2 * linux/arch/x86_64/entry.S
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 * Copyright (C) 2000, 2001, 2002 Andi Kleen SuSE Labs
6 * Copyright (C) 2000 Pavel Machek <pavel@suse.cz>
10 * entry.S contains the system-call and fault low-level handling routines.
12 * Some of this is documented in Documentation/x86/entry_64.txt
14 * NOTE: This code handles signal-recognition, which happens every time
15 * after an interrupt and after each system call.
17 * Normal syscalls and interrupts don't save a full stack frame, this is
18 * only done for syscall tracing, signals or fork/exec et.al.
20 * A note on terminology:
21 * - top of stack: Architecture defined interrupt frame from SS to RIP
22 * at the top of the kernel process stack.
23 * - partial stack frame: partially saved registers up to R11.
24 * - full stack frame: Like partial stack frame, but all register saved.
27 * - CFI macros are used to generate dwarf2 unwind information for better
28 * backtraces. They don't change any code.
29 * - SAVE_ALL/RESTORE_ALL - Save/restore all registers
30 * - SAVE_ARGS/RESTORE_ARGS - Save/restore registers that C functions modify.
31 * There are unfortunately lots of special cases where some registers
32 * not touched. The macro is a big mess that should be cleaned up.
33 * - SAVE_REST/RESTORE_REST - Handle the registers not saved by SAVE_ARGS.
34 * Gives a full stack frame.
35 * - ENTRY/END Define functions in the symbol table.
36 * - FIXUP_TOP_OF_STACK/RESTORE_TOP_OF_STACK - Fix up the hardware stack
37 * frame that is otherwise undefined after a SYSCALL
38 * - TRACE_IRQ_* - Trace hard interrupt state for lock debugging.
39 * - errorentry/paranoidentry/zeroentry - Define exception entry points.
42 #include <linux/linkage.h>
43 #include <asm/segment.h>
44 #include <asm/cache.h>
45 #include <asm/errno.h>
46 #include <asm/dwarf2.h>
47 #include <asm/calling.h>
48 #include <asm/asm-offsets.h>
50 #include <asm/unistd.h>
51 #include <asm/thread_info.h>
52 #include <asm/hw_irq.h>
53 #include <asm/page_types.h>
54 #include <asm/irqflags.h>
55 #include <asm/paravirt.h>
56 #include <asm/ftrace.h>
57 #include <asm/percpu.h>
60 #include <linux/err.h>
62 /* Avoid __ASSEMBLER__'ifying <linux/audit.h> just for this. */
63 #include <linux/elf-em.h>
64 #define AUDIT_ARCH_X86_64 (EM_X86_64|__AUDIT_ARCH_64BIT|__AUDIT_ARCH_LE)
65 #define __AUDIT_ARCH_64BIT 0x80000000
66 #define __AUDIT_ARCH_LE 0x40000000
69 .section .entry.text, "ax"
71 #ifdef CONFIG_FUNCTION_TRACER
73 #ifdef CC_USING_FENTRY
74 # define function_hook __fentry__
76 # define function_hook mcount
79 #ifdef CONFIG_DYNAMIC_FTRACE
85 /* skip is set if stack has been adjusted */
86 .macro ftrace_caller_setup skip=0
87 MCOUNT_SAVE_FRAME \skip
89 /* Load the ftrace_ops into the 3rd parameter */
90 leaq function_trace_op, %rdx
92 /* Load ip into the first parameter */
94 subq $MCOUNT_INSN_SIZE, %rdi
95 /* Load the parent_ip into the second parameter */
96 #ifdef CC_USING_FENTRY
97 movq SS+16(%rsp), %rsi
104 /* Check if tracing was disabled (quick check) */
105 cmpl $0, function_trace_stop
109 /* regs go into 4th parameter (but make it NULL) */
118 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
119 GLOBAL(ftrace_graph_call)
127 ENTRY(ftrace_regs_caller)
128 /* Save the current flags before compare (in SS location)*/
131 /* Check if tracing was disabled (quick check) */
132 cmpl $0, function_trace_stop
133 jne ftrace_restore_flags
135 /* skip=8 to skip flags saved in SS */
136 ftrace_caller_setup 8
138 /* Save the rest of pt_regs */
147 /* Copy saved flags */
149 movq %rcx, EFLAGS(%rsp)
150 /* Kernel segments */
151 movq $__KERNEL_DS, %rcx
153 movq $__KERNEL_CS, %rcx
155 /* Stack - skipping return address */
156 leaq SS+16(%rsp), %rcx
159 /* regs go into 4th parameter */
162 GLOBAL(ftrace_regs_call)
165 /* Copy flags back to SS, to restore them */
166 movq EFLAGS(%rsp), %rax
169 /* Handlers can change the RIP */
171 movq %rax, SS+8(%rsp)
173 /* restore the rest of pt_regs */
182 /* skip=8 to skip flags saved in SS */
183 MCOUNT_RESTORE_FRAME 8
189 ftrace_restore_flags:
193 END(ftrace_regs_caller)
196 #else /* ! CONFIG_DYNAMIC_FTRACE */
199 cmpl $0, function_trace_stop
202 cmpq $ftrace_stub, ftrace_trace_function
205 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
206 cmpq $ftrace_stub, ftrace_graph_return
207 jnz ftrace_graph_caller
209 cmpq $ftrace_graph_entry_stub, ftrace_graph_entry
210 jnz ftrace_graph_caller
220 #ifdef CC_USING_FENTRY
221 movq SS+16(%rsp), %rsi
225 subq $MCOUNT_INSN_SIZE, %rdi
227 call *ftrace_trace_function
233 #endif /* CONFIG_DYNAMIC_FTRACE */
234 #endif /* CONFIG_FUNCTION_TRACER */
236 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
237 ENTRY(ftrace_graph_caller)
240 #ifdef CC_USING_FENTRY
241 leaq SS+16(%rsp), %rdi
242 movq $0, %rdx /* No framepointers needed */
248 subq $MCOUNT_INSN_SIZE, %rsi
250 call prepare_ftrace_return
255 END(ftrace_graph_caller)
257 GLOBAL(return_to_handler)
260 /* Save the return values */
265 call ftrace_return_to_handler
275 #ifndef CONFIG_PREEMPT
276 #define retint_kernel retint_restore_args
279 #ifdef CONFIG_PARAVIRT
280 ENTRY(native_usergs_sysret64)
283 ENDPROC(native_usergs_sysret64)
284 #endif /* CONFIG_PARAVIRT */
287 .macro TRACE_IRQS_IRETQ offset=ARGOFFSET
288 #ifdef CONFIG_TRACE_IRQFLAGS
289 bt $9,EFLAGS-\offset(%rsp) /* interrupts off? */
297 * When dynamic function tracer is enabled it will add a breakpoint
298 * to all locations that it is about to modify, sync CPUs, update
299 * all the code, sync CPUs, then remove the breakpoints. In this time
300 * if lockdep is enabled, it might jump back into the debug handler
301 * outside the updating of the IST protection. (TRACE_IRQS_ON/OFF).
