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>
61 /* Avoid __ASSEMBLER__'ifying <linux/audit.h> just for this. */
62 #include <linux/elf-em.h>
63 #define AUDIT_ARCH_I386 (EM_386|__AUDIT_ARCH_LE)
64 #define __AUDIT_ARCH_LE 0x40000000
66 #ifndef CONFIG_AUDITSYSCALL
67 #define sysenter_audit syscall_trace_entry
68 #define sysexit_audit syscall_exit_work
71 .section .entry.text, "ax"
74 * We use macros for low-level operations which need to be overridden
75 * for paravirtualization. The following will never clobber any registers:
76 * INTERRUPT_RETURN (aka. "iret")
77 * GET_CR0_INTO_EAX (aka. "movl %cr0, %eax")
78 * ENABLE_INTERRUPTS_SYSEXIT (aka "sti; sysexit").
80 * For DISABLE_INTERRUPTS/ENABLE_INTERRUPTS (aka "cli"/"sti"), you must
81 * specify what registers can be overwritten (CLBR_NONE, CLBR_EAX/EDX/ECX/ANY).
82 * Allowing a register to be clobbered can shrink the paravirt replacement
83 * enough to patch inline, increasing performance.
87 #define preempt_stop(clobbers) DISABLE_INTERRUPTS(clobbers); TRACE_IRQS_OFF
89 #define preempt_stop(clobbers)
90 #define resume_kernel restore_all
93 .macro TRACE_IRQS_IRET
94 #ifdef CONFIG_TRACE_IRQFLAGS
95 testl $X86_EFLAGS_IF,PT_EFLAGS(%esp) # interrupts off?
103 * User gs save/restore
105 * %gs is used for userland TLS and kernel only uses it for stack
106 * canary which is required to be at %gs:20 by gcc. Read the comment
107 * at the top of stackprotector.h for more info.
109 * Local labels 98 and 99 are used.
111 #ifdef CONFIG_X86_32_LAZY_GS
113 /* unfortunately push/pop can't be no-op */
118 addl $(4 + \pop), %esp
119 CFI_ADJUST_CFA_OFFSET -(4 + \pop)
124 /* all the rest are no-op */
131 .macro REG_TO_PTGS reg
133 .macro SET_KERNEL_GS reg
136 #else /* CONFIG_X86_32_LAZY_GS */
140 /*CFI_REL_OFFSET gs, 0*/
148 CFI_ADJUST_CFA_OFFSET -\pop
152 .pushsection .fixup, "ax"
156 _ASM_EXTABLE(98b,99b)
160 98: mov PT_GS(%esp), %gs
163 .pushsection .fixup, "ax"
164 99: movl $0, PT_GS(%esp)
167 _ASM_EXTABLE(98b,99b)
172 /*CFI_REGISTER gs, \reg*/
174 .macro REG_TO_PTGS reg
175 movl \reg, PT_GS(%esp)
176 /*CFI_REL_OFFSET gs, PT_GS*/
178 .macro SET_KERNEL_GS reg
179 movl $(__KERNEL_STACK_CANARY), \reg
183 #endif /* CONFIG_X86_32_LAZY_GS */
189 /*CFI_REL_OFFSET fs, 0;*/
191 /*CFI_REL_OFFSET es, 0;*/
193 /*CFI_REL_OFFSET ds, 0;*/
195 CFI_REL_OFFSET eax, 0
197 CFI_REL_OFFSET ebp, 0
199 CFI_REL_OFFSET edi, 0
201 CFI_REL_OFFSET esi, 0
203 CFI_REL_OFFSET edx, 0
205 CFI_REL_OFFSET ecx, 0
207 CFI_REL_OFFSET ebx, 0
208 movl $(__USER_DS), %edx
211 movl $(__KERNEL_PERCPU), %edx
216 .macro RESTORE_INT_REGS
233 .macro RESTORE_REGS pop=0
242 .pushsection .fixup, "ax"
256 .macro RING0_INT_FRAME
260 /*CFI_OFFSET cs, -2*4;*/
264 .macro RING0_EC_FRAME
268 /*CFI_OFFSET cs, -2*4;*/
272 .macro RING0_PTREGS_FRAME
275 CFI_DEF_CFA esp, PT_OLDESP-PT_EBX
276 /*CFI_OFFSET cs, PT_CS-PT_OLDESP;*/
277 CFI_OFFSET eip, PT_EIP-PT_OLDESP
278 /*CFI_OFFSET es, PT_ES-PT_OLDESP;*/
