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>
57 #include <asm/alternative-asm.h>
59 /* Avoid __ASSEMBLER__'ifying <linux/audit.h> just for this. */
60 #include <linux/elf-em.h>
61 #define AUDIT_ARCH_I386 (EM_386|__AUDIT_ARCH_LE)
62 #define __AUDIT_ARCH_LE 0x40000000
64 #ifndef CONFIG_AUDITSYSCALL
65 #define sysenter_audit syscall_trace_entry
66 #define sysexit_audit syscall_exit_work
69 .section .entry.text, "ax"
72 * We use macros for low-level operations which need to be overridden
73 * for paravirtualization. The following will never clobber any registers:
74 * INTERRUPT_RETURN (aka. "iret")
75 * GET_CR0_INTO_EAX (aka. "movl %cr0, %eax")
76 * ENABLE_INTERRUPTS_SYSEXIT (aka "sti; sysexit").
78 * For DISABLE_INTERRUPTS/ENABLE_INTERRUPTS (aka "cli"/"sti"), you must
79 * specify what registers can be overwritten (CLBR_NONE, CLBR_EAX/EDX/ECX/ANY).
80 * Allowing a register to be clobbered can shrink the paravirt replacement
81 * enough to patch inline, increasing performance.
85 #define preempt_stop(clobbers) DISABLE_INTERRUPTS(clobbers); TRACE_IRQS_OFF
87 #define preempt_stop(clobbers)
88 #define resume_kernel restore_all
91 .macro TRACE_IRQS_IRET
92 #ifdef CONFIG_TRACE_IRQFLAGS
93 testl $X86_EFLAGS_IF,PT_EFLAGS(%esp) # interrupts off?
101 #define resume_userspace_sig check_userspace
103 #define resume_userspace_sig resume_userspace
107 * User gs save/restore
109 * %gs is used for userland TLS and kernel only uses it for stack
110 * canary which is required to be at %gs:20 by gcc. Read the comment
111 * at the top of stackprotector.h for more info.
113 * Local labels 98 and 99 are used.
115 #ifdef CONFIG_X86_32_LAZY_GS
117 /* unfortunately push/pop can't be no-op */
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_REL_OFFSET gs, 0*/
152 CFI_ADJUST_CFA_OFFSET -\pop
156 .pushsection .fixup, "ax"
159 .section __ex_table, "a"
166 98: mov PT_GS(%esp), %gs
169 .pushsection .fixup, "ax"
170 99: movl $0, PT_GS(%esp)
172 .section __ex_table, "a"
180 /*CFI_REGISTER gs, \reg*/
182 .macro REG_TO_PTGS reg
183 movl \reg, PT_GS(%esp)
184 /*CFI_REL_OFFSET gs, PT_GS*/
186 .macro SET_KERNEL_GS reg
187 movl $(__KERNEL_STACK_CANARY), \reg
191 #endif /* CONFIG_X86_32_LAZY_GS */
197 /*CFI_REL_OFFSET fs, 0;*/
199 /*CFI_REL_OFFSET es, 0;*/
201 /*CFI_REL_OFFSET ds, 0;*/
203 CFI_REL_OFFSET eax, 0
205 CFI_REL_OFFSET ebp, 0
207 CFI_REL_OFFSET edi, 0
209 CFI_REL_OFFSET esi, 0
211 CFI_REL_OFFSET edx, 0
213 CFI_REL_OFFSET ecx, 0
215 CFI_REL_OFFSET ebx, 0
216 movl $(__USER_DS), %edx
219 movl $(__KERNEL_PERCPU), %edx
224 .macro RESTORE_INT_REGS
241 .macro RESTORE_REGS pop=0
250 .pushsection .fixup, "ax"
257 .section __ex_table, "a"
266 .macro RING0_INT_FRAME
270 /*CFI_OFFSET cs, -2*4;*/
274 .macro RING0_EC_FRAME
278 /*CFI_OFFSET cs, -2*4;*/
282 .macro RING0_PTREGS_FRAME
