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 <linux/err.h>
46 #include <asm/thread_info.h>
47 #include <asm/irqflags.h>
48 #include <asm/errno.h>
49 #include <asm/segment.h>
51 #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>
57 #include <asm/cpufeature.h>
58 #include <asm/alternative-asm.h>
62 /* Avoid __ASSEMBLER__'ifying <linux/audit.h> just for this. */
63 #include <linux/elf-em.h>
64 #define AUDIT_ARCH_I386 (EM_386|__AUDIT_ARCH_LE)
65 #define __AUDIT_ARCH_LE 0x40000000
67 #ifndef CONFIG_AUDITSYSCALL
68 #define sysenter_audit syscall_trace_entry
69 #define sysexit_audit syscall_exit_work
72 .section .entry.text, "ax"
75 * We use macros for low-level operations which need to be overridden
76 * for paravirtualization. The following will never clobber any registers:
77 * INTERRUPT_RETURN (aka. "iret")
78 * GET_CR0_INTO_EAX (aka. "movl %cr0, %eax")
79 * ENABLE_INTERRUPTS_SYSEXIT (aka "sti; sysexit").
81 * For DISABLE_INTERRUPTS/ENABLE_INTERRUPTS (aka "cli"/"sti"), you must
82 * specify what registers can be overwritten (CLBR_NONE, CLBR_EAX/EDX/ECX/ANY).
83 * Allowing a register to be clobbered can shrink the paravirt replacement
84 * enough to patch inline, increasing performance.
88 #define preempt_stop(clobbers) DISABLE_INTERRUPTS(clobbers); TRACE_IRQS_OFF
90 #define preempt_stop(clobbers)
91 #define resume_kernel restore_all
94 .macro TRACE_IRQS_IRET
95 #ifdef CONFIG_TRACE_IRQFLAGS
96 testl $X86_EFLAGS_IF,PT_EFLAGS(%esp) # interrupts off?
104 * User gs save/restore
106 * %gs is used for userland TLS and kernel only uses it for stack
107 * canary which is required to be at %gs:20 by gcc. Read the comment
108 * at the top of stackprotector.h for more info.
110 * Local labels 98 and 99 are used.
112 #ifdef CONFIG_X86_32_LAZY_GS
114 /* unfortunately push/pop can't be no-op */
119 addl $(4 + \pop), %esp
120 CFI_ADJUST_CFA_OFFSET -(4 + \pop)
125 /* all the rest are no-op */
132 .macro REG_TO_PTGS reg
134 .macro SET_KERNEL_GS reg
137 #else /* CONFIG_X86_32_LAZY_GS */
141 /*CFI_REL_OFFSET gs, 0*/
149 CFI_ADJUST_CFA_OFFSET -\pop
153 .pushsection .fixup, "ax"
157 _ASM_EXTABLE(98b,99b)
161 98: mov PT_GS(%esp), %gs
164 .pushsection .fixup, "ax"
165 99: movl $0, PT_GS(%esp)
168 _ASM_EXTABLE(98b,99b)
173 /*CFI_REGISTER gs, \reg*/
175 .macro REG_TO_PTGS reg
176 movl \reg, PT_GS(%esp)
177 /*CFI_REL_OFFSET gs, PT_GS*/
179 .macro SET_KERNEL_GS reg
180 movl $(__KERNEL_STACK_CANARY), \reg
184 #endif /* CONFIG_X86_32_LAZY_GS */
190 /*CFI_REL_OFFSET fs, 0;*/
192 /*CFI_REL_OFFSET es, 0;*/
194 /*CFI_REL_OFFSET ds, 0;*/
196 CFI_REL_OFFSET eax, 0
198 CFI_REL_OFFSET ebp, 0
200 CFI_REL_OFFSET edi, 0
202 CFI_REL_OFFSET esi, 0
204 CFI_REL_OFFSET edx, 0
206 CFI_REL_OFFSET ecx, 0
208 CFI_REL_OFFSET ebx, 0
209 movl $(__USER_DS), %edx
212 movl $(__KERNEL_PERCPU), %edx
217 .macro RESTORE_INT_REGS
234 .macro RESTORE_REGS pop=0
243 .pushsection .fixup, "ax"
257 .macro RING0_INT_FRAME
261 /*CFI_OFFSET cs, -2*4;*/
265 .macro RING0_EC_FRAME
269 /*CFI_OFFSET cs, -2*4;*/