279 /*CFI_OFFSET ds, PT_DS-PT_OLDESP;*/
280 CFI_OFFSET eax, PT_EAX-PT_OLDESP
281 CFI_OFFSET ebp, PT_EBP-PT_OLDESP
282 CFI_OFFSET edi, PT_EDI-PT_OLDESP
283 CFI_OFFSET esi, PT_ESI-PT_OLDESP
284 CFI_OFFSET edx, PT_EDX-PT_OLDESP
285 CFI_OFFSET ecx, PT_ECX-PT_OLDESP
286 CFI_OFFSET ebx, PT_EBX-PT_OLDESP
293 GET_THREAD_INFO(%ebp)
295 pushl_cfi $0x0202 # Reset kernel eflags
301 ENTRY(ret_from_kernel_execve)
303 movl $PTREGS_SIZE, %ecx
306 GET_THREAD_INFO(%ebp)
308 END(ret_from_kernel_execve)
311 * Interrupt exit functions should be protected against kprobes
313 .pushsection .kprobes.text, "ax"
315 * Return to user mode is not as complex as all this looks,
316 * but we want the default path for a system call return to
317 * go as quickly as possible which is why some of this is
318 * less clear than it otherwise should be.
321 # userspace resumption stub bypassing syscall exit tracing
325 preempt_stop(CLBR_ANY)
327 GET_THREAD_INFO(%ebp)
329 movl PT_EFLAGS(%esp), %eax # mix EFLAGS and CS
330 movb PT_CS(%esp), %al
331 andl $(X86_EFLAGS_VM | SEGMENT_RPL_MASK), %eax
334 * We can be coming here from child spawned by kernel_thread().
336 movl PT_CS(%esp), %eax
337 andl $SEGMENT_RPL_MASK, %eax
340 jb resume_kernel # not returning to v8086 or userspace
342 ENTRY(resume_userspace)
344 DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt
345 # setting need_resched or sigpending
346 # between sampling and the iret
348 movl TI_flags(%ebp), %ecx
349 andl $_TIF_WORK_MASK, %ecx # is there any work to be done on
350 # int/exception return?
353 END(ret_from_exception)
355 #ifdef CONFIG_PREEMPT
357 DISABLE_INTERRUPTS(CLBR_ANY)
358 cmpl $0,TI_preempt_count(%ebp) # non-zero preempt_count ?
361 movl TI_flags(%ebp), %ecx # need_resched set ?
362 testb $_TIF_NEED_RESCHED, %cl
364 testl $X86_EFLAGS_IF,PT_EFLAGS(%esp) # interrupts off (exception path) ?
366 call preempt_schedule_irq
372 * End of kprobes section
376 /* SYSENTER_RETURN points to after the "sysenter" instruction in
377 the vsyscall page. See vsyscall-sysentry.S, which defines the symbol. */
379 # sysenter call handler stub
380 ENTRY(ia32_sysenter_target)
384 CFI_REGISTER esp, ebp
385 movl TSS_sysenter_sp0(%esp),%esp
388 * Interrupts are disabled here, but we can't trace it until
389 * enough kernel state to call TRACE_IRQS_OFF can be called - but
390 * we immediately enable interrupts at that point anyway.
393 /*CFI_REL_OFFSET ss, 0*/
395 CFI_REL_OFFSET esp, 0
397 orl $X86_EFLAGS_IF, (%esp)
399 /*CFI_REL_OFFSET cs, 0*/
401 * Push current_thread_info()->sysenter_return to the stack.
402 * A tiny bit of offset fixup is necessary - 4*4 means the 4 words
403 * pushed above; +8 corresponds to copy_thread's esp0 setting.
405 pushl_cfi ((TI_sysenter_return)-THREAD_SIZE+8+4*4)(%esp)
406 CFI_REL_OFFSET eip, 0
410 ENABLE_INTERRUPTS(CLBR_NONE)
413 * Load the potential sixth argument from user stack.
414 * Careful about security.
