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>
59 #include <linux/err.h>
61 /* Avoid __ASSEMBLER__'ifying <linux/audit.h> just for this. */
62 #include <linux/elf-em.h>
63 #define AUDIT_ARCH_X86_64 (EM_X86_64|__AUDIT_ARCH_64BIT|__AUDIT_ARCH_LE)
64 #define __AUDIT_ARCH_64BIT 0x80000000
65 #define __AUDIT_ARCH_LE 0x40000000
68 .section .entry.text, "ax"
70 #ifdef CONFIG_FUNCTION_TRACER
71 #ifdef CONFIG_DYNAMIC_FTRACE
76 /* skip is set if stack has been adjusted */
77 .macro ftrace_caller_setup skip=0
78 MCOUNT_SAVE_FRAME \skip
80 /* Load the ftrace_ops into the 3rd parameter */
81 leaq function_trace_op, %rdx
83 /* Load ip into the first parameter */
85 subq $MCOUNT_INSN_SIZE, %rdi
86 /* Load the parent_ip into the second parameter */
91 /* Check if tracing was disabled (quick check) */
92 cmpl $0, function_trace_stop
96 /* regs go into 4th parameter (but make it NULL) */
105 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
106 GLOBAL(ftrace_graph_call)
114 ENTRY(ftrace_regs_caller)
115 /* Save the current flags before compare (in SS location)*/
118 /* Check if tracing was disabled (quick check) */
119 cmpl $0, function_trace_stop
120 jne ftrace_restore_flags
122 /* skip=8 to skip flags saved in SS */
123 ftrace_caller_setup 8
125 /* Save the rest of pt_regs */
134 /* Copy saved flags */
136 movq %rcx, EFLAGS(%rsp)
137 /* Kernel segments */
138 movq $__KERNEL_DS, %rcx
140 movq $__KERNEL_CS, %rcx
142 /* Stack - skipping return address */
143 leaq SS+16(%rsp), %rcx
146 /* regs go into 4th parameter */
149 GLOBAL(ftrace_regs_call)
152 /* Copy flags back to SS, to restore them */
153 movq EFLAGS(%rsp), %rax
156 /* restore the rest of pt_regs */
165 /* skip=8 to skip flags saved in SS */
166 MCOUNT_RESTORE_FRAME 8
172 ftrace_restore_flags:
176 END(ftrace_regs_caller)
179 #else /* ! CONFIG_DYNAMIC_FTRACE */
181 cmpl $0, function_trace_stop
184 cmpq $ftrace_stub, ftrace_trace_function
187 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
188 cmpq $ftrace_stub, ftrace_graph_return
189 jnz ftrace_graph_caller
191 cmpq $ftrace_graph_entry_stub, ftrace_graph_entry
192 jnz ftrace_graph_caller
203 subq $MCOUNT_INSN_SIZE, %rdi
205 call *ftrace_trace_function
211 #endif /* CONFIG_DYNAMIC_FTRACE */
212 #endif /* CONFIG_FUNCTION_TRACER */
214 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
215 ENTRY(ftrace_graph_caller)
216 cmpl $0, function_trace_stop
224 subq $MCOUNT_INSN_SIZE, %rsi
226 call prepare_ftrace_return
231 END(ftrace_graph_caller)
233 GLOBAL(return_to_handler)
236 /* Save the return values */
241 call ftrace_return_to_handler
251 #ifndef CONFIG_PREEMPT
252 #define retint_kernel retint_restore_args
255 #ifdef CONFIG_PARAVIRT
256 ENTRY(native_usergs_sysret64)
259 ENDPROC(native_usergs_sysret64)
260 #endif /* CONFIG_PARAVIRT */
263 .macro TRACE_IRQS_IRETQ offset=ARGOFFSET
264 #ifdef CONFIG_TRACE_IRQFLAGS
265 bt $9,EFLAGS-\offset(%rsp) /* interrupts off? */
273 * When dynamic function tracer is enabled it will add a breakpoint
274 * to all locations that it is about to modify, sync CPUs, update
275 * all the code, sync CPUs, then remove the breakpoints. In this time
276 * if lockdep is enabled, it might jump back into the debug handler
277 * outside the updating of the IST protection. (TRACE_IRQS_ON/OFF).
279 * We need to change the IDT table before calling TRACE_IRQS_ON/OFF to
280 * make sure the stack pointer does not get reset back to the top
281 * of the debug stack, and instead just reuses the current stack.
283 #if defined(CONFIG_DYNAMIC_FTRACE) && defined(CONFIG_TRACE_IRQFLAGS)
285 .macro TRACE_IRQS_OFF_DEBUG
286 call debug_stack_set_zero
288 call debug_stack_reset
291 .macro TRACE_IRQS_ON_DEBUG
292 call debug_stack_set_zero
294 call debug_stack_reset
297 .macro TRACE_IRQS_IRETQ_DEBUG offset=ARGOFFSET
298 bt $9,EFLAGS-\offset(%rsp) /* interrupts off? */
305 # define TRACE_IRQS_OFF_DEBUG TRACE_IRQS_OFF
306 # define TRACE_IRQS_ON_DEBUG TRACE_IRQS_ON
307 # define TRACE_IRQS_IRETQ_DEBUG TRACE_IRQS_IRETQ
311 * C code is not supposed to know about undefined top of stack. Every time
312 * a C function with an pt_regs argument is called from the SYSCALL based
313 * fast path FIXUP_TOP_OF_STACK is needed.
