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>
56 #include <asm/cpufeature.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
68 .section .entry.text, "ax"
71 * We use macros for low-level operations which need to be overridden
72 * for paravirtualization. The following will never clobber any registers:
73 * INTERRUPT_RETURN (aka. "iret")
74 * GET_CR0_INTO_EAX (aka. "movl %cr0, %eax")
75 * ENABLE_INTERRUPTS_SYSEXIT (aka "sti; sysexit").
77 * For DISABLE_INTERRUPTS/ENABLE_INTERRUPTS (aka "cli"/"sti"), you must
78 * specify what registers can be overwritten (CLBR_NONE, CLBR_EAX/EDX/ECX/ANY).
79 * Allowing a register to be clobbered can shrink the paravirt replacement
80 * enough to patch inline, increasing performance.
83 #define nr_syscalls ((syscall_table_size)/4)
86 #define preempt_stop(clobbers) DISABLE_INTERRUPTS(clobbers); TRACE_IRQS_OFF
88 #define preempt_stop(clobbers)
89 #define resume_kernel restore_all
92 .macro TRACE_IRQS_IRET
93 #ifdef CONFIG_TRACE_IRQFLAGS
94 testl $X86_EFLAGS_IF,PT_EFLAGS(%esp) # interrupts off?
102 #define resume_userspace_sig check_userspace
104 #define resume_userspace_sig resume_userspace
108 * User gs save/restore
110 * %gs is used for userland TLS and kernel only uses it for stack
111 * canary which is required to be at %gs:20 by gcc. Read the comment
112 * at the top of stackprotector.h for more info.
114 * Local labels 98 and 99 are used.
116 #ifdef CONFIG_X86_32_LAZY_GS
118 /* unfortunately push/pop can't be no-op */
123 addl $(4 + \pop), %esp
124 CFI_ADJUST_CFA_OFFSET -(4 + \pop)
129 /* all the rest are no-op */
136 .macro REG_TO_PTGS reg
138 .macro SET_KERNEL_GS reg
141 #else /* CONFIG_X86_32_LAZY_GS */
145 /*CFI_REL_OFFSET gs, 0*/
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_REL_OFFSET fs, 0;*/
200 /*CFI_REL_OFFSET es, 0;*/
202 /*CFI_REL_OFFSET ds, 0;*/
204 CFI_REL_OFFSET eax, 0
206 CFI_REL_OFFSET ebp, 0
208 CFI_REL_OFFSET edi, 0
210 CFI_REL_OFFSET esi, 0
212 CFI_REL_OFFSET edx, 0
214 CFI_REL_OFFSET ecx, 0
216 CFI_REL_OFFSET ebx, 0
217 movl $(__USER_DS), %edx
220 movl $(__KERNEL_PERCPU), %edx
225 .macro RESTORE_INT_REGS
242 .macro RESTORE_REGS pop=0
251 .pushsection .fixup, "ax"
258 .section __ex_table, "a"
267 .macro RING0_INT_FRAME
271 /*CFI_OFFSET cs, -2*4;*/
275 .macro RING0_EC_FRAME
279 /*CFI_OFFSET cs, -2*4;*/
283 .macro RING0_PTREGS_FRAME
286 CFI_DEF_CFA esp, PT_OLDESP-PT_EBX
287 /*CFI_OFFSET cs, PT_CS-PT_OLDESP;*/
288 CFI_OFFSET eip, PT_EIP-PT_OLDESP
289 /*CFI_OFFSET es, PT_ES-PT_OLDESP;*/
290 /*CFI_OFFSET ds, PT_DS-PT_OLDESP;*/
291 CFI_OFFSET eax, PT_EAX-PT_OLDESP
292 CFI_OFFSET ebp, PT_EBP-PT_OLDESP
293 CFI_OFFSET edi, PT_EDI-PT_OLDESP
294 CFI_OFFSET esi, PT_ESI-PT_OLDESP
295 CFI_OFFSET edx, PT_EDX-PT_OLDESP
296 CFI_OFFSET ecx, PT_ECX-PT_OLDESP
297 CFI_OFFSET ebx, PT_EBX-PT_OLDESP
304 GET_THREAD_INFO(%ebp)
306 pushl_cfi $0x0202 # Reset kernel eflags
313 * Interrupt exit functions should be protected against kprobes
315 .pushsection .kprobes.text, "ax"
317 * Return to user mode is not as complex as all this looks,
318 * but we want the default path for a system call return to
319 * go as quickly as possible which is why some of this is
320 * less clear than it otherwise should be.
