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>
61 #include <linux/err.h>
63 /* Avoid __ASSEMBLER__'ifying <linux/audit.h> just for this. */
64 #include <linux/elf-em.h>
65 #define AUDIT_ARCH_X86_64 (EM_X86_64|__AUDIT_ARCH_64BIT|__AUDIT_ARCH_LE)
66 #define __AUDIT_ARCH_64BIT 0x80000000
67 #define __AUDIT_ARCH_LE 0x40000000
70 .section .entry.text, "ax"
72 #ifdef CONFIG_FUNCTION_TRACER
74 #ifdef CC_USING_FENTRY
75 # define function_hook __fentry__
77 # define function_hook mcount
80 #ifdef CONFIG_DYNAMIC_FTRACE
86 /* skip is set if stack has been adjusted */
87 .macro ftrace_caller_setup skip=0
88 MCOUNT_SAVE_FRAME \skip
90 /* Load the ftrace_ops into the 3rd parameter */
91 leaq function_trace_op, %rdx
93 /* Load ip into the first parameter */
95 subq $MCOUNT_INSN_SIZE, %rdi
96 /* Load the parent_ip into the second parameter */
97 #ifdef CC_USING_FENTRY
98 movq SS+16(%rsp), %rsi
105 /* Check if tracing was disabled (quick check) */
106 cmpl $0, function_trace_stop
110 /* regs go into 4th parameter (but make it NULL) */
119 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
120 GLOBAL(ftrace_graph_call)
128 ENTRY(ftrace_regs_caller)
129 /* Save the current flags before compare (in SS location)*/
132 /* Check if tracing was disabled (quick check) */
133 cmpl $0, function_trace_stop
134 jne ftrace_restore_flags
136 /* skip=8 to skip flags saved in SS */
137 ftrace_caller_setup 8
139 /* Save the rest of pt_regs */
148 /* Copy saved flags */
150 movq %rcx, EFLAGS(%rsp)
151 /* Kernel segments */
152 movq $__KERNEL_DS, %rcx
154 movq $__KERNEL_CS, %rcx
156 /* Stack - skipping return address */
157 leaq SS+16(%rsp), %rcx
160 /* regs go into 4th parameter */
163 GLOBAL(ftrace_regs_call)
166 /* Copy flags back to SS, to restore them */
167 movq EFLAGS(%rsp), %rax
170 /* Handlers can change the RIP */
172 movq %rax, SS+8(%rsp)
174 /* restore the rest of pt_regs */
183 /* skip=8 to skip flags saved in SS */
184 MCOUNT_RESTORE_FRAME 8
190 ftrace_restore_flags:
194 END(ftrace_regs_caller)
197 #else /* ! CONFIG_DYNAMIC_FTRACE */
200 cmpl $0, function_trace_stop
203 cmpq $ftrace_stub, ftrace_trace_function
206 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
207 cmpq $ftrace_stub, ftrace_graph_return
208 jnz ftrace_graph_caller
210 cmpq $ftrace_graph_entry_stub, ftrace_graph_entry
211 jnz ftrace_graph_caller
221 #ifdef CC_USING_FENTRY
222 movq SS+16(%rsp), %rsi
226 subq $MCOUNT_INSN_SIZE, %rdi
228 call *ftrace_trace_function
234 #endif /* CONFIG_DYNAMIC_FTRACE */
235 #endif /* CONFIG_FUNCTION_TRACER */
237 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
238 ENTRY(ftrace_graph_caller)
241 #ifdef CC_USING_FENTRY
242 leaq SS+16(%rsp), %rdi
243 movq $0, %rdx /* No framepointers needed */
249 subq $MCOUNT_INSN_SIZE, %rsi
251 call prepare_ftrace_return
256 END(ftrace_graph_caller)
258 GLOBAL(return_to_handler)
261 /* Save the return values */
266 call ftrace_return_to_handler
276 #ifndef CONFIG_PREEMPT
277 #define retint_kernel retint_restore_args
280 #ifdef CONFIG_PARAVIRT
281 ENTRY(native_usergs_sysret64)
284 ENDPROC(native_usergs_sysret64)
285 #endif /* CONFIG_PARAVIRT */
288 .macro TRACE_IRQS_IRETQ offset=ARGOFFSET
289 #ifdef CONFIG_TRACE_IRQFLAGS
290 bt $9,EFLAGS-\offset(%rsp) /* interrupts off? */
298 * When dynamic function tracer is enabled it will add a breakpoint
299 * to all locations that it is about to modify, sync CPUs, update
300 * all the code, sync CPUs, then remove the breakpoints. In this time
301 * if lockdep is enabled, it might jump back into the debug handler
302 * outside the updating of the IST protection. (TRACE_IRQS_ON/OFF).
304 * We need to change the IDT table before calling TRACE_IRQS_ON/OFF to
305 * make sure the stack pointer does not get reset back to the top
306 * of the debug stack, and instead just reuses the current stack.
308 #if defined(CONFIG_DYNAMIC_FTRACE) && defined(CONFIG_TRACE_IRQFLAGS)
310 .macro TRACE_IRQS_OFF_DEBUG
311 call debug_stack_set_zero
313 call debug_stack_reset
316 .macro TRACE_IRQS_ON_DEBUG
317 call debug_stack_set_zero
319 call debug_stack_reset
322 .macro TRACE_IRQS_IRETQ_DEBUG offset=ARGOFFSET
323 bt $9,EFLAGS-\offset(%rsp) /* interrupts off? */
330 # define TRACE_IRQS_OFF_DEBUG TRACE_IRQS_OFF
331 # define TRACE_IRQS_ON_DEBUG TRACE_IRQS_ON
332 # define TRACE_IRQS_IRETQ_DEBUG TRACE_IRQS_IRETQ
336 * C code is not supposed to know about undefined top of stack. Every time
337 * a C function with an pt_regs argument is called from the SYSCALL based
338 * fast path FIXUP_TOP_OF_STACK is needed.
