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, kernel_preemption
376 cmpl $0,TI_preempt_count(%ebp) # non-zero preempt_count ?
379 movl TI_flags(%ebp), %ecx # need_resched set ?
380 testb $_TIF_NEED_RESCHED, %cl
382 testl $X86_EFLAGS_IF,PT_EFLAGS(%esp) # interrupts off (exception path) ?
384 DISABLE_INTERRUPTS(CLBR_ANY)
386 call preempt_schedule_irq
392 /* SYSENTER_RETURN points to after the "sysenter" instruction in
393 the vsyscall page. See vsyscall-sysentry.S, which defines the symbol. */
395 # sysenter call handler stub
396 ENTRY(ia32_sysenter_target)
400 CFI_REGISTER esp, ebp
401 movl TSS_sysenter_sp0(%esp),%esp
404 * Interrupts are disabled here, but we can't trace it until
405 * enough kernel state to call TRACE_IRQS_OFF can be called - but
406 * we immediately enable interrupts at that point anyway.
409 CFI_ADJUST_CFA_OFFSET 4
410 /*CFI_REL_OFFSET ss, 0*/
412 CFI_ADJUST_CFA_OFFSET 4
413 CFI_REL_OFFSET esp, 0
415 orl $X86_EFLAGS_IF, (%esp)
416 CFI_ADJUST_CFA_OFFSET 4
418 CFI_ADJUST_CFA_OFFSET 4
419 /*CFI_REL_OFFSET cs, 0*/
421 * Push current_thread_info()->sysenter_return to the stack.
422 * A tiny bit of offset fixup is necessary - 4*4 means the 4 words
423 * pushed above; +8 corresponds to copy_thread's esp0 setting.
425 pushl (TI_sysenter_return-THREAD_SIZE+8+4*4)(%esp)
426 CFI_ADJUST_CFA_OFFSET 4
427 CFI_REL_OFFSET eip, 0
430 CFI_ADJUST_CFA_OFFSET 4
432 ENABLE_INTERRUPTS(CLBR_NONE)
435 * Load the potential sixth argument from user stack.
436 * Careful about security.
438 cmpl $__PAGE_OFFSET-3,%ebp
441 movl %ebp,PT_EBP(%esp)
442 .section __ex_table,"a"
444 .long 1b,syscall_fault
447 GET_THREAD_INFO(%ebp)
449 testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%ebp)
452 cmpl $(nr_syscalls), %eax
454 call *sys_call_table(,%eax,4)
455 movl %eax,PT_EAX(%esp)
457 DISABLE_INTERRUPTS(CLBR_ANY)
459 movl TI_flags(%ebp), %ecx
460 testl $_TIF_ALLWORK_MASK, %ecx
463 /* if something modifies registers it must also disable sysexit */
464 movl PT_EIP(%esp), %edx
465 movl PT_OLDESP(%esp), %ecx
468 1: mov PT_FS(%esp), %fs
470 ENABLE_INTERRUPTS_SYSEXIT
472 #ifdef CONFIG_AUDITSYSCALL
474 testl $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags(%ebp)
475 jnz syscall_trace_entry
477 CFI_ADJUST_CFA_OFFSET -4
478 /* %esi already in 8(%esp) 6th arg: 4th syscall arg */
479 /* %edx already in 4(%esp) 5th arg: 3rd syscall arg */
480 /* %ecx already in 0(%esp) 4th arg: 2nd syscall arg */
481 movl %ebx,%ecx /* 3rd arg: 1st syscall arg */
482 movl %eax,%edx /* 2nd arg: syscall number */
483 movl $AUDIT_ARCH_I386,%eax /* 1st arg: audit arch */
484 call audit_syscall_entry
486 CFI_ADJUST_CFA_OFFSET 4
487 movl PT_EAX(%esp),%eax /* reload syscall number */
491 testl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT), %ecx
492 jne syscall_exit_work
494 ENABLE_INTERRUPTS(CLBR_ANY)
495 movl %eax,%edx /* second arg, syscall return value */
496 cmpl $0,%eax /* is it < 0? */
497 setl %al /* 1 if so, 0 if not */
498 movzbl %al,%eax /* zero-extend that */
499 inc %eax /* first arg, 0->1(AUDITSC_SUCCESS), 1->2(AUDITSC_FAILURE) */