273 .macro RING0_PTREGS_FRAME
276 CFI_DEF_CFA esp, PT_OLDESP-PT_EBX
277 /*CFI_OFFSET cs, PT_CS-PT_OLDESP;*/
278 CFI_OFFSET eip, PT_EIP-PT_OLDESP
279 /*CFI_OFFSET es, PT_ES-PT_OLDESP;*/
280 /*CFI_OFFSET ds, PT_DS-PT_OLDESP;*/
281 CFI_OFFSET eax, PT_EAX-PT_OLDESP
282 CFI_OFFSET ebp, PT_EBP-PT_OLDESP
283 CFI_OFFSET edi, PT_EDI-PT_OLDESP
284 CFI_OFFSET esi, PT_ESI-PT_OLDESP
285 CFI_OFFSET edx, PT_EDX-PT_OLDESP
286 CFI_OFFSET ecx, PT_ECX-PT_OLDESP
287 CFI_OFFSET ebx, PT_EBX-PT_OLDESP
294 GET_THREAD_INFO(%ebp)
296 pushl_cfi $0x0202 # Reset kernel eflags
303 * Interrupt exit functions should be protected against kprobes
305 .pushsection .kprobes.text, "ax"
307 * Return to user mode is not as complex as all this looks,
308 * but we want the default path for a system call return to
309 * go as quickly as possible which is why some of this is
310 * less clear than it otherwise should be.
313 # userspace resumption stub bypassing syscall exit tracing
317 preempt_stop(CLBR_ANY)
319 GET_THREAD_INFO(%ebp)
321 movl PT_EFLAGS(%esp), %eax # mix EFLAGS and CS
322 movb PT_CS(%esp), %al
323 andl $(X86_EFLAGS_VM | SEGMENT_RPL_MASK), %eax
326 * We can be coming here from a syscall done in the kernel space,
327 * e.g. a failed kernel_execve().
329 movl PT_CS(%esp), %eax
330 andl $SEGMENT_RPL_MASK, %eax
333 jb resume_kernel # not returning to v8086 or userspace
335 ENTRY(resume_userspace)
337 DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt
338 # setting need_resched or sigpending
339 # between sampling and the iret
341 movl TI_flags(%ebp), %ecx
342 andl $_TIF_WORK_MASK, %ecx # is there any work to be done on
343 # int/exception return?
346 END(ret_from_exception)
348 #ifdef CONFIG_PREEMPT
350 DISABLE_INTERRUPTS(CLBR_ANY)
351 cmpl $0,TI_preempt_count(%ebp) # non-zero preempt_count ?
354 movl TI_flags(%ebp), %ecx # need_resched set ?
355 testb $_TIF_NEED_RESCHED, %cl
357 testl $X86_EFLAGS_IF,PT_EFLAGS(%esp) # interrupts off (exception path) ?
359 call preempt_schedule_irq
365 * End of kprobes section
369 /* SYSENTER_RETURN points to after the "sysenter" instruction in
370 the vsyscall page. See vsyscall-sysentry.S, which defines the symbol. */
372 # sysenter call handler stub
373 ENTRY(ia32_sysenter_target)
377 CFI_REGISTER esp, ebp
378 movl TSS_sysenter_sp0(%esp),%esp
381 * Interrupts are disabled here, but we can't trace it until
382 * enough kernel state to call TRACE_IRQS_OFF can be called - but
383 * we immediately enable interrupts at that point anyway.
386 /*CFI_REL_OFFSET ss, 0*/
388 CFI_REL_OFFSET esp, 0
390 orl $X86_EFLAGS_IF, (%esp)
392 /*CFI_REL_OFFSET cs, 0*/
394 * Push current_thread_info()->sysenter_return to the stack.
395 * A tiny bit of offset fixup is necessary - 4*4 means the 4 words
396 * pushed above; +8 corresponds to copy_thread's esp0 setting.
398 pushl_cfi ((TI_sysenter_return)-THREAD_SIZE+8+4*4)(%esp)
399 CFI_REL_OFFSET eip, 0
403 ENABLE_INTERRUPTS(CLBR_NONE)
406 * Load the potential sixth argument from user stack.