339 * RESTORE_TOP_OF_STACK syncs the syscall state after any possible ptregs
343 /* %rsp:at FRAMEEND */
344 .macro FIXUP_TOP_OF_STACK tmp offset=0
345 movq PER_CPU_VAR(old_rsp),\tmp
346 movq \tmp,RSP+\offset(%rsp)
347 movq $__USER_DS,SS+\offset(%rsp)
348 movq $__USER_CS,CS+\offset(%rsp)
349 movq $-1,RCX+\offset(%rsp)
350 movq R11+\offset(%rsp),\tmp /* get eflags */
351 movq \tmp,EFLAGS+\offset(%rsp)
354 .macro RESTORE_TOP_OF_STACK tmp offset=0
355 movq RSP+\offset(%rsp),\tmp
356 movq \tmp,PER_CPU_VAR(old_rsp)
357 movq EFLAGS+\offset(%rsp),\tmp
358 movq \tmp,R11+\offset(%rsp)
361 .macro FAKE_STACK_FRAME child_rip
362 /* push in order ss, rsp, eflags, cs, rip */
364 pushq_cfi $__KERNEL_DS /* ss */
365 /*CFI_REL_OFFSET ss,0*/
366 pushq_cfi %rax /* rsp */
368 pushq_cfi $(X86_EFLAGS_IF|X86_EFLAGS_BIT1) /* eflags - interrupts on */
369 /*CFI_REL_OFFSET rflags,0*/
370 pushq_cfi $__KERNEL_CS /* cs */
371 /*CFI_REL_OFFSET cs,0*/
372 pushq_cfi \child_rip /* rip */
374 pushq_cfi %rax /* orig rax */
377 .macro UNFAKE_STACK_FRAME
379 CFI_ADJUST_CFA_OFFSET -(6*8)
383 * initial frame state for interrupts (and exceptions without error code)
385 .macro EMPTY_FRAME start=1 offset=0
389 CFI_DEF_CFA rsp,8+\offset
391 CFI_DEF_CFA_OFFSET 8+\offset
396 * initial frame state for interrupts (and exceptions without error code)
398 .macro INTR_FRAME start=1 offset=0
399 EMPTY_FRAME \start, SS+8+\offset-RIP
400 /*CFI_REL_OFFSET ss, SS+\offset-RIP*/
401 CFI_REL_OFFSET rsp, RSP+\offset-RIP
402 /*CFI_REL_OFFSET rflags, EFLAGS+\offset-RIP*/
403 /*CFI_REL_OFFSET cs, CS+\offset-RIP*/
404 CFI_REL_OFFSET rip, RIP+\offset-RIP
408 * initial frame state for exceptions with error code (and interrupts
409 * with vector already pushed)
411 .macro XCPT_FRAME start=1 offset=0
412 INTR_FRAME \start, RIP+\offset-ORIG_RAX
413 /*CFI_REL_OFFSET orig_rax, ORIG_RAX-ORIG_RAX*/
417 * frame that enables calling into C.
419 .macro PARTIAL_FRAME start=1 offset=0
420 XCPT_FRAME \start, ORIG_RAX+\offset-ARGOFFSET
421 CFI_REL_OFFSET rdi, RDI+\offset-ARGOFFSET
422 CFI_REL_OFFSET rsi, RSI+\offset-ARGOFFSET
423 CFI_REL_OFFSET rdx, RDX+\offset-ARGOFFSET
424 CFI_REL_OFFSET rcx, RCX+\offset-ARGOFFSET
425 CFI_REL_OFFSET rax, RAX+\offset-ARGOFFSET
426 CFI_REL_OFFSET r8, R8+\offset-ARGOFFSET
427 CFI_REL_OFFSET r9, R9+\offset-ARGOFFSET
428 CFI_REL_OFFSET r10, R10+\offset-ARGOFFSET
429 CFI_REL_OFFSET r11, R11+\offset-ARGOFFSET
433 * frame that enables passing a complete pt_regs to a C function.