323 # userspace resumption stub bypassing syscall exit tracing
327 preempt_stop(CLBR_ANY)
329 GET_THREAD_INFO(%ebp)
331 movl PT_EFLAGS(%esp), %eax # mix EFLAGS and CS
332 movb PT_CS(%esp), %al
333 andl $(X86_EFLAGS_VM | SEGMENT_RPL_MASK), %eax
335 jb resume_kernel # not returning to v8086 or userspace
337 ENTRY(resume_userspace)
339 DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt
340 # setting need_resched or sigpending
341 # between sampling and the iret
343 movl TI_flags(%ebp), %ecx
344 andl $_TIF_WORK_MASK, %ecx # is there any work to be done on
345 # int/exception return?
348 END(ret_from_exception)
350 #ifdef CONFIG_PREEMPT
352 DISABLE_INTERRUPTS(CLBR_ANY)
353 cmpl $0,TI_preempt_count(%ebp) # non-zero preempt_count ?
356 movl TI_flags(%ebp), %ecx # need_resched set ?
357 testb $_TIF_NEED_RESCHED, %cl
359 testl $X86_EFLAGS_IF,PT_EFLAGS(%esp) # interrupts off (exception path) ?
361 call preempt_schedule_irq
367 * End of kprobes section
371 /* SYSENTER_RETURN points to after the "sysenter" instruction in
372 the vsyscall page. See vsyscall-sysentry.S, which defines the symbol. */
374 # sysenter call handler stub
375 ENTRY(ia32_sysenter_target)
379 CFI_REGISTER esp, ebp
380 movl TSS_sysenter_sp0(%esp),%esp
383 * Interrupts are disabled here, but we can't trace it until
384 * enough kernel state to call TRACE_IRQS_OFF can be called - but
385 * we immediately enable interrupts at that point anyway.
388 /*CFI_REL_OFFSET ss, 0*/
390 CFI_REL_OFFSET esp, 0
392 orl $X86_EFLAGS_IF, (%esp)
394 /*CFI_REL_OFFSET cs, 0*/
396 * Push current_thread_info()->sysenter_return to the stack.
397 * A tiny bit of offset fixup is necessary - 4*4 means the 4 words
398 * pushed above; +8 corresponds to copy_thread's esp0 setting.
400 pushl_cfi ((TI_sysenter_return)-THREAD_SIZE+8+4*4)(%esp)
401 CFI_REL_OFFSET eip, 0
405 ENABLE_INTERRUPTS(CLBR_NONE)
408 * Load the potential sixth argument from user stack.
409 * Careful about security.
411 cmpl $__PAGE_OFFSET-3,%ebp
414 movl %ebp,PT_EBP(%esp)
415 .section __ex_table,"a"
417 .long 1b,syscall_fault
420 GET_THREAD_INFO(%ebp)
422 testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%ebp)
425 cmpl $(nr_syscalls), %eax
427 call *sys_call_table(,%eax,4)
428 movl %eax,PT_EAX(%esp)
430 DISABLE_INTERRUPTS(CLBR_ANY)
432 movl TI_flags(%ebp), %ecx
433 testl $_TIF_ALLWORK_MASK, %ecx
436 /* if something modifies registers it must also disable sysexit */
437 movl PT_EIP(%esp), %edx
438 movl PT_OLDESP(%esp), %ecx
441 1: mov PT_FS(%esp), %fs
443 ENABLE_INTERRUPTS_SYSEXIT
445 #ifdef CONFIG_AUDITSYSCALL
447 testl $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags(%ebp)
448 jnz syscall_trace_entry
450 CFI_ADJUST_CFA_OFFSET -4
451 /* %esi already in 8(%esp) 6th arg: 4th syscall arg */
452 /* %edx already in 4(%esp) 5th arg: 3rd syscall arg */
453 /* %ecx already in 0(%esp) 4th arg: 2nd syscall arg */
454 movl %ebx,%ecx /* 3rd arg: 1st syscall arg */
455 movl %eax,%edx /* 2nd arg: syscall number */
456 movl $AUDIT_ARCH_I386,%eax /* 1st arg: audit arch */
457 call audit_syscall_entry
459 movl PT_EAX(%esp),%eax /* reload syscall number */