285 CFI_DEF_CFA esp, PT_OLDESP-PT_EBX
286 /*CFI_OFFSET cs, PT_CS-PT_OLDESP;*/
287 CFI_OFFSET eip, PT_EIP-PT_OLDESP
288 /*CFI_OFFSET es, PT_ES-PT_OLDESP;*/
289 /*CFI_OFFSET ds, PT_DS-PT_OLDESP;*/
290 CFI_OFFSET eax, PT_EAX-PT_OLDESP
291 CFI_OFFSET ebp, PT_EBP-PT_OLDESP
292 CFI_OFFSET edi, PT_EDI-PT_OLDESP
293 CFI_OFFSET esi, PT_ESI-PT_OLDESP
294 CFI_OFFSET edx, PT_EDX-PT_OLDESP
295 CFI_OFFSET ecx, PT_ECX-PT_OLDESP
296 CFI_OFFSET ebx, PT_EBX-PT_OLDESP
303 GET_THREAD_INFO(%ebp)
305 pushl_cfi $0x0202 # Reset kernel eflags
312 * Interrupt exit functions should be protected against kprobes
314 .pushsection .kprobes.text, "ax"
316 * Return to user mode is not as complex as all this looks,
317 * but we want the default path for a system call return to
318 * go as quickly as possible which is why some of this is
319 * less clear than it otherwise should be.
322 # userspace resumption stub bypassing syscall exit tracing
326 preempt_stop(CLBR_ANY)
328 GET_THREAD_INFO(%ebp)
330 movl PT_EFLAGS(%esp), %eax # mix EFLAGS and CS
331 movb PT_CS(%esp), %al
332 andl $(X86_EFLAGS_VM | SEGMENT_RPL_MASK), %eax
334 jb resume_kernel # not returning to v8086 or userspace
336 ENTRY(resume_userspace)
338 DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt
339 # setting need_resched or sigpending
340 # between sampling and the iret
342 movl TI_flags(%ebp), %ecx
343 andl $_TIF_WORK_MASK, %ecx # is there any work to be done on
344 # int/exception return?
347 END(ret_from_exception)
349 #ifdef CONFIG_PREEMPT
351 DISABLE_INTERRUPTS(CLBR_ANY)
352 cmpl $0,TI_preempt_count(%ebp) # non-zero preempt_count ?
355 movl TI_flags(%ebp), %ecx # need_resched set ?
356 testb $_TIF_NEED_RESCHED, %cl
358 testl $X86_EFLAGS_IF,PT_EFLAGS(%esp) # interrupts off (exception path) ?
360 call preempt_schedule_irq
366 * End of kprobes section
370 /* SYSENTER_RETURN points to after the "sysenter" instruction in
371 the vsyscall page. See vsyscall-sysentry.S, which defines the symbol. */
373 # sysenter call handler stub
374 ENTRY(ia32_sysenter_target)
378 CFI_REGISTER esp, ebp
379 movl TSS_sysenter_sp0(%esp),%esp
382 * Interrupts are disabled here, but we can't trace it until
383 * enough kernel state to call TRACE_IRQS_OFF can be called - but
384 * we immediately enable interrupts at that point anyway.
387 /*CFI_REL_OFFSET ss, 0*/
389 CFI_REL_OFFSET esp, 0
391 orl $X86_EFLAGS_IF, (%esp)
393 /*CFI_REL_OFFSET cs, 0*/
395 * Push current_thread_info()->sysenter_return to the stack.
396 * A tiny bit of offset fixup is necessary - 4*4 means the 4 words
397 * pushed above; +8 corresponds to copy_thread's esp0 setting.
399 pushl_cfi ((TI_sysenter_return)-THREAD_SIZE+8+4*4)(%esp)
400 CFI_REL_OFFSET eip, 0
404 ENABLE_INTERRUPTS(CLBR_NONE)
407 * Load the potential sixth argument from user stack.