314 * RESTORE_TOP_OF_STACK syncs the syscall state after any possible ptregs
318 /* %rsp:at FRAMEEND */
319 .macro FIXUP_TOP_OF_STACK tmp offset=0
320 movq PER_CPU_VAR(old_rsp),\tmp
321 movq \tmp,RSP+\offset(%rsp)
322 movq $__USER_DS,SS+\offset(%rsp)
323 movq $__USER_CS,CS+\offset(%rsp)
324 movq $-1,RCX+\offset(%rsp)
325 movq R11+\offset(%rsp),\tmp /* get eflags */
326 movq \tmp,EFLAGS+\offset(%rsp)
329 .macro RESTORE_TOP_OF_STACK tmp offset=0
330 movq RSP+\offset(%rsp),\tmp
331 movq \tmp,PER_CPU_VAR(old_rsp)
332 movq EFLAGS+\offset(%rsp),\tmp
333 movq \tmp,R11+\offset(%rsp)
336 .macro FAKE_STACK_FRAME child_rip
337 /* push in order ss, rsp, eflags, cs, rip */
339 pushq_cfi $__KERNEL_DS /* ss */
340 /*CFI_REL_OFFSET ss,0*/
341 pushq_cfi %rax /* rsp */
343 pushq_cfi $(X86_EFLAGS_IF|X86_EFLAGS_BIT1) /* eflags - interrupts on */
344 /*CFI_REL_OFFSET rflags,0*/
345 pushq_cfi $__KERNEL_CS /* cs */
346 /*CFI_REL_OFFSET cs,0*/
347 pushq_cfi \child_rip /* rip */
349 pushq_cfi %rax /* orig rax */
352 .macro UNFAKE_STACK_FRAME
354 CFI_ADJUST_CFA_OFFSET -(6*8)
358 * initial frame state for interrupts (and exceptions without error code)
360 .macro EMPTY_FRAME start=1 offset=0
364 CFI_DEF_CFA rsp,8+\offset
366 CFI_DEF_CFA_OFFSET 8+\offset
371 * initial frame state for interrupts (and exceptions without error code)
373 .macro INTR_FRAME start=1 offset=0
374 EMPTY_FRAME \start, SS+8+\offset-RIP
375 /*CFI_REL_OFFSET ss, SS+\offset-RIP*/
376 CFI_REL_OFFSET rsp, RSP+\offset-RIP
377 /*CFI_REL_OFFSET rflags, EFLAGS+\offset-RIP*/
378 /*CFI_REL_OFFSET cs, CS+\offset-RIP*/
379 CFI_REL_OFFSET rip, RIP+\offset-RIP
383 * initial frame state for exceptions with error code (and interrupts
384 * with vector already pushed)
386 .macro XCPT_FRAME start=1 offset=0
387 INTR_FRAME \start, RIP+\offset-ORIG_RAX
388 /*CFI_REL_OFFSET orig_rax, ORIG_RAX-ORIG_RAX*/
392 * frame that enables calling into C.
394 .macro PARTIAL_FRAME start=1 offset=0
395 XCPT_FRAME \start, ORIG_RAX+\offset-ARGOFFSET
396 CFI_REL_OFFSET rdi, RDI+\offset-ARGOFFSET
397 CFI_REL_OFFSET rsi, RSI+\offset-ARGOFFSET
398 CFI_REL_OFFSET rdx, RDX+\offset-ARGOFFSET
399 CFI_REL_OFFSET rcx, RCX+\offset-ARGOFFSET
400 CFI_REL_OFFSET rax, RAX+\offset-ARGOFFSET
401 CFI_REL_OFFSET r8, R8+\offset-ARGOFFSET
402 CFI_REL_OFFSET r9, R9+\offset-ARGOFFSET
403 CFI_REL_OFFSET r10, R10+\offset-ARGOFFSET
404 CFI_REL_OFFSET r11, R11+\offset-ARGOFFSET
408 * frame that enables passing a complete pt_regs to a C function.
410 .macro DEFAULT_FRAME start=1 offset=0
411 PARTIAL_FRAME \start, R11+\offset-R15
412 CFI_REL_OFFSET rbx, RBX+\offset
413 CFI_REL_OFFSET rbp, RBP+\offset
414 CFI_REL_OFFSET r12, R12+\offset
415 CFI_REL_OFFSET r13, R13+\offset
416 CFI_REL_OFFSET r14, R14+\offset
417 CFI_REL_OFFSET r15, R15+\offset
420 /* save partial stack frame */
423 /* start from rbp in pt_regs and jump over */
424 movq_cfi rdi, RDI-RBP
425 movq_cfi rsi, RSI-RBP
426 movq_cfi rdx, RDX-RBP
427 movq_cfi rcx, RCX-RBP
428 movq_cfi rax, RAX-RBP
431 movq_cfi r10, R10-RBP
432 movq_cfi r11, R11-RBP
434 /* Save rbp so that we can unwind from get_irq_regs() */
437 /* Save previous stack value */
440 leaq -RBP(%rsp),%rdi /* arg1 for handler */
441 testl $3, CS-RBP(%rsi)
445 * irq_count is used to check if a CPU is already on an interrupt stack
446 * or not. While this is essentially redundant with preempt_count it is
447 * a little cheaper to use a separate counter in the PDA (short of
448 * moving irq_enter into assembly, which would be too much work)
450 1: incl PER_CPU_VAR(irq_count)
451 cmovzq PER_CPU_VAR(irq_stack_ptr),%rsp
452 CFI_DEF_CFA_REGISTER rsi
454 /* Store previous stack value */
456 CFI_ESCAPE 0x0f /* DW_CFA_def_cfa_expression */, 6, \
457 0x77 /* DW_OP_breg7 */, 0, \
458 0x06 /* DW_OP_deref */, \
459 0x08 /* DW_OP_const1u */, SS+8-RBP, \
460 0x22 /* DW_OP_plus */
461 /* We entered an interrupt context - irqs are off: */
466 PARTIAL_FRAME 1 REST_SKIP+8
467 movq 5*8+16(%rsp), %r11 /* save return address */
474 movq %r11, 8(%rsp) /* return address */
475 FIXUP_TOP_OF_STACK %r11, 16
480 /* save complete stack frame */
481 .pushsection .kprobes.text, "ax"
501 movl $MSR_GS_BASE,%ecx
504 js 1f /* negative -> in kernel */
513 * A newly forked process directly context switches into this address.
515 * rdi: prev task we switched from
520 LOCK ; btr $TIF_FORK,TI_flags(%r8)
522 pushq_cfi kernel_eflags(%rip)
523 popfq_cfi # reset kernel eflags
525 call schedule_tail # rdi: 'prev' task parameter
527 GET_THREAD_INFO(%rcx)
531 testl $3, CS-ARGOFFSET(%rsp) # from kernel_thread?
