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 * Interrupt exit functions should be protected against kprobes
339 .pushsection .kprobes.text, "ax"
341 * Return to user mode is not as complex as all this looks,
342 * but we want the default path for a system call return to
343 * go as quickly as possible which is why some of this is
344 * less clear than it otherwise should be.
347 # userspace resumption stub bypassing syscall exit tracing
351 preempt_stop(CLBR_ANY)
353 GET_THREAD_INFO(%ebp)
355 movl PT_EFLAGS(%esp), %eax # mix EFLAGS and CS
356 movb PT_CS(%esp), %al
357 andl $(X86_EFLAGS_VM | SEGMENT_RPL_MASK), %eax
359 jb resume_kernel # not returning to v8086 or userspace
361 ENTRY(resume_userspace)
363 DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt
364 # setting need_resched or sigpending
365 # between sampling and the iret
367 movl TI_flags(%ebp), %ecx
368 andl $_TIF_WORK_MASK, %ecx # is there any work to be done on
369 # int/exception return?
372 END(ret_from_exception)
374 #ifdef CONFIG_PREEMPT
376 DISABLE_INTERRUPTS(CLBR_ANY)
377 cmpl $0,TI_preempt_count(%ebp) # non-zero preempt_count ?
380 movl TI_flags(%ebp), %ecx # need_resched set ?
381 testb $_TIF_NEED_RESCHED, %cl
383 testl $X86_EFLAGS_IF,PT_EFLAGS(%esp) # interrupts off (exception path) ?
385 call preempt_schedule_irq
391 * End of kprobes section
395 /* SYSENTER_RETURN points to after the "sysenter" instruction in
396 the vsyscall page. See vsyscall-sysentry.S, which defines the symbol. */
398 # sysenter call handler stub
399 ENTRY(ia32_sysenter_target)
403 CFI_REGISTER esp, ebp
404 movl TSS_sysenter_sp0(%esp),%esp
407 * Interrupts are disabled here, but we can't trace it until
408 * enough kernel state to call TRACE_IRQS_OFF can be called - but
409 * we immediately enable interrupts at that point anyway.
412 CFI_ADJUST_CFA_OFFSET 4
413 /*CFI_REL_OFFSET ss, 0*/
415 CFI_ADJUST_CFA_OFFSET 4
416 CFI_REL_OFFSET esp, 0
418 orl $X86_EFLAGS_IF, (%esp)
419 CFI_ADJUST_CFA_OFFSET 4
421 CFI_ADJUST_CFA_OFFSET 4
422 /*CFI_REL_OFFSET cs, 0*/
424 * Push current_thread_info()->sysenter_return to the stack.
425 * A tiny bit of offset fixup is necessary - 4*4 means the 4 words
426 * pushed above; +8 corresponds to copy_thread's esp0 setting.
428 pushl (TI_sysenter_return-THREAD_SIZE+8+4*4)(%esp)
429 CFI_ADJUST_CFA_OFFSET 4
430 CFI_REL_OFFSET eip, 0
433 CFI_ADJUST_CFA_OFFSET 4
435 ENABLE_INTERRUPTS(CLBR_NONE)
438 * Load the potential sixth argument from user stack.
439 * Careful about security.