408 * Careful about security.
410 cmpl $__PAGE_OFFSET-3,%ebp
413 movl %ebp,PT_EBP(%esp)
414 .section __ex_table,"a"
416 .long 1b,syscall_fault
419 GET_THREAD_INFO(%ebp)
421 testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%ebp)
424 cmpl $(NR_syscalls), %eax
426 call *sys_call_table(,%eax,4)
427 movl %eax,PT_EAX(%esp)
429 DISABLE_INTERRUPTS(CLBR_ANY)
431 movl TI_flags(%ebp), %ecx
432 testl $_TIF_ALLWORK_MASK, %ecx
435 /* if something modifies registers it must also disable sysexit */
436 movl PT_EIP(%esp), %edx
437 movl PT_OLDESP(%esp), %ecx
440 1: mov PT_FS(%esp), %fs
442 ENABLE_INTERRUPTS_SYSEXIT
444 #ifdef CONFIG_AUDITSYSCALL
446 testl $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags(%ebp)
447 jnz syscall_trace_entry
449 CFI_ADJUST_CFA_OFFSET -4
450 /* %esi already in 8(%esp) 6th arg: 4th syscall arg */
451 /* %edx already in 4(%esp) 5th arg: 3rd syscall arg */
452 /* %ecx already in 0(%esp) 4th arg: 2nd syscall arg */
453 movl %ebx,%ecx /* 3rd arg: 1st syscall arg */
454 movl %eax,%edx /* 2nd arg: syscall number */
455 movl $AUDIT_ARCH_I386,%eax /* 1st arg: audit arch */
456 call audit_syscall_entry
458 movl PT_EAX(%esp),%eax /* reload syscall number */
462 testl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT), %ecx
463 jne syscall_exit_work
465 ENABLE_INTERRUPTS(CLBR_ANY)
466 movl %eax,%edx /* second arg, syscall return value */
467 cmpl $0,%eax /* is it < 0? */
468 setl %al /* 1 if so, 0 if not */
469 movzbl %al,%eax /* zero-extend that */
470 inc %eax /* first arg, 0->1(AUDITSC_SUCCESS), 1->2(AUDITSC_FAILURE) */
471 call audit_syscall_exit
472 DISABLE_INTERRUPTS(CLBR_ANY)
474 movl TI_flags(%ebp), %ecx
475 testl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT), %ecx
476 jne syscall_exit_work
477 movl PT_EAX(%esp),%eax /* reload syscall return value */
482 .pushsection .fixup,"ax"
483 2: movl $0,PT_FS(%esp)
485 .section __ex_table,"a"
490 ENDPROC(ia32_sysenter_target)
493 * syscall stub including irq exit should be protected against kprobes
495 .pushsection .kprobes.text, "ax"
496 # system call handler stub
498 RING0_INT_FRAME # can't unwind into user space anyway
499 pushl_cfi %eax # save orig_eax
501 GET_THREAD_INFO(%ebp)
502 # system call tracing in operation / emulation
503 testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%ebp)
504 jnz syscall_trace_entry
505 cmpl $(NR_syscalls), %eax
508 call *sys_call_table(,%eax,4)
509 movl %eax,PT_EAX(%esp) # store the return value
512 DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt
513 # setting need_resched or sigpending
514 # between sampling and the iret
516 movl TI_flags(%ebp), %ecx
517 testl $_TIF_ALLWORK_MASK, %ecx # current->work
518 jne syscall_exit_work
523 movl PT_EFLAGS(%esp), %eax # mix EFLAGS, SS and CS
524 # Warning: PT_OLDSS(%esp) contains the wrong/random values if we
525 # are returning to the kernel.
526 # See comments in process.c:copy_thread() for details.