407 * Careful about security.
409 cmpl $__PAGE_OFFSET-3,%ebp
414 movl %ebp,PT_EBP(%esp)
415 _ASM_EXTABLE(1b,syscall_fault)
417 GET_THREAD_INFO(%ebp)
419 testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%ebp)
422 cmpl $(NR_syscalls), %eax
424 call *sys_call_table(,%eax,4)
425 movl %eax,PT_EAX(%esp)
427 DISABLE_INTERRUPTS(CLBR_ANY)
429 movl TI_flags(%ebp), %ecx
430 testl $_TIF_ALLWORK_MASK, %ecx
433 /* if something modifies registers it must also disable sysexit */
434 movl PT_EIP(%esp), %edx
435 movl PT_OLDESP(%esp), %ecx
438 1: mov PT_FS(%esp), %fs
440 ENABLE_INTERRUPTS_SYSEXIT
442 #ifdef CONFIG_AUDITSYSCALL
444 testl $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags(%ebp)
445 jnz syscall_trace_entry
447 CFI_ADJUST_CFA_OFFSET -4
448 /* %esi already in 8(%esp) 6th arg: 4th syscall arg */
449 /* %edx already in 4(%esp) 5th arg: 3rd syscall arg */
450 /* %ecx already in 0(%esp) 4th arg: 2nd syscall arg */
451 movl %ebx,%ecx /* 3rd arg: 1st syscall arg */
452 movl %eax,%edx /* 2nd arg: syscall number */
453 movl $AUDIT_ARCH_I386,%eax /* 1st arg: audit arch */
454 call __audit_syscall_entry
456 movl PT_EAX(%esp),%eax /* reload syscall number */
460 testl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT), %ecx
461 jne syscall_exit_work
463 ENABLE_INTERRUPTS(CLBR_ANY)
464 movl %eax,%edx /* second arg, syscall return value */
465 cmpl $-MAX_ERRNO,%eax /* is it an error ? */
466 setbe %al /* 1 if so, 0 if not */
467 movzbl %al,%eax /* zero-extend that */
468 call __audit_syscall_exit
469 DISABLE_INTERRUPTS(CLBR_ANY)
471 movl TI_flags(%ebp), %ecx
472 testl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT), %ecx
473 jne syscall_exit_work
474 movl PT_EAX(%esp),%eax /* reload syscall return value */
479 .pushsection .fixup,"ax"
480 2: movl $0,PT_FS(%esp)
485 ENDPROC(ia32_sysenter_target)
488 * syscall stub including irq exit should be protected against kprobes
490 .pushsection .kprobes.text, "ax"
491 # system call handler stub
493 RING0_INT_FRAME # can't unwind into user space anyway
495 pushl_cfi %eax # save orig_eax
497 GET_THREAD_INFO(%ebp)
498 # system call tracing in operation / emulation
499 testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%ebp)
500 jnz syscall_trace_entry
501 cmpl $(NR_syscalls), %eax
504 call *sys_call_table(,%eax,4)
505 movl %eax,PT_EAX(%esp) # store the return value
508 DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt
509 # setting need_resched or sigpending
510 # between sampling and the iret
512 movl TI_flags(%ebp), %ecx
513 testl $_TIF_ALLWORK_MASK, %ecx # current->work
514 jne syscall_exit_work
519 movl PT_EFLAGS(%esp), %eax # mix EFLAGS, SS and CS
520 # Warning: PT_OLDSS(%esp) contains the wrong/random values if we
521 # are returning to the kernel.
522 # See comments in process.c:copy_thread() for details.
523 movb PT_OLDSS(%esp), %ah
524 movb PT_CS(%esp), %al
525 andl $(X86_EFLAGS_VM | (SEGMENT_TI_MASK << 8) | SEGMENT_RPL_MASK), %eax
526 cmpl $((SEGMENT_LDT << 8) | USER_RPL), %eax
528 je ldt_ss # returning to user-space with LDT SS
530 RESTORE_REGS 4 # skip orig_eax/error_code
535 pushl $0 # no error code
539 _ASM_EXTABLE(irq_return,iret_exc)
543 larl PT_OLDSS(%esp), %eax
545 testl $0x00400000, %eax # returning to 32bit stack?