303 * We need to change the IDT table before calling TRACE_IRQS_ON/OFF to
304 * make sure the stack pointer does not get reset back to the top
305 * of the debug stack, and instead just reuses the current stack.
307 #if defined(CONFIG_DYNAMIC_FTRACE) && defined(CONFIG_TRACE_IRQFLAGS)
309 .macro TRACE_IRQS_OFF_DEBUG
310 call debug_stack_set_zero
312 call debug_stack_reset
315 .macro TRACE_IRQS_ON_DEBUG
316 call debug_stack_set_zero
318 call debug_stack_reset
321 .macro TRACE_IRQS_IRETQ_DEBUG offset=ARGOFFSET
322 bt $9,EFLAGS-\offset(%rsp) /* interrupts off? */
329 # define TRACE_IRQS_OFF_DEBUG TRACE_IRQS_OFF
330 # define TRACE_IRQS_ON_DEBUG TRACE_IRQS_ON
331 # define TRACE_IRQS_IRETQ_DEBUG TRACE_IRQS_IRETQ
335 * C code is not supposed to know about undefined top of stack. Every time
336 * a C function with an pt_regs argument is called from the SYSCALL based
337 * fast path FIXUP_TOP_OF_STACK is needed.
338 * RESTORE_TOP_OF_STACK syncs the syscall state after any possible ptregs
342 /* %rsp:at FRAMEEND */
343 .macro FIXUP_TOP_OF_STACK tmp offset=0
344 movq PER_CPU_VAR(old_rsp),\tmp
345 movq \tmp,RSP+\offset(%rsp)
346 movq $__USER_DS,SS+\offset(%rsp)
347 movq $__USER_CS,CS+\offset(%rsp)
348 movq $-1,RCX+\offset(%rsp)
349 movq R11+\offset(%rsp),\tmp /* get eflags */
350 movq \tmp,EFLAGS+\offset(%rsp)
353 .macro RESTORE_TOP_OF_STACK tmp offset=0
354 movq RSP+\offset(%rsp),\tmp
355 movq \tmp,PER_CPU_VAR(old_rsp)
356 movq EFLAGS+\offset(%rsp),\tmp
357 movq \tmp,R11+\offset(%rsp)
360 .macro FAKE_STACK_FRAME child_rip
361 /* push in order ss, rsp, eflags, cs, rip */
363 pushq_cfi $__KERNEL_DS /* ss */
364 /*CFI_REL_OFFSET ss,0*/
365 pushq_cfi %rax /* rsp */
367 pushq_cfi $(X86_EFLAGS_IF|X86_EFLAGS_BIT1) /* eflags - interrupts on */
368 /*CFI_REL_OFFSET rflags,0*/
369 pushq_cfi $__KERNEL_CS /* cs */
370 /*CFI_REL_OFFSET cs,0*/
371 pushq_cfi \child_rip /* rip */
373 pushq_cfi %rax /* orig rax */
376 .macro UNFAKE_STACK_FRAME
378 CFI_ADJUST_CFA_OFFSET -(6*8)
382 * initial frame state for interrupts (and exceptions without error code)
384 .macro EMPTY_FRAME start=1 offset=0
388 CFI_DEF_CFA rsp,8+\offset
390 CFI_DEF_CFA_OFFSET 8+\offset
395 * initial frame state for interrupts (and exceptions without error code)
397 .macro INTR_FRAME start=1 offset=0
398 EMPTY_FRAME \start, SS+8+\offset-RIP
399 /*CFI_REL_OFFSET ss, SS+\offset-RIP*/
400 CFI_REL_OFFSET rsp, RSP+\offset-RIP
401 /*CFI_REL_OFFSET rflags, EFLAGS+\offset-RIP*/
402 /*CFI_REL_OFFSET cs, CS+\offset-RIP*/
403 CFI_REL_OFFSET rip, RIP+\offset-RIP
407 * initial frame state for exceptions with error code (and interrupts
408 * with vector already pushed)
410 .macro XCPT_FRAME start=1 offset=0
411 INTR_FRAME \start, RIP+\offset-ORIG_RAX
412 /*CFI_REL_OFFSET orig_rax, ORIG_RAX-ORIG_RAX*/
416 * frame that enables calling into C.
418 .macro PARTIAL_FRAME start=1 offset=0
419 XCPT_FRAME \start, ORIG_RAX+\offset-ARGOFFSET
420 CFI_REL_OFFSET rdi, RDI+\offset-ARGOFFSET
421 CFI_REL_OFFSET rsi, RSI+\offset-ARGOFFSET
422 CFI_REL_OFFSET rdx, RDX+\offset-ARGOFFSET
423 CFI_REL_OFFSET rcx, RCX+\offset-ARGOFFSET
424 CFI_REL_OFFSET rax, RAX+\offset-ARGOFFSET
425 CFI_REL_OFFSET r8, R8+\offset-ARGOFFSET
426 CFI_REL_OFFSET r9, R9+\offset-ARGOFFSET
427 CFI_REL_OFFSET r10, R10+\offset-ARGOFFSET
428 CFI_REL_OFFSET r11, R11+\offset-ARGOFFSET
432 * frame that enables passing a complete pt_regs to a C function.