435 .macro DEFAULT_FRAME start=1 offset=0
436 PARTIAL_FRAME \start, R11+\offset-R15
437 CFI_REL_OFFSET rbx, RBX+\offset
438 CFI_REL_OFFSET rbp, RBP+\offset
439 CFI_REL_OFFSET r12, R12+\offset
440 CFI_REL_OFFSET r13, R13+\offset
441 CFI_REL_OFFSET r14, R14+\offset
442 CFI_REL_OFFSET r15, R15+\offset
445 /* save partial stack frame */
448 /* start from rbp in pt_regs and jump over */
449 movq_cfi rdi, (RDI-RBP)
450 movq_cfi rsi, (RSI-RBP)
451 movq_cfi rdx, (RDX-RBP)
452 movq_cfi rcx, (RCX-RBP)
453 movq_cfi rax, (RAX-RBP)
454 movq_cfi r8, (R8-RBP)
455 movq_cfi r9, (R9-RBP)
456 movq_cfi r10, (R10-RBP)
457 movq_cfi r11, (R11-RBP)
459 /* Save rbp so that we can unwind from get_irq_regs() */
462 /* Save previous stack value */
465 leaq -RBP(%rsp),%rdi /* arg1 for handler */
466 testl $3, CS-RBP(%rsi)
470 * irq_count is used to check if a CPU is already on an interrupt stack
471 * or not. While this is essentially redundant with preempt_count it is
472 * a little cheaper to use a separate counter in the PDA (short of
473 * moving irq_enter into assembly, which would be too much work)
475 1: incl PER_CPU_VAR(irq_count)
476 cmovzq PER_CPU_VAR(irq_stack_ptr),%rsp
477 CFI_DEF_CFA_REGISTER rsi
479 /* Store previous stack value */
481 CFI_ESCAPE 0x0f /* DW_CFA_def_cfa_expression */, 6, \
482 0x77 /* DW_OP_breg7 */, 0, \
483 0x06 /* DW_OP_deref */, \
484 0x08 /* DW_OP_const1u */, SS+8-RBP, \
485 0x22 /* DW_OP_plus */
486 /* We entered an interrupt context - irqs are off: */
491 PARTIAL_FRAME 1 (REST_SKIP+8)
492 movq 5*8+16(%rsp), %r11 /* save return address */
499 movq %r11, 8(%rsp) /* return address */
500 FIXUP_TOP_OF_STACK %r11, 16
505 /* save complete stack frame */
506 .pushsection .kprobes.text, "ax"
526 movl $MSR_GS_BASE,%ecx
529 js 1f /* negative -> in kernel */
538 * A newly forked process directly context switches into this address.
540 * rdi: prev task we switched from
545 LOCK ; btr $TIF_FORK,TI_flags(%r8)
548 popfq_cfi # reset kernel eflags
550 call schedule_tail # rdi: 'prev' task parameter
552 GET_THREAD_INFO(%rcx)
556 testl $3, CS-ARGOFFSET(%rsp) # from kernel_thread?
557 jz retint_restore_args
559 testl $_TIF_IA32, TI_flags(%rcx) # 32-bit compat task needs IRET
560 jnz int_ret_from_sys_call
562 RESTORE_TOP_OF_STACK %rdi, -ARGOFFSET
563 jmp ret_from_sys_call # go to the SYSRET fastpath
569 * System call entry. Up to 6 arguments in registers are supported.
571 * SYSCALL does not save anything on the stack and does not change the
572 * stack pointer. However, it does mask the flags register for us, so
573 * CLD and CLAC are not needed.
578 * rax system call number
580 * rcx return address for syscall/sysret, C arg3
583 * r10 arg3 (--> moved to rcx for C)
586 * r11 eflags for syscall/sysret, temporary for C
587 * r12-r15,rbp,rbx saved by C code, not touched.
589 * Interrupts are off on entry.
590 * Only called from user space.
592 * XXX if we had a free scratch register we could save the RSP into the stack frame
593 * and report it properly in ps. Unfortunately we haven't.
595 * When user can change the frames always force IRET. That is because
596 * it deals with uncanonical addresses better. SYSRET has trouble
597 * with them due to bugs in both AMD and Intel CPUs.
603 CFI_DEF_CFA rsp,KERNEL_STACK_OFFSET
605 /*CFI_REGISTER rflags,r11*/
608 * A hypervisor implementation might want to use a label
609 * after the swapgs, so that it can do the swapgs
610 * for the guest and jump here on syscall.
612 GLOBAL(system_call_after_swapgs)
614 movq %rsp,PER_CPU_VAR(old_rsp)
615 movq PER_CPU_VAR(kernel_stack),%rsp
617 * No need to follow this irqs off/on section - it's straight
620 ENABLE_INTERRUPTS(CLBR_NONE)
622 movq %rax,ORIG_RAX-ARGOFFSET(%rsp)
623 movq %rcx,RIP-ARGOFFSET(%rsp)
624 CFI_REL_OFFSET rip,RIP-ARGOFFSET
625 testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET)
627 system_call_fastpath:
628 #if __SYSCALL_MASK == ~0
629 cmpq $__NR_syscall_max,%rax
631 andl $__SYSCALL_MASK,%eax
632 cmpl $__NR_syscall_max,%eax
636 call *sys_call_table(,%rax,8) # XXX: rip relative
637 movq %rax,RAX-ARGOFFSET(%rsp)
639 * Syscall return path ending with SYSRET (fast path)
640 * Has incomplete stack frame and undefined top of stack.
643 movl $_TIF_ALLWORK_MASK,%edi
647 DISABLE_INTERRUPTS(CLBR_NONE)
649 movl TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET),%edx
654 * sysretq will re-enable interrupts:
657 movq RIP-ARGOFFSET(%rsp),%rcx
659 RESTORE_ARGS 1,-ARG_SKIP,0
660 /*CFI_REGISTER rflags,r11*/
661 movq PER_CPU_VAR(old_rsp), %rsp
665 /* Handle reschedules */
666 /* edx: work, edi: workmask */
668 bt $TIF_NEED_RESCHED,%edx
671 ENABLE_INTERRUPTS(CLBR_NONE)
677 /* Handle a signal */
680 ENABLE_INTERRUPTS(CLBR_NONE)
681 #ifdef CONFIG_AUDITSYSCALL
682 bt $TIF_SYSCALL_AUDIT,%edx
686 * We have a signal, or exit tracing or single-step.
687 * These all wind up with the iret return path anyway,
688 * so just join that path right now.
690 FIXUP_TOP_OF_STACK %r11, -ARGOFFSET
691 jmp int_check_syscall_exit_work
694 movq $-ENOSYS,RAX-ARGOFFSET(%rsp)
695 jmp ret_from_sys_call
697 #ifdef CONFIG_AUDITSYSCALL
699 * Fast path for syscall audit without full syscall trace.
