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: TI_sysenter_return
399 * is relative to thread_info, which is at the bottom of the
400 * kernel stack page. 4*4 means the 4 words pushed above;
401 * TOP_OF_KERNEL_STACK_PADDING takes us to the top of the stack;
402 * and THREAD_SIZE takes us to the bottom.
404 pushl_cfi ((TI_sysenter_return) - THREAD_SIZE + TOP_OF_KERNEL_STACK_PADDING + 4*4)(%esp)
405 CFI_REL_OFFSET eip, 0
409 ENABLE_INTERRUPTS(CLBR_NONE)
412 * Load the potential sixth argument from user stack.
413 * Careful about security.
415 cmpl $__PAGE_OFFSET-3,%ebp
420 movl %ebp,PT_EBP(%esp)
421 _ASM_EXTABLE(1b,syscall_fault)
423 GET_THREAD_INFO(%ebp)
425 testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%ebp)
428 cmpl $(NR_syscalls), %eax
430 call *sys_call_table(,%eax,4)
432 movl %eax,PT_EAX(%esp)
434 DISABLE_INTERRUPTS(CLBR_ANY)
436 movl TI_flags(%ebp), %ecx
437 testl $_TIF_ALLWORK_MASK, %ecx
440 /* if something modifies registers it must also disable sysexit */
441 movl PT_EIP(%esp), %edx
442 movl PT_OLDESP(%esp), %ecx
445 1: mov PT_FS(%esp), %fs
447 ENABLE_INTERRUPTS_SYSEXIT
449 #ifdef CONFIG_AUDITSYSCALL
451 testl $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags(%ebp)
452 jnz syscall_trace_entry
453 /* movl PT_EAX(%esp), %eax already set, syscall number: 1st arg to audit */
454 movl PT_EBX(%esp), %edx /* ebx/a0: 2nd arg to audit */
455 /* movl PT_ECX(%esp), %ecx already set, a1: 3nd arg to audit */
456 pushl_cfi PT_ESI(%esp) /* a3: 5th arg */
457 pushl_cfi PT_EDX+4(%esp) /* a2: 4th arg */
458 call __audit_syscall_entry
459 popl_cfi %ecx /* get that remapped edx off the stack */
460 popl_cfi %ecx /* get that remapped esi off the stack */
461 movl PT_EAX(%esp),%eax /* reload syscall number */
465 testl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT), %ecx
466 jnz 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 jnz 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)
492 # system call handler stub
494 RING0_INT_FRAME # can't unwind into user space anyway
496 pushl_cfi %eax # save orig_eax
498 GET_THREAD_INFO(%ebp)
499 # system call tracing in operation / emulation
500 testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%ebp)
501 jnz syscall_trace_entry
502 cmpl $(NR_syscalls), %eax
505 call *sys_call_table(,%eax,4)
507 movl %eax,PT_EAX(%esp) # store the return value
510 DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt
511 # setting need_resched or sigpending
512 # between sampling and the iret
514 movl TI_flags(%ebp), %ecx
515 testl $_TIF_ALLWORK_MASK, %ecx # current->work
516 jnz syscall_exit_work
521 #ifdef CONFIG_X86_ESPFIX32
522 movl PT_EFLAGS(%esp), %eax # mix EFLAGS, SS and CS
523 # Warning: PT_OLDSS(%esp) contains the wrong/random values if we
524 # are returning to the kernel.
525 # See comments in process.c:copy_thread() for details.
526 movb PT_OLDSS(%esp), %ah
527 movb PT_CS(%esp), %al
528 andl $(X86_EFLAGS_VM | (SEGMENT_TI_MASK << 8) | SEGMENT_RPL_MASK), %eax
529 cmpl $((SEGMENT_LDT << 8) | USER_RPL), %eax
531 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)
545 #ifdef CONFIG_X86_ESPFIX32
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
593 # perform work that needs to be done immediately before resumption
595 RING0_PTREGS_FRAME # can't unwind into user space anyway
597 testb $_TIF_NEED_RESCHED, %cl
602 DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt
603 # setting need_resched or sigpending
604 # between sampling and the iret
606 movl TI_flags(%ebp), %ecx
607 andl $_TIF_WORK_MASK, %ecx # is there any work to be done other
608 # than syscall tracing?