434 .macro DEFAULT_FRAME start=1 offset=0
435 PARTIAL_FRAME \start, R11+\offset-R15
436 CFI_REL_OFFSET rbx, RBX+\offset
437 CFI_REL_OFFSET rbp, RBP+\offset
438 CFI_REL_OFFSET r12, R12+\offset
439 CFI_REL_OFFSET r13, R13+\offset
440 CFI_REL_OFFSET r14, R14+\offset
441 CFI_REL_OFFSET r15, R15+\offset
444 /* save partial stack frame */
447 /* start from rbp in pt_regs and jump over */
448 movq_cfi rdi, (RDI-RBP)
449 movq_cfi rsi, (RSI-RBP)
450 movq_cfi rdx, (RDX-RBP)
451 movq_cfi rcx, (RCX-RBP)
452 movq_cfi rax, (RAX-RBP)
453 movq_cfi r8, (R8-RBP)
454 movq_cfi r9, (R9-RBP)
455 movq_cfi r10, (R10-RBP)
456 movq_cfi r11, (R11-RBP)
458 /* Save rbp so that we can unwind from get_irq_regs() */
461 /* Save previous stack value */
464 leaq -RBP(%rsp),%rdi /* arg1 for handler */
465 testl $3, CS-RBP(%rsi)
469 * irq_count is used to check if a CPU is already on an interrupt stack
470 * or not. While this is essentially redundant with preempt_count it is
471 * a little cheaper to use a separate counter in the PDA (short of
472 * moving irq_enter into assembly, which would be too much work)
474 1: incl PER_CPU_VAR(irq_count)
475 cmovzq PER_CPU_VAR(irq_stack_ptr),%rsp
476 CFI_DEF_CFA_REGISTER rsi
478 /* Store previous stack value */
480 CFI_ESCAPE 0x0f /* DW_CFA_def_cfa_expression */, 6, \
481 0x77 /* DW_OP_breg7 */, 0, \
482 0x06 /* DW_OP_deref */, \
483 0x08 /* DW_OP_const1u */, SS+8-RBP, \
484 0x22 /* DW_OP_plus */
485 /* We entered an interrupt context - irqs are off: */
490 PARTIAL_FRAME 1 (REST_SKIP+8)
491 movq 5*8+16(%rsp), %r11 /* save return address */
498 movq %r11, 8(%rsp) /* return address */
499 FIXUP_TOP_OF_STACK %r11, 16
504 /* save complete stack frame */
505 .pushsection .kprobes.text, "ax"
525 movl $MSR_GS_BASE,%ecx
528 js 1f /* negative -> in kernel */
537 * A newly forked process directly context switches into this address.
539 * rdi: prev task we switched from
544 LOCK ; btr $TIF_FORK,TI_flags(%r8)
547 popfq_cfi # reset kernel eflags
549 call schedule_tail # rdi: 'prev' task parameter
551 GET_THREAD_INFO(%rcx)
555 testl $3, CS-ARGOFFSET(%rsp) # from kernel_thread?
556 jz retint_restore_args
558 testl $_TIF_IA32, TI_flags(%rcx) # 32-bit compat task needs IRET
559 jnz int_ret_from_sys_call
561 RESTORE_TOP_OF_STACK %rdi, -ARGOFFSET
562 jmp ret_from_sys_call # go to the SYSRET fastpath
568 * System call entry. Up to 6 arguments in registers are supported.
570 * SYSCALL does not save anything on the stack and does not change the
576 * rax system call number
578 * rcx return address for syscall/sysret, C arg3
581 * r10 arg3 (--> moved to rcx for C)
584 * r11 eflags for syscall/sysret, temporary for C
585 * r12-r15,rbp,rbx saved by C code, not touched.
587 * Interrupts are off on entry.
588 * Only called from user space.
590 * XXX if we had a free scratch register we could save the RSP into the stack frame
591 * and report it properly in ps. Unfortunately we haven't.
593 * When user can change the frames always force IRET. That is because
594 * it deals with uncanonical addresses better. SYSRET has trouble
595 * with them due to bugs in both AMD and Intel CPUs.
601 CFI_DEF_CFA rsp,KERNEL_STACK_OFFSET
603 /*CFI_REGISTER rflags,r11*/
606 * A hypervisor implementation might want to use a label
607 * after the swapgs, so that it can do the swapgs
608 * for the guest and jump here on syscall.
610 GLOBAL(system_call_after_swapgs)
612 movq %rsp,PER_CPU_VAR(old_rsp)
613 movq PER_CPU_VAR(kernel_stack),%rsp
615 * No need to follow this irqs off/on section - it's straight
618 ENABLE_INTERRUPTS(CLBR_NONE)
620 movq %rax,ORIG_RAX-ARGOFFSET(%rsp)
621 movq %rcx,RIP-ARGOFFSET(%rsp)
622 CFI_REL_OFFSET rip,RIP-ARGOFFSET
623 testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET)
625 system_call_fastpath:
626 #if __SYSCALL_MASK == ~0
627 cmpq $__NR_syscall_max,%rax
629 andl $__SYSCALL_MASK,%eax
630 cmpl $__NR_syscall_max,%eax
634 call *sys_call_table(,%rax,8) # XXX: rip relative
635 movq %rax,RAX-ARGOFFSET(%rsp)
637 * Syscall return path ending with SYSRET (fast path)
638 * Has incomplete stack frame and undefined top of stack.
641 movl $_TIF_ALLWORK_MASK,%edi
645 DISABLE_INTERRUPTS(CLBR_NONE)
647 movl TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET),%edx
652 * sysretq will re-enable interrupts:
655 movq RIP-ARGOFFSET(%rsp),%rcx
657 RESTORE_ARGS 1,-ARG_SKIP,0
658 /*CFI_REGISTER rflags,r11*/
659 movq PER_CPU_VAR(old_rsp), %rsp
663 /* Handle reschedules */
664 /* edx: work, edi: workmask */
666 bt $TIF_NEED_RESCHED,%edx
669 ENABLE_INTERRUPTS(CLBR_NONE)
675 /* Handle a signal */
678 ENABLE_INTERRUPTS(CLBR_NONE)
679 #ifdef CONFIG_AUDITSYSCALL
680 bt $TIF_SYSCALL_AUDIT,%edx
684 * We have a signal, or exit tracing or single-step.
685 * These all wind up with the iret return path anyway,
686 * so just join that path right now.
688 FIXUP_TOP_OF_STACK %r11, -ARGOFFSET
689 jmp int_check_syscall_exit_work
692 movq $-ENOSYS,RAX-ARGOFFSET(%rsp)
693 jmp ret_from_sys_call
695 #ifdef CONFIG_AUDITSYSCALL
697 * Fast path for syscall audit without full syscall trace.
698 * We just call __audit_syscall_entry() directly, and then
699 * jump back to the normal fast path.
