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
302 ENTRY(ret_from_kernel_thread)
306 GET_THREAD_INFO(%ebp)
308 pushl_cfi $0x0202 # Reset kernel eflags
310 movl PT_EBP(%esp),%eax
315 ENDPROC(ret_from_kernel_thread)
318 * Return to user mode is not as complex as all this looks,
319 * but we want the default path for a system call return to
320 * go as quickly as possible which is why some of this is
321 * less clear than it otherwise should be.
324 # userspace resumption stub bypassing syscall exit tracing
328 preempt_stop(CLBR_ANY)
330 GET_THREAD_INFO(%ebp)
332 movl PT_EFLAGS(%esp), %eax # mix EFLAGS and CS
333 movb PT_CS(%esp), %al
334 andl $(X86_EFLAGS_VM | SEGMENT_RPL_MASK), %eax
337 * We can be coming here from child spawned by kernel_thread().
339 movl PT_CS(%esp), %eax
340 andl $SEGMENT_RPL_MASK, %eax
343 jb resume_kernel # not returning to v8086 or userspace
345 ENTRY(resume_userspace)
347 DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt
348 # setting need_resched or sigpending
349 # between sampling and the iret
351 movl TI_flags(%ebp), %ecx
352 andl $_TIF_WORK_MASK, %ecx # is there any work to be done on
353 # int/exception return?
356 END(ret_from_exception)
358 #ifdef CONFIG_PREEMPT
360 DISABLE_INTERRUPTS(CLBR_ANY)
362 cmpl $0,PER_CPU_VAR(__preempt_count)
364 testl $X86_EFLAGS_IF,PT_EFLAGS(%esp) # interrupts off (exception path) ?
366 call preempt_schedule_irq
372 /* SYSENTER_RETURN points to after the "sysenter" instruction in
373 the vsyscall page. See vsyscall-sysentry.S, which defines the symbol. */
375 # sysenter call handler stub
376 ENTRY(ia32_sysenter_target)
380 CFI_REGISTER esp, ebp
381 movl TSS_sysenter_sp0(%esp),%esp
384 * Interrupts are disabled here, but we can't trace it until
385 * enough kernel state to call TRACE_IRQS_OFF can be called - but
386 * we immediately enable interrupts at that point anyway.
389 /*CFI_REL_OFFSET ss, 0*/
391 CFI_REL_OFFSET esp, 0
393 orl $X86_EFLAGS_IF, (%esp)
395 /*CFI_REL_OFFSET cs, 0*/
397 * Push current_thread_info()->sysenter_return to the stack.
398 * A tiny bit of offset fixup is necessary - 4*4 means the 4 words
399 * pushed above; +8 corresponds to copy_thread's esp0 setting.
401 pushl_cfi ((TI_sysenter_return)-THREAD_SIZE+8+4*4)(%esp)
402 CFI_REL_OFFSET eip, 0
406 ENABLE_INTERRUPTS(CLBR_NONE)
409 * Load the potential sixth argument from user stack.
410 * Careful about security.