500 call audit_syscall_exit
501 DISABLE_INTERRUPTS(CLBR_ANY)
503 movl TI_flags(%ebp), %ecx
504 testl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT), %ecx
505 jne syscall_exit_work
506 movl PT_EAX(%esp),%eax /* reload syscall return value */
511 .pushsection .fixup,"ax"
512 2: movl $0,PT_FS(%esp)
514 .section __ex_table,"a"
519 ENDPROC(ia32_sysenter_target)
521 # system call handler stub
523 RING0_INT_FRAME # can't unwind into user space anyway
524 pushl %eax # save orig_eax
525 CFI_ADJUST_CFA_OFFSET 4
527 GET_THREAD_INFO(%ebp)
528 # system call tracing in operation / emulation
529 testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%ebp)
530 jnz syscall_trace_entry
531 cmpl $(nr_syscalls), %eax
534 call *sys_call_table(,%eax,4)
535 movl %eax,PT_EAX(%esp) # store the return value
538 DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt
539 # setting need_resched or sigpending
540 # between sampling and the iret
542 movl TI_flags(%ebp), %ecx
543 testl $_TIF_ALLWORK_MASK, %ecx # current->work
544 jne syscall_exit_work
547 movl PT_EFLAGS(%esp), %eax # mix EFLAGS, SS and CS
548 # Warning: PT_OLDSS(%esp) contains the wrong/random values if we
549 # are returning to the kernel.
550 # See comments in process.c:copy_thread() for details.
551 movb PT_OLDSS(%esp), %ah
552 movb PT_CS(%esp), %al
553 andl $(X86_EFLAGS_VM | (SEGMENT_TI_MASK << 8) | SEGMENT_RPL_MASK), %eax
554 cmpl $((SEGMENT_LDT << 8) | USER_RPL), %eax
556 je ldt_ss # returning to user-space with LDT SS
559 restore_nocheck_notrace:
560 RESTORE_REGS 4 # skip orig_eax/error_code
561 CFI_ADJUST_CFA_OFFSET -4
566 pushl $0 # no error code
570 .section __ex_table,"a"
572 .long irq_return,iret_exc
577 larl PT_OLDSS(%esp), %eax
579 testl $0x00400000, %eax # returning to 32bit stack?
580 jnz restore_nocheck # allright, normal return
582 #ifdef CONFIG_PARAVIRT
584 * The kernel can't run on a non-flat stack if paravirt mode
585 * is active. Rather than try to fixup the high bits of
586 * ESP, bypass this code entirely. This may break DOSemu
587 * and/or Wine support in a paravirt VM, although the option
588 * is still available to implement the setting of the high
589 * 16-bits in the INTERRUPT_RETURN paravirt-op.
591 cmpl $0, pv_info+PARAVIRT_enabled
595 /* If returning to userspace with 16bit stack,
596 * try to fix the higher word of ESP, as the CPU
598 * This is an "official" bug of all the x86-compatible
599 * CPUs, which we can try to work around to make
600 * dosemu and wine happy. */
601 movl PT_OLDESP(%esp), %eax
603 call patch_espfix_desc
605 CFI_ADJUST_CFA_OFFSET 4
607 CFI_ADJUST_CFA_OFFSET 4
608 DISABLE_INTERRUPTS(CLBR_EAX)
611 CFI_ADJUST_CFA_OFFSET -8
616 # perform work that needs to be done immediately before resumption
618 RING0_PTREGS_FRAME # can't unwind into user space anyway
620 testl $(_TIF_NEED_RESCHED), %ecx
625 DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt
626 # setting need_resched or sigpending
627 # between sampling and the iret
628 movl TI_flags(%ebp), %ecx
629 andl $_TIF_WORK_MASK, %ecx # is there any work to be done other
630 # than syscall tracing?