546 jnz restore_nocheck # allright, normal return
548 #ifdef CONFIG_PARAVIRT
550 * The kernel can't run on a non-flat stack if paravirt mode
551 * is active. Rather than try to fixup the high bits of
552 * ESP, bypass this code entirely. This may break DOSemu
553 * and/or Wine support in a paravirt VM, although the option
554 * is still available to implement the setting of the high
555 * 16-bits in the INTERRUPT_RETURN paravirt-op.
557 cmpl $0, pv_info+PARAVIRT_enabled
562 * Setup and switch to ESPFIX stack
564 * We're returning to userspace with a 16 bit stack. The CPU will not
565 * restore the high word of ESP for us on executing iret... This is an
566 * "official" bug of all the x86-compatible CPUs, which we can work
567 * around to make dosemu and wine happy. We do this by preloading the
568 * high word of ESP with the high word of the userspace ESP while
569 * compensating for the offset by changing to the ESPFIX segment with
570 * a base address that matches for the difference.
572 #define GDT_ESPFIX_SS PER_CPU_VAR(gdt_page) + (GDT_ENTRY_ESPFIX_SS * 8)
573 mov %esp, %edx /* load kernel esp */
574 mov PT_OLDESP(%esp), %eax /* load userspace esp */
575 mov %dx, %ax /* eax: new kernel esp */
576 sub %eax, %edx /* offset (low word is 0) */
578 mov %dl, GDT_ESPFIX_SS + 4 /* bits 16..23 */
579 mov %dh, GDT_ESPFIX_SS + 7 /* bits 24..31 */
580 pushl_cfi $__ESPFIX_SS
581 pushl_cfi %eax /* new kernel esp */
582 /* Disable interrupts, but do not irqtrace this section: we
583 * will soon execute iret and the tracer was already set to
584 * the irqstate after the iret */
585 DISABLE_INTERRUPTS(CLBR_EAX)
586 lss (%esp), %esp /* switch to espfix segment */
587 CFI_ADJUST_CFA_OFFSET -8
592 # perform work that needs to be done immediately before resumption
594 RING0_PTREGS_FRAME # can't unwind into user space anyway
596 testb $_TIF_NEED_RESCHED, %cl
601 DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt
602 # setting need_resched or sigpending
603 # between sampling and the iret
605 movl TI_flags(%ebp), %ecx
606 andl $_TIF_WORK_MASK, %ecx # is there any work to be done other
607 # than syscall tracing?
609 testb $_TIF_NEED_RESCHED, %cl
612 work_notifysig: # deal with pending signals and
613 # notify-resume requests
615 testl $X86_EFLAGS_VM, PT_EFLAGS(%esp)
617 jne work_notifysig_v86 # returning to kernel-space or
620 ENABLE_INTERRUPTS(CLBR_NONE)
621 movb PT_CS(%esp), %bl
622 andb $SEGMENT_RPL_MASK, %bl
626 call do_notify_resume
631 pushl_cfi %ecx # save ti_flags for do_notify_resume
632 call save_v86_state # %eax contains pt_regs pointer
639 ENABLE_INTERRUPTS(CLBR_NONE)
640 movb PT_CS(%esp), %bl
641 andb $SEGMENT_RPL_MASK, %bl
645 call do_notify_resume
649 # perform syscall exit tracing
652 movl $-ENOSYS,PT_EAX(%esp)
654 call syscall_trace_enter
655 /* What it returned is what we'll actually use. */
656 cmpl $(NR_syscalls), %eax
659 END(syscall_trace_entry)
661 # perform syscall exit tracing
664 testl $_TIF_WORK_SYSCALL_EXIT, %ecx
667 ENABLE_INTERRUPTS(CLBR_ANY) # could let syscall_trace_leave() call
670 call syscall_trace_leave
672 END(syscall_exit_work)
675 RING0_INT_FRAME # can't unwind into user space anyway
678 GET_THREAD_INFO(%ebp)
679 movl $-EFAULT,PT_EAX(%esp)
684 movl $-ENOSYS,PT_EAX(%esp)
689 * End of kprobes section
694 * System calls that need a pt_regs pointer.