441 cmpl $__PAGE_OFFSET-3,%ebp
444 movl %ebp,PT_EBP(%esp)
445 .section __ex_table,"a"
447 .long 1b,syscall_fault
450 GET_THREAD_INFO(%ebp)
452 testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%ebp)
455 cmpl $(nr_syscalls), %eax
457 call *sys_call_table(,%eax,4)
458 movl %eax,PT_EAX(%esp)
460 DISABLE_INTERRUPTS(CLBR_ANY)
462 movl TI_flags(%ebp), %ecx
463 testl $_TIF_ALLWORK_MASK, %ecx
466 /* if something modifies registers it must also disable sysexit */
467 movl PT_EIP(%esp), %edx
468 movl PT_OLDESP(%esp), %ecx
471 1: mov PT_FS(%esp), %fs
473 ENABLE_INTERRUPTS_SYSEXIT
475 #ifdef CONFIG_AUDITSYSCALL
477 testl $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags(%ebp)
478 jnz syscall_trace_entry
480 CFI_ADJUST_CFA_OFFSET -4
481 /* %esi already in 8(%esp) 6th arg: 4th syscall arg */
482 /* %edx already in 4(%esp) 5th arg: 3rd syscall arg */
483 /* %ecx already in 0(%esp) 4th arg: 2nd syscall arg */
484 movl %ebx,%ecx /* 3rd arg: 1st syscall arg */
485 movl %eax,%edx /* 2nd arg: syscall number */
486 movl $AUDIT_ARCH_I386,%eax /* 1st arg: audit arch */
487 call audit_syscall_entry
489 CFI_ADJUST_CFA_OFFSET 4
490 movl PT_EAX(%esp),%eax /* reload syscall number */
494 testl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT), %ecx
495 jne syscall_exit_work
497 ENABLE_INTERRUPTS(CLBR_ANY)
498 movl %eax,%edx /* second arg, syscall return value */
499 cmpl $0,%eax /* is it < 0? */
500 setl %al /* 1 if so, 0 if not */
501 movzbl %al,%eax /* zero-extend that */
502 inc %eax /* first arg, 0->1(AUDITSC_SUCCESS), 1->2(AUDITSC_FAILURE) */
503 call audit_syscall_exit
504 DISABLE_INTERRUPTS(CLBR_ANY)
506 movl TI_flags(%ebp), %ecx
507 testl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT), %ecx
508 jne syscall_exit_work
509 movl PT_EAX(%esp),%eax /* reload syscall return value */
514 .pushsection .fixup,"ax"
515 2: movl $0,PT_FS(%esp)
517 .section __ex_table,"a"
522 ENDPROC(ia32_sysenter_target)
525 * syscall stub including irq exit should be protected against kprobes
527 .pushsection .kprobes.text, "ax"
528 # system call handler stub
530 RING0_INT_FRAME # can't unwind into user space anyway
531 pushl %eax # save orig_eax
532 CFI_ADJUST_CFA_OFFSET 4
534 GET_THREAD_INFO(%ebp)
535 # system call tracing in operation / emulation
536 testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%ebp)
537 jnz syscall_trace_entry
538 cmpl $(nr_syscalls), %eax
541 call *sys_call_table(,%eax,4)
542 movl %eax,PT_EAX(%esp) # store the return value
545 DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt
546 # setting need_resched or sigpending
547 # between sampling and the iret
549 movl TI_flags(%ebp), %ecx
550 testl $_TIF_ALLWORK_MASK, %ecx # current->work
551 jne syscall_exit_work
556 movl PT_EFLAGS(%esp), %eax # mix EFLAGS, SS and CS
557 # Warning: PT_OLDSS(%esp) contains the wrong/random values if we
558 # are returning to the kernel.
559 # See comments in process.c:copy_thread() for details.
560 movb PT_OLDSS(%esp), %ah
561 movb PT_CS(%esp), %al
562 andl $(X86_EFLAGS_VM | (SEGMENT_TI_MASK << 8) | SEGMENT_RPL_MASK), %eax
563 cmpl $((SEGMENT_LDT << 8) | USER_RPL), %eax
565 je ldt_ss # returning to user-space with LDT SS
567 RESTORE_REGS 4 # skip orig_eax/error_code
568 CFI_ADJUST_CFA_OFFSET -4
573 pushl $0 # no error code
577 .section __ex_table,"a"
579 .long irq_return,iret_exc
584 larl PT_OLDSS(%esp), %eax
586 testl $0x00400000, %eax # returning to 32bit stack?
587 jnz restore_nocheck # allright, normal return
589 #ifdef CONFIG_PARAVIRT
591 * The kernel can't run on a non-flat stack if paravirt mode
592 * is active. Rather than try to fixup the high bits of
593 * ESP, bypass this code entirely. This may break DOSemu
594 * and/or Wine support in a paravirt VM, although the option
595 * is still available to implement the setting of the high
596 * 16-bits in the INTERRUPT_RETURN paravirt-op.