632 testl $(_TIF_NEED_RESCHED), %ecx
635 work_notifysig: # deal with pending signals and
636 # notify-resume requests
638 testl $X86_EFLAGS_VM, PT_EFLAGS(%esp)
640 jne work_notifysig_v86 # returning to kernel-space or
643 call do_notify_resume
644 jmp resume_userspace_sig
648 pushl %ecx # save ti_flags for do_notify_resume
649 CFI_ADJUST_CFA_OFFSET 4
650 call save_v86_state # %eax contains pt_regs pointer
652 CFI_ADJUST_CFA_OFFSET -4
658 call do_notify_resume
659 jmp resume_userspace_sig
662 # perform syscall exit tracing
665 movl $-ENOSYS,PT_EAX(%esp)
667 call syscall_trace_enter
668 /* What it returned is what we'll actually use. */
669 cmpl $(nr_syscalls), %eax
672 END(syscall_trace_entry)
674 # perform syscall exit tracing
677 testl $_TIF_WORK_SYSCALL_EXIT, %ecx
680 ENABLE_INTERRUPTS(CLBR_ANY) # could let syscall_trace_leave() call
683 call syscall_trace_leave
685 END(syscall_exit_work)
688 RING0_INT_FRAME # can't unwind into user space anyway
690 GET_THREAD_INFO(%ebp)
691 movl $-EFAULT,PT_EAX(%esp)
696 movl $-ENOSYS,PT_EAX(%esp)
702 * System calls that need a pt_regs pointer.
704 #define PTREGSCALL(name) \
715 PTREGSCALL(sigaltstack)
716 PTREGSCALL(sigreturn)
717 PTREGSCALL(rt_sigreturn)
721 .macro FIXUP_ESPFIX_STACK
722 /* since we are on a wrong stack, we cant make it a C code :( */
723 PER_CPU(gdt_page, %ebx)
724 GET_DESC_BASE(GDT_ENTRY_ESPFIX_SS, %ebx, %eax, %ax, %al, %ah)
727 CFI_ADJUST_CFA_OFFSET 4
729 CFI_ADJUST_CFA_OFFSET 4
731 CFI_ADJUST_CFA_OFFSET -8
733 .macro UNWIND_ESPFIX_STACK
735 /* see if on espfix stack */
736 cmpw $__ESPFIX_SS, %ax
738 movl $__KERNEL_DS, %eax
741 /* switch to normal stack */
747 * Build the entry stubs and pointer table with some assembler magic.
748 * We pack 7 stubs into a single 32-byte chunk, which will fit in a
749 * single cache line on all modern x86 implementations.
751 .section .init.rodata,"a"
755 .p2align CONFIG_X86_L1_CACHE_SHIFT
756 ENTRY(irq_entries_start)
758 vector=FIRST_EXTERNAL_VECTOR
759 .rept (NR_VECTORS-FIRST_EXTERNAL_VECTOR+6)/7
762 .if vector < NR_VECTORS
763 .if vector <> FIRST_EXTERNAL_VECTOR
764 CFI_ADJUST_CFA_OFFSET -4
766 1: pushl $(~vector+0x80) /* Note: always in signed byte range */
767 CFI_ADJUST_CFA_OFFSET 4
768 .if ((vector-FIRST_EXTERNAL_VECTOR)%7) <> 6
777 2: jmp common_interrupt
779 END(irq_entries_start)
786 * the CPU automatically disables interrupts when executing an IRQ vector,
787 * so IRQ-flags tracing has to follow that:
789 .p2align CONFIG_X86_L1_CACHE_SHIFT
791 addl $-0x80,(%esp) /* Adjust vector into the [-256,-1] range */
797 ENDPROC(common_interrupt)
800 #define BUILD_INTERRUPT3(name, nr, fn) \
804 CFI_ADJUST_CFA_OFFSET 4; \
813 #define BUILD_INTERRUPT(name, nr) BUILD_INTERRUPT3(name, nr, smp_##name)