412 cmpl $__PAGE_OFFSET-3,%ebp
417 movl %ebp,PT_EBP(%esp)
418 _ASM_EXTABLE(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)
429 movl %eax,PT_EAX(%esp)
431 DISABLE_INTERRUPTS(CLBR_ANY)
433 movl TI_flags(%ebp), %ecx
434 testl $_TIF_ALLWORK_MASK, %ecx
437 /* if something modifies registers it must also disable sysexit */
438 movl PT_EIP(%esp), %edx
439 movl PT_OLDESP(%esp), %ecx
442 1: mov PT_FS(%esp), %fs
444 ENABLE_INTERRUPTS_SYSEXIT
446 #ifdef CONFIG_AUDITSYSCALL
448 testl $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags(%ebp)
449 jnz syscall_trace_entry
451 CFI_ADJUST_CFA_OFFSET -4
452 /* %esi already in 8(%esp) 6th arg: 4th syscall arg */
453 /* %edx already in 4(%esp) 5th arg: 3rd syscall arg */
454 /* %ecx already in 0(%esp) 4th arg: 2nd syscall arg */
455 movl %ebx,%ecx /* 3rd arg: 1st syscall arg */
456 movl %eax,%edx /* 2nd arg: syscall number */
457 movl $AUDIT_ARCH_I386,%eax /* 1st arg: audit arch */
458 call __audit_syscall_entry
460 movl PT_EAX(%esp),%eax /* reload syscall number */
464 testl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT), %ecx
465 jne syscall_exit_work
467 ENABLE_INTERRUPTS(CLBR_ANY)
468 movl %eax,%edx /* second arg, syscall return value */
469 cmpl $-MAX_ERRNO,%eax /* is it an error ? */
470 setbe %al /* 1 if so, 0 if not */
471 movzbl %al,%eax /* zero-extend that */
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)
489 ENDPROC(ia32_sysenter_target)
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)
506 movl %eax,PT_EAX(%esp) # store the return value
509 DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt
510 # setting need_resched or sigpending
511 # between sampling and the iret
513 movl TI_flags(%ebp), %ecx
514 testl $_TIF_ALLWORK_MASK, %ecx # current->work
515 jne syscall_exit_work
520 #ifdef CONFIG_X86_ESPFIX32
521 movl PT_EFLAGS(%esp), %eax # mix EFLAGS, SS and CS
522 # Warning: PT_OLDSS(%esp) contains the wrong/random values if we
523 # are returning to the kernel.
524 # See comments in process.c:copy_thread() for details.
525 movb PT_OLDSS(%esp), %ah
526 movb PT_CS(%esp), %al
527 andl $(X86_EFLAGS_VM | (SEGMENT_TI_MASK << 8) | SEGMENT_RPL_MASK), %eax
528 cmpl $((SEGMENT_LDT << 8) | USER_RPL), %eax
530 je ldt_ss # returning to user-space with LDT SS
533 RESTORE_REGS 4 # skip orig_eax/error_code
538 pushl $0 # no error code
542 _ASM_EXTABLE(irq_return,iret_exc)
544 #ifdef CONFIG_X86_ESPFIX32
547 #ifdef CONFIG_PARAVIRT
549 * The kernel can't run on a non-flat stack if paravirt mode
550 * is active. Rather than try to fixup the high bits of
551 * ESP, bypass this code entirely. This may break DOSemu
552 * and/or Wine support in a paravirt VM, although the option
553 * is still available to implement the setting of the high
554 * 16-bits in the INTERRUPT_RETURN paravirt-op.
556 cmpl $0, pv_info+PARAVIRT_enabled
561 * Setup and switch to ESPFIX stack
563 * We're returning to userspace with a 16 bit stack. The CPU will not
564 * restore the high word of ESP for us on executing iret... This is an
565 * "official" bug of all the x86-compatible CPUs, which we can work
566 * around to make dosemu and wine happy. We do this by preloading the
567 * high word of ESP with the high word of the userspace ESP while
568 * compensating for the offset by changing to the ESPFIX segment with
569 * a base address that matches for the difference.
571 #define GDT_ESPFIX_SS PER_CPU_VAR(gdt_page) + (GDT_ENTRY_ESPFIX_SS * 8)
572 mov %esp, %edx /* load kernel esp */
573 mov PT_OLDESP(%esp), %eax /* load userspace esp */
574 mov %dx, %ax /* eax: new kernel esp */
575 sub %eax, %edx /* offset (low word is 0) */
577 mov %dl, GDT_ESPFIX_SS + 4 /* bits 16..23 */
578 mov %dh, GDT_ESPFIX_SS + 7 /* bits 24..31 */
579 pushl_cfi $__ESPFIX_SS
580 pushl_cfi %eax /* new kernel esp */
581 /* Disable interrupts, but do not irqtrace this section: we
582 * will soon execute iret and the tracer was already set to
583 * the irqstate after the iret */
584 DISABLE_INTERRUPTS(CLBR_EAX)
585 lss (%esp), %esp /* switch to espfix segment */
586 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
624 ENABLE_INTERRUPTS(CLBR_NONE)
625 movb PT_CS(%esp), %bl
626 andb $SEGMENT_RPL_MASK, %bl
630 call do_notify_resume
636 pushl_cfi %ecx # save ti_flags for do_notify_resume
637 call save_v86_state # %eax contains pt_regs pointer
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
673 GET_THREAD_INFO(%ebp)
674 movl $-EFAULT,PT_EAX(%esp)
680 jmp syscall_after_call
685 jmp sysenter_after_call
689 .macro FIXUP_ESPFIX_STACK
691 * Switch back for ESPFIX stack to the normal zerobased stack
693 * We can't call C functions using the ESPFIX stack. This code reads
694 * the high word of the segment base from the GDT and swiches to the
695 * normal stack and adjusts ESP with the matching offset.