702 movq %r10,%r9 /* 6th arg: 4th syscall arg */
703 movq %rdx,%r8 /* 5th arg: 3rd syscall arg */
704 movq %rsi,%rcx /* 4th arg: 2nd syscall arg */
705 movq %rdi,%rdx /* 3rd arg: 1st syscall arg */
706 movq %rax,%rsi /* 2nd arg: syscall number */
707 movl $AUDIT_ARCH_X86_64,%edi /* 1st arg: audit arch */
708 call __audit_syscall_entry
709 LOAD_ARGS 0 /* reload call-clobbered registers */
710 jmp system_call_fastpath
713 * Return fast path for syscall audit. Call __audit_syscall_exit()
714 * directly and then jump back to the fast path with TIF_SYSCALL_AUDIT
718 movq RAX-ARGOFFSET(%rsp),%rsi /* second arg, syscall return value */
719 cmpq $-MAX_ERRNO,%rsi /* is it < -MAX_ERRNO? */
720 setbe %al /* 1 if so, 0 if not */
721 movzbl %al,%edi /* zero-extend that into %edi */
722 call __audit_syscall_exit
723 movl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT),%edi
725 #endif /* CONFIG_AUDITSYSCALL */
727 /* Do syscall tracing */
729 #ifdef CONFIG_AUDITSYSCALL
730 testl $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET)
734 movq $-ENOSYS,RAX(%rsp) /* ptrace can change this for a bad syscall */
735 FIXUP_TOP_OF_STACK %rdi
737 call syscall_trace_enter
739 * Reload arg registers from stack in case ptrace changed them.
740 * We don't reload %rax because syscall_trace_enter() returned
741 * the value it wants us to use in the table lookup.
743 LOAD_ARGS ARGOFFSET, 1
745 #if __SYSCALL_MASK == ~0
746 cmpq $__NR_syscall_max,%rax
748 andl $__SYSCALL_MASK,%eax
749 cmpl $__NR_syscall_max,%eax
751 ja int_ret_from_sys_call /* RAX(%rsp) set to -ENOSYS above */
752 movq %r10,%rcx /* fixup for C */
753 call *sys_call_table(,%rax,8)
754 movq %rax,RAX-ARGOFFSET(%rsp)
755 /* Use IRET because user could have changed frame */
758 * Syscall return path ending with IRET.
759 * Has correct top of stack, but partial stack frame.
761 GLOBAL(int_ret_from_sys_call)
762 DISABLE_INTERRUPTS(CLBR_NONE)
764 movl $_TIF_ALLWORK_MASK,%edi
765 /* edi: mask to check */
766 GLOBAL(int_with_check)
768 GET_THREAD_INFO(%rcx)
769 movl TI_flags(%rcx),%edx
772 andl $~TS_COMPAT,TI_status(%rcx)
775 /* Either reschedule or signal or syscall exit tracking needed. */
776 /* First do a reschedule test. */
777 /* edx: work, edi: workmask */
779 bt $TIF_NEED_RESCHED,%edx
782 ENABLE_INTERRUPTS(CLBR_NONE)
786 DISABLE_INTERRUPTS(CLBR_NONE)
790 /* handle signals and tracing -- both require a full stack frame */
793 ENABLE_INTERRUPTS(CLBR_NONE)
794 int_check_syscall_exit_work:
796 /* Check for syscall exit trace */
797 testl $_TIF_WORK_SYSCALL_EXIT,%edx
800 leaq 8(%rsp),%rdi # &ptregs -> arg1
801 call syscall_trace_leave
803 andl $~(_TIF_WORK_SYSCALL_EXIT|_TIF_SYSCALL_EMU),%edi
807 testl $_TIF_DO_NOTIFY_MASK,%edx
809 movq %rsp,%rdi # &ptregs -> arg1
810 xorl %esi,%esi # oldset -> arg2
811 call do_notify_resume
812 1: movl $_TIF_WORK_MASK,%edi
815 DISABLE_INTERRUPTS(CLBR_NONE)
822 * Certain special system calls that need to save a complete full stack frame.
824 .macro PTREGSCALL label,func,arg
826 PARTIAL_FRAME 1 8 /* offset 8: return address */
827 subq $REST_SKIP, %rsp
828 CFI_ADJUST_CFA_OFFSET REST_SKIP
830 DEFAULT_FRAME 0 8 /* offset 8: return address */
831 leaq 8(%rsp), \arg /* pt_regs pointer */
833 jmp ptregscall_common
838 PTREGSCALL stub_clone, sys_clone, %r8
839 PTREGSCALL stub_fork, sys_fork, %rdi
840 PTREGSCALL stub_vfork, sys_vfork, %rdi
841 PTREGSCALL stub_sigaltstack, sys_sigaltstack, %rdx
842 PTREGSCALL stub_iopl, sys_iopl, %rsi
844 ENTRY(ptregscall_common)
845 DEFAULT_FRAME 1 8 /* offset 8: return address */
846 RESTORE_TOP_OF_STACK %r11, 8
847 movq_cfi_restore R15+8, r15
848 movq_cfi_restore R14+8, r14
849 movq_cfi_restore R13+8, r13
850 movq_cfi_restore R12+8, r12
851 movq_cfi_restore RBP+8, rbp
852 movq_cfi_restore RBX+8, rbx
853 ret $REST_SKIP /* pop extended registers */
855 END(ptregscall_common)
862 FIXUP_TOP_OF_STACK %r11
865 RESTORE_TOP_OF_STACK %r11
868 jmp int_ret_from_sys_call
873 * sigreturn is special because it needs to restore all registers on return.
874 * This cannot be done with SYSRET, so use the IRET return path instead.
876 ENTRY(stub_rt_sigreturn)
882 FIXUP_TOP_OF_STACK %r11
883 call sys_rt_sigreturn
884 movq %rax,RAX(%rsp) # fixme, this could be done at the higher layer
886 jmp int_ret_from_sys_call
888 END(stub_rt_sigreturn)
890 #ifdef CONFIG_X86_X32_ABI
891 PTREGSCALL stub_x32_sigaltstack, sys32_sigaltstack, %rdx
893 ENTRY(stub_x32_rt_sigreturn)
899 FIXUP_TOP_OF_STACK %r11
900 call sys32_x32_rt_sigreturn
901 movq %rax,RAX(%rsp) # fixme, this could be done at the higher layer
903 jmp int_ret_from_sys_call
905 END(stub_x32_rt_sigreturn)
907 ENTRY(stub_x32_execve)
912 FIXUP_TOP_OF_STACK %r11
915 RESTORE_TOP_OF_STACK %r11
918 jmp int_ret_from_sys_call
925 * Build the entry stubs and pointer table with some assembler magic.