532 jz retint_restore_args
534 testl $_TIF_IA32, TI_flags(%rcx) # 32-bit compat task needs IRET
535 jnz int_ret_from_sys_call
537 RESTORE_TOP_OF_STACK %rdi, -ARGOFFSET
538 jmp ret_from_sys_call # go to the SYSRET fastpath
544 * System call entry. Up to 6 arguments in registers are supported.
546 * SYSCALL does not save anything on the stack and does not change the
552 * rax system call number
554 * rcx return address for syscall/sysret, C arg3
557 * r10 arg3 (--> moved to rcx for C)
560 * r11 eflags for syscall/sysret, temporary for C
561 * r12-r15,rbp,rbx saved by C code, not touched.
563 * Interrupts are off on entry.
564 * Only called from user space.
566 * XXX if we had a free scratch register we could save the RSP into the stack frame
567 * and report it properly in ps. Unfortunately we haven't.
569 * When user can change the frames always force IRET. That is because
570 * it deals with uncanonical addresses better. SYSRET has trouble
571 * with them due to bugs in both AMD and Intel CPUs.
577 CFI_DEF_CFA rsp,KERNEL_STACK_OFFSET
579 /*CFI_REGISTER rflags,r11*/
582 * A hypervisor implementation might want to use a label
583 * after the swapgs, so that it can do the swapgs
584 * for the guest and jump here on syscall.
586 GLOBAL(system_call_after_swapgs)
588 movq %rsp,PER_CPU_VAR(old_rsp)
589 movq PER_CPU_VAR(kernel_stack),%rsp
591 * No need to follow this irqs off/on section - it's straight
594 ENABLE_INTERRUPTS(CLBR_NONE)
596 movq %rax,ORIG_RAX-ARGOFFSET(%rsp)
597 movq %rcx,RIP-ARGOFFSET(%rsp)
598 CFI_REL_OFFSET rip,RIP-ARGOFFSET
599 testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET)
601 system_call_fastpath:
602 #if __SYSCALL_MASK == ~0
603 cmpq $__NR_syscall_max,%rax
605 andl $__SYSCALL_MASK,%eax
606 cmpl $__NR_syscall_max,%eax
610 call *sys_call_table(,%rax,8) # XXX: rip relative
611 movq %rax,RAX-ARGOFFSET(%rsp)
613 * Syscall return path ending with SYSRET (fast path)
614 * Has incomplete stack frame and undefined top of stack.
617 movl $_TIF_ALLWORK_MASK,%edi
621 DISABLE_INTERRUPTS(CLBR_NONE)
623 movl TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET),%edx
628 * sysretq will re-enable interrupts:
631 movq RIP-ARGOFFSET(%rsp),%rcx
633 RESTORE_ARGS 1,-ARG_SKIP,0
634 /*CFI_REGISTER rflags,r11*/
635 movq PER_CPU_VAR(old_rsp), %rsp
639 /* Handle reschedules */
640 /* edx: work, edi: workmask */
642 bt $TIF_NEED_RESCHED,%edx
645 ENABLE_INTERRUPTS(CLBR_NONE)
651 /* Handle a signal */
654 ENABLE_INTERRUPTS(CLBR_NONE)
655 #ifdef CONFIG_AUDITSYSCALL
656 bt $TIF_SYSCALL_AUDIT,%edx
660 * We have a signal, or exit tracing or single-step.
661 * These all wind up with the iret return path anyway,
662 * so just join that path right now.
664 FIXUP_TOP_OF_STACK %r11, -ARGOFFSET
665 jmp int_check_syscall_exit_work
668 movq $-ENOSYS,RAX-ARGOFFSET(%rsp)
669 jmp ret_from_sys_call
671 #ifdef CONFIG_AUDITSYSCALL
673 * Fast path for syscall audit without full syscall trace.
674 * We just call __audit_syscall_entry() directly, and then
675 * jump back to the normal fast path.
678 movq %r10,%r9 /* 6th arg: 4th syscall arg */
679 movq %rdx,%r8 /* 5th arg: 3rd syscall arg */
680 movq %rsi,%rcx /* 4th arg: 2nd syscall arg */
681 movq %rdi,%rdx /* 3rd arg: 1st syscall arg */
682 movq %rax,%rsi /* 2nd arg: syscall number */
683 movl $AUDIT_ARCH_X86_64,%edi /* 1st arg: audit arch */
684 call __audit_syscall_entry
685 LOAD_ARGS 0 /* reload call-clobbered registers */
686 jmp system_call_fastpath
689 * Return fast path for syscall audit. Call __audit_syscall_exit()
690 * directly and then jump back to the fast path with TIF_SYSCALL_AUDIT
694 movq RAX-ARGOFFSET(%rsp),%rsi /* second arg, syscall return value */
695 cmpq $-MAX_ERRNO,%rsi /* is it < -MAX_ERRNO? */
696 setbe %al /* 1 if so, 0 if not */
697 movzbl %al,%edi /* zero-extend that into %edi */
698 call __audit_syscall_exit
699 movl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT),%edi
701 #endif /* CONFIG_AUDITSYSCALL */
703 /* Do syscall tracing */
705 #ifdef CONFIG_AUDITSYSCALL
706 testl $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET)
710 movq $-ENOSYS,RAX(%rsp) /* ptrace can change this for a bad syscall */
711 FIXUP_TOP_OF_STACK %rdi
713 call syscall_trace_enter
715 * Reload arg registers from stack in case ptrace changed them.
716 * We don't reload %rax because syscall_trace_enter() returned
717 * the value it wants us to use in the table lookup.
719 LOAD_ARGS ARGOFFSET, 1
721 #if __SYSCALL_MASK == ~0
722 cmpq $__NR_syscall_max,%rax
724 andl $__SYSCALL_MASK,%eax
725 cmpl $__NR_syscall_max,%eax
727 ja int_ret_from_sys_call /* RAX(%rsp) set to -ENOSYS above */
728 movq %r10,%rcx /* fixup for C */
729 call *sys_call_table(,%rax,8)
730 movq %rax,RAX-ARGOFFSET(%rsp)
731 /* Use IRET because user could have changed frame */
734 * Syscall return path ending with IRET.