697 #ifdef CONFIG_X86_ESPFIX32
698 /* fixup the stack */
699 mov GDT_ESPFIX_SS + 4, %al /* bits 16..23 */
700 mov GDT_ESPFIX_SS + 7, %ah /* bits 24..31 */
702 addl %esp, %eax /* the adjusted stack pointer */
703 pushl_cfi $__KERNEL_DS
705 lss (%esp), %esp /* switch to the normal stack segment */
706 CFI_ADJUST_CFA_OFFSET -8
709 .macro UNWIND_ESPFIX_STACK
710 #ifdef CONFIG_X86_ESPFIX32
712 /* see if on espfix stack */
713 cmpw $__ESPFIX_SS, %ax
715 movl $__KERNEL_DS, %eax
718 /* switch to normal stack */
725 * Build the entry stubs and pointer table with some assembler magic.
726 * We pack 7 stubs into a single 32-byte chunk, which will fit in a
727 * single cache line on all modern x86 implementations.
729 .section .init.rodata,"a"
731 .section .entry.text, "ax"
733 .p2align CONFIG_X86_L1_CACHE_SHIFT
734 ENTRY(irq_entries_start)
736 vector=FIRST_EXTERNAL_VECTOR
737 .rept (NR_VECTORS-FIRST_EXTERNAL_VECTOR+6)/7
740 .if vector < NR_VECTORS
741 .if vector <> FIRST_EXTERNAL_VECTOR
742 CFI_ADJUST_CFA_OFFSET -4
744 1: pushl_cfi $(~vector+0x80) /* Note: always in signed byte range */
745 .if ((vector-FIRST_EXTERNAL_VECTOR)%7) <> 6
750 .section .entry.text, "ax"
754 2: jmp common_interrupt
756 END(irq_entries_start)
763 * the CPU automatically disables interrupts when executing an IRQ vector,
764 * so IRQ-flags tracing has to follow that:
766 .p2align CONFIG_X86_L1_CACHE_SHIFT
769 addl $-0x80,(%esp) /* Adjust vector into the [-256,-1] range */
775 ENDPROC(common_interrupt)
778 #define BUILD_INTERRUPT3(name, nr, fn) \
792 #ifdef CONFIG_TRACING
793 #define TRACE_BUILD_INTERRUPT(name, nr) \
794 BUILD_INTERRUPT3(trace_##name, nr, smp_trace_##name)
796 #define TRACE_BUILD_INTERRUPT(name, nr)
799 #define BUILD_INTERRUPT(name, nr) \
800 BUILD_INTERRUPT3(name, nr, smp_##name); \
801 TRACE_BUILD_INTERRUPT(name, nr)
803 /* The include is where all of the SMP etc. interrupts come from */
804 #include <asm/entry_arch.h>
806 ENTRY(coprocessor_error)
810 pushl_cfi $do_coprocessor_error
813 END(coprocessor_error)
815 ENTRY(simd_coprocessor_error)
819 #ifdef CONFIG_X86_INVD_BUG
820 /* AMD 486 bug: invd from userspace calls exception 19 instead of #GP */
821 661: pushl_cfi $do_general_protection
823 .section .altinstructions,"a"
824 altinstruction_entry 661b, 663f, X86_FEATURE_XMM, 662b-661b, 664f-663f
826 .section .altinstr_replacement,"ax"
827 663: pushl $do_simd_coprocessor_error
831 pushl_cfi $do_simd_coprocessor_error
835 END(simd_coprocessor_error)
837 ENTRY(device_not_available)
840 pushl_cfi $-1 # mark this as an int
841 pushl_cfi $do_device_not_available
844 END(device_not_available)
846 #ifdef CONFIG_PARAVIRT
849 _ASM_EXTABLE(native_iret, iret_exc)
852 ENTRY(native_irq_enable_sysexit)
855 END(native_irq_enable_sysexit)
862 pushl_cfi $do_overflow
880 pushl_cfi $do_invalid_op
885 ENTRY(coprocessor_segment_overrun)
889 pushl_cfi $do_coprocessor_segment_overrun
892 END(coprocessor_segment_overrun)
897 pushl_cfi $do_invalid_TSS
902 ENTRY(segment_not_present)