926 * We pack 7 stubs into a single 32-byte chunk, which will fit in a
927 * single cache line on all modern x86 implementations.
929 .section .init.rodata,"a"
933 .p2align CONFIG_X86_L1_CACHE_SHIFT
934 ENTRY(irq_entries_start)
936 vector=FIRST_EXTERNAL_VECTOR
937 .rept (NR_VECTORS-FIRST_EXTERNAL_VECTOR+6)/7
940 .if vector < NR_VECTORS
941 .if vector <> FIRST_EXTERNAL_VECTOR
942 CFI_ADJUST_CFA_OFFSET -8
944 1: pushq_cfi $(~vector+0x80) /* Note: always in signed byte range */
945 .if ((vector-FIRST_EXTERNAL_VECTOR)%7) <> 6
954 2: jmp common_interrupt
957 END(irq_entries_start)
964 * Interrupt entry/exit.
966 * Interrupt entry points save only callee clobbered registers in fast path.
968 * Entry runs with interrupts off.
971 /* 0(%rsp): ~(interrupt number) */
972 .macro interrupt func
973 /* reserve pt_regs for scratch regs and rbp */
974 subq $ORIG_RAX-RBP, %rsp
975 CFI_ADJUST_CFA_OFFSET ORIG_RAX-RBP
981 * Interrupt entry/exit should be protected against kprobes
983 .pushsection .kprobes.text, "ax"
985 * The interrupt stubs push (~vector+0x80) onto the stack and
986 * then jump to common_interrupt.
988 .p2align CONFIG_X86_L1_CACHE_SHIFT
991 addq $-0x80,(%rsp) /* Adjust vector to [-256,-1] range */
993 /* 0(%rsp): old_rsp-ARGOFFSET */
995 DISABLE_INTERRUPTS(CLBR_NONE)
997 decl PER_CPU_VAR(irq_count)
999 /* Restore saved previous stack */
1001 CFI_DEF_CFA rsi,SS+8-RBP /* reg/off reset after def_cfa_expr */
1002 leaq ARGOFFSET-RBP(%rsi), %rsp
1003 CFI_DEF_CFA_REGISTER rsp
1004 CFI_ADJUST_CFA_OFFSET RBP-ARGOFFSET
1007 GET_THREAD_INFO(%rcx)
1008 testl $3,CS-ARGOFFSET(%rsp)
1011 /* Interrupt came from user space */
1013 * Has a correct top of stack, but a partial stack frame
1014 * %rcx: thread info. Interrupts off.
1016 retint_with_reschedule:
1017 movl $_TIF_WORK_MASK,%edi
1019 LOCKDEP_SYS_EXIT_IRQ
1020 movl TI_flags(%rcx),%edx
1025 retint_swapgs: /* return to user-space */
1027 * The iretq could re-enable interrupts:
1029 DISABLE_INTERRUPTS(CLBR_ANY)
1034 retint_restore_args: /* return to kernel space */
1035 DISABLE_INTERRUPTS(CLBR_ANY)
1037 * The iretq could re-enable interrupts:
1045 _ASM_EXTABLE(irq_return, bad_iret)
1047 #ifdef CONFIG_PARAVIRT
1050 _ASM_EXTABLE(native_iret, bad_iret)
1053 .section .fixup,"ax"
1056 * The iret traps when the %cs or %ss being restored is bogus.
1057 * We've lost the original trap vector and error code.
1058 * #GPF is the most likely one to get for an invalid selector.
1059 * So pretend we completed the iret and took the #GPF in user mode.
1061 * We are now running with the kernel GS after exception recovery.
1062 * But error_entry expects us to have user GS to match the user %cs,
1068 jmp general_protection
1072 /* edi: workmask, edx: work */
1075 bt $TIF_NEED_RESCHED,%edx
1078 ENABLE_INTERRUPTS(CLBR_NONE)
1082 GET_THREAD_INFO(%rcx)
1083 DISABLE_INTERRUPTS(CLBR_NONE)
1088 testl $_TIF_DO_NOTIFY_MASK,%edx
1091 ENABLE_INTERRUPTS(CLBR_NONE)
1093 movq $-1,ORIG_RAX(%rsp)
1094 xorl %esi,%esi # oldset
1095 movq %rsp,%rdi # &pt_regs
1096 call do_notify_resume
1098 DISABLE_INTERRUPTS(CLBR_NONE)
1100 GET_THREAD_INFO(%rcx)
1101 jmp retint_with_reschedule
1103 #ifdef CONFIG_PREEMPT
1104 /* Returning to kernel space. Check if we need preemption */
1105 /* rcx: threadinfo. interrupts off. */
1106 ENTRY(retint_kernel)
1107 cmpl $0,TI_preempt_count(%rcx)
1108 jnz retint_restore_args
1109 bt $TIF_NEED_RESCHED,TI_flags(%rcx)
1110 jnc retint_restore_args
1111 bt $9,EFLAGS-ARGOFFSET(%rsp) /* interrupts off? */
1112 jnc retint_restore_args
1113 call preempt_schedule_irq
1118 END(common_interrupt)
1120 * End of kprobes section
1127 .macro apicinterrupt num sym do_sym
1139 apicinterrupt IRQ_MOVE_CLEANUP_VECTOR \
1140 irq_move_cleanup_interrupt smp_irq_move_cleanup_interrupt
1141 apicinterrupt REBOOT_VECTOR \
1142 reboot_interrupt smp_reboot_interrupt
1145 #ifdef CONFIG_X86_UV
1146 apicinterrupt UV_BAU_MESSAGE \
1147 uv_bau_message_intr1 uv_bau_message_interrupt
1149 apicinterrupt LOCAL_TIMER_VECTOR \
1150 apic_timer_interrupt smp_apic_timer_interrupt
1151 apicinterrupt X86_PLATFORM_IPI_VECTOR \
1152 x86_platform_ipi smp_x86_platform_ipi
1154 apicinterrupt THRESHOLD_APIC_VECTOR \
1155 threshold_interrupt smp_threshold_interrupt
1156 apicinterrupt THERMAL_APIC_VECTOR \
1157 thermal_interrupt smp_thermal_interrupt
1160 apicinterrupt CALL_FUNCTION_SINGLE_VECTOR \
1161 call_function_single_interrupt smp_call_function_single_interrupt
1162 apicinterrupt CALL_FUNCTION_VECTOR \
1163 call_function_interrupt smp_call_function_interrupt
1164 apicinterrupt RESCHEDULE_VECTOR \
1165 reschedule_interrupt smp_reschedule_interrupt
1168 apicinterrupt ERROR_APIC_VECTOR \
1169 error_interrupt smp_error_interrupt
1170 apicinterrupt SPURIOUS_APIC_VECTOR \
1171 spurious_interrupt smp_spurious_interrupt
1173 #ifdef CONFIG_IRQ_WORK
1174 apicinterrupt IRQ_WORK_VECTOR \
1175 irq_work_interrupt smp_irq_work_interrupt
1179 * Exception entry points.