416 cmpl $__PAGE_OFFSET-3,%ebp
419 movl %ebp,PT_EBP(%esp)
420 _ASM_EXTABLE(1b,syscall_fault)
422 GET_THREAD_INFO(%ebp)
424 testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%ebp)
427 cmpl $(NR_syscalls), %eax
429 call *sys_call_table(,%eax,4)
430 movl %eax,PT_EAX(%esp)
432 DISABLE_INTERRUPTS(CLBR_ANY)
434 movl TI_flags(%ebp), %ecx
435 testl $_TIF_ALLWORK_MASK, %ecx
438 /* if something modifies registers it must also disable sysexit */
439 movl PT_EIP(%esp), %edx
440 movl PT_OLDESP(%esp), %ecx
443 1: mov PT_FS(%esp), %fs
445 ENABLE_INTERRUPTS_SYSEXIT
447 #ifdef CONFIG_AUDITSYSCALL
449 testl $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags(%ebp)
450 jnz syscall_trace_entry
452 CFI_ADJUST_CFA_OFFSET -4
453 /* %esi already in 8(%esp) 6th arg: 4th syscall arg */
454 /* %edx already in 4(%esp) 5th arg: 3rd syscall arg */
455 /* %ecx already in 0(%esp) 4th arg: 2nd syscall arg */
456 movl %ebx,%ecx /* 3rd arg: 1st syscall arg */
457 movl %eax,%edx /* 2nd arg: syscall number */
458 movl $AUDIT_ARCH_I386,%eax /* 1st arg: audit arch */
459 call __audit_syscall_entry
461 movl PT_EAX(%esp),%eax /* reload syscall number */
465 testl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT), %ecx
466 jne syscall_exit_work
468 ENABLE_INTERRUPTS(CLBR_ANY)
469 movl %eax,%edx /* second arg, syscall return value */
470 cmpl $-MAX_ERRNO,%eax /* is it an error ? */
471 setbe %al /* 1 if so, 0 if not */
472 movzbl %al,%eax /* zero-extend that */
473 call __audit_syscall_exit
474 DISABLE_INTERRUPTS(CLBR_ANY)
476 movl TI_flags(%ebp), %ecx
477 testl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT), %ecx
478 jne syscall_exit_work
479 movl PT_EAX(%esp),%eax /* reload syscall return value */
484 .pushsection .fixup,"ax"
485 2: movl $0,PT_FS(%esp)
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 _ASM_EXTABLE(irq_return,iret_exc)
547 larl PT_OLDSS(%esp), %eax
549 testl $0x00400000, %eax # returning to 32bit stack?
550 jnz restore_nocheck # allright, normal return
552 #ifdef CONFIG_PARAVIRT
554 * The kernel can't run on a non-flat stack if paravirt mode
555 * is active. Rather than try to fixup the high bits of
556 * ESP, bypass this code entirely. This may break DOSemu
557 * and/or Wine support in a paravirt VM, although the option
558 * is still available to implement the setting of the high
559 * 16-bits in the INTERRUPT_RETURN paravirt-op.
561 cmpl $0, pv_info+PARAVIRT_enabled
566 * Setup and switch to ESPFIX stack
568 * We're returning to userspace with a 16 bit stack. The CPU will not
569 * restore the high word of ESP for us on executing iret... This is an
570 * "official" bug of all the x86-compatible CPUs, which we can work
571 * around to make dosemu and wine happy. We do this by preloading the
572 * high word of ESP with the high word of the userspace ESP while
573 * compensating for the offset by changing to the ESPFIX segment with
574 * a base address that matches for the difference.
576 #define GDT_ESPFIX_SS PER_CPU_VAR(gdt_page) + (GDT_ENTRY_ESPFIX_SS * 8)
577 mov %esp, %edx /* load kernel esp */
578 mov PT_OLDESP(%esp), %eax /* load userspace esp */
579 mov %dx, %ax /* eax: new kernel esp */
580 sub %eax, %edx /* offset (low word is 0) */
582 mov %dl, GDT_ESPFIX_SS + 4 /* bits 16..23 */
583 mov %dh, GDT_ESPFIX_SS + 7 /* bits 24..31 */
584 pushl_cfi $__ESPFIX_SS
585 pushl_cfi %eax /* new kernel esp */
586 /* Disable interrupts, but do not irqtrace this section: we
587 * will soon execute iret and the tracer was already set to
588 * the irqstate after the iret */
589 DISABLE_INTERRUPTS(CLBR_EAX)
590 lss (%esp), %esp /* switch to espfix segment */
591 CFI_ADJUST_CFA_OFFSET -8
596 # perform work that needs to be done immediately before resumption
598 RING0_PTREGS_FRAME # can't unwind into user space anyway
600 testb $_TIF_NEED_RESCHED, %cl
605 DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt
606 # setting need_resched or sigpending
607 # between sampling and the iret
609 movl TI_flags(%ebp), %ecx
610 andl $_TIF_WORK_MASK, %ecx # is there any work to be done other
611 # than syscall tracing?