598 cmpl $0, pv_info+PARAVIRT_enabled
603 * Setup and switch to ESPFIX stack
605 * We're returning to userspace with a 16 bit stack. The CPU will not
606 * restore the high word of ESP for us on executing iret... This is an
607 * "official" bug of all the x86-compatible CPUs, which we can work
608 * around to make dosemu and wine happy. We do this by preloading the
609 * high word of ESP with the high word of the userspace ESP while
610 * compensating for the offset by changing to the ESPFIX segment with
611 * a base address that matches for the difference.
613 mov %esp, %edx /* load kernel esp */
614 mov PT_OLDESP(%esp), %eax /* load userspace esp */
615 mov %dx, %ax /* eax: new kernel esp */
616 sub %eax, %edx /* offset (low word is 0) */
617 PER_CPU(gdt_page, %ebx)
619 mov %dl, GDT_ENTRY_ESPFIX_SS * 8 + 4(%ebx) /* bits 16..23 */
620 mov %dh, GDT_ENTRY_ESPFIX_SS * 8 + 7(%ebx) /* bits 24..31 */
622 CFI_ADJUST_CFA_OFFSET 4
623 push %eax /* new kernel esp */
624 CFI_ADJUST_CFA_OFFSET 4
625 /* Disable interrupts, but do not irqtrace this section: we
626 * will soon execute iret and the tracer was already set to
627 * the irqstate after the iret */
628 DISABLE_INTERRUPTS(CLBR_EAX)
629 lss (%esp), %esp /* switch to espfix segment */
630 CFI_ADJUST_CFA_OFFSET -8
635 # perform work that needs to be done immediately before resumption
637 RING0_PTREGS_FRAME # can't unwind into user space anyway
639 testb $_TIF_NEED_RESCHED, %cl
644 DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt
645 # setting need_resched or sigpending
646 # between sampling and the iret
648 movl TI_flags(%ebp), %ecx
649 andl $_TIF_WORK_MASK, %ecx # is there any work to be done other
650 # than syscall tracing?
652 testb $_TIF_NEED_RESCHED, %cl
655 work_notifysig: # deal with pending signals and
656 # notify-resume requests
658 testl $X86_EFLAGS_VM, PT_EFLAGS(%esp)
660 jne work_notifysig_v86 # returning to kernel-space or
663 call do_notify_resume
664 jmp resume_userspace_sig
668 pushl %ecx # save ti_flags for do_notify_resume
669 CFI_ADJUST_CFA_OFFSET 4
670 call save_v86_state # %eax contains pt_regs pointer
672 CFI_ADJUST_CFA_OFFSET -4
678 call do_notify_resume
679 jmp resume_userspace_sig
682 # perform syscall exit tracing
685 movl $-ENOSYS,PT_EAX(%esp)
687 call syscall_trace_enter
688 /* What it returned is what we'll actually use. */
689 cmpl $(nr_syscalls), %eax
692 END(syscall_trace_entry)
694 # perform syscall exit tracing
697 testl $_TIF_WORK_SYSCALL_EXIT, %ecx
700 ENABLE_INTERRUPTS(CLBR_ANY) # could let syscall_trace_leave() call
703 call syscall_trace_leave
705 END(syscall_exit_work)
708 RING0_INT_FRAME # can't unwind into user space anyway
710 GET_THREAD_INFO(%ebp)
711 movl $-EFAULT,PT_EAX(%esp)
716 movl $-ENOSYS,PT_EAX(%esp)
721 * End of kprobes section
726 * System calls that need a pt_regs pointer.
728 #define PTREGSCALL(name) \
739 PTREGSCALL(sigaltstack)
740 PTREGSCALL(sigreturn)
741 PTREGSCALL(rt_sigreturn)
745 .macro FIXUP_ESPFIX_STACK
747 * Switch back for ESPFIX stack to the normal zerobased stack
749 * We can't call C functions using the ESPFIX stack. This code reads
750 * the high word of the segment base from the GDT and swiches to the
751 * normal stack and adjusts ESP with the matching offset.