610 testb $_TIF_NEED_RESCHED, %cl
613 work_notifysig: # deal with pending signals and
614 # notify-resume requests
616 testl $X86_EFLAGS_VM, PT_EFLAGS(%esp)
618 jnz work_notifysig_v86 # returning to kernel-space or
625 ENABLE_INTERRUPTS(CLBR_NONE)
626 movb PT_CS(%esp), %bl
627 andb $SEGMENT_RPL_MASK, %bl
631 call do_notify_resume
637 pushl_cfi %ecx # save ti_flags for do_notify_resume
638 call save_v86_state # %eax contains pt_regs pointer
645 # perform syscall exit tracing
648 movl $-ENOSYS,PT_EAX(%esp)
650 call syscall_trace_enter
651 /* What it returned is what we'll actually use. */
652 cmpl $(NR_syscalls), %eax
655 END(syscall_trace_entry)
657 # perform syscall exit tracing
660 testl $_TIF_WORK_SYSCALL_EXIT, %ecx
663 ENABLE_INTERRUPTS(CLBR_ANY) # could let syscall_trace_leave() call
666 call syscall_trace_leave
668 END(syscall_exit_work)
671 RING0_INT_FRAME # can't unwind into user space anyway
674 GET_THREAD_INFO(%ebp)
675 movl $-EFAULT,PT_EAX(%esp)
681 jmp syscall_after_call
686 jmp sysenter_after_call
690 .macro FIXUP_ESPFIX_STACK
692 * Switch back for ESPFIX stack to the normal zerobased stack
694 * We can't call C functions using the ESPFIX stack. This code reads
695 * the high word of the segment base from the GDT and swiches to the
696 * normal stack and adjusts ESP with the matching offset.
698 #ifdef CONFIG_X86_ESPFIX32
699 /* fixup the stack */
700 mov GDT_ESPFIX_SS + 4, %al /* bits 16..23 */
701 mov GDT_ESPFIX_SS + 7, %ah /* bits 24..31 */
703 addl %esp, %eax /* the adjusted stack pointer */
704 pushl_cfi $__KERNEL_DS
706 lss (%esp), %esp /* switch to the normal stack segment */
707 CFI_ADJUST_CFA_OFFSET -8
710 .macro UNWIND_ESPFIX_STACK
711 #ifdef CONFIG_X86_ESPFIX32
713 /* see if on espfix stack */
714 cmpw $__ESPFIX_SS, %ax
716 movl $__KERNEL_DS, %eax
719 /* switch to normal stack */
726 * Build the entry stubs with some assembler magic.
727 * We pack 1 stub into every 8-byte block.
730 ENTRY(irq_entries_start)
732 vector=FIRST_EXTERNAL_VECTOR
733 .rept (FIRST_SYSTEM_VECTOR - FIRST_EXTERNAL_VECTOR)
734 pushl_cfi $(~vector+0x80) /* Note: always in signed byte range */
737 CFI_ADJUST_CFA_OFFSET -4
740 END(irq_entries_start)
743 * the CPU automatically disables interrupts when executing an IRQ vector,
744 * so IRQ-flags tracing has to follow that:
746 .p2align CONFIG_X86_L1_CACHE_SHIFT
749 addl $-0x80,(%esp) /* Adjust vector into the [-256,-1] range */
755 ENDPROC(common_interrupt)
758 #define BUILD_INTERRUPT3(name, nr, fn) \
772 #ifdef CONFIG_TRACING
773 #define TRACE_BUILD_INTERRUPT(name, nr) \
774 BUILD_INTERRUPT3(trace_##name, nr, smp_trace_##name)
776 #define TRACE_BUILD_INTERRUPT(name, nr)
779 #define BUILD_INTERRUPT(name, nr) \
780 BUILD_INTERRUPT3(name, nr, smp_##name); \
781 TRACE_BUILD_INTERRUPT(name, nr)
783 /* The include is where all of the SMP etc. interrupts come from */
784 #include <asm/entry_arch.h>
786 ENTRY(coprocessor_error)
790 pushl_cfi $do_coprocessor_error
793 END(coprocessor_error)
795 ENTRY(simd_coprocessor_error)
799 #ifdef CONFIG_X86_INVD_BUG
800 /* AMD 486 bug: invd from userspace calls exception 19 instead of #GP */
801 ALTERNATIVE "pushl_cfi $do_general_protection", \
802 "pushl $do_simd_coprocessor_error", \
805 pushl_cfi $do_simd_coprocessor_error
809 END(simd_coprocessor_error)
811 ENTRY(device_not_available)
814 pushl_cfi $-1 # mark this as an int
815 pushl_cfi $do_device_not_available