735 * Has correct top of stack, but partial stack frame.
737 GLOBAL(int_ret_from_sys_call)
738 DISABLE_INTERRUPTS(CLBR_NONE)
740 movl $_TIF_ALLWORK_MASK,%edi
741 /* edi: mask to check */
742 GLOBAL(int_with_check)
744 GET_THREAD_INFO(%rcx)
745 movl TI_flags(%rcx),%edx
748 andl $~TS_COMPAT,TI_status(%rcx)
751 /* Either reschedule or signal or syscall exit tracking needed. */
752 /* First do a reschedule test. */
753 /* edx: work, edi: workmask */
755 bt $TIF_NEED_RESCHED,%edx
758 ENABLE_INTERRUPTS(CLBR_NONE)
762 DISABLE_INTERRUPTS(CLBR_NONE)
766 /* handle signals and tracing -- both require a full stack frame */
769 ENABLE_INTERRUPTS(CLBR_NONE)
770 int_check_syscall_exit_work:
772 /* Check for syscall exit trace */
773 testl $_TIF_WORK_SYSCALL_EXIT,%edx
776 leaq 8(%rsp),%rdi # &ptregs -> arg1
777 call syscall_trace_leave
779 andl $~(_TIF_WORK_SYSCALL_EXIT|_TIF_SYSCALL_EMU),%edi
783 testl $_TIF_DO_NOTIFY_MASK,%edx
785 movq %rsp,%rdi # &ptregs -> arg1
786 xorl %esi,%esi # oldset -> arg2
787 call do_notify_resume
788 1: movl $_TIF_WORK_MASK,%edi
791 DISABLE_INTERRUPTS(CLBR_NONE)
798 * Certain special system calls that need to save a complete full stack frame.
800 .macro PTREGSCALL label,func,arg
802 PARTIAL_FRAME 1 8 /* offset 8: return address */
803 subq $REST_SKIP, %rsp
804 CFI_ADJUST_CFA_OFFSET REST_SKIP
806 DEFAULT_FRAME 0 8 /* offset 8: return address */
807 leaq 8(%rsp), \arg /* pt_regs pointer */
809 jmp ptregscall_common
814 PTREGSCALL stub_clone, sys_clone, %r8
815 PTREGSCALL stub_fork, sys_fork, %rdi
816 PTREGSCALL stub_vfork, sys_vfork, %rdi
817 PTREGSCALL stub_sigaltstack, sys_sigaltstack, %rdx
818 PTREGSCALL stub_iopl, sys_iopl, %rsi
820 ENTRY(ptregscall_common)
821 DEFAULT_FRAME 1 8 /* offset 8: return address */
822 RESTORE_TOP_OF_STACK %r11, 8
823 movq_cfi_restore R15+8, r15
824 movq_cfi_restore R14+8, r14
825 movq_cfi_restore R13+8, r13
826 movq_cfi_restore R12+8, r12
827 movq_cfi_restore RBP+8, rbp
828 movq_cfi_restore RBX+8, rbx
829 ret $REST_SKIP /* pop extended registers */
831 END(ptregscall_common)
838 FIXUP_TOP_OF_STACK %r11
841 RESTORE_TOP_OF_STACK %r11
844 jmp int_ret_from_sys_call
849 * sigreturn is special because it needs to restore all registers on return.
850 * This cannot be done with SYSRET, so use the IRET return path instead.
852 ENTRY(stub_rt_sigreturn)
858 FIXUP_TOP_OF_STACK %r11
859 call sys_rt_sigreturn
860 movq %rax,RAX(%rsp) # fixme, this could be done at the higher layer
862 jmp int_ret_from_sys_call
864 END(stub_rt_sigreturn)
866 #ifdef CONFIG_X86_X32_ABI
867 PTREGSCALL stub_x32_sigaltstack, sys32_sigaltstack, %rdx
869 ENTRY(stub_x32_rt_sigreturn)
875 FIXUP_TOP_OF_STACK %r11
876 call sys32_x32_rt_sigreturn
877 movq %rax,RAX(%rsp) # fixme, this could be done at the higher layer
879 jmp int_ret_from_sys_call
881 END(stub_x32_rt_sigreturn)
883 ENTRY(stub_x32_execve)
888 FIXUP_TOP_OF_STACK %r11
891 RESTORE_TOP_OF_STACK %r11
894 jmp int_ret_from_sys_call
901 * Build the entry stubs and pointer table with some assembler magic.
902 * We pack 7 stubs into a single 32-byte chunk, which will fit in a
903 * single cache line on all modern x86 implementations.
905 .section .init.rodata,"a"
909 .p2align CONFIG_X86_L1_CACHE_SHIFT
910 ENTRY(irq_entries_start)
912 vector=FIRST_EXTERNAL_VECTOR
913 .rept (NR_VECTORS-FIRST_EXTERNAL_VECTOR+6)/7
916 .if vector < NR_VECTORS
917 .if vector <> FIRST_EXTERNAL_VECTOR
918 CFI_ADJUST_CFA_OFFSET -8
920 1: pushq_cfi $(~vector+0x80) /* Note: always in signed byte range */
921 .if ((vector-FIRST_EXTERNAL_VECTOR)%7) <> 6
930 2: jmp common_interrupt
933 END(irq_entries_start)
940 * Interrupt entry/exit.
942 * Interrupt entry points save only callee clobbered registers in fast path.
944 * Entry runs with interrupts off.
947 /* 0(%rsp): ~(interrupt number) */
948 .macro interrupt func
949 /* reserve pt_regs for scratch regs and rbp */
950 subq $ORIG_RAX-RBP, %rsp
951 CFI_ADJUST_CFA_OFFSET ORIG_RAX-RBP
957 * Interrupt entry/exit should be protected against kprobes
959 .pushsection .kprobes.text, "ax"
961 * The interrupt stubs push (~vector+0x80) onto the stack and
962 * then jump to common_interrupt.