753 /* fixup the stack */
754 PER_CPU(gdt_page, %ebx)
755 mov GDT_ENTRY_ESPFIX_SS * 8 + 4(%ebx), %al /* bits 16..23 */
756 mov GDT_ENTRY_ESPFIX_SS * 8 + 7(%ebx), %ah /* bits 24..31 */
758 addl %esp, %eax /* the adjusted stack pointer */
760 CFI_ADJUST_CFA_OFFSET 4
762 CFI_ADJUST_CFA_OFFSET 4
763 lss (%esp), %esp /* switch to the normal stack segment */
764 CFI_ADJUST_CFA_OFFSET -8
766 .macro UNWIND_ESPFIX_STACK
768 /* see if on espfix stack */
769 cmpw $__ESPFIX_SS, %ax
771 movl $__KERNEL_DS, %eax
774 /* switch to normal stack */
780 * Build the entry stubs and pointer table with some assembler magic.
781 * We pack 7 stubs into a single 32-byte chunk, which will fit in a
782 * single cache line on all modern x86 implementations.
784 .section .init.rodata,"a"
788 .p2align CONFIG_X86_L1_CACHE_SHIFT
789 ENTRY(irq_entries_start)
791 vector=FIRST_EXTERNAL_VECTOR
792 .rept (NR_VECTORS-FIRST_EXTERNAL_VECTOR+6)/7
795 .if vector < NR_VECTORS
796 .if vector <> FIRST_EXTERNAL_VECTOR
797 CFI_ADJUST_CFA_OFFSET -4
799 1: pushl $(~vector+0x80) /* Note: always in signed byte range */
800 CFI_ADJUST_CFA_OFFSET 4
801 .if ((vector-FIRST_EXTERNAL_VECTOR)%7) <> 6
810 2: jmp common_interrupt
812 END(irq_entries_start)
819 * the CPU automatically disables interrupts when executing an IRQ vector,
820 * so IRQ-flags tracing has to follow that:
822 .p2align CONFIG_X86_L1_CACHE_SHIFT
824 addl $-0x80,(%esp) /* Adjust vector into the [-256,-1] range */
830 ENDPROC(common_interrupt)
834 * Irq entries should be protected against kprobes
836 .pushsection .kprobes.text, "ax"
837 #define BUILD_INTERRUPT3(name, nr, fn) \
841 CFI_ADJUST_CFA_OFFSET 4; \
850 #define BUILD_INTERRUPT(name, nr) BUILD_INTERRUPT3(name, nr, smp_##name)
852 /* The include is where all of the SMP etc. interrupts come from */
853 #include <asm/entry_arch.h>
855 ENTRY(coprocessor_error)
858 CFI_ADJUST_CFA_OFFSET 4
859 pushl $do_coprocessor_error
860 CFI_ADJUST_CFA_OFFSET 4
863 END(coprocessor_error)
865 ENTRY(simd_coprocessor_error)
868 CFI_ADJUST_CFA_OFFSET 4
869 pushl $do_simd_coprocessor_error
870 CFI_ADJUST_CFA_OFFSET 4
873 END(simd_coprocessor_error)
875 ENTRY(device_not_available)
877 pushl $-1 # mark this as an int
878 CFI_ADJUST_CFA_OFFSET 4
879 pushl $do_device_not_available
880 CFI_ADJUST_CFA_OFFSET 4
883 END(device_not_available)
885 #ifdef CONFIG_PARAVIRT
888 .section __ex_table,"a"
890 .long native_iret, iret_exc
894 ENTRY(native_irq_enable_sysexit)
897 END(native_irq_enable_sysexit)
903 CFI_ADJUST_CFA_OFFSET 4
905 CFI_ADJUST_CFA_OFFSET 4
913 CFI_ADJUST_CFA_OFFSET 4
915 CFI_ADJUST_CFA_OFFSET 4
923 CFI_ADJUST_CFA_OFFSET 4