463 testl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT), %ecx
464 jne syscall_exit_work
466 ENABLE_INTERRUPTS(CLBR_ANY)
467 movl %eax,%edx /* second arg, syscall return value */
468 cmpl $0,%eax /* is it < 0? */
469 setl %al /* 1 if so, 0 if not */
470 movzbl %al,%eax /* zero-extend that */
471 inc %eax /* first arg, 0->1(AUDITSC_SUCCESS), 1->2(AUDITSC_FAILURE) */
472 call audit_syscall_exit
473 DISABLE_INTERRUPTS(CLBR_ANY)
475 movl TI_flags(%ebp), %ecx
476 testl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT), %ecx
477 jne syscall_exit_work
478 movl PT_EAX(%esp),%eax /* reload syscall return value */
483 .pushsection .fixup,"ax"
484 2: movl $0,PT_FS(%esp)
486 .section __ex_table,"a"
491 ENDPROC(ia32_sysenter_target)
494 * syscall stub including irq exit should be protected against kprobes
496 .pushsection .kprobes.text, "ax"
497 # system call handler stub
499 RING0_INT_FRAME # can't unwind into user space anyway
500 pushl_cfi %eax # save orig_eax
502 GET_THREAD_INFO(%ebp)
503 # system call tracing in operation / emulation
504 testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%ebp)
505 jnz syscall_trace_entry
506 cmpl $(nr_syscalls), %eax
509 call *sys_call_table(,%eax,4)
510 movl %eax,PT_EAX(%esp) # store the return value
513 DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt
514 # setting need_resched or sigpending
515 # between sampling and the iret
517 movl TI_flags(%ebp), %ecx
518 testl $_TIF_ALLWORK_MASK, %ecx # current->work
519 jne syscall_exit_work
524 movl PT_EFLAGS(%esp), %eax # mix EFLAGS, SS and CS
525 # Warning: PT_OLDSS(%esp) contains the wrong/random values if we
526 # are returning to the kernel.
527 # See comments in process.c:copy_thread() for details.
528 movb PT_OLDSS(%esp), %ah
529 movb PT_CS(%esp), %al
530 andl $(X86_EFLAGS_VM | (SEGMENT_TI_MASK << 8) | SEGMENT_RPL_MASK), %eax
531 cmpl $((SEGMENT_LDT << 8) | USER_RPL), %eax
533 je ldt_ss # returning to user-space with LDT SS
535 RESTORE_REGS 4 # skip orig_eax/error_code
540 pushl $0 # no error code
544 .section __ex_table,"a"
546 .long irq_return,iret_exc
551 larl PT_OLDSS(%esp), %eax
553 testl $0x00400000, %eax # returning to 32bit stack?
554 jnz restore_nocheck # allright, normal return
556 #ifdef CONFIG_PARAVIRT
558 * The kernel can't run on a non-flat stack if paravirt mode
559 * is active. Rather than try to fixup the high bits of
560 * ESP, bypass this code entirely. This may break DOSemu
561 * and/or Wine support in a paravirt VM, although the option
562 * is still available to implement the setting of the high
563 * 16-bits in the INTERRUPT_RETURN paravirt-op.
565 cmpl $0, pv_info+PARAVIRT_enabled
570 * Setup and switch to ESPFIX stack
572 * We're returning to userspace with a 16 bit stack. The CPU will not
573 * restore the high word of ESP for us on executing iret... This is an
574 * "official" bug of all the x86-compatible CPUs, which we can work
575 * around to make dosemu and wine happy. We do this by preloading the
576 * high word of ESP with the high word of the userspace ESP while
577 * compensating for the offset by changing to the ESPFIX segment with
578 * a base address that matches for the difference.