964 .p2align CONFIG_X86_L1_CACHE_SHIFT
967 addq $-0x80,(%rsp) /* Adjust vector to [-256,-1] range */
969 /* 0(%rsp): old_rsp-ARGOFFSET */
971 DISABLE_INTERRUPTS(CLBR_NONE)
973 decl PER_CPU_VAR(irq_count)
975 /* Restore saved previous stack */
977 CFI_DEF_CFA rsi,SS+8-RBP /* reg/off reset after def_cfa_expr */
978 leaq ARGOFFSET-RBP(%rsi), %rsp
979 CFI_DEF_CFA_REGISTER rsp
980 CFI_ADJUST_CFA_OFFSET RBP-ARGOFFSET
983 GET_THREAD_INFO(%rcx)
984 testl $3,CS-ARGOFFSET(%rsp)
987 /* Interrupt came from user space */
989 * Has a correct top of stack, but a partial stack frame
990 * %rcx: thread info. Interrupts off.
992 retint_with_reschedule:
993 movl $_TIF_WORK_MASK,%edi
996 movl TI_flags(%rcx),%edx
1001 retint_swapgs: /* return to user-space */
1003 * The iretq could re-enable interrupts:
1005 DISABLE_INTERRUPTS(CLBR_ANY)
1010 retint_restore_args: /* return to kernel space */
1011 DISABLE_INTERRUPTS(CLBR_ANY)
1013 * The iretq could re-enable interrupts:
1021 _ASM_EXTABLE(irq_return, bad_iret)
1023 #ifdef CONFIG_PARAVIRT
1026 _ASM_EXTABLE(native_iret, bad_iret)
1029 .section .fixup,"ax"
1032 * The iret traps when the %cs or %ss being restored is bogus.
1033 * We've lost the original trap vector and error code.
1034 * #GPF is the most likely one to get for an invalid selector.
1035 * So pretend we completed the iret and took the #GPF in user mode.
1037 * We are now running with the kernel GS after exception recovery.
1038 * But error_entry expects us to have user GS to match the user %cs,
1044 jmp general_protection
1048 /* edi: workmask, edx: work */
1051 bt $TIF_NEED_RESCHED,%edx
1054 ENABLE_INTERRUPTS(CLBR_NONE)
1058 GET_THREAD_INFO(%rcx)
1059 DISABLE_INTERRUPTS(CLBR_NONE)
1064 testl $_TIF_DO_NOTIFY_MASK,%edx
1067 ENABLE_INTERRUPTS(CLBR_NONE)
1069 movq $-1,ORIG_RAX(%rsp)
1070 xorl %esi,%esi # oldset
1071 movq %rsp,%rdi # &pt_regs
1072 call do_notify_resume
1074 DISABLE_INTERRUPTS(CLBR_NONE)
1076 GET_THREAD_INFO(%rcx)
1077 jmp retint_with_reschedule
1079 #ifdef CONFIG_PREEMPT
1080 /* Returning to kernel space. Check if we need preemption */
1081 /* rcx: threadinfo. interrupts off. */
1082 ENTRY(retint_kernel)
1083 cmpl $0,TI_preempt_count(%rcx)
1084 jnz retint_restore_args
1085 bt $TIF_NEED_RESCHED,TI_flags(%rcx)
1086 jnc retint_restore_args
1087 bt $9,EFLAGS-ARGOFFSET(%rsp) /* interrupts off? */
1088 jnc retint_restore_args
1089 call preempt_schedule_irq
1094 END(common_interrupt)
1096 * End of kprobes section
1103 .macro apicinterrupt num sym do_sym
1115 apicinterrupt IRQ_MOVE_CLEANUP_VECTOR \
1116 irq_move_cleanup_interrupt smp_irq_move_cleanup_interrupt
1117 apicinterrupt REBOOT_VECTOR \
1118 reboot_interrupt smp_reboot_interrupt
1121 #ifdef CONFIG_X86_UV
1122 apicinterrupt UV_BAU_MESSAGE \
1123 uv_bau_message_intr1 uv_bau_message_interrupt
1125 apicinterrupt LOCAL_TIMER_VECTOR \
1126 apic_timer_interrupt smp_apic_timer_interrupt
1127 apicinterrupt X86_PLATFORM_IPI_VECTOR \
1128 x86_platform_ipi smp_x86_platform_ipi
1133 .irp idx,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15, \
1134 16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31
1135 .if NUM_INVALIDATE_TLB_VECTORS > \idx
1136 ENTRY(invalidate_interrupt\idx)
1137 pushq_cfi $~(INVALIDATE_TLB_VECTOR_START+\idx)
1138 jmp .Lcommon_invalidate_interrupt0
1139 CFI_ADJUST_CFA_OFFSET -8
1140 END(invalidate_interrupt\idx)
1144 apicinterrupt INVALIDATE_TLB_VECTOR_START, \
1145 invalidate_interrupt0, smp_invalidate_interrupt
1148 apicinterrupt THRESHOLD_APIC_VECTOR \
1149 threshold_interrupt smp_threshold_interrupt
1150 apicinterrupt THERMAL_APIC_VECTOR \
1151 thermal_interrupt smp_thermal_interrupt
1154 apicinterrupt CALL_FUNCTION_SINGLE_VECTOR \
1155 call_function_single_interrupt smp_call_function_single_interrupt
1156 apicinterrupt CALL_FUNCTION_VECTOR \
1157 call_function_interrupt smp_call_function_interrupt
1158 apicinterrupt RESCHEDULE_VECTOR \
1159 reschedule_interrupt smp_reschedule_interrupt
1162 apicinterrupt ERROR_APIC_VECTOR \
1163 error_interrupt smp_error_interrupt
1164 apicinterrupt SPURIOUS_APIC_VECTOR \
1165 spurious_interrupt smp_spurious_interrupt
1167 #ifdef CONFIG_IRQ_WORK
1168 apicinterrupt IRQ_WORK_VECTOR \
1169 irq_work_interrupt smp_irq_work_interrupt
1173 * Exception entry points.