527 movb PT_OLDSS(%esp), %ah
528 movb PT_CS(%esp), %al
529 andl $(X86_EFLAGS_VM | (SEGMENT_TI_MASK << 8) | SEGMENT_RPL_MASK), %eax
530 cmpl $((SEGMENT_LDT << 8) | USER_RPL), %eax
532 je ldt_ss # returning to user-space with LDT SS
534 RESTORE_REGS 4 # skip orig_eax/error_code
539 pushl $0 # no error code
543 .section __ex_table,"a"
545 .long irq_return,iret_exc
550 larl PT_OLDSS(%esp), %eax
552 testl $0x00400000, %eax # returning to 32bit stack?
553 jnz restore_nocheck # allright, normal return
555 #ifdef CONFIG_PARAVIRT
557 * The kernel can't run on a non-flat stack if paravirt mode
558 * is active. Rather than try to fixup the high bits of
559 * ESP, bypass this code entirely. This may break DOSemu
560 * and/or Wine support in a paravirt VM, although the option
561 * is still available to implement the setting of the high
562 * 16-bits in the INTERRUPT_RETURN paravirt-op.
564 cmpl $0, pv_info+PARAVIRT_enabled
569 * Setup and switch to ESPFIX stack
571 * We're returning to userspace with a 16 bit stack. The CPU will not
572 * restore the high word of ESP for us on executing iret... This is an
573 * "official" bug of all the x86-compatible CPUs, which we can work
574 * around to make dosemu and wine happy. We do this by preloading the
575 * high word of ESP with the high word of the userspace ESP while
576 * compensating for the offset by changing to the ESPFIX segment with
577 * a base address that matches for the difference.
579 #define GDT_ESPFIX_SS PER_CPU_VAR(gdt_page) + (GDT_ENTRY_ESPFIX_SS * 8)
580 mov %esp, %edx /* load kernel esp */
581 mov PT_OLDESP(%esp), %eax /* load userspace esp */
582 mov %dx, %ax /* eax: new kernel esp */
583 sub %eax, %edx /* offset (low word is 0) */
585 mov %dl, GDT_ESPFIX_SS + 4 /* bits 16..23 */
586 mov %dh, GDT_ESPFIX_SS + 7 /* bits 24..31 */
587 pushl_cfi $__ESPFIX_SS
588 pushl_cfi %eax /* new kernel esp */
589 /* Disable interrupts, but do not irqtrace this section: we
590 * will soon execute iret and the tracer was already set to
591 * the irqstate after the iret */
592 DISABLE_INTERRUPTS(CLBR_EAX)
593 lss (%esp), %esp /* switch to espfix segment */
594 CFI_ADJUST_CFA_OFFSET -8
599 # perform work that needs to be done immediately before resumption
601 RING0_PTREGS_FRAME # can't unwind into user space anyway
603 testb $_TIF_NEED_RESCHED, %cl
608 DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt
609 # setting need_resched or sigpending
610 # between sampling and the iret
612 movl TI_flags(%ebp), %ecx
613 andl $_TIF_WORK_MASK, %ecx # is there any work to be done other
614 # than syscall tracing?
616 testb $_TIF_NEED_RESCHED, %cl
619 work_notifysig: # deal with pending signals and
620 # notify-resume requests
622 testl $X86_EFLAGS_VM, PT_EFLAGS(%esp)
624 jne work_notifysig_v86 # returning to kernel-space or
627 call do_notify_resume
628 jmp resume_userspace_sig
632 pushl_cfi %ecx # save ti_flags for do_notify_resume
633 call save_v86_state # %eax contains pt_regs pointer
640 call do_notify_resume
641 jmp resume_userspace_sig
644 # perform syscall exit tracing
647 movl $-ENOSYS,PT_EAX(%esp)
649 call syscall_trace_enter
650 /* What it returned is what we'll actually use. */
651 cmpl $(NR_syscalls), %eax
654 END(syscall_trace_entry)
656 # perform syscall exit tracing
659 testl $_TIF_WORK_SYSCALL_EXIT, %ecx
662 ENABLE_INTERRUPTS(CLBR_ANY) # could let syscall_trace_leave() call
665 call syscall_trace_leave
667 END(syscall_exit_work)
670 RING0_INT_FRAME # can't unwind into user space anyway
672 GET_THREAD_INFO(%ebp)
673 movl $-EFAULT,PT_EAX(%esp)
678 movl $-ENOSYS,PT_EAX(%esp)
683 * End of kprobes section
688 * System calls that need a pt_regs pointer.