925 CFI_ADJUST_CFA_OFFSET 4
930 ENTRY(coprocessor_segment_overrun)
933 CFI_ADJUST_CFA_OFFSET 4
934 pushl $do_coprocessor_segment_overrun
935 CFI_ADJUST_CFA_OFFSET 4
938 END(coprocessor_segment_overrun)
942 pushl $do_invalid_TSS
943 CFI_ADJUST_CFA_OFFSET 4
948 ENTRY(segment_not_present)
950 pushl $do_segment_not_present
951 CFI_ADJUST_CFA_OFFSET 4
954 END(segment_not_present)
958 pushl $do_stack_segment
959 CFI_ADJUST_CFA_OFFSET 4
964 ENTRY(alignment_check)
966 pushl $do_alignment_check
967 CFI_ADJUST_CFA_OFFSET 4
974 pushl $0 # no error code
975 CFI_ADJUST_CFA_OFFSET 4
976 pushl $do_divide_error
977 CFI_ADJUST_CFA_OFFSET 4
982 #ifdef CONFIG_X86_MCE
986 CFI_ADJUST_CFA_OFFSET 4
987 pushl machine_check_vector
988 CFI_ADJUST_CFA_OFFSET 4
994 ENTRY(spurious_interrupt_bug)
997 CFI_ADJUST_CFA_OFFSET 4
998 pushl $do_spurious_interrupt_bug
999 CFI_ADJUST_CFA_OFFSET 4
1002 END(spurious_interrupt_bug)
1004 * End of kprobes section
1008 ENTRY(kernel_thread_helper)
1009 pushl $0 # fake return address for unwinder
1013 CFI_ADJUST_CFA_OFFSET 4
1016 CFI_ADJUST_CFA_OFFSET 4
1018 ud2 # padding for call trace
1020 ENDPROC(kernel_thread_helper)
1023 /* Xen doesn't set %esp to be precisely what the normal sysenter
1024 entrypoint expects, so fix it up before using the normal path. */
1025 ENTRY(xen_sysenter_target)
1027 addl $5*4, %esp /* remove xen-provided frame */
1028 CFI_ADJUST_CFA_OFFSET -5*4
1029 jmp sysenter_past_esp
1032 ENTRY(xen_hypervisor_callback)
1035 CFI_ADJUST_CFA_OFFSET 4
1039 /* Check to see if we got the event in the critical
1040 region in xen_iret_direct, after we've reenabled
1041 events and checked for pending events. This simulates
1042 iret instruction's behaviour where it delivers a
1043 pending interrupt when enabling interrupts. */
1044 movl PT_EIP(%esp),%eax
1045 cmpl $xen_iret_start_crit,%eax
1047 cmpl $xen_iret_end_crit,%eax
1050 jmp xen_iret_crit_fixup
1052 ENTRY(xen_do_upcall)
1054 call xen_evtchn_do_upcall
1057 ENDPROC(xen_hypervisor_callback)
1059 # Hypervisor uses this for application faults while it executes.
1060 # We get here for two reasons:
1061 # 1. Fault while reloading DS, ES, FS or GS
1062 # 2. Fault while executing IRET
1063 # Category 1 we fix up by reattempting the load, and zeroing the segment
1064 # register if the load fails.
1065 # Category 2 we fix up by jumping to do_iret_error. We cannot use the
1066 # normal Linux return path in this case because if we use the IRET hypercall
1067 # to pop the stack frame we end up in an infinite loop of failsafe callbacks.
1068 # We distinguish between categories by maintaining a status value in EAX.