905 pushl_cfi $do_segment_not_present
908 END(segment_not_present)
913 pushl_cfi $do_stack_segment
918 ENTRY(alignment_check)
921 pushl_cfi $do_alignment_check
929 pushl_cfi $0 # no error code
930 pushl_cfi $do_divide_error
935 #ifdef CONFIG_X86_MCE
940 pushl_cfi machine_check_vector
946 ENTRY(spurious_interrupt_bug)
950 pushl_cfi $do_spurious_interrupt_bug
953 END(spurious_interrupt_bug)
956 /* Xen doesn't set %esp to be precisely what the normal sysenter
957 entrypoint expects, so fix it up before using the normal path. */
958 ENTRY(xen_sysenter_target)
960 addl $5*4, %esp /* remove xen-provided frame */
961 CFI_ADJUST_CFA_OFFSET -5*4
962 jmp sysenter_past_esp
965 ENTRY(xen_hypervisor_callback)
967 pushl_cfi $-1 /* orig_ax = -1 => not a system call */
971 /* Check to see if we got the event in the critical
972 region in xen_iret_direct, after we've reenabled
973 events and checked for pending events. This simulates
974 iret instruction's behaviour where it delivers a
975 pending interrupt when enabling interrupts. */
976 movl PT_EIP(%esp),%eax
977 cmpl $xen_iret_start_crit,%eax
979 cmpl $xen_iret_end_crit,%eax
982 jmp xen_iret_crit_fixup
986 call xen_evtchn_do_upcall
989 ENDPROC(xen_hypervisor_callback)
991 # Hypervisor uses this for application faults while it executes.
992 # We get here for two reasons:
993 # 1. Fault while reloading DS, ES, FS or GS
994 # 2. Fault while executing IRET
995 # Category 1 we fix up by reattempting the load, and zeroing the segment
996 # register if the load fails.
997 # Category 2 we fix up by jumping to do_iret_error. We cannot use the
998 # normal Linux return path in this case because if we use the IRET hypercall
999 # to pop the stack frame we end up in an infinite loop of failsafe callbacks.
1000 # We distinguish between categories by maintaining a status value in EAX.
1001 ENTRY(xen_failsafe_callback)
1009 /* EAX == 0 => Category 1 (Bad segment)
1010 EAX != 0 => Category 2 (Bad IRET) */
1014 CFI_ADJUST_CFA_OFFSET -16
1017 5: pushl_cfi $-1 /* orig_ax = -1 => not a system call */
1019 jmp ret_from_exception
1022 .section .fixup,"ax"
1040 ENDPROC(xen_failsafe_callback)
1042 BUILD_INTERRUPT3(xen_hvm_callback_vector, HYPERVISOR_CALLBACK_VECTOR,
1043 xen_evtchn_do_upcall)
1045 #endif /* CONFIG_XEN */
1047 #if IS_ENABLED(CONFIG_HYPERV)
1049 BUILD_INTERRUPT3(hyperv_callback_vector, HYPERVISOR_CALLBACK_VECTOR,
1050 hyperv_vector_handler)
1052 #endif /* CONFIG_HYPERV */
1054 #ifdef CONFIG_FUNCTION_TRACER
1055 #ifdef CONFIG_DYNAMIC_FTRACE
1061 ENTRY(ftrace_caller)
1065 pushl $0 /* Pass NULL as regs pointer */
1066 movl 4*4(%esp), %eax
1067 movl 0x4(%ebp), %edx
1068 movl function_trace_op, %ecx
1069 subl $MCOUNT_INSN_SIZE, %eax
1075 addl $4,%esp /* skip NULL pointer */
1080 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1081 .globl ftrace_graph_call
1091 ENTRY(ftrace_regs_caller)
1092 pushf /* push flags before compare (in cs location) */
1095 * i386 does not save SS and ESP when coming from kernel.