696 #define PTREGSCALL0(name) \
697 ENTRY(ptregs_##name) ; \
700 ENDPROC(ptregs_##name)
702 #define PTREGSCALL1(name) \
703 ENTRY(ptregs_##name) ; \
705 movl (PT_EBX+4)(%esp),%eax; \
707 ENDPROC(ptregs_##name)
709 #define PTREGSCALL2(name) \
710 ENTRY(ptregs_##name) ; \
712 movl (PT_ECX+4)(%esp),%edx; \
713 movl (PT_EBX+4)(%esp),%eax; \
715 ENDPROC(ptregs_##name)
717 #define PTREGSCALL3(name) \
718 ENTRY(ptregs_##name) ; \
722 movl PT_EDX(%eax),%ecx; \
723 movl PT_ECX(%eax),%edx; \
724 movl PT_EBX(%eax),%eax; \
727 CFI_ADJUST_CFA_OFFSET -4; \
730 ENDPROC(ptregs_##name)
736 PTREGSCALL2(sigaltstack)
737 PTREGSCALL0(sigreturn)
738 PTREGSCALL0(rt_sigreturn)
742 /* Clone is an oddball. The 4th arg is in %edi */
747 pushl_cfi PT_EDI(%eax)
748 movl PT_EDX(%eax),%ecx
749 movl PT_ECX(%eax),%edx
750 movl PT_EBX(%eax),%eax
753 CFI_ADJUST_CFA_OFFSET -8
756 ENDPROC(ptregs_clone)
758 .macro FIXUP_ESPFIX_STACK
760 * Switch back for ESPFIX stack to the normal zerobased stack
762 * We can't call C functions using the ESPFIX stack. This code reads
763 * the high word of the segment base from the GDT and swiches to the
764 * normal stack and adjusts ESP with the matching offset.
766 /* fixup the stack */
767 mov GDT_ESPFIX_SS + 4, %al /* bits 16..23 */
768 mov GDT_ESPFIX_SS + 7, %ah /* bits 24..31 */
770 addl %esp, %eax /* the adjusted stack pointer */
771 pushl_cfi $__KERNEL_DS
773 lss (%esp), %esp /* switch to the normal stack segment */
774 CFI_ADJUST_CFA_OFFSET -8
776 .macro UNWIND_ESPFIX_STACK
778 /* see if on espfix stack */
779 cmpw $__ESPFIX_SS, %ax
781 movl $__KERNEL_DS, %eax
784 /* switch to normal stack */
790 * Build the entry stubs and pointer table with some assembler magic.
791 * We pack 7 stubs into a single 32-byte chunk, which will fit in a
792 * single cache line on all modern x86 implementations.
794 .section .init.rodata,"a"
796 .section .entry.text, "ax"
798 .p2align CONFIG_X86_L1_CACHE_SHIFT
799 ENTRY(irq_entries_start)
801 vector=FIRST_EXTERNAL_VECTOR
802 .rept (NR_VECTORS-FIRST_EXTERNAL_VECTOR+6)/7
805 .if vector < NR_VECTORS
806 .if vector <> FIRST_EXTERNAL_VECTOR
807 CFI_ADJUST_CFA_OFFSET -4
809 1: pushl_cfi $(~vector+0x80) /* Note: always in signed byte range */
810 .if ((vector-FIRST_EXTERNAL_VECTOR)%7) <> 6
815 .section .entry.text, "ax"
819 2: jmp common_interrupt
821 END(irq_entries_start)
828 * the CPU automatically disables interrupts when executing an IRQ vector,
829 * so IRQ-flags tracing has to follow that:
831 .p2align CONFIG_X86_L1_CACHE_SHIFT
834 addl $-0x80,(%esp) /* Adjust vector into the [-256,-1] range */
840 ENDPROC(common_interrupt)
844 * Irq entries should be protected against kprobes
846 .pushsection .kprobes.text, "ax"
847 #define BUILD_INTERRUPT3(name, nr, fn) \
860 #define BUILD_INTERRUPT(name, nr) BUILD_INTERRUPT3(name, nr, smp_##name)
862 /* The include is where all of the SMP etc. interrupts come from */
863 #include <asm/entry_arch.h>
865 ENTRY(coprocessor_error)
869 pushl_cfi $do_coprocessor_error