690 #define PTREGSCALL0(name) \
691 ENTRY(ptregs_##name) ; \
694 ENDPROC(ptregs_##name)
696 #define PTREGSCALL1(name) \
697 ENTRY(ptregs_##name) ; \
699 movl (PT_EBX+4)(%esp),%eax; \
701 ENDPROC(ptregs_##name)
703 #define PTREGSCALL2(name) \
704 ENTRY(ptregs_##name) ; \
706 movl (PT_ECX+4)(%esp),%edx; \
707 movl (PT_EBX+4)(%esp),%eax; \
709 ENDPROC(ptregs_##name)
711 #define PTREGSCALL3(name) \
712 ENTRY(ptregs_##name) ; \
716 movl PT_EDX(%eax),%ecx; \
717 movl PT_ECX(%eax),%edx; \
718 movl PT_EBX(%eax),%eax; \
721 CFI_ADJUST_CFA_OFFSET -4; \
724 ENDPROC(ptregs_##name)
730 PTREGSCALL2(sigaltstack)
731 PTREGSCALL0(sigreturn)
732 PTREGSCALL0(rt_sigreturn)
736 /* Clone is an oddball. The 4th arg is in %edi */
741 pushl_cfi PT_EDI(%eax)
742 movl PT_EDX(%eax),%ecx
743 movl PT_ECX(%eax),%edx
744 movl PT_EBX(%eax),%eax
747 CFI_ADJUST_CFA_OFFSET -8
750 ENDPROC(ptregs_clone)
752 .macro FIXUP_ESPFIX_STACK
754 * Switch back for ESPFIX stack to the normal zerobased stack
756 * We can't call C functions using the ESPFIX stack. This code reads
757 * the high word of the segment base from the GDT and swiches to the
758 * normal stack and adjusts ESP with the matching offset.
760 /* fixup the stack */
761 mov GDT_ESPFIX_SS + 4, %al /* bits 16..23 */
762 mov GDT_ESPFIX_SS + 7, %ah /* bits 24..31 */
764 addl %esp, %eax /* the adjusted stack pointer */
765 pushl_cfi $__KERNEL_DS
767 lss (%esp), %esp /* switch to the normal stack segment */
768 CFI_ADJUST_CFA_OFFSET -8
770 .macro UNWIND_ESPFIX_STACK
772 /* see if on espfix stack */
773 cmpw $__ESPFIX_SS, %ax
775 movl $__KERNEL_DS, %eax
778 /* switch to normal stack */
784 * Build the entry stubs and pointer table with some assembler magic.
785 * We pack 7 stubs into a single 32-byte chunk, which will fit in a
786 * single cache line on all modern x86 implementations.