1069 ENTRY(xen_failsafe_callback)
1072 CFI_ADJUST_CFA_OFFSET 4
1080 CFI_ADJUST_CFA_OFFSET -4
1082 CFI_ADJUST_CFA_OFFSET -16
1085 jmp iret_exc # EAX != 0 => Category 2 (Bad IRET)
1086 5: pushl $0 # EAX == 0 => Category 1 (Bad segment)
1087 CFI_ADJUST_CFA_OFFSET 4
1089 jmp ret_from_exception
1092 .section .fixup,"ax"
1106 .section __ex_table,"a"
1113 ENDPROC(xen_failsafe_callback)
1115 #endif /* CONFIG_XEN */
1117 #ifdef CONFIG_FUNCTION_TRACER
1118 #ifdef CONFIG_DYNAMIC_FTRACE
1124 ENTRY(ftrace_caller)
1125 cmpl $0, function_trace_stop
1131 movl 0xc(%esp), %eax
1132 movl 0x4(%ebp), %edx
1133 subl $MCOUNT_INSN_SIZE, %eax
1142 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1143 .globl ftrace_graph_call
1153 #else /* ! CONFIG_DYNAMIC_FTRACE */
1156 cmpl $0, function_trace_stop
1159 cmpl $ftrace_stub, ftrace_trace_function
1161 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1162 cmpl $ftrace_stub, ftrace_graph_return
1163 jnz ftrace_graph_caller
1165 cmpl $ftrace_graph_entry_stub, ftrace_graph_entry
1166 jnz ftrace_graph_caller
1172 /* taken from glibc */
1177 movl 0xc(%esp), %eax
1178 movl 0x4(%ebp), %edx
1179 subl $MCOUNT_INSN_SIZE, %eax
1181 call *ftrace_trace_function
1188 #endif /* CONFIG_DYNAMIC_FTRACE */
1189 #endif /* CONFIG_FUNCTION_TRACER */
1191 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1192 ENTRY(ftrace_graph_caller)
1193 cmpl $0, function_trace_stop
1199 movl 0xc(%esp), %edx
1202 subl $MCOUNT_INSN_SIZE, %edx
1203 call prepare_ftrace_return
1208 END(ftrace_graph_caller)
1210 .globl return_to_handler
1215 call ftrace_return_to_handler
1222 .section .rodata,"a"
1223 #include "syscall_table_32.S"
1225 syscall_table_size=(.-sys_call_table)
1228 * Some functions should be protected against kprobes
1230 .pushsection .kprobes.text, "ax"
1234 pushl $do_page_fault
1235 CFI_ADJUST_CFA_OFFSET 4
1238 /* the function address is in %gs's slot on the stack */
1240 CFI_ADJUST_CFA_OFFSET 4
1241 /*CFI_REL_OFFSET fs, 0*/
1243 CFI_ADJUST_CFA_OFFSET 4
1244 /*CFI_REL_OFFSET es, 0*/
1246 CFI_ADJUST_CFA_OFFSET 4
1247 /*CFI_REL_OFFSET ds, 0*/
1249 CFI_ADJUST_CFA_OFFSET 4
1250 CFI_REL_OFFSET eax, 0
1252 CFI_ADJUST_CFA_OFFSET 4
1253 CFI_REL_OFFSET ebp, 0
1255 CFI_ADJUST_CFA_OFFSET 4
1256 CFI_REL_OFFSET edi, 0
1258 CFI_ADJUST_CFA_OFFSET 4
1259 CFI_REL_OFFSET esi, 0
1261 CFI_ADJUST_CFA_OFFSET 4
1262 CFI_REL_OFFSET edx, 0
1264 CFI_ADJUST_CFA_OFFSET 4
1265 CFI_REL_OFFSET ecx, 0
1267 CFI_ADJUST_CFA_OFFSET 4
1268 CFI_REL_OFFSET ebx, 0
1270 movl $(__KERNEL_PERCPU), %ecx
1274 movl PT_GS(%esp), %edi # get the function address
1275 movl PT_ORIG_EAX(%esp), %edx # get the error code
1276 movl $-1, PT_ORIG_EAX(%esp) # no syscall to restart
1279 movl $(__USER_DS), %ecx
1283 movl %esp,%eax # pt_regs pointer
1285 jmp ret_from_exception
1290 * Debug traps and NMI can happen at the one SYSENTER instruction
1291 * that sets up the real kernel stack. Check here, since we can't
1292 * allow the wrong stack to be used.
1294 * "TSS_sysenter_sp0+12" is because the NMI/debug handler will have
1295 * already pushed 3 words if it hits on the sysenter instruction:
1296 * eflags, cs and eip.