580 #define GDT_ESPFIX_SS PER_CPU_VAR(gdt_page) + (GDT_ENTRY_ESPFIX_SS * 8)
581 mov %esp, %edx /* load kernel esp */
582 mov PT_OLDESP(%esp), %eax /* load userspace esp */
583 mov %dx, %ax /* eax: new kernel esp */
584 sub %eax, %edx /* offset (low word is 0) */
586 mov %dl, GDT_ESPFIX_SS + 4 /* bits 16..23 */
587 mov %dh, GDT_ESPFIX_SS + 7 /* bits 24..31 */
588 pushl_cfi $__ESPFIX_SS
589 pushl_cfi %eax /* new kernel esp */
590 /* Disable interrupts, but do not irqtrace this section: we
591 * will soon execute iret and the tracer was already set to
592 * the irqstate after the iret */
593 DISABLE_INTERRUPTS(CLBR_EAX)
594 lss (%esp), %esp /* switch to espfix segment */
595 CFI_ADJUST_CFA_OFFSET -8
600 # perform work that needs to be done immediately before resumption
602 RING0_PTREGS_FRAME # can't unwind into user space anyway
604 testb $_TIF_NEED_RESCHED, %cl
609 DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt
610 # setting need_resched or sigpending
611 # between sampling and the iret
613 movl TI_flags(%ebp), %ecx
614 andl $_TIF_WORK_MASK, %ecx # is there any work to be done other
615 # than syscall tracing?
617 testb $_TIF_NEED_RESCHED, %cl
620 work_notifysig: # deal with pending signals and
621 # notify-resume requests
623 testl $X86_EFLAGS_VM, PT_EFLAGS(%esp)
625 jne work_notifysig_v86 # returning to kernel-space or
628 call do_notify_resume
629 jmp resume_userspace_sig
633 pushl_cfi %ecx # save ti_flags for do_notify_resume
634 call save_v86_state # %eax contains pt_regs pointer
641 call do_notify_resume
642 jmp resume_userspace_sig
645 # perform syscall exit tracing
648 movl $-ENOSYS,PT_EAX(%esp)
650 call syscall_trace_enter
651 /* What it returned is what we'll actually use. */
652 cmpl $(nr_syscalls), %eax
655 END(syscall_trace_entry)
657 # perform syscall exit tracing
660 testl $_TIF_WORK_SYSCALL_EXIT, %ecx
663 ENABLE_INTERRUPTS(CLBR_ANY) # could let syscall_trace_leave() call
666 call syscall_trace_leave
668 END(syscall_exit_work)
671 RING0_INT_FRAME # can't unwind into user space anyway
673 GET_THREAD_INFO(%ebp)
674 movl $-EFAULT,PT_EAX(%esp)
679 movl $-ENOSYS,PT_EAX(%esp)
684 * End of kprobes section
689 * System calls that need a pt_regs pointer.
691 #define PTREGSCALL0(name) \
697 #define PTREGSCALL1(name) \
701 movl (PT_EBX+4)(%esp),%eax; \
704 #define PTREGSCALL2(name) \
708 movl (PT_ECX+4)(%esp),%edx; \
709 movl (PT_EBX+4)(%esp),%eax; \
712 #define PTREGSCALL3(name) \
718 movl PT_EDX(%eax),%ecx; \
719 movl PT_ECX(%eax),%edx; \
720 movl PT_EBX(%eax),%eax; \
723 CFI_ADJUST_CFA_OFFSET -4; \
726 ENDPROC(ptregs_##name)
732 PTREGSCALL2(sigaltstack)
733 PTREGSCALL0(sigreturn)
734 PTREGSCALL0(rt_sigreturn)
738 /* Clone is an oddball. The 4th arg is in %edi */
744 pushl_cfi PT_EDI(%eax)
745 movl PT_EDX(%eax),%ecx
746 movl PT_ECX(%eax),%edx
747 movl PT_EBX(%eax),%eax
750 CFI_ADJUST_CFA_OFFSET -8
753 ENDPROC(ptregs_clone)
755 .macro FIXUP_ESPFIX_STACK
757 * Switch back for ESPFIX stack to the normal zerobased stack
759 * We can't call C functions using the ESPFIX stack. This code reads
760 * the high word of the segment base from the GDT and swiches to the
761 * normal stack and adjusts ESP with the matching offset.