788 .section .init.rodata,"a"
790 .section .entry.text, "ax"
792 .p2align CONFIG_X86_L1_CACHE_SHIFT
793 ENTRY(irq_entries_start)
795 vector=FIRST_EXTERNAL_VECTOR
796 .rept (NR_VECTORS-FIRST_EXTERNAL_VECTOR+6)/7
799 .if vector < NR_VECTORS
800 .if vector <> FIRST_EXTERNAL_VECTOR
801 CFI_ADJUST_CFA_OFFSET -4
803 1: pushl_cfi $(~vector+0x80) /* Note: always in signed byte range */
804 .if ((vector-FIRST_EXTERNAL_VECTOR)%7) <> 6
809 .section .entry.text, "ax"
813 2: jmp common_interrupt
815 END(irq_entries_start)
822 * the CPU automatically disables interrupts when executing an IRQ vector,
823 * so IRQ-flags tracing has to follow that:
825 .p2align CONFIG_X86_L1_CACHE_SHIFT
827 addl $-0x80,(%esp) /* Adjust vector into the [-256,-1] range */
833 ENDPROC(common_interrupt)
837 * Irq entries should be protected against kprobes
839 .pushsection .kprobes.text, "ax"
840 #define BUILD_INTERRUPT3(name, nr, fn) \
852 #define BUILD_INTERRUPT(name, nr) BUILD_INTERRUPT3(name, nr, smp_##name)
854 /* The include is where all of the SMP etc. interrupts come from */
855 #include <asm/entry_arch.h>
857 ENTRY(coprocessor_error)
860 pushl_cfi $do_coprocessor_error
863 END(coprocessor_error)
865 ENTRY(simd_coprocessor_error)
868 #ifdef CONFIG_X86_INVD_BUG
869 /* AMD 486 bug: invd from userspace calls exception 19 instead of #GP */
870 661: pushl_cfi $do_general_protection
872 .section .altinstructions,"a"
873 altinstruction_entry 661b, 663f, X86_FEATURE_XMM, 662b-661b, 664f-663f
875 .section .altinstr_replacement,"ax"
876 663: pushl $do_simd_coprocessor_error
880 pushl_cfi $do_simd_coprocessor_error
884 END(simd_coprocessor_error)
886 ENTRY(device_not_available)
888 pushl_cfi $-1 # mark this as an int
889 pushl_cfi $do_device_not_available
892 END(device_not_available)
894 #ifdef CONFIG_PARAVIRT
897 .section __ex_table,"a"
899 .long native_iret, iret_exc
903 ENTRY(native_irq_enable_sysexit)
906 END(native_irq_enable_sysexit)
912 pushl_cfi $do_overflow
928 pushl_cfi $do_invalid_op
933 ENTRY(coprocessor_segment_overrun)
936 pushl_cfi $do_coprocessor_segment_overrun
939 END(coprocessor_segment_overrun)
943 pushl_cfi $do_invalid_TSS
948 ENTRY(segment_not_present)
950 pushl_cfi $do_segment_not_present
953 END(segment_not_present)
957 pushl_cfi $do_stack_segment
962 ENTRY(alignment_check)
964 pushl_cfi $do_alignment_check
971 pushl_cfi $0 # no error code
972 pushl_cfi $do_divide_error
977 #ifdef CONFIG_X86_MCE
981 pushl_cfi machine_check_vector
987 ENTRY(spurious_interrupt_bug)
990 pushl_cfi $do_spurious_interrupt_bug
993 END(spurious_interrupt_bug)
995 * End of kprobes section
999 ENTRY(kernel_thread_helper)
1000 pushl $0 # fake return address for unwinder
1005 ud2 # padding for call trace
1007 ENDPROC(kernel_thread_helper)
1010 /* Xen doesn't set %esp to be precisely what the normal sysenter
1011 entrypoint expects, so fix it up before using the normal path. */
1012 ENTRY(xen_sysenter_target)
1014 addl $5*4, %esp /* remove xen-provided frame */
1015 CFI_ADJUST_CFA_OFFSET -5*4
1016 jmp sysenter_past_esp
1019 ENTRY(xen_hypervisor_callback)
1025 /* Check to see if we got the event in the critical
1026 region in xen_iret_direct, after we've reenabled
1027 events and checked for pending events. This simulates
1028 iret instruction's behaviour where it delivers a
1029 pending interrupt when enabling interrupts. */
1030 movl PT_EIP(%esp),%eax
1031 cmpl $xen_iret_start_crit,%eax
1033 cmpl $xen_iret_end_crit,%eax
1036 jmp xen_iret_crit_fixup
1038 ENTRY(xen_do_upcall)
1040 call xen_evtchn_do_upcall
1043 ENDPROC(xen_hypervisor_callback)
1045 # Hypervisor uses this for application faults while it executes.