872 END(coprocessor_error)
874 ENTRY(simd_coprocessor_error)
878 #ifdef CONFIG_X86_INVD_BUG
879 /* AMD 486 bug: invd from userspace calls exception 19 instead of #GP */
880 661: pushl_cfi $do_general_protection
882 .section .altinstructions,"a"
883 altinstruction_entry 661b, 663f, X86_FEATURE_XMM, 662b-661b, 664f-663f
885 .section .altinstr_replacement,"ax"
886 663: pushl $do_simd_coprocessor_error
890 pushl_cfi $do_simd_coprocessor_error
894 END(simd_coprocessor_error)
896 ENTRY(device_not_available)
899 pushl_cfi $-1 # mark this as an int
900 pushl_cfi $do_device_not_available
903 END(device_not_available)
905 #ifdef CONFIG_PARAVIRT
908 _ASM_EXTABLE(native_iret, iret_exc)
911 ENTRY(native_irq_enable_sysexit)
914 END(native_irq_enable_sysexit)
921 pushl_cfi $do_overflow
939 pushl_cfi $do_invalid_op
944 ENTRY(coprocessor_segment_overrun)
948 pushl_cfi $do_coprocessor_segment_overrun
951 END(coprocessor_segment_overrun)
956 pushl_cfi $do_invalid_TSS
961 ENTRY(segment_not_present)
964 pushl_cfi $do_segment_not_present
967 END(segment_not_present)
972 pushl_cfi $do_stack_segment
977 ENTRY(alignment_check)
980 pushl_cfi $do_alignment_check
988 pushl_cfi $0 # no error code
989 pushl_cfi $do_divide_error
994 #ifdef CONFIG_X86_MCE
999 pushl_cfi machine_check_vector
1005 ENTRY(spurious_interrupt_bug)
1009 pushl_cfi $do_spurious_interrupt_bug
1012 END(spurious_interrupt_bug)
1014 * End of kprobes section
1018 ENTRY(kernel_thread_helper)
1019 pushl $0 # fake return address for unwinder
1024 ud2 # padding for call trace
1026 ENDPROC(kernel_thread_helper)
1029 /* Xen doesn't set %esp to be precisely what the normal sysenter
1030 entrypoint expects, so fix it up before using the normal path. */
1031 ENTRY(xen_sysenter_target)
1033 addl $5*4, %esp /* remove xen-provided frame */
1034 CFI_ADJUST_CFA_OFFSET -5*4
1035 jmp sysenter_past_esp
1038 ENTRY(xen_hypervisor_callback)
1044 /* Check to see if we got the event in the critical
1045 region in xen_iret_direct, after we've reenabled
1046 events and checked for pending events. This simulates
1047 iret instruction's behaviour where it delivers a
1048 pending interrupt when enabling interrupts. */
1049 movl PT_EIP(%esp),%eax
1050 cmpl $xen_iret_start_crit,%eax
1052 cmpl $xen_iret_end_crit,%eax
1055 jmp xen_iret_crit_fixup
1057 ENTRY(xen_do_upcall)
1059 call xen_evtchn_do_upcall
1062 ENDPROC(xen_hypervisor_callback)
1064 # Hypervisor uses this for application faults while it executes.
1065 # We get here for two reasons:
1066 # 1. Fault while reloading DS, ES, FS or GS
1067 # 2. Fault while executing IRET
1068 # Category 1 we fix up by reattempting the load, and zeroing the segment
1069 # register if the load fails.
1070 # Category 2 we fix up by jumping to do_iret_error. We cannot use the
1071 # normal Linux return path in this case because if we use the IRET hypercall
1072 # to pop the stack frame we end up in an infinite loop of failsafe callbacks.
1073 # We distinguish between categories by maintaining a status value in EAX.