1175 .macro zeroentry sym do_sym
1178 PARAVIRT_ADJUST_EXCEPTION_FRAME
1179 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1180 subq $ORIG_RAX-R15, %rsp
1181 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1184 movq %rsp,%rdi /* pt_regs pointer */
1185 xorl %esi,%esi /* no error code */
1187 jmp error_exit /* %ebx: no swapgs flag */
1192 .macro paranoidzeroentry sym do_sym
1195 PARAVIRT_ADJUST_EXCEPTION_FRAME
1196 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1197 subq $ORIG_RAX-R15, %rsp
1198 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1201 movq %rsp,%rdi /* pt_regs pointer */
1202 xorl %esi,%esi /* no error code */
1204 jmp paranoid_exit /* %ebx: no swapgs flag */
1209 #define INIT_TSS_IST(x) PER_CPU_VAR(init_tss) + (TSS_ist + ((x) - 1) * 8)
1210 .macro paranoidzeroentry_ist sym do_sym ist
1213 PARAVIRT_ADJUST_EXCEPTION_FRAME
1214 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1215 subq $ORIG_RAX-R15, %rsp
1216 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1218 TRACE_IRQS_OFF_DEBUG
1219 movq %rsp,%rdi /* pt_regs pointer */
1220 xorl %esi,%esi /* no error code */
1221 subq $EXCEPTION_STKSZ, INIT_TSS_IST(\ist)
1223 addq $EXCEPTION_STKSZ, INIT_TSS_IST(\ist)
1224 jmp paranoid_exit /* %ebx: no swapgs flag */
1229 .macro errorentry sym do_sym
1232 PARAVIRT_ADJUST_EXCEPTION_FRAME
1233 subq $ORIG_RAX-R15, %rsp
1234 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1237 movq %rsp,%rdi /* pt_regs pointer */
1238 movq ORIG_RAX(%rsp),%rsi /* get error code */
1239 movq $-1,ORIG_RAX(%rsp) /* no syscall to restart */
1241 jmp error_exit /* %ebx: no swapgs flag */
1246 /* error code is on the stack already */
1247 .macro paranoiderrorentry sym do_sym
1250 PARAVIRT_ADJUST_EXCEPTION_FRAME
1251 subq $ORIG_RAX-R15, %rsp
1252 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1256 movq %rsp,%rdi /* pt_regs pointer */
1257 movq ORIG_RAX(%rsp),%rsi /* get error code */
1258 movq $-1,ORIG_RAX(%rsp) /* no syscall to restart */
1260 jmp paranoid_exit /* %ebx: no swapgs flag */
1265 zeroentry divide_error do_divide_error
1266 zeroentry overflow do_overflow
1267 zeroentry bounds do_bounds
1268 zeroentry invalid_op do_invalid_op
1269 zeroentry device_not_available do_device_not_available
1270 paranoiderrorentry double_fault do_double_fault
1271 zeroentry coprocessor_segment_overrun do_coprocessor_segment_overrun
1272 errorentry invalid_TSS do_invalid_TSS
1273 errorentry segment_not_present do_segment_not_present
1274 zeroentry spurious_interrupt_bug do_spurious_interrupt_bug
1275 zeroentry coprocessor_error do_coprocessor_error
1276 errorentry alignment_check do_alignment_check
1277 zeroentry simd_coprocessor_error do_simd_coprocessor_error
1280 /* Reload gs selector with exception handling */
1281 /* edi: new selector */
1282 ENTRY(native_load_gs_index)
1285 DISABLE_INTERRUPTS(CLBR_ANY & ~CLBR_RDI)
1289 2: mfence /* workaround */
1294 END(native_load_gs_index)
1296 _ASM_EXTABLE(gs_change,bad_gs)
1297 .section .fixup,"ax"
1298 /* running with kernelgs */
1300 SWAPGS /* switch back to user gs */
1306 ENTRY(kernel_thread_helper)
1307 pushq $0 # fake return address
1310 * Here we are in the child and the registers are set as they were
1311 * at kernel_thread() invocation in the parent.
1317 ud2 # padding for call trace
1319 END(kernel_thread_helper)
1322 * execve(). This function needs to use IRET, not SYSRET, to set up all state properly.
1324 * C extern interface:
1325 * extern long execve(const char *name, char **argv, char **envp)
1327 * asm input arguments:
1328 * rdi: name, rsi: argv, rdx: envp
1330 * We want to fallback into:
1331 * extern long sys_execve(const char *name, char **argv,char **envp, struct pt_regs *regs)
1333 * do_sys_execve asm fallback arguments:
1334 * rdi: name, rsi: argv, rdx: envp, rcx: fake frame on the stack
1336 ENTRY(kernel_execve)
1342 movq %rax, RAX(%rsp)
1345 je int_ret_from_sys_call
1352 /* Call softirq on interrupt stack. Interrupts are off. */
1356 CFI_REL_OFFSET rbp,0
1358 CFI_DEF_CFA_REGISTER rbp
1359 incl PER_CPU_VAR(irq_count)
1360 cmove PER_CPU_VAR(irq_stack_ptr),%rsp
1361 push %rbp # backlink for old unwinder
1365 CFI_DEF_CFA_REGISTER rsp
1366 CFI_ADJUST_CFA_OFFSET -8
1367 decl PER_CPU_VAR(irq_count)
1373 zeroentry xen_hypervisor_callback xen_do_hypervisor_callback
1376 * A note on the "critical region" in our callback handler.
1377 * We want to avoid stacking callback handlers due to events occurring
1378 * during handling of the last event. To do this, we keep events disabled
1379 * until we've done all processing. HOWEVER, we must enable events before
1380 * popping the stack frame (can't be done atomically) and so it would still
1381 * be possible to get enough handler activations to overflow the stack.
1382 * Although unlikely, bugs of that kind are hard to track down, so we'd
1383 * like to avoid the possibility.
1384 * So, on entry to the handler we detect whether we interrupted an
1385 * existing activation in its critical region -- if so, we pop the current
1386 * activation and restart the handler using the previous one.