1046 # We get here for two reasons:
1047 # 1. Fault while reloading DS, ES, FS or GS
1048 # 2. Fault while executing IRET
1049 # Category 1 we fix up by reattempting the load, and zeroing the segment
1050 # register if the load fails.
1051 # Category 2 we fix up by jumping to do_iret_error. We cannot use the
1052 # normal Linux return path in this case because if we use the IRET hypercall
1053 # to pop the stack frame we end up in an infinite loop of failsafe callbacks.
1054 # We distinguish between categories by maintaining a status value in EAX.
1055 ENTRY(xen_failsafe_callback)
1066 CFI_ADJUST_CFA_OFFSET -16
1069 jmp iret_exc # EAX != 0 => Category 2 (Bad IRET)
1070 5: pushl_cfi $0 # EAX == 0 => Category 1 (Bad segment)
1072 jmp ret_from_exception
1075 .section .fixup,"ax"
1089 .section __ex_table,"a"
1096 ENDPROC(xen_failsafe_callback)
1098 BUILD_INTERRUPT3(xen_hvm_callback_vector, XEN_HVM_EVTCHN_CALLBACK,
1099 xen_evtchn_do_upcall)
1101 #endif /* CONFIG_XEN */
1103 #ifdef CONFIG_FUNCTION_TRACER
1104 #ifdef CONFIG_DYNAMIC_FTRACE
1110 ENTRY(ftrace_caller)
1111 cmpl $0, function_trace_stop
1117 movl 0xc(%esp), %eax
1118 movl 0x4(%ebp), %edx
1119 subl $MCOUNT_INSN_SIZE, %eax
1128 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1129 .globl ftrace_graph_call
1139 #else /* ! CONFIG_DYNAMIC_FTRACE */
1142 cmpl $0, function_trace_stop
1145 cmpl $ftrace_stub, ftrace_trace_function
1147 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1148 cmpl $ftrace_stub, ftrace_graph_return
1149 jnz ftrace_graph_caller
1151 cmpl $ftrace_graph_entry_stub, ftrace_graph_entry
1152 jnz ftrace_graph_caller
1158 /* taken from glibc */
1163 movl 0xc(%esp), %eax
1164 movl 0x4(%ebp), %edx
1165 subl $MCOUNT_INSN_SIZE, %eax
1167 call *ftrace_trace_function
1174 #endif /* CONFIG_DYNAMIC_FTRACE */
1175 #endif /* CONFIG_FUNCTION_TRACER */
1177 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1178 ENTRY(ftrace_graph_caller)
1179 cmpl $0, function_trace_stop
1185 movl 0xc(%esp), %edx
1188 subl $MCOUNT_INSN_SIZE, %edx
1189 call prepare_ftrace_return
1194 END(ftrace_graph_caller)
1196 .globl return_to_handler
1201 call ftrace_return_to_handler
1209 * Some functions should be protected against kprobes
1211 .pushsection .kprobes.text, "ax"
1215 pushl_cfi $do_page_fault
1218 /* the function address is in %gs's slot on the stack */
1220 /*CFI_REL_OFFSET fs, 0*/
1222 /*CFI_REL_OFFSET es, 0*/
1224 /*CFI_REL_OFFSET ds, 0*/
1226 CFI_REL_OFFSET eax, 0
1228 CFI_REL_OFFSET ebp, 0
1230 CFI_REL_OFFSET edi, 0
1232 CFI_REL_OFFSET esi, 0
1234 CFI_REL_OFFSET edx, 0
1236 CFI_REL_OFFSET ecx, 0
1238 CFI_REL_OFFSET ebx, 0
1240 movl $(__KERNEL_PERCPU), %ecx
1244 movl PT_GS(%esp), %edi # get the function address
1245 movl PT_ORIG_EAX(%esp), %edx # get the error code
1246 movl $-1, PT_ORIG_EAX(%esp) # no syscall to restart
1249 movl $(__USER_DS), %ecx
1253 movl %esp,%eax # pt_regs pointer
1255 jmp ret_from_exception
1260 * Debug traps and NMI can happen at the one SYSENTER instruction
1261 * that sets up the real kernel stack. Check here, since we can't
1262 * allow the wrong stack to be used.