815 /* The include is where all of the SMP etc. interrupts come from */
816 #include <asm/entry_arch.h>
818 ENTRY(coprocessor_error)
821 CFI_ADJUST_CFA_OFFSET 4
822 pushl $do_coprocessor_error
823 CFI_ADJUST_CFA_OFFSET 4
826 END(coprocessor_error)
828 ENTRY(simd_coprocessor_error)
831 CFI_ADJUST_CFA_OFFSET 4
832 pushl $do_simd_coprocessor_error
833 CFI_ADJUST_CFA_OFFSET 4
836 END(simd_coprocessor_error)
838 ENTRY(device_not_available)
840 pushl $-1 # mark this as an int
841 CFI_ADJUST_CFA_OFFSET 4
842 pushl $do_device_not_available
843 CFI_ADJUST_CFA_OFFSET 4
846 END(device_not_available)
848 #ifdef CONFIG_PARAVIRT
851 .section __ex_table,"a"
853 .long native_iret, iret_exc
857 ENTRY(native_irq_enable_sysexit)
860 END(native_irq_enable_sysexit)
866 CFI_ADJUST_CFA_OFFSET 4
868 CFI_ADJUST_CFA_OFFSET 4
876 CFI_ADJUST_CFA_OFFSET 4
878 CFI_ADJUST_CFA_OFFSET 4
886 CFI_ADJUST_CFA_OFFSET 4
888 CFI_ADJUST_CFA_OFFSET 4
893 ENTRY(coprocessor_segment_overrun)
896 CFI_ADJUST_CFA_OFFSET 4
897 pushl $do_coprocessor_segment_overrun
898 CFI_ADJUST_CFA_OFFSET 4
901 END(coprocessor_segment_overrun)
905 pushl $do_invalid_TSS
906 CFI_ADJUST_CFA_OFFSET 4
911 ENTRY(segment_not_present)
913 pushl $do_segment_not_present
914 CFI_ADJUST_CFA_OFFSET 4
917 END(segment_not_present)
921 pushl $do_stack_segment
922 CFI_ADJUST_CFA_OFFSET 4
927 ENTRY(alignment_check)
929 pushl $do_alignment_check
930 CFI_ADJUST_CFA_OFFSET 4
937 pushl $0 # no error code
938 CFI_ADJUST_CFA_OFFSET 4
939 pushl $do_divide_error
940 CFI_ADJUST_CFA_OFFSET 4
945 #ifdef CONFIG_X86_MCE
949 CFI_ADJUST_CFA_OFFSET 4
950 pushl machine_check_vector
951 CFI_ADJUST_CFA_OFFSET 4
957 ENTRY(spurious_interrupt_bug)
960 CFI_ADJUST_CFA_OFFSET 4
961 pushl $do_spurious_interrupt_bug
962 CFI_ADJUST_CFA_OFFSET 4
965 END(spurious_interrupt_bug)
967 ENTRY(kernel_thread_helper)
968 pushl $0 # fake return address for unwinder
972 CFI_ADJUST_CFA_OFFSET 4
975 CFI_ADJUST_CFA_OFFSET 4
977 ud2 # padding for call trace
979 ENDPROC(kernel_thread_helper)
982 /* Xen doesn't set %esp to be precisely what the normal sysenter
983 entrypoint expects, so fix it up before using the normal path. */
984 ENTRY(xen_sysenter_target)
986 addl $5*4, %esp /* remove xen-provided frame */
987 CFI_ADJUST_CFA_OFFSET -5*4
988 jmp sysenter_past_esp
991 ENTRY(xen_hypervisor_callback)
994 CFI_ADJUST_CFA_OFFSET 4
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)
1031 CFI_ADJUST_CFA_OFFSET 4
1039 CFI_ADJUST_CFA_OFFSET -4
1041 CFI_ADJUST_CFA_OFFSET -16
1044 jmp iret_exc # EAX != 0 => Category 2 (Bad IRET)
1045 5: pushl $0 # EAX == 0 => Category 1 (Bad segment)
1046 CFI_ADJUST_CFA_OFFSET 4
1048 jmp ret_from_exception
1051 .section .fixup,"ax"
1065 .section __ex_table,"a"
1072 ENDPROC(xen_failsafe_callback)
1074 #endif /* CONFIG_XEN */