700 * We just call __audit_syscall_entry() directly, and then
701 * jump back to the normal fast path.
704 movq %r10,%r9 /* 6th arg: 4th syscall arg */
705 movq %rdx,%r8 /* 5th arg: 3rd syscall arg */
706 movq %rsi,%rcx /* 4th arg: 2nd syscall arg */
707 movq %rdi,%rdx /* 3rd arg: 1st syscall arg */
708 movq %rax,%rsi /* 2nd arg: syscall number */
709 movl $AUDIT_ARCH_X86_64,%edi /* 1st arg: audit arch */
710 call __audit_syscall_entry
711 LOAD_ARGS 0 /* reload call-clobbered registers */
712 jmp system_call_fastpath
715 * Return fast path for syscall audit. Call __audit_syscall_exit()
716 * directly and then jump back to the fast path with TIF_SYSCALL_AUDIT
720 movq RAX-ARGOFFSET(%rsp),%rsi /* second arg, syscall return value */
721 cmpq $-MAX_ERRNO,%rsi /* is it < -MAX_ERRNO? */
722 setbe %al /* 1 if so, 0 if not */
723 movzbl %al,%edi /* zero-extend that into %edi */
724 call __audit_syscall_exit
725 movl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT),%edi
727 #endif /* CONFIG_AUDITSYSCALL */
729 /* Do syscall tracing */
731 #ifdef CONFIG_AUDITSYSCALL
732 testl $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET)
736 movq $-ENOSYS,RAX(%rsp) /* ptrace can change this for a bad syscall */
737 FIXUP_TOP_OF_STACK %rdi
739 call syscall_trace_enter
741 * Reload arg registers from stack in case ptrace changed them.
742 * We don't reload %rax because syscall_trace_enter() returned
743 * the value it wants us to use in the table lookup.
745 LOAD_ARGS ARGOFFSET, 1
747 #if __SYSCALL_MASK == ~0
748 cmpq $__NR_syscall_max,%rax
750 andl $__SYSCALL_MASK,%eax
751 cmpl $__NR_syscall_max,%eax
753 ja int_ret_from_sys_call /* RAX(%rsp) set to -ENOSYS above */
754 movq %r10,%rcx /* fixup for C */
755 call *sys_call_table(,%rax,8)
756 movq %rax,RAX-ARGOFFSET(%rsp)
757 /* Use IRET because user could have changed frame */
760 * Syscall return path ending with IRET.
761 * Has correct top of stack, but partial stack frame.
763 GLOBAL(int_ret_from_sys_call)
764 DISABLE_INTERRUPTS(CLBR_NONE)
766 movl $_TIF_ALLWORK_MASK,%edi
767 /* edi: mask to check */
768 GLOBAL(int_with_check)
770 GET_THREAD_INFO(%rcx)
771 movl TI_flags(%rcx),%edx
774 andl $~TS_COMPAT,TI_status(%rcx)
777 /* Either reschedule or signal or syscall exit tracking needed. */
778 /* First do a reschedule test. */
779 /* edx: work, edi: workmask */
781 bt $TIF_NEED_RESCHED,%edx
784 ENABLE_INTERRUPTS(CLBR_NONE)
788 DISABLE_INTERRUPTS(CLBR_NONE)
792 /* handle signals and tracing -- both require a full stack frame */
795 ENABLE_INTERRUPTS(CLBR_NONE)
796 int_check_syscall_exit_work:
798 /* Check for syscall exit trace */
799 testl $_TIF_WORK_SYSCALL_EXIT,%edx
802 leaq 8(%rsp),%rdi # &ptregs -> arg1
803 call syscall_trace_leave
805 andl $~(_TIF_WORK_SYSCALL_EXIT|_TIF_SYSCALL_EMU),%edi
809 testl $_TIF_DO_NOTIFY_MASK,%edx
811 movq %rsp,%rdi # &ptregs -> arg1
812 xorl %esi,%esi # oldset -> arg2
813 call do_notify_resume
814 1: movl $_TIF_WORK_MASK,%edi
817 DISABLE_INTERRUPTS(CLBR_NONE)
824 * Certain special system calls that need to save a complete full stack frame.
826 .macro PTREGSCALL label,func,arg
828 PARTIAL_FRAME 1 8 /* offset 8: return address */
829 subq $REST_SKIP, %rsp
830 CFI_ADJUST_CFA_OFFSET REST_SKIP
832 DEFAULT_FRAME 0 8 /* offset 8: return address */
833 leaq 8(%rsp), \arg /* pt_regs pointer */
835 jmp ptregscall_common
840 PTREGSCALL stub_clone, sys_clone, %r8
841 PTREGSCALL stub_fork, sys_fork, %rdi
842 PTREGSCALL stub_vfork, sys_vfork, %rdi
843 PTREGSCALL stub_sigaltstack, sys_sigaltstack, %rdx
844 PTREGSCALL stub_iopl, sys_iopl, %rsi
846 ENTRY(ptregscall_common)
847 DEFAULT_FRAME 1 8 /* offset 8: return address */
848 RESTORE_TOP_OF_STACK %r11, 8
849 movq_cfi_restore R15+8, r15
850 movq_cfi_restore R14+8, r14
851 movq_cfi_restore R13+8, r13
852 movq_cfi_restore R12+8, r12
853 movq_cfi_restore RBP+8, rbp
854 movq_cfi_restore RBX+8, rbx
855 ret $REST_SKIP /* pop extended registers */
857 END(ptregscall_common)
864 FIXUP_TOP_OF_STACK %r11
867 RESTORE_TOP_OF_STACK %r11
870 jmp int_ret_from_sys_call
875 * sigreturn is special because it needs to restore all registers on return.