1181 .macro zeroentry sym do_sym
1184 PARAVIRT_ADJUST_EXCEPTION_FRAME
1185 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1186 subq $ORIG_RAX-R15, %rsp
1187 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1190 movq %rsp,%rdi /* pt_regs pointer */
1191 xorl %esi,%esi /* no error code */
1193 jmp error_exit /* %ebx: no swapgs flag */
1198 .macro paranoidzeroentry sym do_sym
1201 PARAVIRT_ADJUST_EXCEPTION_FRAME
1202 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1203 subq $ORIG_RAX-R15, %rsp
1204 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1207 movq %rsp,%rdi /* pt_regs pointer */
1208 xorl %esi,%esi /* no error code */
1210 jmp paranoid_exit /* %ebx: no swapgs flag */
1215 #define INIT_TSS_IST(x) PER_CPU_VAR(init_tss) + (TSS_ist + ((x) - 1) * 8)
1216 .macro paranoidzeroentry_ist sym do_sym ist
1219 PARAVIRT_ADJUST_EXCEPTION_FRAME
1220 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1221 subq $ORIG_RAX-R15, %rsp
1222 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1224 TRACE_IRQS_OFF_DEBUG
1225 movq %rsp,%rdi /* pt_regs pointer */
1226 xorl %esi,%esi /* no error code */
1227 subq $EXCEPTION_STKSZ, INIT_TSS_IST(\ist)
1229 addq $EXCEPTION_STKSZ, INIT_TSS_IST(\ist)
1230 jmp paranoid_exit /* %ebx: no swapgs flag */
1235 .macro errorentry sym do_sym
1238 PARAVIRT_ADJUST_EXCEPTION_FRAME
1239 subq $ORIG_RAX-R15, %rsp
1240 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1243 movq %rsp,%rdi /* pt_regs pointer */
1244 movq ORIG_RAX(%rsp),%rsi /* get error code */
1245 movq $-1,ORIG_RAX(%rsp) /* no syscall to restart */
1247 jmp error_exit /* %ebx: no swapgs flag */
1252 /* error code is on the stack already */
1253 .macro paranoiderrorentry sym do_sym
1256 PARAVIRT_ADJUST_EXCEPTION_FRAME
1257 subq $ORIG_RAX-R15, %rsp
1258 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1262 movq %rsp,%rdi /* pt_regs pointer */
1263 movq ORIG_RAX(%rsp),%rsi /* get error code */
1264 movq $-1,ORIG_RAX(%rsp) /* no syscall to restart */
1266 jmp paranoid_exit /* %ebx: no swapgs flag */
1271 zeroentry divide_error do_divide_error
1272 zeroentry overflow do_overflow
1273 zeroentry bounds do_bounds
1274 zeroentry invalid_op do_invalid_op
1275 zeroentry device_not_available do_device_not_available
1276 paranoiderrorentry double_fault do_double_fault
1277 zeroentry coprocessor_segment_overrun do_coprocessor_segment_overrun
1278 errorentry invalid_TSS do_invalid_TSS
1279 errorentry segment_not_present do_segment_not_present
1280 zeroentry spurious_interrupt_bug do_spurious_interrupt_bug
1281 zeroentry coprocessor_error do_coprocessor_error
1282 errorentry alignment_check do_alignment_check
1283 zeroentry simd_coprocessor_error do_simd_coprocessor_error
1286 /* Reload gs selector with exception handling */
1287 /* edi: new selector */
1288 ENTRY(native_load_gs_index)
1291 DISABLE_INTERRUPTS(CLBR_ANY & ~CLBR_RDI)
1295 2: mfence /* workaround */
1300 END(native_load_gs_index)
1302 _ASM_EXTABLE(gs_change,bad_gs)
1303 .section .fixup,"ax"
1304 /* running with kernelgs */
1306 SWAPGS /* switch back to user gs */
1312 ENTRY(kernel_thread_helper)
1313 pushq $0 # fake return address
1316 * Here we are in the child and the registers are set as they were
1317 * at kernel_thread() invocation in the parent.
1323 ud2 # padding for call trace
1325 END(kernel_thread_helper)
1328 * execve(). This function needs to use IRET, not SYSRET, to set up all state properly.
1330 * C extern interface:
1331 * extern long execve(const char *name, char **argv, char **envp)
1333 * asm input arguments:
1334 * rdi: name, rsi: argv, rdx: envp
1336 * We want to fallback into:
1337 * extern long sys_execve(const char *name, char **argv,char **envp, struct pt_regs *regs)
1339 * do_sys_execve asm fallback arguments:
1340 * rdi: name, rsi: argv, rdx: envp, rcx: fake frame on the stack
1342 ENTRY(kernel_execve)
1348 movq %rax, RAX(%rsp)
1351 je int_ret_from_sys_call
1358 /* Call softirq on interrupt stack. Interrupts are off. */
1362 CFI_REL_OFFSET rbp,0
1364 CFI_DEF_CFA_REGISTER rbp
1365 incl PER_CPU_VAR(irq_count)
1366 cmove PER_CPU_VAR(irq_stack_ptr),%rsp
1367 push %rbp # backlink for old unwinder
1371 CFI_DEF_CFA_REGISTER rsp
1372 CFI_ADJUST_CFA_OFFSET -8
1373 decl PER_CPU_VAR(irq_count)
1379 zeroentry xen_hypervisor_callback xen_do_hypervisor_callback
1382 * A note on the "critical region" in our callback handler.
1383 * We want to avoid stacking callback handlers due to events occurring
1384 * during handling of the last event. To do this, we keep events disabled
1385 * until we've done all processing. HOWEVER, we must enable events before
1386 * popping the stack frame (can't be done atomically) and so it would still
1387 * be possible to get enough handler activations to overflow the stack.
1388 * Although unlikely, bugs of that kind are hard to track down, so we'd
1389 * like to avoid the possibility.