763 /* fixup the stack */
764 mov GDT_ESPFIX_SS + 4, %al /* bits 16..23 */
765 mov GDT_ESPFIX_SS + 7, %ah /* bits 24..31 */
767 addl %esp, %eax /* the adjusted stack pointer */
768 pushl_cfi $__KERNEL_DS
770 lss (%esp), %esp /* switch to the normal stack segment */
771 CFI_ADJUST_CFA_OFFSET -8
773 .macro UNWIND_ESPFIX_STACK
775 /* see if on espfix stack */
776 cmpw $__ESPFIX_SS, %ax
778 movl $__KERNEL_DS, %eax
781 /* switch to normal stack */
787 * Build the entry stubs and pointer table with some assembler magic.
788 * We pack 7 stubs into a single 32-byte chunk, which will fit in a
789 * single cache line on all modern x86 implementations.
791 .section .init.rodata,"a"
793 .section .entry.text, "ax"
795 .p2align CONFIG_X86_L1_CACHE_SHIFT
796 ENTRY(irq_entries_start)
798 vector=FIRST_EXTERNAL_VECTOR
799 .rept (NR_VECTORS-FIRST_EXTERNAL_VECTOR+6)/7
802 .if vector < NR_VECTORS
803 .if vector <> FIRST_EXTERNAL_VECTOR
804 CFI_ADJUST_CFA_OFFSET -4
806 1: pushl_cfi $(~vector+0x80) /* Note: always in signed byte range */
807 .if ((vector-FIRST_EXTERNAL_VECTOR)%7) <> 6
812 .section .entry.text, "ax"
816 2: jmp common_interrupt
818 END(irq_entries_start)
825 * the CPU automatically disables interrupts when executing an IRQ vector,
826 * so IRQ-flags tracing has to follow that:
828 .p2align CONFIG_X86_L1_CACHE_SHIFT
830 addl $-0x80,(%esp) /* Adjust vector into the [-256,-1] range */
836 ENDPROC(common_interrupt)
840 * Irq entries should be protected against kprobes
842 .pushsection .kprobes.text, "ax"
843 #define BUILD_INTERRUPT3(name, nr, fn) \
855 #define BUILD_INTERRUPT(name, nr) BUILD_INTERRUPT3(name, nr, smp_##name)
857 /* The include is where all of the SMP etc. interrupts come from */
858 #include <asm/entry_arch.h>
860 ENTRY(coprocessor_error)
863 pushl_cfi $do_coprocessor_error
866 END(coprocessor_error)
868 ENTRY(simd_coprocessor_error)
871 #ifdef CONFIG_X86_INVD_BUG
872 /* AMD 486 bug: invd from userspace calls exception 19 instead of #GP */
873 661: pushl_cfi $do_general_protection
875 .section .altinstructions,"a"
879 .word X86_FEATURE_XMM
883 .section .altinstr_replacement,"ax"
884 663: pushl $do_simd_coprocessor_error
888 pushl_cfi $do_simd_coprocessor_error
892 END(simd_coprocessor_error)
894 ENTRY(device_not_available)
896 pushl_cfi $-1 # mark this as an int
897 pushl_cfi $do_device_not_available
900 END(device_not_available)
902 #ifdef CONFIG_PARAVIRT
905 .section __ex_table,"a"
907 .long native_iret, iret_exc
911 ENTRY(native_irq_enable_sysexit)
914 END(native_irq_enable_sysexit)
920 pushl_cfi $do_overflow
936 pushl_cfi $do_invalid_op
941 ENTRY(coprocessor_segment_overrun)
944 pushl_cfi $do_coprocessor_segment_overrun
947 END(coprocessor_segment_overrun)
951 pushl_cfi $do_invalid_TSS
956 ENTRY(segment_not_present)
958 pushl_cfi $do_segment_not_present
961 END(segment_not_present)
965 pushl_cfi $do_stack_segment
970 ENTRY(alignment_check)
972 pushl_cfi $do_alignment_check
979 pushl_cfi $0 # no error code
980 pushl_cfi $do_divide_error
985 #ifdef CONFIG_X86_MCE
989 pushl_cfi machine_check_vector
995 ENTRY(spurious_interrupt_bug)
998 pushl_cfi $do_spurious_interrupt_bug
1001 END(spurious_interrupt_bug)
1003 * End of kprobes section
1007 ENTRY(kernel_thread_helper)
1008 pushl $0 # fake return address for unwinder
1013 ud2 # padding for call trace
1015 ENDPROC(kernel_thread_helper)
1018 /* Xen doesn't set %esp to be precisely what the normal sysenter
1019 entrypoint expects, so fix it up before using the normal path. */
1020 ENTRY(xen_sysenter_target)
1022 addl $5*4, %esp /* remove xen-provided frame */
1023 CFI_ADJUST_CFA_OFFSET -5*4
1024 jmp sysenter_past_esp
1027 ENTRY(xen_hypervisor_callback)
1033 /* Check to see if we got the event in the critical
1034 region in xen_iret_direct, after we've reenabled
1035 events and checked for pending events. This simulates
1036 iret instruction's behaviour where it delivers a
1037 pending interrupt when enabling interrupts. */
1038 movl PT_EIP(%esp),%eax
1039 cmpl $xen_iret_start_crit,%eax
1041 cmpl $xen_iret_end_crit,%eax
1044 jmp xen_iret_crit_fixup
1046 ENTRY(xen_do_upcall)
1048 call xen_evtchn_do_upcall
1051 ENDPROC(xen_hypervisor_callback)
1053 # Hypervisor uses this for application faults while it executes.