818 END(device_not_available)
820 #ifdef CONFIG_PARAVIRT
823 _ASM_EXTABLE(native_iret, iret_exc)
826 ENTRY(native_irq_enable_sysexit)
829 END(native_irq_enable_sysexit)
836 pushl_cfi $do_overflow
854 pushl_cfi $do_invalid_op
859 ENTRY(coprocessor_segment_overrun)
863 pushl_cfi $do_coprocessor_segment_overrun
866 END(coprocessor_segment_overrun)
871 pushl_cfi $do_invalid_TSS
876 ENTRY(segment_not_present)
879 pushl_cfi $do_segment_not_present
882 END(segment_not_present)
887 pushl_cfi $do_stack_segment
892 ENTRY(alignment_check)
895 pushl_cfi $do_alignment_check
903 pushl_cfi $0 # no error code
904 pushl_cfi $do_divide_error
909 #ifdef CONFIG_X86_MCE
914 pushl_cfi machine_check_vector
920 ENTRY(spurious_interrupt_bug)
924 pushl_cfi $do_spurious_interrupt_bug
927 END(spurious_interrupt_bug)
930 /* Xen doesn't set %esp to be precisely what the normal sysenter
931 entrypoint expects, so fix it up before using the normal path. */
932 ENTRY(xen_sysenter_target)
934 addl $5*4, %esp /* remove xen-provided frame */
935 CFI_ADJUST_CFA_OFFSET -5*4
936 jmp sysenter_past_esp
939 ENTRY(xen_hypervisor_callback)
941 pushl_cfi $-1 /* orig_ax = -1 => not a system call */
945 /* Check to see if we got the event in the critical
946 region in xen_iret_direct, after we've reenabled
947 events and checked for pending events. This simulates
948 iret instruction's behaviour where it delivers a
949 pending interrupt when enabling interrupts. */
950 movl PT_EIP(%esp),%eax
951 cmpl $xen_iret_start_crit,%eax
953 cmpl $xen_iret_end_crit,%eax
956 jmp xen_iret_crit_fixup
960 call xen_evtchn_do_upcall
961 #ifndef CONFIG_PREEMPT
962 call xen_maybe_preempt_hcall
966 ENDPROC(xen_hypervisor_callback)
968 # Hypervisor uses this for application faults while it executes.
969 # We get here for two reasons:
970 # 1. Fault while reloading DS, ES, FS or GS
971 # 2. Fault while executing IRET
972 # Category 1 we fix up by reattempting the load, and zeroing the segment
973 # register if the load fails.
974 # Category 2 we fix up by jumping to do_iret_error. We cannot use the
975 # normal Linux return path in this case because if we use the IRET hypercall
976 # to pop the stack frame we end up in an infinite loop of failsafe callbacks.
977 # We distinguish between categories by maintaining a status value in EAX.
978 ENTRY(xen_failsafe_callback)
986 /* EAX == 0 => Category 1 (Bad segment)
987 EAX != 0 => Category 2 (Bad IRET) */
991 CFI_ADJUST_CFA_OFFSET -16
994 5: pushl_cfi $-1 /* orig_ax = -1 => not a system call */
996 jmp ret_from_exception
1017 ENDPROC(xen_failsafe_callback)
1019 BUILD_INTERRUPT3(xen_hvm_callback_vector, HYPERVISOR_CALLBACK_VECTOR,
1020 xen_evtchn_do_upcall)
1022 #endif /* CONFIG_XEN */
1024 #if IS_ENABLED(CONFIG_HYPERV)
1026 BUILD_INTERRUPT3(hyperv_callback_vector, HYPERVISOR_CALLBACK_VECTOR,
1027 hyperv_vector_handler)
1029 #endif /* CONFIG_HYPERV */
1031 #ifdef CONFIG_FUNCTION_TRACER
1032 #ifdef CONFIG_DYNAMIC_FTRACE
1038 ENTRY(ftrace_caller)
1042 pushl $0 /* Pass NULL as regs pointer */
1043 movl 4*4(%esp), %eax
1044 movl 0x4(%ebp), %edx
1045 movl function_trace_op, %ecx
1046 subl $MCOUNT_INSN_SIZE, %eax
1052 addl $4,%esp /* skip NULL pointer */
1057 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1058 .globl ftrace_graph_call
1068 ENTRY(ftrace_regs_caller)
1069 pushf /* push flags before compare (in cs location) */
1072 * i386 does not save SS and ESP when coming from kernel.