1074 ENTRY(xen_failsafe_callback)
1085 CFI_ADJUST_CFA_OFFSET -16
1088 jmp iret_exc # EAX != 0 => Category 2 (Bad IRET)
1089 5: pushl_cfi $0 # EAX == 0 => Category 1 (Bad segment)
1091 jmp ret_from_exception
1094 .section .fixup,"ax"
1112 ENDPROC(xen_failsafe_callback)
1114 BUILD_INTERRUPT3(xen_hvm_callback_vector, XEN_HVM_EVTCHN_CALLBACK,
1115 xen_evtchn_do_upcall)
1117 #endif /* CONFIG_XEN */
1119 #ifdef CONFIG_FUNCTION_TRACER
1120 #ifdef CONFIG_DYNAMIC_FTRACE
1126 ENTRY(ftrace_caller)
1127 cmpl $0, function_trace_stop
1133 pushl $0 /* Pass NULL as regs pointer */
1134 movl 4*4(%esp), %eax
1135 movl 0x4(%ebp), %edx
1136 leal function_trace_op, %ecx
1137 subl $MCOUNT_INSN_SIZE, %eax
1143 addl $4,%esp /* skip NULL pointer */
1148 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1149 .globl ftrace_graph_call
1159 ENTRY(ftrace_regs_caller)
1160 pushf /* push flags before compare (in cs location) */
1161 cmpl $0, function_trace_stop
1162 jne ftrace_restore_flags
1165 * i386 does not save SS and ESP when coming from kernel.
1166 * Instead, to get sp, ®s->sp is used (see ptrace.h).
1167 * Unfortunately, that means eflags must be at the same location
1168 * as the current return ip is. We move the return ip into the
1169 * ip location, and move flags into the return ip location.
1171 pushl 4(%esp) /* save return ip into ip slot */
1173 pushl $0 /* Load 0 into orig_ax */
1186 movl 13*4(%esp), %eax /* Get the saved flags */
1187 movl %eax, 14*4(%esp) /* Move saved flags into regs->flags location */
1188 /* clobbering return ip */
1189 movl $__KERNEL_CS,13*4(%esp)
1191 movl 12*4(%esp), %eax /* Load ip (1st parameter) */
1192 subl $MCOUNT_INSN_SIZE, %eax /* Adjust ip */
1193 movl 0x4(%ebp), %edx /* Load parent ip (2nd parameter) */
1194 leal function_trace_op, %ecx /* Save ftrace_pos in 3rd parameter */
1195 pushl %esp /* Save pt_regs as 4th parameter */
1197 GLOBAL(ftrace_regs_call)
1200 addl $4, %esp /* Skip pt_regs */
1201 movl 14*4(%esp), %eax /* Move flags back into cs */
1202 movl %eax, 13*4(%esp) /* Needed to keep addl from modifying flags */
1203 movl 12*4(%esp), %eax /* Get return ip from regs->ip */
1204 movl %eax, 14*4(%esp) /* Put return ip back for ret */
1217 addl $8, %esp /* Skip orig_ax and ip */
1218 popf /* Pop flags at end (no addl to corrupt flags) */
1221 ftrace_restore_flags:
1224 #else /* ! CONFIG_DYNAMIC_FTRACE */
1227 cmpl $0, function_trace_stop
1230 cmpl $ftrace_stub, ftrace_trace_function
1232 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1233 cmpl $ftrace_stub, ftrace_graph_return
1234 jnz ftrace_graph_caller
1236 cmpl $ftrace_graph_entry_stub, ftrace_graph_entry
1237 jnz ftrace_graph_caller
1243 /* taken from glibc */
1248 movl 0xc(%esp), %eax
1249 movl 0x4(%ebp), %edx
1250 subl $MCOUNT_INSN_SIZE, %eax
1252 call *ftrace_trace_function
1259 #endif /* CONFIG_DYNAMIC_FTRACE */
1260 #endif /* CONFIG_FUNCTION_TRACER */
1262 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1263 ENTRY(ftrace_graph_caller)
1267 movl 0xc(%esp), %edx
1270 subl $MCOUNT_INSN_SIZE, %edx
1271 call prepare_ftrace_return
1276 END(ftrace_graph_caller)
1278 .globl return_to_handler
1283 call ftrace_return_to_handler
1291 * Some functions should be protected against kprobes
1293 .pushsection .kprobes.text, "ax"
1298 pushl_cfi $do_page_fault
1301 /* the function address is in %gs's slot on the stack */
1303 /*CFI_REL_OFFSET fs, 0*/
1305 /*CFI_REL_OFFSET es, 0*/
1307 /*CFI_REL_OFFSET ds, 0*/
1309 CFI_REL_OFFSET eax, 0
1311 CFI_REL_OFFSET ebp, 0
1313 CFI_REL_OFFSET edi, 0
1315 CFI_REL_OFFSET esi, 0
1317 CFI_REL_OFFSET edx, 0
1319 CFI_REL_OFFSET ecx, 0
1321 CFI_REL_OFFSET ebx, 0
1323 movl $(__KERNEL_PERCPU), %ecx
1327 movl PT_GS(%esp), %edi # get the function address
1328 movl PT_ORIG_EAX(%esp), %edx # get the error code
1329 movl $-1, PT_ORIG_EAX(%esp) # no syscall to restart
1332 movl $(__USER_DS), %ecx
1336 movl %esp,%eax # pt_regs pointer
1338 jmp ret_from_exception
1343 * Debug traps and NMI can happen at the one SYSENTER instruction
1344 * that sets up the real kernel stack. Check here, since we can't
1345 * allow the wrong stack to be used.