1054 # We get here for two reasons:
1055 # 1. Fault while reloading DS, ES, FS or GS
1056 # 2. Fault while executing IRET
1057 # Category 1 we fix up by reattempting the load, and zeroing the segment
1058 # register if the load fails.
1059 # Category 2 we fix up by jumping to do_iret_error. We cannot use the
1060 # normal Linux return path in this case because if we use the IRET hypercall
1061 # to pop the stack frame we end up in an infinite loop of failsafe callbacks.
1062 # We distinguish between categories by maintaining a status value in EAX.
1063 ENTRY(xen_failsafe_callback)
1074 CFI_ADJUST_CFA_OFFSET -16
1077 jmp iret_exc # EAX != 0 => Category 2 (Bad IRET)
1078 5: pushl_cfi $0 # EAX == 0 => Category 1 (Bad segment)
1080 jmp ret_from_exception
1083 .section .fixup,"ax"
1097 .section __ex_table,"a"
1104 ENDPROC(xen_failsafe_callback)
1106 BUILD_INTERRUPT3(xen_hvm_callback_vector, XEN_HVM_EVTCHN_CALLBACK,
1107 xen_evtchn_do_upcall)
1109 #endif /* CONFIG_XEN */
1111 #ifdef CONFIG_FUNCTION_TRACER
1112 #ifdef CONFIG_DYNAMIC_FTRACE
1118 ENTRY(ftrace_caller)
1119 cmpl $0, function_trace_stop
1125 movl 0xc(%esp), %eax
1126 movl 0x4(%ebp), %edx
1127 subl $MCOUNT_INSN_SIZE, %eax
1136 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1137 .globl ftrace_graph_call
1147 #else /* ! CONFIG_DYNAMIC_FTRACE */
1150 cmpl $0, function_trace_stop
1153 cmpl $ftrace_stub, ftrace_trace_function
1155 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1156 cmpl $ftrace_stub, ftrace_graph_return
1157 jnz ftrace_graph_caller
1159 cmpl $ftrace_graph_entry_stub, ftrace_graph_entry
1160 jnz ftrace_graph_caller
1166 /* taken from glibc */
1171 movl 0xc(%esp), %eax
1172 movl 0x4(%ebp), %edx
1173 subl $MCOUNT_INSN_SIZE, %eax
1175 call *ftrace_trace_function
1182 #endif /* CONFIG_DYNAMIC_FTRACE */
1183 #endif /* CONFIG_FUNCTION_TRACER */
1185 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1186 ENTRY(ftrace_graph_caller)
1187 cmpl $0, function_trace_stop
1193 movl 0xc(%esp), %edx
1196 subl $MCOUNT_INSN_SIZE, %edx
1197 call prepare_ftrace_return
1202 END(ftrace_graph_caller)
1204 .globl return_to_handler
1209 call ftrace_return_to_handler
1216 .section .rodata,"a"
1217 #include "syscall_table_32.S"
1219 syscall_table_size=(.-sys_call_table)
1222 * Some functions should be protected against kprobes
1224 .pushsection .kprobes.text, "ax"
1228 pushl_cfi $do_page_fault
1231 /* the function address is in %gs's slot on the stack */
1233 /*CFI_REL_OFFSET fs, 0*/
1235 /*CFI_REL_OFFSET es, 0*/
1237 /*CFI_REL_OFFSET ds, 0*/
1239 CFI_REL_OFFSET eax, 0
1241 CFI_REL_OFFSET ebp, 0
1243 CFI_REL_OFFSET edi, 0
1245 CFI_REL_OFFSET esi, 0
1247 CFI_REL_OFFSET edx, 0
1249 CFI_REL_OFFSET ecx, 0
1251 CFI_REL_OFFSET ebx, 0
1253 movl $(__KERNEL_PERCPU), %ecx
1257 movl PT_GS(%esp), %edi # get the function address
1258 movl PT_ORIG_EAX(%esp), %edx # get the error code
1259 movl $-1, PT_ORIG_EAX(%esp) # no syscall to restart
1262 movl $(__USER_DS), %ecx
1266 movl %esp,%eax # pt_regs pointer
1268 jmp ret_from_exception
1273 * Debug traps and NMI can happen at the one SYSENTER instruction
1274 * that sets up the real kernel stack. Check here, since we can't
1275 * allow the wrong stack to be used.