876 * This cannot be done with SYSRET, so use the IRET return path instead.
878 ENTRY(stub_rt_sigreturn)
884 FIXUP_TOP_OF_STACK %r11
885 call sys_rt_sigreturn
886 movq %rax,RAX(%rsp) # fixme, this could be done at the higher layer
888 jmp int_ret_from_sys_call
890 END(stub_rt_sigreturn)
892 #ifdef CONFIG_X86_X32_ABI
893 PTREGSCALL stub_x32_sigaltstack, sys32_sigaltstack, %rdx
895 ENTRY(stub_x32_rt_sigreturn)
901 FIXUP_TOP_OF_STACK %r11
902 call sys32_x32_rt_sigreturn
903 movq %rax,RAX(%rsp) # fixme, this could be done at the higher layer
905 jmp int_ret_from_sys_call
907 END(stub_x32_rt_sigreturn)
909 ENTRY(stub_x32_execve)
914 FIXUP_TOP_OF_STACK %r11
917 RESTORE_TOP_OF_STACK %r11
920 jmp int_ret_from_sys_call
927 * Build the entry stubs and pointer table with some assembler magic.
928 * We pack 7 stubs into a single 32-byte chunk, which will fit in a
929 * single cache line on all modern x86 implementations.
931 .section .init.rodata,"a"
935 .p2align CONFIG_X86_L1_CACHE_SHIFT
936 ENTRY(irq_entries_start)
938 vector=FIRST_EXTERNAL_VECTOR
939 .rept (NR_VECTORS-FIRST_EXTERNAL_VECTOR+6)/7
942 .if vector < NR_VECTORS
943 .if vector <> FIRST_EXTERNAL_VECTOR
944 CFI_ADJUST_CFA_OFFSET -8
946 1: pushq_cfi $(~vector+0x80) /* Note: always in signed byte range */
947 .if ((vector-FIRST_EXTERNAL_VECTOR)%7) <> 6
956 2: jmp common_interrupt
959 END(irq_entries_start)
966 * Interrupt entry/exit.
968 * Interrupt entry points save only callee clobbered registers in fast path.
970 * Entry runs with interrupts off.
973 /* 0(%rsp): ~(interrupt number) */
974 .macro interrupt func
975 /* reserve pt_regs for scratch regs and rbp */
976 subq $ORIG_RAX-RBP, %rsp
977 CFI_ADJUST_CFA_OFFSET ORIG_RAX-RBP
983 * Interrupt entry/exit should be protected against kprobes
985 .pushsection .kprobes.text, "ax"
987 * The interrupt stubs push (~vector+0x80) onto the stack and
988 * then jump to common_interrupt.
990 .p2align CONFIG_X86_L1_CACHE_SHIFT
994 addq $-0x80,(%rsp) /* Adjust vector to [-256,-1] range */
996 /* 0(%rsp): old_rsp-ARGOFFSET */
998 DISABLE_INTERRUPTS(CLBR_NONE)
1000 decl PER_CPU_VAR(irq_count)
1002 /* Restore saved previous stack */
1004 CFI_DEF_CFA rsi,SS+8-RBP /* reg/off reset after def_cfa_expr */
1005 leaq ARGOFFSET-RBP(%rsi), %rsp
1006 CFI_DEF_CFA_REGISTER rsp
1007 CFI_ADJUST_CFA_OFFSET RBP-ARGOFFSET
1010 GET_THREAD_INFO(%rcx)
1011 testl $3,CS-ARGOFFSET(%rsp)
1014 /* Interrupt came from user space */
1016 * Has a correct top of stack, but a partial stack frame
1017 * %rcx: thread info. Interrupts off.
1019 retint_with_reschedule:
1020 movl $_TIF_WORK_MASK,%edi
1022 LOCKDEP_SYS_EXIT_IRQ
1023 movl TI_flags(%rcx),%edx
1028 retint_swapgs: /* return to user-space */
1030 * The iretq could re-enable interrupts:
1032 DISABLE_INTERRUPTS(CLBR_ANY)
1037 retint_restore_args: /* return to kernel space */
1038 DISABLE_INTERRUPTS(CLBR_ANY)
1040 * The iretq could re-enable interrupts:
1048 _ASM_EXTABLE(irq_return, bad_iret)
1050 #ifdef CONFIG_PARAVIRT
1053 _ASM_EXTABLE(native_iret, bad_iret)
1056 .section .fixup,"ax"
1059 * The iret traps when the %cs or %ss being restored is bogus.
1060 * We've lost the original trap vector and error code.
1061 * #GPF is the most likely one to get for an invalid selector.
1062 * So pretend we completed the iret and took the #GPF in user mode.
1064 * We are now running with the kernel GS after exception recovery.