1076 #ifdef CONFIG_FUNCTION_TRACER
1077 #ifdef CONFIG_DYNAMIC_FTRACE
1083 ENTRY(ftrace_caller)
1084 cmpl $0, function_trace_stop
1090 movl 0xc(%esp), %eax
1091 movl 0x4(%ebp), %edx
1092 subl $MCOUNT_INSN_SIZE, %eax
1101 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1102 .globl ftrace_graph_call
1112 #else /* ! CONFIG_DYNAMIC_FTRACE */
1115 cmpl $0, function_trace_stop
1118 cmpl $ftrace_stub, ftrace_trace_function
1120 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1121 cmpl $ftrace_stub, ftrace_graph_return
1122 jnz ftrace_graph_caller
1124 cmpl $ftrace_graph_entry_stub, ftrace_graph_entry
1125 jnz ftrace_graph_caller
1131 /* taken from glibc */
1136 movl 0xc(%esp), %eax
1137 movl 0x4(%ebp), %edx
1138 subl $MCOUNT_INSN_SIZE, %eax
1140 call *ftrace_trace_function
1147 #endif /* CONFIG_DYNAMIC_FTRACE */
1148 #endif /* CONFIG_FUNCTION_TRACER */
1150 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1151 ENTRY(ftrace_graph_caller)
1152 cmpl $0, function_trace_stop
1158 movl 0xc(%esp), %edx
1160 subl $MCOUNT_INSN_SIZE, %edx
1161 call prepare_ftrace_return
1166 END(ftrace_graph_caller)
1168 .globl return_to_handler
1174 call ftrace_return_to_handler
1175 movl %eax, 0xc(%esp)
1182 .section .rodata,"a"
1183 #include "syscall_table_32.S"
1185 syscall_table_size=(.-sys_call_table)
1188 * Some functions should be protected against kprobes
1190 .pushsection .kprobes.text, "ax"
1194 pushl $do_page_fault
1195 CFI_ADJUST_CFA_OFFSET 4
1198 /* the function address is in %gs's slot on the stack */
1200 CFI_ADJUST_CFA_OFFSET 4
1201 /*CFI_REL_OFFSET fs, 0*/
1203 CFI_ADJUST_CFA_OFFSET 4
1204 /*CFI_REL_OFFSET es, 0*/
1206 CFI_ADJUST_CFA_OFFSET 4
1207 /*CFI_REL_OFFSET ds, 0*/
1209 CFI_ADJUST_CFA_OFFSET 4
1210 CFI_REL_OFFSET eax, 0
1212 CFI_ADJUST_CFA_OFFSET 4
1213 CFI_REL_OFFSET ebp, 0
1215 CFI_ADJUST_CFA_OFFSET 4
1216 CFI_REL_OFFSET edi, 0
1218 CFI_ADJUST_CFA_OFFSET 4
1219 CFI_REL_OFFSET esi, 0
1221 CFI_ADJUST_CFA_OFFSET 4
1222 CFI_REL_OFFSET edx, 0
1224 CFI_ADJUST_CFA_OFFSET 4
1225 CFI_REL_OFFSET ecx, 0
1227 CFI_ADJUST_CFA_OFFSET 4
1228 CFI_REL_OFFSET ebx, 0
1230 movl $(__KERNEL_PERCPU), %ecx
1234 movl PT_GS(%esp), %edi # get the function address
1235 movl PT_ORIG_EAX(%esp), %edx # get the error code
1236 movl $-1, PT_ORIG_EAX(%esp) # no syscall to restart
1239 movl $(__USER_DS), %ecx
1243 movl %esp,%eax # pt_regs pointer
1245 jmp ret_from_exception
1250 * Debug traps and NMI can happen at the one SYSENTER instruction
1251 * that sets up the real kernel stack. Check here, since we can't
1252 * allow the wrong stack to be used.
1254 * "TSS_sysenter_sp0+12" is because the NMI/debug handler will have
1255 * already pushed 3 words if it hits on the sysenter instruction:
1256 * eflags, cs and eip.