1390 * So, on entry to the handler we detect whether we interrupted an
1391 * existing activation in its critical region -- if so, we pop the current
1392 * activation and restart the handler using the previous one.
1394 ENTRY(xen_do_hypervisor_callback) # do_hypervisor_callback(struct *pt_regs)
1397 * Since we don't modify %rdi, evtchn_do_upall(struct *pt_regs) will
1398 * see the correct pointer to the pt_regs
1400 movq %rdi, %rsp # we don't return, adjust the stack frame
1403 11: incl PER_CPU_VAR(irq_count)
1405 CFI_DEF_CFA_REGISTER rbp
1406 cmovzq PER_CPU_VAR(irq_stack_ptr),%rsp
1407 pushq %rbp # backlink for old unwinder
1408 call xen_evtchn_do_upcall
1410 CFI_DEF_CFA_REGISTER rsp
1411 decl PER_CPU_VAR(irq_count)
1414 END(xen_do_hypervisor_callback)
1417 * Hypervisor uses this for application faults while it executes.
1418 * We get here for two reasons:
1419 * 1. Fault while reloading DS, ES, FS or GS
1420 * 2. Fault while executing IRET
1421 * Category 1 we do not need to fix up as Xen has already reloaded all segment
1422 * registers that could be reloaded and zeroed the others.
1423 * Category 2 we fix up by killing the current process. We cannot use the
1424 * normal Linux return path in this case because if we use the IRET hypercall
1425 * to pop the stack frame we end up in an infinite loop of failsafe callbacks.
1426 * We distinguish between categories by comparing each saved segment register
1427 * with its current contents: any discrepancy means we in category 1.
1429 ENTRY(xen_failsafe_callback)
1431 /*CFI_REL_OFFSET gs,GS*/
1432 /*CFI_REL_OFFSET fs,FS*/
1433 /*CFI_REL_OFFSET es,ES*/
1434 /*CFI_REL_OFFSET ds,DS*/
1435 CFI_REL_OFFSET r11,8
1436 CFI_REL_OFFSET rcx,0
1450 /* All segments match their saved values => Category 2 (Bad IRET). */
1456 CFI_ADJUST_CFA_OFFSET -0x30
1457 pushq_cfi $0 /* RIP */
1460 jmp general_protection
1462 1: /* Segment mismatch => Category 1 (Bad segment). Retry the IRET. */
1468 CFI_ADJUST_CFA_OFFSET -0x30
1473 END(xen_failsafe_callback)
1475 apicinterrupt XEN_HVM_EVTCHN_CALLBACK \
1476 xen_hvm_callback_vector xen_evtchn_do_upcall
1478 #endif /* CONFIG_XEN */
1481 * Some functions should be protected against kprobes
1483 .pushsection .kprobes.text, "ax"
1485 paranoidzeroentry_ist debug do_debug DEBUG_STACK
1486 paranoidzeroentry_ist int3 do_int3 DEBUG_STACK
1487 paranoiderrorentry stack_segment do_stack_segment
1489 zeroentry xen_debug do_debug
1490 zeroentry xen_int3 do_int3
1491 errorentry xen_stack_segment do_stack_segment
1493 errorentry general_protection do_general_protection
1494 errorentry page_fault do_page_fault
1495 #ifdef CONFIG_KVM_GUEST
1496 errorentry async_page_fault do_async_page_fault
1498 #ifdef CONFIG_X86_MCE
1499 paranoidzeroentry machine_check *machine_check_vector(%rip)
1503 * "Paranoid" exit path from exception stack.
1504 * Paranoid because this is used by NMIs and cannot take
1505 * any kernel state for granted.
1506 * We don't do kernel preemption checks here, because only
1507 * NMI should be common and it does not enable IRQs and
1508 * cannot get reschedule ticks.
1510 * "trace" is 0 for the NMI handler only, because irq-tracing
1511 * is fundamentally NMI-unsafe. (we cannot change the soft and
1512 * hard flags at once, atomically)
1515 /* ebx: no swapgs flag */
1516 ENTRY(paranoid_exit)
1518 DISABLE_INTERRUPTS(CLBR_NONE)
1519 TRACE_IRQS_OFF_DEBUG
1520 testl %ebx,%ebx /* swapgs needed? */
1521 jnz paranoid_restore
1523 jnz paranoid_userspace
1530 TRACE_IRQS_IRETQ_DEBUG 0
1534 GET_THREAD_INFO(%rcx)
1535 movl TI_flags(%rcx),%ebx
1536 andl $_TIF_WORK_MASK,%ebx
1538 movq %rsp,%rdi /* &pt_regs */
1540 movq %rax,%rsp /* switch stack for scheduling */
1541 testl $_TIF_NEED_RESCHED,%ebx
1542 jnz paranoid_schedule
1543 movl %ebx,%edx /* arg3: thread flags */
1545 ENABLE_INTERRUPTS(CLBR_NONE)
1546 xorl %esi,%esi /* arg2: oldset */
1547 movq %rsp,%rdi /* arg1: &pt_regs */
1548 call do_notify_resume
1549 DISABLE_INTERRUPTS(CLBR_NONE)
1551 jmp paranoid_userspace
1554 ENABLE_INTERRUPTS(CLBR_ANY)
1556 DISABLE_INTERRUPTS(CLBR_ANY)
1558 jmp paranoid_userspace
1563 * Exception entry point. This expects an error code/orig_rax on the stack.
1564 * returns in "no swapgs flag" in %ebx.
1568 CFI_ADJUST_CFA_OFFSET 15*8
1569 /* oldrax contains error code */
1588 je error_kernelspace
1596 * There are two places in the kernel that can potentially fault with
1597 * usergs. Handle them here. The exception handlers after iret run with
1598 * kernel gs again, so don't set the user space flag. B stepping K8s
1599 * sometimes report an truncated RIP for IRET exceptions returning to
1600 * compat mode. Check for these here too.
1604 leaq irq_return(%rip),%rcx
1605 cmpq %rcx,RIP+8(%rsp)
1607 movl %ecx,%eax /* zero extend */
1608 cmpq %rax,RIP+8(%rsp)
1610 cmpq $gs_change,RIP+8(%rsp)
1615 /* Fix truncated RIP */
1616 movq %rcx,RIP+8(%rsp)
1622 /* ebx: no swapgs flag (1: don't need swapgs, 0: need it) */
1627 DISABLE_INTERRUPTS(CLBR_NONE)
1629 GET_THREAD_INFO(%rcx)
1632 LOCKDEP_SYS_EXIT_IRQ
1633 movl TI_flags(%rcx),%edx
1634 movl $_TIF_WORK_MASK,%edi
1642 * Test if a given stack is an NMI stack or not.
