 4
1274 CFI_REL_OFFSET eip, 0
1279 cmpl $ia32_sysenter_target,(%esp)
1280 jne debug_stack_correct
1281 FIX_STACK 12, debug_stack_correct, debug_esp_fix_insn
1282 debug_stack_correct:
1283 pushl $-1 # mark this as an int
1284 CFI_ADJUST_CFA_OFFSET 4
1287 xorl %edx,%edx # error code 0
1288 movl %esp,%eax # pt_regs pointer
1290 jmp ret_from_exception
1295 * NMI is doubly nasty. It can happen _while_ we're handling
1296 * a debug fault, and the debug fault hasn't yet been able to
1297 * clear up the stack. So we first check whether we got an
1298 * NMI on the sysenter entry path, but after that we need to
1299 * check whether we got an NMI on the debug path where the debug
1300 * fault happened on the sysenter path.
1305 CFI_ADJUST_CFA_OFFSET 4
1307 cmpw $__ESPFIX_SS, %ax
1309 CFI_ADJUST_CFA_OFFSET -4
1311 cmpl $ia32_sysenter_target,(%esp)
1314 CFI_ADJUST_CFA_OFFSET 4
1316 /* Do not access memory above the end of our stack page,
1317 * it might not exist.
1319 andl $(THREAD_SIZE-1),%eax
1320 cmpl $(THREAD_SIZE-20),%eax
1322 CFI_ADJUST_CFA_OFFSET -4
1323 jae nmi_stack_correct
1324 cmpl $ia32_sysenter_target,12(%esp)
1325 je nmi_debug_stack_check
1327 /* We have a RING0_INT_FRAME here */
1329 CFI_ADJUST_CFA_OFFSET 4
1331 xorl %edx,%edx # zero error code
1332 movl %esp,%eax # pt_regs pointer
1334 jmp restore_nocheck_notrace
1339 FIX_STACK 12, nmi_stack_correct, 1
1340 jmp nmi_stack_correct
1342 nmi_debug_stack_check:
1343 /* We have a RING0_INT_FRAME here */
1344 cmpw $__KERNEL_CS,16(%esp)
1345 jne nmi_stack_correct
1347 jb nmi_stack_correct
1348 cmpl $debug_esp_fix_insn,(%esp)
1349 ja nmi_stack_correct
1350 FIX_STACK 24, nmi_stack_correct, 1
1351 jmp nmi_stack_correct
1354 /* We have a RING0_INT_FRAME here.
1356 * create the pointer to lss back
1359 CFI_ADJUST_CFA_OFFSET 4
1361 CFI_ADJUST_CFA_OFFSET 4
1363 /* copy the iret frame of 12 bytes */
1366 CFI_ADJUST_CFA_OFFSET 4
1369 CFI_ADJUST_CFA_OFFSET 4
1371 FIXUP_ESPFIX_STACK # %eax == %esp
1372 xorl %edx,%edx # zero error code
1375 lss 12+4(%esp), %esp # back to espfix stack
1376 CFI_ADJUST_CFA_OFFSET -24
1383 pushl $-1 # mark this as an int
1384 CFI_ADJUST_CFA_OFFSET 4
1387 xorl %edx,%edx # zero error code
1388 movl %esp,%eax # pt_regs pointer
1390 jmp ret_from_exception
1394 ENTRY(general_protection)
1396 pushl $do_general_protection
1397 CFI_ADJUST_CFA_OFFSET 4
1400 END(general_protection)
1403 * End of kprobes section