1388 ENTRY(xen_do_hypervisor_callback) # do_hypervisor_callback(struct *pt_regs)
1391 * Since we don't modify %rdi, evtchn_do_upall(struct *pt_regs) will
1392 * see the correct pointer to the pt_regs
1394 movq %rdi, %rsp # we don't return, adjust the stack frame
1397 11: incl PER_CPU_VAR(irq_count)
1399 CFI_DEF_CFA_REGISTER rbp
1400 cmovzq PER_CPU_VAR(irq_stack_ptr),%rsp
1401 pushq %rbp # backlink for old unwinder
1402 call xen_evtchn_do_upcall
1404 CFI_DEF_CFA_REGISTER rsp
1405 decl PER_CPU_VAR(irq_count)
1408 END(xen_do_hypervisor_callback)
1411 * Hypervisor uses this for application faults while it executes.
1412 * We get here for two reasons:
1413 * 1. Fault while reloading DS, ES, FS or GS
1414 * 2. Fault while executing IRET
1415 * Category 1 we do not need to fix up as Xen has already reloaded all segment
1416 * registers that could be reloaded and zeroed the others.
1417 * Category 2 we fix up by killing the current process. We cannot use the
1418 * normal Linux return path in this case because if we use the IRET hypercall
1419 * to pop the stack frame we end up in an infinite loop of failsafe callbacks.
1420 * We distinguish between categories by comparing each saved segment register
1421 * with its current contents: any discrepancy means we in category 1.
1423 ENTRY(xen_failsafe_callback)
1425 /*CFI_REL_OFFSET gs,GS*/
1426 /*CFI_REL_OFFSET fs,FS*/
1427 /*CFI_REL_OFFSET es,ES*/
1428 /*CFI_REL_OFFSET ds,DS*/
1429 CFI_REL_OFFSET r11,8
1430 CFI_REL_OFFSET rcx,0
1444 /* All segments match their saved values => Category 2 (Bad IRET). */
1450 CFI_ADJUST_CFA_OFFSET -0x30
1451 pushq_cfi $0 /* RIP */
1454 jmp general_protection
1456 1: /* Segment mismatch => Category 1 (Bad segment). Retry the IRET. */
1462 CFI_ADJUST_CFA_OFFSET -0x30
1467 END(xen_failsafe_callback)
1469 apicinterrupt XEN_HVM_EVTCHN_CALLBACK \
1470 xen_hvm_callback_vector xen_evtchn_do_upcall
1472 #endif /* CONFIG_XEN */
1475 * Some functions should be protected against kprobes
1477 .pushsection .kprobes.text, "ax"
1479 paranoidzeroentry_ist debug do_debug DEBUG_STACK
1480 paranoidzeroentry_ist int3 do_int3 DEBUG_STACK
1481 paranoiderrorentry stack_segment do_stack_segment
1483 zeroentry xen_debug do_debug
1484 zeroentry xen_int3 do_int3
1485 errorentry xen_stack_segment do_stack_segment
1487 errorentry general_protection do_general_protection
1488 errorentry page_fault do_page_fault
1489 #ifdef CONFIG_KVM_GUEST
1490 errorentry async_page_fault do_async_page_fault
1492 #ifdef CONFIG_X86_MCE
1493 paranoidzeroentry machine_check *machine_check_vector(%rip)
1497 * "Paranoid" exit path from exception stack.
1498 * Paranoid because this is used by NMIs and cannot take
1499 * any kernel state for granted.
1500 * We don't do kernel preemption checks here, because only
1501 * NMI should be common and it does not enable IRQs and
1502 * cannot get reschedule ticks.
1504 * "trace" is 0 for the NMI handler only, because irq-tracing
1505 * is fundamentally NMI-unsafe. (we cannot change the soft and
1506 * hard flags at once, atomically)
1509 /* ebx: no swapgs flag */
1510 ENTRY(paranoid_exit)
1512 DISABLE_INTERRUPTS(CLBR_NONE)
1513 TRACE_IRQS_OFF_DEBUG
1514 testl %ebx,%ebx /* swapgs needed? */
1515 jnz paranoid_restore
1517 jnz paranoid_userspace
1524 TRACE_IRQS_IRETQ_DEBUG 0
1528 GET_THREAD_INFO(%rcx)
1529 movl TI_flags(%rcx),%ebx
1530 andl $_TIF_WORK_MASK,%ebx
1532 movq %rsp,%rdi /* &pt_regs */
1534 movq %rax,%rsp /* switch stack for scheduling */
1535 testl $_TIF_NEED_RESCHED,%ebx
1536 jnz paranoid_schedule
1537 movl %ebx,%edx /* arg3: thread flags */
1539 ENABLE_INTERRUPTS(CLBR_NONE)
1540 xorl %esi,%esi /* arg2: oldset */
1541 movq %rsp,%rdi /* arg1: &pt_regs */
1542 call do_notify_resume
1543 DISABLE_INTERRUPTS(CLBR_NONE)
1545 jmp paranoid_userspace
1548 ENABLE_INTERRUPTS(CLBR_ANY)
1550 DISABLE_INTERRUPTS(CLBR_ANY)
1552 jmp paranoid_userspace
1557 * Exception entry point. This expects an error code/orig_rax on the stack.
1558 * returns in "no swapgs flag" in %ebx.
1562 CFI_ADJUST_CFA_OFFSET 15*8
1563 /* oldrax contains error code */
1582 je error_kernelspace
1590 * There are two places in the kernel that can potentially fault with
1591 * usergs. Handle them here. The exception handlers after iret run with
1592 * kernel gs again, so don't set the user space flag. B stepping K8s
1593 * sometimes report an truncated RIP for IRET exceptions returning to
1594 * compat mode. Check for these here too.
1598 leaq irq_return(%rip),%rcx
1599 cmpq %rcx,RIP+8(%rsp)
1601 movl %ecx,%eax /* zero extend */
1602 cmpq %rax,RIP+8(%rsp)
1604 cmpq $gs_change,RIP+8(%rsp)
1609 /* Fix truncated RIP */
1610 movq %rcx,RIP+8(%rsp)
1616 /* ebx: no swapgs flag (1: don't need swapgs, 0: need it) */
1621 DISABLE_INTERRUPTS(CLBR_NONE)
1623 GET_THREAD_INFO(%rcx)
1626 LOCKDEP_SYS_EXIT_IRQ
1627 movl TI_flags(%rcx),%edx
1628 movl $_TIF_WORK_MASK,%edi
1636 * Test if a given stack is an NMI stack or not.