1065 * But error_entry expects us to have user GS to match the user %cs,
1071 jmp general_protection
1075 /* edi: workmask, edx: work */
1078 bt $TIF_NEED_RESCHED,%edx
1081 ENABLE_INTERRUPTS(CLBR_NONE)
1085 GET_THREAD_INFO(%rcx)
1086 DISABLE_INTERRUPTS(CLBR_NONE)
1091 testl $_TIF_DO_NOTIFY_MASK,%edx
1094 ENABLE_INTERRUPTS(CLBR_NONE)
1096 movq $-1,ORIG_RAX(%rsp)
1097 xorl %esi,%esi # oldset
1098 movq %rsp,%rdi # &pt_regs
1099 call do_notify_resume
1101 DISABLE_INTERRUPTS(CLBR_NONE)
1103 GET_THREAD_INFO(%rcx)
1104 jmp retint_with_reschedule
1106 #ifdef CONFIG_PREEMPT
1107 /* Returning to kernel space. Check if we need preemption */
1108 /* rcx: threadinfo. interrupts off. */
1109 ENTRY(retint_kernel)
1110 cmpl $0,TI_preempt_count(%rcx)
1111 jnz retint_restore_args
1112 bt $TIF_NEED_RESCHED,TI_flags(%rcx)
1113 jnc retint_restore_args
1114 bt $9,EFLAGS-ARGOFFSET(%rsp) /* interrupts off? */
1115 jnc retint_restore_args
1116 call preempt_schedule_irq
1121 END(common_interrupt)
1123 * End of kprobes section
1130 .macro apicinterrupt num sym do_sym
1143 apicinterrupt IRQ_MOVE_CLEANUP_VECTOR \
1144 irq_move_cleanup_interrupt smp_irq_move_cleanup_interrupt
1145 apicinterrupt REBOOT_VECTOR \
1146 reboot_interrupt smp_reboot_interrupt
1149 #ifdef CONFIG_X86_UV
1150 apicinterrupt UV_BAU_MESSAGE \
1151 uv_bau_message_intr1 uv_bau_message_interrupt
1153 apicinterrupt LOCAL_TIMER_VECTOR \
1154 apic_timer_interrupt smp_apic_timer_interrupt
1155 apicinterrupt X86_PLATFORM_IPI_VECTOR \
1156 x86_platform_ipi smp_x86_platform_ipi
1158 apicinterrupt THRESHOLD_APIC_VECTOR \
1159 threshold_interrupt smp_threshold_interrupt
1160 apicinterrupt THERMAL_APIC_VECTOR \
1161 thermal_interrupt smp_thermal_interrupt
1164 apicinterrupt CALL_FUNCTION_SINGLE_VECTOR \
1165 call_function_single_interrupt smp_call_function_single_interrupt
1166 apicinterrupt CALL_FUNCTION_VECTOR \
1167 call_function_interrupt smp_call_function_interrupt
1168 apicinterrupt RESCHEDULE_VECTOR \
1169 reschedule_interrupt smp_reschedule_interrupt
1172 apicinterrupt ERROR_APIC_VECTOR \
1173 error_interrupt smp_error_interrupt
1174 apicinterrupt SPURIOUS_APIC_VECTOR \
1175 spurious_interrupt smp_spurious_interrupt
1177 #ifdef CONFIG_IRQ_WORK
1178 apicinterrupt IRQ_WORK_VECTOR \
1179 irq_work_interrupt smp_irq_work_interrupt
1183 * Exception entry points.
1185 .macro zeroentry sym do_sym
1189 PARAVIRT_ADJUST_EXCEPTION_FRAME
1190 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1191 subq $ORIG_RAX-R15, %rsp
1192 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1195 movq %rsp,%rdi /* pt_regs pointer */
1196 xorl %esi,%esi /* no error code */
1198 jmp error_exit /* %ebx: no swapgs flag */
1203 .macro paranoidzeroentry sym do_sym
1207 PARAVIRT_ADJUST_EXCEPTION_FRAME
1208 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1209 subq $ORIG_RAX-R15, %rsp
1210 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1213 movq %rsp,%rdi /* pt_regs pointer */
1214 xorl %esi,%esi /* no error code */
1216 jmp paranoid_exit /* %ebx: no swapgs flag */
1221 #define INIT_TSS_IST(x) PER_CPU_VAR(init_tss) + (TSS_ist + ((x) - 1) * 8)
1222 .macro paranoidzeroentry_ist sym do_sym ist
1226 PARAVIRT_ADJUST_EXCEPTION_FRAME
1227 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1228 subq $ORIG_RAX-R15, %rsp
1229 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1231 TRACE_IRQS_OFF_DEBUG
1232 movq %rsp,%rdi /* pt_regs pointer */
1233 xorl %esi,%esi /* no error code */
1234 subq $EXCEPTION_STKSZ, INIT_TSS_IST(\ist)
1236 addq $EXCEPTION_STKSZ, INIT_TSS_IST(\ist)
1237 jmp paranoid_exit /* %ebx: no swapgs flag */
1242 .macro errorentry sym do_sym
1246 PARAVIRT_ADJUST_EXCEPTION_FRAME
1247 subq $ORIG_RAX-R15, %rsp
1248 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1251 movq %rsp,%rdi /* pt_regs pointer */
1252 movq ORIG_RAX(%rsp),%rsi /* get error code */
1253 movq $-1,ORIG_RAX(%rsp) /* no syscall to restart */
1255 jmp error_exit /* %ebx: no swapgs flag */
1260 /* error code is on the stack already */
1261 .macro paranoiderrorentry sym do_sym
1265 PARAVIRT_ADJUST_EXCEPTION_FRAME
1266 subq $ORIG_RAX-R15, %rsp
1267 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1271 movq %rsp,%rdi /* pt_regs pointer */
1272 movq ORIG_RAX(%rsp),%rsi /* get error code */
1273 movq $-1,ORIG_RAX(%rsp) /* no syscall to restart */
1275 jmp paranoid_exit /* %ebx: no swapgs flag */
1280 zeroentry divide_error do_divide_error
1281 zeroentry overflow do_overflow
1282 zeroentry bounds do_bounds
1283 zeroentry invalid_op do_invalid_op
1284 zeroentry device_not_available do_device_not_available
1285 paranoiderrorentry double_fault do_double_fault
1286 zeroentry coprocessor_segment_overrun do_coprocessor_segment_overrun
1287 errorentry invalid_TSS do_invalid_TSS
1288 errorentry segment_not_present do_segment_not_present
1289 zeroentry spurious_interrupt_bug do_spurious_interrupt_bug
1290 zeroentry coprocessor_error do_coprocessor_error
1291 errorentry alignment_check do_alignment_check
1292 zeroentry simd_coprocessor_error do_simd_coprocessor_error
1295 /* Reload gs selector with exception handling */
1296 /* edi: new selector */
1297 ENTRY(native_load_gs_index)
1300 DISABLE_INTERRUPTS(CLBR_ANY & ~CLBR_RDI)
1304 2: mfence /* workaround */
1309 END(native_load_gs_index)
1311 _ASM_EXTABLE(gs_change,bad_gs)
1312 .section .fixup,"ax"
1313 /* running with kernelgs */
1315 SWAPGS /* switch back to user gs */
1321 ENTRY(kernel_thread_helper)
1322 pushq $0 # fake return address
1325 * Here we are in the child and the registers are set as they were
1326 * at kernel_thread() invocation in the parent.