1277 * "TSS_sysenter_sp0+12" is because the NMI/debug handler will have
1278 * already pushed 3 words if it hits on the sysenter instruction:
1279 * eflags, cs and eip.
1281 * We just load the right stack, and push the three (known) values
1282 * by hand onto the new stack - while updating the return eip past
1283 * the instruction that would have done it for sysenter.
1285 .macro FIX_STACK offset ok label
1286 cmpw $__KERNEL_CS, 4(%esp)
1289 movl TSS_sysenter_sp0 + \offset(%esp), %esp
1293 pushl_cfi $__KERNEL_CS
1294 pushl_cfi $sysenter_past_esp
1295 CFI_REL_OFFSET eip, 0
1300 cmpl $ia32_sysenter_target,(%esp)
1301 jne debug_stack_correct
1302 FIX_STACK 12, debug_stack_correct, debug_esp_fix_insn
1303 debug_stack_correct:
1304 pushl_cfi $-1 # mark this as an int
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.
1326 cmpw $__ESPFIX_SS, %ax
1329 cmpl $ia32_sysenter_target,(%esp)
1333 /* Do not access memory above the end of our stack page,
1334 * it might not exist.
1336 andl $(THREAD_SIZE-1),%eax
1337 cmpl $(THREAD_SIZE-20),%eax
1339 jae nmi_stack_correct
1340 cmpl $ia32_sysenter_target,12(%esp)
1341 je nmi_debug_stack_check
1343 /* We have a RING0_INT_FRAME here */
1346 xorl %edx,%edx # zero error code
1347 movl %esp,%eax # pt_regs pointer
1349 jmp restore_all_notrace
1354 FIX_STACK 12, nmi_stack_correct, 1
1355 jmp nmi_stack_correct
1357 nmi_debug_stack_check:
1358 /* We have a RING0_INT_FRAME here */
1359 cmpw $__KERNEL_CS,16(%esp)
1360 jne nmi_stack_correct
1362 jb nmi_stack_correct
1363 cmpl $debug_esp_fix_insn,(%esp)
1364 ja nmi_stack_correct
1365 FIX_STACK 24, nmi_stack_correct, 1
1366 jmp nmi_stack_correct
1369 /* We have a RING0_INT_FRAME here.
1371 * create the pointer to lss back
1376 /* copy the iret frame of 12 bytes */
1382 FIXUP_ESPFIX_STACK # %eax == %esp
1383 xorl %edx,%edx # zero error code
1386 lss 12+4(%esp), %esp # back to espfix stack
1387 CFI_ADJUST_CFA_OFFSET -24
1394 pushl_cfi $-1 # mark this as an int
1397 xorl %edx,%edx # zero error code
1398 movl %esp,%eax # pt_regs pointer
1400 jmp ret_from_exception
1404 ENTRY(general_protection)
1406 pushl_cfi $do_general_protection
1409 END(general_protection)
1411 #ifdef CONFIG_KVM_GUEST
1412 ENTRY(async_page_fault)
1414 pushl_cfi $do_async_page_fault
1417 END(async_page_fault)
1421 * End of kprobes section