1644 .macro test_in_nmi reg stack nmi_ret normal_ret
1647 subq $EXCEPTION_STKSZ, %\reg
1653 /* runs on exception stack */
1656 PARAVIRT_ADJUST_EXCEPTION_FRAME
1658 * We allow breakpoints in NMIs. If a breakpoint occurs, then
1659 * the iretq it performs will take us out of NMI context.
1660 * This means that we can have nested NMIs where the next
1661 * NMI is using the top of the stack of the previous NMI. We
1662 * can't let it execute because the nested NMI will corrupt the
1663 * stack of the previous NMI. NMI handlers are not re-entrant
1666 * To handle this case we do the following:
1667 * Check the a special location on the stack that contains
1668 * a variable that is set when NMIs are executing.
1669 * The interrupted task's stack is also checked to see if it
1671 * If the variable is not set and the stack is not the NMI
1673 * o Set the special variable on the stack
1674 * o Copy the interrupt frame into a "saved" location on the stack
1675 * o Copy the interrupt frame into a "copy" location on the stack
1676 * o Continue processing the NMI
1677 * If the variable is set or the previous stack is the NMI stack:
1678 * o Modify the "copy" location to jump to the repeate_nmi
1679 * o return back to the first NMI
1681 * Now on exit of the first NMI, we first clear the stack variable
1682 * The NMI stack will tell any nested NMIs at that point that it is
1683 * nested. Then we pop the stack normally with iret, and if there was
1684 * a nested NMI that updated the copy interrupt stack frame, a
1685 * jump will be made to the repeat_nmi code that will handle the second
1689 /* Use %rdx as out temp variable throughout */
1691 CFI_REL_OFFSET rdx, 0
1694 * If %cs was not the kernel segment, then the NMI triggered in user
1695 * space, which means it is definitely not nested.
1697 cmpl $__KERNEL_CS, 16(%rsp)
1701 * Check the special variable on the stack to see if NMIs are
1708 * Now test if the previous stack was an NMI stack.
1709 * We need the double check. We check the NMI stack to satisfy the
1710 * race when the first NMI clears the variable before returning.
1711 * We check the variable because the first NMI could be in a
1712 * breakpoint routine using a breakpoint stack.
1715 test_in_nmi rdx, 4*8(%rsp), nested_nmi, first_nmi
1720 * Do nothing if we interrupted the fixup in repeat_nmi.
1721 * It's about to repeat the NMI handler, so we are fine
1722 * with ignoring this one.
1724 movq $repeat_nmi, %rdx
1727 movq $end_repeat_nmi, %rdx
1732 /* Set up the interrupted NMIs stack to jump to repeat_nmi */
1733 leaq -6*8(%rsp), %rdx
1735 CFI_ADJUST_CFA_OFFSET 6*8
1736 pushq_cfi $__KERNEL_DS
1739 pushq_cfi $__KERNEL_CS
1740 pushq_cfi $repeat_nmi
1742 /* Put stack back */
1744 CFI_ADJUST_CFA_OFFSET -11*8
1750 /* No need to check faults here */
1756 * Because nested NMIs will use the pushed location that we
1757 * stored in rdx, we must keep that space available.
1758 * Here's what our stack frame will look like:
1759 * +-------------------------+
1761 * | original Return RSP |
1762 * | original RFLAGS |
1765 * +-------------------------+
1766 * | temp storage for rdx |
1767 * +-------------------------+
1768 * | NMI executing variable |
1769 * +-------------------------+
1771 * | Saved Return RSP |
1775 * +-------------------------+
1777 * | copied Return RSP |
1781 * +-------------------------+
1783 * +-------------------------+
1785 * The saved stack frame is used to fix up the copied stack frame
1786 * that a nested NMI may change to make the interrupted NMI iret jump
1787 * to the repeat_nmi. The original stack frame and the temp storage
1788 * is also used by nested NMIs and can not be trusted on exit.
1790 /* Do not pop rdx, nested NMIs will corrupt that part of the stack */
1794 /* Set the NMI executing variable on the stack. */
1797 /* Copy the stack frame to the Saved frame */
1801 CFI_DEF_CFA_OFFSET SS+8-RIP
1803 /* Everything up to here is safe from nested NMIs */
1806 * If there was a nested NMI, the first NMI's iret will return
1807 * here. But NMIs are still enabled and we can take another
1808 * nested NMI. The nested NMI checks the interrupted RIP to see
1809 * if it is between repeat_nmi and end_repeat_nmi, and if so
1810 * it will just return, as we are about to repeat an NMI anyway.
1811 * This makes it safe to copy to the stack frame that a nested
1816 * Update the stack variable to say we are still in NMI (the update
1817 * is benign for the non-repeat case, where 1 was pushed just above
1818 * to this very stack slot).
1822 /* Make another copy, this one may be modified by nested NMIs */
1826 CFI_DEF_CFA_OFFSET SS+8-RIP
1830 * Everything below this point can be preempted by a nested
1831 * NMI if the first NMI took an exception and reset our iret stack
1832 * so that we repeat another NMI.
1834 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1835 subq $ORIG_RAX-R15, %rsp
1836 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1838 * Use save_paranoid to handle SWAPGS, but no need to use paranoid_exit
1839 * as we should not be calling schedule in NMI context.
1840 * Even with normal interrupts enabled. An NMI should not be
1841 * setting NEED_RESCHED or anything that normal interrupts and
1842 * exceptions might do.
1848 * Save off the CR2 register. If we take a page fault in the NMI then
1849 * it could corrupt the CR2 value. If the NMI preempts a page fault
1850 * handler before it was able to read the CR2 register, and then the
1851 * NMI itself takes a page fault, the page fault that was preempted
1852 * will read the information from the NMI page fault and not the
1853 * origin fault. Save it off and restore it if it changes.
1854 * Use the r12 callee-saved register.
1858 /* paranoidentry do_nmi, 0; without TRACE_IRQS_OFF */
1863 /* Did the NMI take a page fault? Restore cr2 if it did */
1870 testl %ebx,%ebx /* swapgs needed? */
1876 /* Clear the NMI executing stack variable */
1882 ENTRY(ignore_sysret)
1890 * End of kprobes section