1073 * Instead, to get sp, ®s->sp is used (see ptrace.h).
1074 * Unfortunately, that means eflags must be at the same location
1075 * as the current return ip is. We move the return ip into the
1076 * ip location, and move flags into the return ip location.
1078 pushl 4(%esp) /* save return ip into ip slot */
1080 pushl $0 /* Load 0 into orig_ax */
1093 movl 13*4(%esp), %eax /* Get the saved flags */
1094 movl %eax, 14*4(%esp) /* Move saved flags into regs->flags location */
1095 /* clobbering return ip */
1096 movl $__KERNEL_CS,13*4(%esp)
1098 movl 12*4(%esp), %eax /* Load ip (1st parameter) */
1099 subl $MCOUNT_INSN_SIZE, %eax /* Adjust ip */
1100 movl 0x4(%ebp), %edx /* Load parent ip (2nd parameter) */
1101 movl function_trace_op, %ecx /* Save ftrace_pos in 3rd parameter */
1102 pushl %esp /* Save pt_regs as 4th parameter */
1104 GLOBAL(ftrace_regs_call)
1107 addl $4, %esp /* Skip pt_regs */
1108 movl 14*4(%esp), %eax /* Move flags back into cs */
1109 movl %eax, 13*4(%esp) /* Needed to keep addl from modifying flags */
1110 movl 12*4(%esp), %eax /* Get return ip from regs->ip */
1111 movl %eax, 14*4(%esp) /* Put return ip back for ret */
1124 addl $8, %esp /* Skip orig_ax and ip */
1125 popf /* Pop flags at end (no addl to corrupt flags) */
1130 #else /* ! CONFIG_DYNAMIC_FTRACE */
1133 cmpl $__PAGE_OFFSET, %esp
1134 jb ftrace_stub /* Paging not enabled yet? */
1136 cmpl $ftrace_stub, ftrace_trace_function
1138 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1139 cmpl $ftrace_stub, ftrace_graph_return
1140 jnz ftrace_graph_caller
1142 cmpl $ftrace_graph_entry_stub, ftrace_graph_entry
1143 jnz ftrace_graph_caller
1149 /* taken from glibc */
1154 movl 0xc(%esp), %eax
1155 movl 0x4(%ebp), %edx
1156 subl $MCOUNT_INSN_SIZE, %eax
1158 call *ftrace_trace_function
1165 #endif /* CONFIG_DYNAMIC_FTRACE */
1166 #endif /* CONFIG_FUNCTION_TRACER */
1168 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1169 ENTRY(ftrace_graph_caller)
1173 movl 0xc(%esp), %eax
1176 subl $MCOUNT_INSN_SIZE, %eax
1177 call prepare_ftrace_return
1182 END(ftrace_graph_caller)
1184 .globl return_to_handler
1189 call ftrace_return_to_handler
1196 #ifdef CONFIG_TRACING
1197 ENTRY(trace_page_fault)
1200 pushl_cfi $trace_do_page_fault
1203 END(trace_page_fault)
1209 pushl_cfi $do_page_fault
1212 /* the function address is in %gs's slot on the stack */
1214 /*CFI_REL_OFFSET fs, 0*/
1216 /*CFI_REL_OFFSET es, 0*/
1218 /*CFI_REL_OFFSET ds, 0*/
1227 movl $(__KERNEL_PERCPU), %ecx
1231 movl PT_GS(%esp), %edi # get the function address
1232 movl PT_ORIG_EAX(%esp), %edx # get the error code
1233 movl $-1, PT_ORIG_EAX(%esp) # no syscall to restart
1236 movl $(__USER_DS), %ecx
1240 movl %esp,%eax # pt_regs pointer
1242 jmp ret_from_exception
1247 * Debug traps and NMI can happen at the one SYSENTER instruction
1248 * that sets up the real kernel stack. Check here, since we can't
1249 * allow the wrong stack to be used.