1258 * We just load the right stack, and push the three (known) values
1259 * by hand onto the new stack - while updating the return eip past
1260 * the instruction that would have done it for sysenter.
1262 .macro FIX_STACK offset ok label
1263 cmpw $__KERNEL_CS, 4(%esp)
1266 movl TSS_sysenter_sp0 + \offset(%esp), %esp
1270 CFI_ADJUST_CFA_OFFSET 4
1272 CFI_ADJUST_CFA_OFFSET 4
1273 pushl $sysenter_past_esp
1274 CFI_ADJUST_CFA_OFFSET 4
1275 CFI_REL_OFFSET eip, 0
1280 cmpl $ia32_sysenter_target,(%esp)
1281 jne debug_stack_correct
1282 FIX_STACK 12, debug_stack_correct, debug_esp_fix_insn
1283 debug_stack_correct:
1284 pushl $-1 # mark this as an int
1285 CFI_ADJUST_CFA_OFFSET 4
1288 xorl %edx,%edx # error code 0
1289 movl %esp,%eax # pt_regs pointer
1291 jmp ret_from_exception
1296 * NMI is doubly nasty. It can happen _while_ we're handling
1297 * a debug fault, and the debug fault hasn't yet been able to
1298 * clear up the stack. So we first check whether we got an
1299 * NMI on the sysenter entry path, but after that we need to
1300 * check whether we got an NMI on the debug path where the debug
1301 * fault happened on the sysenter path.
1306 CFI_ADJUST_CFA_OFFSET 4
1308 cmpw $__ESPFIX_SS, %ax
1310 CFI_ADJUST_CFA_OFFSET -4
1312 cmpl $ia32_sysenter_target,(%esp)
1315 CFI_ADJUST_CFA_OFFSET 4
1317 /* Do not access memory above the end of our stack page,
1318 * it might not exist.
1320 andl $(THREAD_SIZE-1),%eax
1321 cmpl $(THREAD_SIZE-20),%eax
1323 CFI_ADJUST_CFA_OFFSET -4
1324 jae nmi_stack_correct
1325 cmpl $ia32_sysenter_target,12(%esp)
1326 je nmi_debug_stack_check
1328 /* We have a RING0_INT_FRAME here */
1330 CFI_ADJUST_CFA_OFFSET 4
1332 xorl %edx,%edx # zero error code
1333 movl %esp,%eax # pt_regs pointer
1335 jmp restore_nocheck_notrace
1340 FIX_STACK 12, nmi_stack_correct, 1
1341 jmp nmi_stack_correct
1343 nmi_debug_stack_check:
1344 /* We have a RING0_INT_FRAME here */
1345 cmpw $__KERNEL_CS,16(%esp)
1346 jne nmi_stack_correct
1348 jb nmi_stack_correct
1349 cmpl $debug_esp_fix_insn,(%esp)
1350 ja nmi_stack_correct
1351 FIX_STACK 24, nmi_stack_correct, 1
1352 jmp nmi_stack_correct
1355 /* We have a RING0_INT_FRAME here.
1357 * create the pointer to lss back
1360 CFI_ADJUST_CFA_OFFSET 4
1362 CFI_ADJUST_CFA_OFFSET 4
1364 /* copy the iret frame of 12 bytes */
1367 CFI_ADJUST_CFA_OFFSET 4
1370 CFI_ADJUST_CFA_OFFSET 4
1372 FIXUP_ESPFIX_STACK # %eax == %esp
1373 xorl %edx,%edx # zero error code
1376 lss 12+4(%esp), %esp # back to espfix stack
1377 CFI_ADJUST_CFA_OFFSET -24
1384 pushl $-1 # mark this as an int
1385 CFI_ADJUST_CFA_OFFSET 4
1388 xorl %edx,%edx # zero error code
1389 movl %esp,%eax # pt_regs pointer
1391 jmp ret_from_exception
1395 ENTRY(general_protection)
1397 pushl $do_general_protection
1398 CFI_ADJUST_CFA_OFFSET 4
1401 END(general_protection)
1404 * End of kprobes section