613 testb $_TIF_NEED_RESCHED, %cl
616 work_notifysig: # deal with pending signals and
617 # notify-resume requests
619 testl $X86_EFLAGS_VM, PT_EFLAGS(%esp)
621 jne work_notifysig_v86 # returning to kernel-space or
628 ENABLE_INTERRUPTS(CLBR_NONE)
629 movb PT_CS(%esp), %bl
630 andb $SEGMENT_RPL_MASK, %bl
634 call do_notify_resume
640 pushl_cfi %ecx # save ti_flags for do_notify_resume
641 call save_v86_state # %eax contains pt_regs pointer
648 # perform syscall exit tracing
651 movl $-ENOSYS,PT_EAX(%esp)
653 call syscall_trace_enter
654 /* What it returned is what we'll actually use. */
655 cmpl $(NR_syscalls), %eax
658 END(syscall_trace_entry)
660 # perform syscall exit tracing
663 testl $_TIF_WORK_SYSCALL_EXIT, %ecx
666 ENABLE_INTERRUPTS(CLBR_ANY) # could let syscall_trace_leave() call
669 call syscall_trace_leave
671 END(syscall_exit_work)
674 RING0_INT_FRAME # can't unwind into user space anyway
676 GET_THREAD_INFO(%ebp)
677 movl $-EFAULT,PT_EAX(%esp)
682 movl $-ENOSYS,PT_EAX(%esp)
687 * End of kprobes section
692 * System calls that need a pt_regs pointer.
694 #define PTREGSCALL0(name) \
695 ENTRY(ptregs_##name) ; \
698 ENDPROC(ptregs_##name)
700 #define PTREGSCALL1(name) \
701 ENTRY(ptregs_##name) ; \
703 movl (PT_EBX+4)(%esp),%eax; \
705 ENDPROC(ptregs_##name)
707 #define PTREGSCALL2(name) \
708 ENTRY(ptregs_##name) ; \
710 movl (PT_ECX+4)(%esp),%edx; \
711 movl (PT_EBX+4)(%esp),%eax; \
713 ENDPROC(ptregs_##name)
715 #define PTREGSCALL3(name) \
716 ENTRY(ptregs_##name) ; \
720 movl PT_EDX(%eax),%ecx; \
721 movl PT_ECX(%eax),%edx; \
722 movl PT_EBX(%eax),%eax; \
725 CFI_ADJUST_CFA_OFFSET -4; \
728 ENDPROC(ptregs_##name)
733 PTREGSCALL2(sigaltstack)
734 PTREGSCALL0(sigreturn)
735 PTREGSCALL0(rt_sigreturn)
739 /* 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"
876 altinstruction_entry 661b, 663f, X86_FEATURE_XMM, 662b-661b, 664f-663f
878 .section .altinstr_replacement,"ax"
879 663: pushl $do_simd_coprocessor_error
883 pushl_cfi $do_simd_coprocessor_error
887 END(simd_coprocessor_error)
889 ENTRY(device_not_available)
891 pushl_cfi $-1 # mark this as an int
892 pushl_cfi $do_device_not_available
895 END(device_not_available)
897 #ifdef CONFIG_PARAVIRT
900 _ASM_EXTABLE(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"
1093 ENDPROC(xen_failsafe_callback)
1095 BUILD_INTERRUPT3(xen_hvm_callback_vector, XEN_HVM_EVTCHN_CALLBACK,
1096 xen_evtchn_do_upcall)
1098 #endif /* CONFIG_XEN */
1100 #ifdef CONFIG_FUNCTION_TRACER
1101 #ifdef CONFIG_DYNAMIC_FTRACE
1107 ENTRY(ftrace_caller)
1108 cmpl $0, function_trace_stop
1114 pushl $0 /* Pass NULL as regs pointer */
1115 movl 4*4(%esp), %eax
1116 movl 0x4(%ebp), %edx
1117 leal function_trace_op, %ecx
1118 subl $MCOUNT_INSN_SIZE, %eax
1124 addl $4,%esp /* skip NULL pointer */
1129 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1130 .globl ftrace_graph_call
1140 ENTRY(ftrace_regs_caller)
1141 pushf /* push flags before compare (in cs location) */
1142 cmpl $0, function_trace_stop
1143 jne ftrace_restore_flags
1146 * i386 does not save SS and ESP when coming from kernel.