1096 * Instead, to get sp, ®s->sp is used (see ptrace.h).
1097 * Unfortunately, that means eflags must be at the same location
1098 * as the current return ip is. We move the return ip into the
1099 * ip location, and move flags into the return ip location.
1101 pushl 4(%esp) /* save return ip into ip slot */
1103 pushl $0 /* Load 0 into orig_ax */
1116 movl 13*4(%esp), %eax /* Get the saved flags */
1117 movl %eax, 14*4(%esp) /* Move saved flags into regs->flags location */
1118 /* clobbering return ip */
1119 movl $__KERNEL_CS,13*4(%esp)
1121 movl 12*4(%esp), %eax /* Load ip (1st parameter) */
1122 subl $MCOUNT_INSN_SIZE, %eax /* Adjust ip */
1123 movl 0x4(%ebp), %edx /* Load parent ip (2nd parameter) */
1124 movl function_trace_op, %ecx /* Save ftrace_pos in 3rd parameter */
1125 pushl %esp /* Save pt_regs as 4th parameter */
1127 GLOBAL(ftrace_regs_call)
1130 addl $4, %esp /* Skip pt_regs */
1131 movl 14*4(%esp), %eax /* Move flags back into cs */
1132 movl %eax, 13*4(%esp) /* Needed to keep addl from modifying flags */
1133 movl 12*4(%esp), %eax /* Get return ip from regs->ip */
1134 movl %eax, 14*4(%esp) /* Put return ip back for ret */
1147 addl $8, %esp /* Skip orig_ax and ip */
1148 popf /* Pop flags at end (no addl to corrupt flags) */
1153 #else /* ! CONFIG_DYNAMIC_FTRACE */
1156 cmpl $__PAGE_OFFSET, %esp
1157 jb ftrace_stub /* Paging not enabled yet? */
1159 cmpl $ftrace_stub, ftrace_trace_function
1161 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1162 cmpl $ftrace_stub, ftrace_graph_return
1163 jnz ftrace_graph_caller
1165 cmpl $ftrace_graph_entry_stub, ftrace_graph_entry
1166 jnz ftrace_graph_caller
1172 /* taken from glibc */
1177 movl 0xc(%esp), %eax
1178 movl 0x4(%ebp), %edx
1179 subl $MCOUNT_INSN_SIZE, %eax
1181 call *ftrace_trace_function
1188 #endif /* CONFIG_DYNAMIC_FTRACE */
1189 #endif /* CONFIG_FUNCTION_TRACER */
1191 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1192 ENTRY(ftrace_graph_caller)
1196 movl 0xc(%esp), %edx
1199 subl $MCOUNT_INSN_SIZE, %edx
1200 call prepare_ftrace_return
1205 END(ftrace_graph_caller)
1207 .globl return_to_handler
1212 call ftrace_return_to_handler
1219 #ifdef CONFIG_TRACING
1220 ENTRY(trace_page_fault)
1223 pushl_cfi $trace_do_page_fault
1226 END(trace_page_fault)
1232 pushl_cfi $do_page_fault
1235 /* the function address is in %gs's slot on the stack */
1237 /*CFI_REL_OFFSET fs, 0*/
1239 /*CFI_REL_OFFSET es, 0*/
1241 /*CFI_REL_OFFSET ds, 0*/
1243 CFI_REL_OFFSET eax, 0
1245 CFI_REL_OFFSET ebp, 0
1247 CFI_REL_OFFSET edi, 0
1249 CFI_REL_OFFSET esi, 0
1251 CFI_REL_OFFSET edx, 0
1253 CFI_REL_OFFSET ecx, 0
1255 CFI_REL_OFFSET ebx, 0
1257 movl $(__KERNEL_PERCPU), %ecx
1261 movl PT_GS(%esp), %edi # get the function address
1262 movl PT_ORIG_EAX(%esp), %edx # get the error code
1263 movl $-1, PT_ORIG_EAX(%esp) # no syscall to restart
1266 movl $(__USER_DS), %ecx
1270 movl %esp,%eax # pt_regs pointer
1272 jmp ret_from_exception
1277 * Debug traps and NMI can happen at the one SYSENTER instruction
1278 * that sets up the real kernel stack. Check here, since we can't
1279 * allow the wrong stack to be used.