1638 .macro test_in_nmi reg stack nmi_ret normal_ret
1641 subq $EXCEPTION_STKSZ, %\reg
1647 /* runs on exception stack */
1650 PARAVIRT_ADJUST_EXCEPTION_FRAME
1652 * We allow breakpoints in NMIs. If a breakpoint occurs, then
1653 * the iretq it performs will take us out of NMI context.
1654 * This means that we can have nested NMIs where the next
1655 * NMI is using the top of the stack of the previous NMI. We
1656 * can't let it execute because the nested NMI will corrupt the
1657 * stack of the previous NMI. NMI handlers are not re-entrant
1660 * To handle this case we do the following:
1661 * Check the a special location on the stack that contains
1662 * a variable that is set when NMIs are executing.
1663 * The interrupted task's stack is also checked to see if it
1665 * If the variable is not set and the stack is not the NMI
1667 * o Set the special variable on the stack
1668 * o Copy the interrupt frame into a "saved" location on the stack
1669 * o Copy the interrupt frame into a "copy" location on the stack
1670 * o Continue processing the NMI
1671 * If the variable is set or the previous stack is the NMI stack:
1672 * o Modify the "copy" location to jump to the repeate_nmi
1673 * o return back to the first NMI
1675 * Now on exit of the first NMI, we first clear the stack variable
1676 * The NMI stack will tell any nested NMIs at that point that it is
1677 * nested. Then we pop the stack normally with iret, and if there was
1678 * a nested NMI that updated the copy interrupt stack frame, a
1679 * jump will be made to the repeat_nmi code that will handle the second
1683 /* Use %rdx as out temp variable throughout */
1685 CFI_REL_OFFSET rdx, 0
1688 * If %cs was not the kernel segment, then the NMI triggered in user
1689 * space, which means it is definitely not nested.
1691 cmpl $__KERNEL_CS, 16(%rsp)
1695 * Check the special variable on the stack to see if NMIs are
1702 * Now test if the previous stack was an NMI stack.
1703 * We need the double check. We check the NMI stack to satisfy the
1704 * race when the first NMI clears the variable before returning.
1705 * We check the variable because the first NMI could be in a
1706 * breakpoint routine using a breakpoint stack.
1709 test_in_nmi rdx, 4*8(%rsp), nested_nmi, first_nmi
1714 * Do nothing if we interrupted the fixup in repeat_nmi.
1715 * It's about to repeat the NMI handler, so we are fine
1716 * with ignoring this one.
1718 movq $repeat_nmi, %rdx
1721 movq $end_repeat_nmi, %rdx
1726 /* Set up the interrupted NMIs stack to jump to repeat_nmi */
1727 leaq -6*8(%rsp), %rdx
1729 CFI_ADJUST_CFA_OFFSET 6*8
1730 pushq_cfi $__KERNEL_DS
1733 pushq_cfi $__KERNEL_CS
1734 pushq_cfi $repeat_nmi
1736 /* Put stack back */
1738 CFI_ADJUST_CFA_OFFSET -11*8
1744 /* No need to check faults here */
1750 * Because nested NMIs will use the pushed location that we
1751 * stored in rdx, we must keep that space available.
1752 * Here's what our stack frame will look like:
1753 * +-------------------------+
1755 * | original Return RSP |
1756 * | original RFLAGS |
1759 * +-------------------------+
1760 * | temp storage for rdx |
1761 * +-------------------------+
1762 * | NMI executing variable |
1763 * +-------------------------+
1765 * | Saved Return RSP |
1769 * +-------------------------+
1771 * | copied Return RSP |
1775 * +-------------------------+
1777 * +-------------------------+
1779 * The saved stack frame is used to fix up the copied stack frame
1780 * that a nested NMI may change to make the interrupted NMI iret jump
1781 * to the repeat_nmi. The original stack frame and the temp storage
1782 * is also used by nested NMIs and can not be trusted on exit.
1784 /* Do not pop rdx, nested NMIs will corrupt that part of the stack */
1788 /* Set the NMI executing variable on the stack. */
1791 /* Copy the stack frame to the Saved frame */
1795 CFI_DEF_CFA_OFFSET SS+8-RIP
1797 /* Everything up to here is safe from nested NMIs */
1800 * If there was a nested NMI, the first NMI's iret will return
1801 * here. But NMIs are still enabled and we can take another
1802 * nested NMI. The nested NMI checks the interrupted RIP to see
1803 * if it is between repeat_nmi and end_repeat_nmi, and if so
1804 * it will just return, as we are about to repeat an NMI anyway.
1805 * This makes it safe to copy to the stack frame that a nested
1810 * Update the stack variable to say we are still in NMI (the update
1811 * is benign for the non-repeat case, where 1 was pushed just above
1812 * to this very stack slot).
1816 /* Make another copy, this one may be modified by nested NMIs */
1820 CFI_DEF_CFA_OFFSET SS+8-RIP
1824 * Everything below this point can be preempted by a nested
1825 * NMI if the first NMI took an exception and reset our iret stack
1826 * so that we repeat another NMI.
1828 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1829 subq $ORIG_RAX-R15, %rsp
1830 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1832 * Use save_paranoid to handle SWAPGS, but no need to use paranoid_exit
1833 * as we should not be calling schedule in NMI context.
1834 * Even with normal interrupts enabled. An NMI should not be
1835 * setting NEED_RESCHED or anything that normal interrupts and
1836 * exceptions might do.
1840 /* paranoidentry do_nmi, 0; without TRACE_IRQS_OFF */
1844 testl %ebx,%ebx /* swapgs needed? */
1850 /* Clear the NMI executing stack variable */
1856 ENTRY(ignore_sysret)
1864 * End of kprobes section