1147 * Instead, to get sp, ®s->sp is used (see ptrace.h).
1148 * Unfortunately, that means eflags must be at the same location
1149 * as the current return ip is. We move the return ip into the
1150 * ip location, and move flags into the return ip location.
1152 pushl 4(%esp) /* save return ip into ip slot */
1153 subl $MCOUNT_INSN_SIZE, (%esp) /* Adjust ip */
1155 pushl $0 /* Load 0 into orig_ax */
1168 movl 13*4(%esp), %eax /* Get the saved flags */
1169 movl %eax, 14*4(%esp) /* Move saved flags into regs->flags location */
1170 /* clobbering return ip */
1171 movl $__KERNEL_CS,13*4(%esp)
1173 movl 12*4(%esp), %eax /* Load ip (1st parameter) */
1174 movl 0x4(%ebp), %edx /* Load parent ip (2nd parameter) */
1175 leal function_trace_op, %ecx /* Save ftrace_pos in 3rd parameter */
1176 pushl %esp /* Save pt_regs as 4th parameter */
1178 GLOBAL(ftrace_regs_call)
1181 addl $4, %esp /* Skip pt_regs */
1182 movl 14*4(%esp), %eax /* Move flags back into cs */
1183 movl %eax, 13*4(%esp) /* Needed to keep addl from modifying flags */
1184 movl 12*4(%esp), %eax /* Get return ip from regs->ip */
1185 addl $MCOUNT_INSN_SIZE, %eax
1186 movl %eax, 14*4(%esp) /* Put return ip back for ret */
1199 addl $8, %esp /* Skip orig_ax and ip */
1200 popf /* Pop flags at end (no addl to corrupt flags) */
1203 ftrace_restore_flags:
1206 #else /* ! CONFIG_DYNAMIC_FTRACE */
1209 cmpl $0, function_trace_stop
1212 cmpl $ftrace_stub, ftrace_trace_function
1214 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1215 cmpl $ftrace_stub, ftrace_graph_return
1216 jnz ftrace_graph_caller
1218 cmpl $ftrace_graph_entry_stub, ftrace_graph_entry
1219 jnz ftrace_graph_caller
1225 /* taken from glibc */
1230 movl 0xc(%esp), %eax
1231 movl 0x4(%ebp), %edx
1232 subl $MCOUNT_INSN_SIZE, %eax
1234 call *ftrace_trace_function
1241 #endif /* CONFIG_DYNAMIC_FTRACE */
1242 #endif /* CONFIG_FUNCTION_TRACER */
1244 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1245 ENTRY(ftrace_graph_caller)
1249 movl 0xc(%esp), %edx
1252 subl $MCOUNT_INSN_SIZE, %edx
1253 call prepare_ftrace_return
1258 END(ftrace_graph_caller)
1260 .globl return_to_handler
1265 call ftrace_return_to_handler
1273 * Some functions should be protected against kprobes
1275 .pushsection .kprobes.text, "ax"
1279 pushl_cfi $do_page_fault
1282 /* the function address is in %gs's slot on the stack */
1284 /*CFI_REL_OFFSET fs, 0*/
1286 /*CFI_REL_OFFSET es, 0*/
1288 /*CFI_REL_OFFSET ds, 0*/
1290 CFI_REL_OFFSET eax, 0
1292 CFI_REL_OFFSET ebp, 0
1294 CFI_REL_OFFSET edi, 0
1296 CFI_REL_OFFSET esi, 0
1298 CFI_REL_OFFSET edx, 0
1300 CFI_REL_OFFSET ecx, 0
1302 CFI_REL_OFFSET ebx, 0
1304 movl $(__KERNEL_PERCPU), %ecx
1308 movl PT_GS(%esp), %edi # get the function address
1309 movl PT_ORIG_EAX(%esp), %edx # get the error code
1310 movl $-1, PT_ORIG_EAX(%esp) # no syscall to restart
1313 movl $(__USER_DS), %ecx
1317 movl %esp,%eax # pt_regs pointer
1319 jmp ret_from_exception
1324 * Debug traps and NMI can happen at the one SYSENTER instruction
1325 * that sets up the real kernel stack. Check here, since we can't
1326 * allow the wrong stack to be used.