1281 * "TSS_sysenter_sp0+12" is because the NMI/debug handler will have
1282 * already pushed 3 words if it hits on the sysenter instruction:
1283 * eflags, cs and eip.
1285 * We just load the right stack, and push the three (known) values
1286 * by hand onto the new stack - while updating the return eip past
1287 * the instruction that would have done it for sysenter.
1289 .macro FIX_STACK offset ok label
1290 cmpw $__KERNEL_CS, 4(%esp)
1293 movl TSS_sysenter_sp0 + \offset(%esp), %esp
1297 pushl_cfi $__KERNEL_CS
1298 pushl_cfi $sysenter_past_esp
1299 CFI_REL_OFFSET eip, 0
1305 cmpl $ia32_sysenter_target,(%esp)
1306 jne debug_stack_correct
1307 FIX_STACK 12, debug_stack_correct, debug_esp_fix_insn
1308 debug_stack_correct:
1309 pushl_cfi $-1 # mark this as an int
1312 xorl %edx,%edx # error code 0
1313 movl %esp,%eax # pt_regs pointer
1315 jmp ret_from_exception
1320 * NMI is doubly nasty. It can happen _while_ we're handling
1321 * a debug fault, and the debug fault hasn't yet been able to
1322 * clear up the stack. So we first check whether we got an
1323 * NMI on the sysenter entry path, but after that we need to
1324 * check whether we got an NMI on the debug path where the debug
1325 * fault happened on the sysenter path.
1330 #ifdef CONFIG_X86_ESPFIX32
1333 cmpw $__ESPFIX_SS, %ax
1337 cmpl $ia32_sysenter_target,(%esp)
1341 /* Do not access memory above the end of our stack page,
1342 * it might not exist.
1344 andl $(THREAD_SIZE-1),%eax
1345 cmpl $(THREAD_SIZE-20),%eax
1347 jae nmi_stack_correct
1348 cmpl $ia32_sysenter_target,12(%esp)
1349 je nmi_debug_stack_check
1351 /* We have a RING0_INT_FRAME here */
1354 xorl %edx,%edx # zero error code
1355 movl %esp,%eax # pt_regs pointer
1357 jmp restore_all_notrace
1362 FIX_STACK 12, nmi_stack_correct, 1
1363 jmp nmi_stack_correct
1365 nmi_debug_stack_check:
1366 /* We have a RING0_INT_FRAME here */
1367 cmpw $__KERNEL_CS,16(%esp)
1368 jne nmi_stack_correct
1370 jb nmi_stack_correct
1371 cmpl $debug_esp_fix_insn,(%esp)
1372 ja nmi_stack_correct
1373 FIX_STACK 24, nmi_stack_correct, 1
1374 jmp nmi_stack_correct
1376 #ifdef CONFIG_X86_ESPFIX32
1378 /* We have a RING0_INT_FRAME here.
1380 * create the pointer to lss back
1385 /* copy the iret frame of 12 bytes */
1391 FIXUP_ESPFIX_STACK # %eax == %esp
1392 xorl %edx,%edx # zero error code
1395 lss 12+4(%esp), %esp # back to espfix stack
1396 CFI_ADJUST_CFA_OFFSET -24
1405 pushl_cfi $-1 # mark this as an int
1408 xorl %edx,%edx # zero error code
1409 movl %esp,%eax # pt_regs pointer
1411 jmp ret_from_exception
1415 ENTRY(general_protection)
1417 pushl_cfi $do_general_protection
1420 END(general_protection)
1422 #ifdef CONFIG_KVM_GUEST
1423 ENTRY(async_page_fault)
1426 pushl_cfi $do_async_page_fault
1429 END(async_page_fault)