1347 * "TSS_sysenter_sp0+12" is because the NMI/debug handler will have
1348 * already pushed 3 words if it hits on the sysenter instruction:
1349 * eflags, cs and eip.
1351 * We just load the right stack, and push the three (known) values
1352 * by hand onto the new stack - while updating the return eip past
1353 * the instruction that would have done it for sysenter.
1355 .macro FIX_STACK offset ok label
1356 cmpw $__KERNEL_CS, 4(%esp)
1359 movl TSS_sysenter_sp0 + \offset(%esp), %esp
1363 pushl_cfi $__KERNEL_CS
1364 pushl_cfi $sysenter_past_esp
1365 CFI_REL_OFFSET eip, 0
1371 cmpl $ia32_sysenter_target,(%esp)
1372 jne debug_stack_correct
1373 FIX_STACK 12, debug_stack_correct, debug_esp_fix_insn
1374 debug_stack_correct:
1375 pushl_cfi $-1 # mark this as an int
1378 xorl %edx,%edx # error code 0
1379 movl %esp,%eax # pt_regs pointer
1381 jmp ret_from_exception
1386 * NMI is doubly nasty. It can happen _while_ we're handling
1387 * a debug fault, and the debug fault hasn't yet been able to
1388 * clear up the stack. So we first check whether we got an
1389 * NMI on the sysenter entry path, but after that we need to
1390 * check whether we got an NMI on the debug path where the debug
1391 * fault happened on the sysenter path.
1398 cmpw $__ESPFIX_SS, %ax
1401 cmpl $ia32_sysenter_target,(%esp)
1405 /* Do not access memory above the end of our stack page,
1406 * it might not exist.
1408 andl $(THREAD_SIZE-1),%eax
1409 cmpl $(THREAD_SIZE-20),%eax
1411 jae nmi_stack_correct
1412 cmpl $ia32_sysenter_target,12(%esp)
1413 je nmi_debug_stack_check
1415 /* We have a RING0_INT_FRAME here */
1418 xorl %edx,%edx # zero error code
1419 movl %esp,%eax # pt_regs pointer
1421 jmp restore_all_notrace
1426 FIX_STACK 12, nmi_stack_correct, 1
1427 jmp nmi_stack_correct
1429 nmi_debug_stack_check:
1430 /* We have a RING0_INT_FRAME here */
1431 cmpw $__KERNEL_CS,16(%esp)
1432 jne nmi_stack_correct
1434 jb nmi_stack_correct
1435 cmpl $debug_esp_fix_insn,(%esp)
1436 ja nmi_stack_correct
1437 FIX_STACK 24, nmi_stack_correct, 1
1438 jmp nmi_stack_correct
1441 /* We have a RING0_INT_FRAME here.
1443 * create the pointer to lss back
1448 /* copy the iret frame of 12 bytes */
1454 FIXUP_ESPFIX_STACK # %eax == %esp
1455 xorl %edx,%edx # zero error code
1458 lss 12+4(%esp), %esp # back to espfix stack
1459 CFI_ADJUST_CFA_OFFSET -24
1467 pushl_cfi $-1 # mark this as an int
1470 xorl %edx,%edx # zero error code
1471 movl %esp,%eax # pt_regs pointer
1473 jmp ret_from_exception
1477 ENTRY(general_protection)
1479 pushl_cfi $do_general_protection
1482 END(general_protection)
1484 #ifdef CONFIG_KVM_GUEST
1485 ENTRY(async_page_fault)
1488 pushl_cfi $do_async_page_fault
1491 END(async_page_fault)
1495 * End of kprobes section