1328 * "TSS_sysenter_sp0+12" is because the NMI/debug handler will have
1329 * already pushed 3 words if it hits on the sysenter instruction:
1330 * eflags, cs and eip.
1332 * We just load the right stack, and push the three (known) values
1333 * by hand onto the new stack - while updating the return eip past
1334 * the instruction that would have done it for sysenter.
1336 .macro FIX_STACK offset ok label
1337 cmpw $__KERNEL_CS, 4(%esp)
1340 movl TSS_sysenter_sp0 + \offset(%esp), %esp
1344 pushl_cfi $__KERNEL_CS
1345 pushl_cfi $sysenter_past_esp
1346 CFI_REL_OFFSET eip, 0
1351 cmpl $ia32_sysenter_target,(%esp)
1352 jne debug_stack_correct
1353 FIX_STACK 12, debug_stack_correct, debug_esp_fix_insn
1354 debug_stack_correct:
1355 pushl_cfi $-1 # mark this as an int
1358 xorl %edx,%edx # error code 0
1359 movl %esp,%eax # pt_regs pointer
1361 jmp ret_from_exception
1366 * NMI is doubly nasty. It can happen _while_ we're handling
1367 * a debug fault, and the debug fault hasn't yet been able to
1368 * clear up the stack. So we first check whether we got an
1369 * NMI on the sysenter entry path, but after that we need to
1370 * check whether we got an NMI on the debug path where the debug
1371 * fault happened on the sysenter path.
1377 cmpw $__ESPFIX_SS, %ax
1380 cmpl $ia32_sysenter_target,(%esp)
1384 /* Do not access memory above the end of our stack page,
1385 * it might not exist.
1387 andl $(THREAD_SIZE-1),%eax
1388 cmpl $(THREAD_SIZE-20),%eax
1390 jae nmi_stack_correct
1391 cmpl $ia32_sysenter_target,12(%esp)
1392 je nmi_debug_stack_check
1394 /* We have a RING0_INT_FRAME here */
1397 xorl %edx,%edx # zero error code
1398 movl %esp,%eax # pt_regs pointer
1400 jmp restore_all_notrace
1405 FIX_STACK 12, nmi_stack_correct, 1
1406 jmp nmi_stack_correct
1408 nmi_debug_stack_check:
1409 /* We have a RING0_INT_FRAME here */
1410 cmpw $__KERNEL_CS,16(%esp)
1411 jne nmi_stack_correct
1413 jb nmi_stack_correct
1414 cmpl $debug_esp_fix_insn,(%esp)
1415 ja nmi_stack_correct
1416 FIX_STACK 24, nmi_stack_correct, 1
1417 jmp nmi_stack_correct
1420 /* We have a RING0_INT_FRAME here.
1422 * create the pointer to lss back
1427 /* copy the iret frame of 12 bytes */
1433 FIXUP_ESPFIX_STACK # %eax == %esp
1434 xorl %edx,%edx # zero error code
1437 lss 12+4(%esp), %esp # back to espfix stack
1438 CFI_ADJUST_CFA_OFFSET -24
1445 pushl_cfi $-1 # mark this as an int
1448 xorl %edx,%edx # zero error code
1449 movl %esp,%eax # pt_regs pointer
1451 jmp ret_from_exception
1455 ENTRY(general_protection)
1457 pushl_cfi $do_general_protection
1460 END(general_protection)
1462 #ifdef CONFIG_KVM_GUEST
1463 ENTRY(async_page_fault)
1465 pushl_cfi $do_async_page_fault
1468 END(async_page_fault)
1472 * End of kprobes section