1332 ud2 # padding for call trace
1334 END(kernel_thread_helper)
1337 * execve(). This function needs to use IRET, not SYSRET, to set up all state properly.
1339 * C extern interface:
1340 * extern long execve(const char *name, char **argv, char **envp)
1342 * asm input arguments:
1343 * rdi: name, rsi: argv, rdx: envp
1345 * We want to fallback into:
1346 * extern long sys_execve(const char *name, char **argv,char **envp, struct pt_regs *regs)
1348 * do_sys_execve asm fallback arguments:
1349 * rdi: name, rsi: argv, rdx: envp, rcx: fake frame on the stack
1351 ENTRY(kernel_execve)
1357 movq %rax, RAX(%rsp)
1360 je int_ret_from_sys_call
1367 /* Call softirq on interrupt stack. Interrupts are off. */
1371 CFI_REL_OFFSET rbp,0
1373 CFI_DEF_CFA_REGISTER rbp
1374 incl PER_CPU_VAR(irq_count)
1375 cmove PER_CPU_VAR(irq_stack_ptr),%rsp
1376 push %rbp # backlink for old unwinder
1380 CFI_DEF_CFA_REGISTER rsp
1381 CFI_ADJUST_CFA_OFFSET -8
1382 decl PER_CPU_VAR(irq_count)
1388 zeroentry xen_hypervisor_callback xen_do_hypervisor_callback
1391 * A note on the "critical region" in our callback handler.
1392 * We want to avoid stacking callback handlers due to events occurring
1393 * during handling of the last event. To do this, we keep events disabled
1394 * until we've done all processing. HOWEVER, we must enable events before
1395 * popping the stack frame (can't be done atomically) and so it would still
1396 * be possible to get enough handler activations to overflow the stack.
1397 * Although unlikely, bugs of that kind are hard to track down, so we'd
1398 * like to avoid the possibility.
1399 * So, on entry to the handler we detect whether we interrupted an
1400 * existing activation in its critical region -- if so, we pop the current
1401 * activation and restart the handler using the previous one.
1403 ENTRY(xen_do_hypervisor_callback) # do_hypervisor_callback(struct *pt_regs)
1406 * Since we don't modify %rdi, evtchn_do_upall(struct *pt_regs) will
1407 * see the correct pointer to the pt_regs
1409 movq %rdi, %rsp # we don't return, adjust the stack frame
1412 11: incl PER_CPU_VAR(irq_count)
1414 CFI_DEF_CFA_REGISTER rbp
1415 cmovzq PER_CPU_VAR(irq_stack_ptr),%rsp
1416 pushq %rbp # backlink for old unwinder
1417 call xen_evtchn_do_upcall
1419 CFI_DEF_CFA_REGISTER rsp
1420 decl PER_CPU_VAR(irq_count)
1423 END(xen_do_hypervisor_callback)
1426 * Hypervisor uses this for application faults while it executes.
1427 * We get here for two reasons:
1428 * 1. Fault while reloading DS, ES, FS or GS
1429 * 2. Fault while executing IRET
1430 * Category 1 we do not need to fix up as Xen has already reloaded all segment
1431 * registers that could be reloaded and zeroed the others.
1432 * Category 2 we fix up by killing the current process. We cannot use the
1433 * normal Linux return path in this case because if we use the IRET hypercall
1434 * to pop the stack frame we end up in an infinite loop of failsafe callbacks.
1435 * We distinguish between categories by comparing each saved segment register
1436 * with its current contents: any discrepancy means we in category 1.