1251 * "TSS_sysenter_sp0+12" is because the NMI/debug handler will have
1252 * already pushed 3 words if it hits on the sysenter instruction:
1253 * eflags, cs and eip.
1255 * We just load the right stack, and push the three (known) values
1256 * by hand onto the new stack - while updating the return eip past
1257 * the instruction that would have done it for sysenter.
1259 .macro FIX_STACK offset ok label
1260 cmpw $__KERNEL_CS, 4(%esp)
1263 movl TSS_sysenter_sp0 + \offset(%esp), %esp
1267 pushl_cfi $__KERNEL_CS
1268 pushl_cfi $sysenter_past_esp
1269 CFI_REL_OFFSET eip, 0
1275 cmpl $ia32_sysenter_target,(%esp)
1276 jne debug_stack_correct
1277 FIX_STACK 12, debug_stack_correct, debug_esp_fix_insn
1278 debug_stack_correct:
1279 pushl_cfi $-1 # mark this as an int
1282 xorl %edx,%edx # error code 0
1283 movl %esp,%eax # pt_regs pointer
1285 jmp ret_from_exception
1290 * NMI is doubly nasty. It can happen _while_ we're handling
1291 * a debug fault, and the debug fault hasn't yet been able to
1292 * clear up the stack. So we first check whether we got an
1293 * NMI on the sysenter entry path, but after that we need to
1294 * check whether we got an NMI on the debug path where the debug
1295 * fault happened on the sysenter path.
1300 #ifdef CONFIG_X86_ESPFIX32
1303 cmpw $__ESPFIX_SS, %ax
1307 cmpl $ia32_sysenter_target,(%esp)
1311 /* Do not access memory above the end of our stack page,
1312 * it might not exist.
1314 andl $(THREAD_SIZE-1),%eax
1315 cmpl $(THREAD_SIZE-20),%eax
1317 jae nmi_stack_correct
1318 cmpl $ia32_sysenter_target,12(%esp)
1319 je nmi_debug_stack_check
1321 /* We have a RING0_INT_FRAME here */
1324 xorl %edx,%edx # zero error code
1325 movl %esp,%eax # pt_regs pointer
1327 jmp restore_all_notrace
1332 FIX_STACK 12, nmi_stack_correct, 1
1333 jmp nmi_stack_correct
1335 nmi_debug_stack_check:
1336 /* We have a RING0_INT_FRAME here */
1337 cmpw $__KERNEL_CS,16(%esp)
1338 jne nmi_stack_correct
1340 jb nmi_stack_correct
1341 cmpl $debug_esp_fix_insn,(%esp)
1342 ja nmi_stack_correct
1343 FIX_STACK 24, nmi_stack_correct, 1
1344 jmp nmi_stack_correct
1346 #ifdef CONFIG_X86_ESPFIX32
1348 /* We have a RING0_INT_FRAME here.
1350 * create the pointer to lss back
1355 /* copy the iret frame of 12 bytes */
1361 FIXUP_ESPFIX_STACK # %eax == %esp
1362 xorl %edx,%edx # zero error code
1365 lss 12+4(%esp), %esp # back to espfix stack
1366 CFI_ADJUST_CFA_OFFSET -24
1375 pushl_cfi $-1 # mark this as an int
1378 xorl %edx,%edx # zero error code
1379 movl %esp,%eax # pt_regs pointer
1381 jmp ret_from_exception
1385 ENTRY(general_protection)
1387 pushl_cfi $do_general_protection
1390 END(general_protection)
1392 #ifdef CONFIG_KVM_GUEST
1393 ENTRY(async_page_fault)
1396 pushl_cfi $do_async_page_fault
1399 END(async_page_fault)