1264 * "TSS_sysenter_sp0+12" is because the NMI/debug handler will have
1265 * already pushed 3 words if it hits on the sysenter instruction:
1266 * eflags, cs and eip.
1268 * We just load the right stack, and push the three (known) values
1269 * by hand onto the new stack - while updating the return eip past
1270 * the instruction that would have done it for sysenter.
1272 .macro FIX_STACK offset ok label
1273 cmpw $__KERNEL_CS, 4(%esp)
1276 movl TSS_sysenter_sp0 + \offset(%esp), %esp
1280 pushl_cfi $__KERNEL_CS
1281 pushl_cfi $sysenter_past_esp
1282 CFI_REL_OFFSET eip, 0
1287 cmpl $ia32_sysenter_target,(%esp)
1288 jne debug_stack_correct
1289 FIX_STACK 12, debug_stack_correct, debug_esp_fix_insn
1290 debug_stack_correct:
1291 pushl_cfi $-1 # mark this as an int
1294 xorl %edx,%edx # error code 0
1295 movl %esp,%eax # pt_regs pointer
1297 jmp ret_from_exception
1302 * NMI is doubly nasty. It can happen _while_ we're handling
1303 * a debug fault, and the debug fault hasn't yet been able to
1304 * clear up the stack. So we first check whether we got an
1305 * NMI on the sysenter entry path, but after that we need to
1306 * check whether we got an NMI on the debug path where the debug
1307 * fault happened on the sysenter path.
1313 cmpw $__ESPFIX_SS, %ax
1316 cmpl $ia32_sysenter_target,(%esp)
1320 /* Do not access memory above the end of our stack page,
1321 * it might not exist.
1323 andl $(THREAD_SIZE-1),%eax
1324 cmpl $(THREAD_SIZE-20),%eax
1326 jae nmi_stack_correct
1327 cmpl $ia32_sysenter_target,12(%esp)
1328 je nmi_debug_stack_check
1330 /* We have a RING0_INT_FRAME here */
1333 xorl %edx,%edx # zero error code
1334 movl %esp,%eax # pt_regs pointer
1336 jmp restore_all_notrace
1341 FIX_STACK 12, nmi_stack_correct, 1
1342 jmp nmi_stack_correct
1344 nmi_debug_stack_check:
1345 /* We have a RING0_INT_FRAME here */
1346 cmpw $__KERNEL_CS,16(%esp)
1347 jne nmi_stack_correct
1349 jb nmi_stack_correct
1350 cmpl $debug_esp_fix_insn,(%esp)
1351 ja nmi_stack_correct
1352 FIX_STACK 24, nmi_stack_correct, 1
1353 jmp nmi_stack_correct
1356 /* We have a RING0_INT_FRAME here.
1358 * create the pointer to lss back
1363 /* copy the iret frame of 12 bytes */
1369 FIXUP_ESPFIX_STACK # %eax == %esp
1370 xorl %edx,%edx # zero error code
1373 lss 12+4(%esp), %esp # back to espfix stack
1374 CFI_ADJUST_CFA_OFFSET -24
1381 pushl_cfi $-1 # mark this as an int
1384 xorl %edx,%edx # zero error code
1385 movl %esp,%eax # pt_regs pointer
1387 jmp ret_from_exception
1391 ENTRY(general_protection)
1393 pushl_cfi $do_general_protection
1396 END(general_protection)
1398 #ifdef CONFIG_KVM_GUEST
1399 ENTRY(async_page_fault)
1401 pushl_cfi $do_async_page_fault
1404 END(async_page_fault)
1408 * End of kprobes section