1298 * We just load the right stack, and push the three (known) values
1299 * by hand onto the new stack - while updating the return eip past
1300 * the instruction that would have done it for sysenter.
1302 .macro FIX_STACK offset ok label
1303 cmpw $__KERNEL_CS, 4(%esp)
1306 movl TSS_sysenter_sp0 + \offset(%esp), %esp
1310 CFI_ADJUST_CFA_OFFSET 4
1312 CFI_ADJUST_CFA_OFFSET 4
1313 pushl $sysenter_past_esp
1314 CFI_ADJUST_CFA_OFFSET 4
1315 CFI_REL_OFFSET eip, 0
1320 cmpl $ia32_sysenter_target,(%esp)
1321 jne debug_stack_correct
1322 FIX_STACK 12, debug_stack_correct, debug_esp_fix_insn
1323 debug_stack_correct:
1324 pushl $-1 # mark this as an int
1325 CFI_ADJUST_CFA_OFFSET 4
1328 xorl %edx,%edx # error code 0
1329 movl %esp,%eax # pt_regs pointer
1331 jmp ret_from_exception
1336 * NMI is doubly nasty. It can happen _while_ we're handling
1337 * a debug fault, and the debug fault hasn't yet been able to
1338 * clear up the stack. So we first check whether we got an
1339 * NMI on the sysenter entry path, but after that we need to
1340 * check whether we got an NMI on the debug path where the debug
1341 * fault happened on the sysenter path.
1346 CFI_ADJUST_CFA_OFFSET 4
1348 cmpw $__ESPFIX_SS, %ax
1350 CFI_ADJUST_CFA_OFFSET -4
1352 cmpl $ia32_sysenter_target,(%esp)
1355 CFI_ADJUST_CFA_OFFSET 4
1357 /* Do not access memory above the end of our stack page,
1358 * it might not exist.
1360 andl $(THREAD_SIZE-1),%eax
1361 cmpl $(THREAD_SIZE-20),%eax
1363 CFI_ADJUST_CFA_OFFSET -4
1364 jae nmi_stack_correct
1365 cmpl $ia32_sysenter_target,12(%esp)
1366 je nmi_debug_stack_check
1368 /* We have a RING0_INT_FRAME here */
1370 CFI_ADJUST_CFA_OFFSET 4
1372 xorl %edx,%edx # zero error code
1373 movl %esp,%eax # pt_regs pointer
1375 jmp restore_all_notrace
1380 FIX_STACK 12, nmi_stack_correct, 1
1381 jmp nmi_stack_correct
1383 nmi_debug_stack_check:
1384 /* We have a RING0_INT_FRAME here */
1385 cmpw $__KERNEL_CS,16(%esp)
1386 jne nmi_stack_correct
1388 jb nmi_stack_correct
1389 cmpl $debug_esp_fix_insn,(%esp)
1390 ja nmi_stack_correct
1391 FIX_STACK 24, nmi_stack_correct, 1
1392 jmp nmi_stack_correct
1395 /* We have a RING0_INT_FRAME here.
1397 * create the pointer to lss back
1400 CFI_ADJUST_CFA_OFFSET 4
1402 CFI_ADJUST_CFA_OFFSET 4
1404 /* copy the iret frame of 12 bytes */
1407 CFI_ADJUST_CFA_OFFSET 4
1410 CFI_ADJUST_CFA_OFFSET 4
1412 FIXUP_ESPFIX_STACK # %eax == %esp
1413 xorl %edx,%edx # zero error code
1416 lss 12+4(%esp), %esp # back to espfix stack
1417 CFI_ADJUST_CFA_OFFSET -24
1424 pushl $-1 # mark this as an int
1425 CFI_ADJUST_CFA_OFFSET 4
1428 xorl %edx,%edx # zero error code
1429 movl %esp,%eax # pt_regs pointer
1431 jmp ret_from_exception
1435 ENTRY(general_protection)
1437 pushl $do_general_protection
1438 CFI_ADJUST_CFA_OFFSET 4
1441 END(general_protection)
1444 * End of kprobes section