1438 ENTRY(xen_failsafe_callback)
1440 /*CFI_REL_OFFSET gs,GS*/
1441 /*CFI_REL_OFFSET fs,FS*/
1442 /*CFI_REL_OFFSET es,ES*/
1443 /*CFI_REL_OFFSET ds,DS*/
1444 CFI_REL_OFFSET r11,8
1445 CFI_REL_OFFSET rcx,0
1459 /* All segments match their saved values => Category 2 (Bad IRET). */
1465 CFI_ADJUST_CFA_OFFSET -0x30
1466 pushq_cfi $0 /* RIP */
1469 jmp general_protection
1471 1: /* Segment mismatch => Category 1 (Bad segment). Retry the IRET. */
1477 CFI_ADJUST_CFA_OFFSET -0x30
1482 END(xen_failsafe_callback)
1484 apicinterrupt XEN_HVM_EVTCHN_CALLBACK \
1485 xen_hvm_callback_vector xen_evtchn_do_upcall
1487 #endif /* CONFIG_XEN */
1490 * Some functions should be protected against kprobes
1492 .pushsection .kprobes.text, "ax"
1494 paranoidzeroentry_ist debug do_debug DEBUG_STACK
1495 paranoidzeroentry_ist int3 do_int3 DEBUG_STACK
1496 paranoiderrorentry stack_segment do_stack_segment
1498 zeroentry xen_debug do_debug
1499 zeroentry xen_int3 do_int3
1500 errorentry xen_stack_segment do_stack_segment
1502 errorentry general_protection do_general_protection
1503 errorentry page_fault do_page_fault
1504 #ifdef CONFIG_KVM_GUEST
1505 errorentry async_page_fault do_async_page_fault
1507 #ifdef CONFIG_X86_MCE
1508 paranoidzeroentry machine_check *machine_check_vector(%rip)
1512 * "Paranoid" exit path from exception stack.
1513 * Paranoid because this is used by NMIs and cannot take
1514 * any kernel state for granted.
1515 * We don't do kernel preemption checks here, because only
1516 * NMI should be common and it does not enable IRQs and
1517 * cannot get reschedule ticks.
1519 * "trace" is 0 for the NMI handler only, because irq-tracing
1520 * is fundamentally NMI-unsafe. (we cannot change the soft and
1521 * hard flags at once, atomically)
1524 /* ebx: no swapgs flag */
1525 ENTRY(paranoid_exit)
1527 DISABLE_INTERRUPTS(CLBR_NONE)
1528 TRACE_IRQS_OFF_DEBUG
1529 testl %ebx,%ebx /* swapgs needed? */
1530 jnz paranoid_restore
1532 jnz paranoid_userspace
1539 TRACE_IRQS_IRETQ_DEBUG 0
1543 GET_THREAD_INFO(%rcx)
1544 movl TI_flags(%rcx),%ebx
1545 andl $_TIF_WORK_MASK,%ebx
1547 movq %rsp,%rdi /* &pt_regs */
1549 movq %rax,%rsp /* switch stack for scheduling */
1550 testl $_TIF_NEED_RESCHED,%ebx
1551 jnz paranoid_schedule
1552 movl %ebx,%edx /* arg3: thread flags */
1554 ENABLE_INTERRUPTS(CLBR_NONE)
1555 xorl %esi,%esi /* arg2: oldset */
1556 movq %rsp,%rdi /* arg1: &pt_regs */
1557 call do_notify_resume
1558 DISABLE_INTERRUPTS(CLBR_NONE)
1560 jmp paranoid_userspace
1563 ENABLE_INTERRUPTS(CLBR_ANY)
1565 DISABLE_INTERRUPTS(CLBR_ANY)
1567 jmp paranoid_userspace
1572 * Exception entry point. This expects an error code/orig_rax on the stack.
1573 * returns in "no swapgs flag" in %ebx.
1577 CFI_ADJUST_CFA_OFFSET 15*8
1578 /* oldrax contains error code */
1597 je error_kernelspace
1605 * There are two places in the kernel that can potentially fault with
1606 * usergs. Handle them here. The exception handlers after iret run with
1607 * kernel gs again, so don't set the user space flag. B stepping K8s
1608 * sometimes report an truncated RIP for IRET exceptions returning to
1609 * compat mode. Check for these here too.
1613 leaq irq_return(%rip),%rcx
1614 cmpq %rcx,RIP+8(%rsp)
1616 movl %ecx,%eax /* zero extend */
1617 cmpq %rax,RIP+8(%rsp)
1619 cmpq $gs_change,RIP+8(%rsp)
1624 /* Fix truncated RIP */
1625 movq %rcx,RIP+8(%rsp)
1631 /* ebx: no swapgs flag (1: don't need swapgs, 0: need it) */
1636 DISABLE_INTERRUPTS(CLBR_NONE)
1638 GET_THREAD_INFO(%rcx)
1641 LOCKDEP_SYS_EXIT_IRQ
1642 movl TI_flags(%rcx),%edx
1643 movl $_TIF_WORK_MASK,%edi
1651 * Test if a given stack is an NMI stack or not.
1653 .macro test_in_nmi reg stack nmi_ret normal_ret
1656 subq $EXCEPTION_STKSZ, %\reg
1662 /* runs on exception stack */
1665 PARAVIRT_ADJUST_EXCEPTION_FRAME
1667 * We allow breakpoints in NMIs. If a breakpoint occurs, then
1668 * the iretq it performs will take us out of NMI context.
1669 * This means that we can have nested NMIs where the next
1670 * NMI is using the top of the stack of the previous NMI. We
1671 * can't let it execute because the nested NMI will corrupt the
1672 * stack of the previous NMI. NMI handlers are not re-entrant
1675 * To handle this case we do the following:
1676 * Check the a special location on the stack that contains
1677 * a variable that is set when NMIs are executing.
1678 * The interrupted task's stack is also checked to see if it
1680 * If the variable is not set and the stack is not the NMI
1682 * o Set the special variable on the stack
1683 * o Copy the interrupt frame into a "saved" location on the stack
1684 * o Copy the interrupt frame into a "copy" location on the stack
1685 * o Continue processing the NMI
1686 * If the variable is set or the previous stack is the NMI stack:
1687 * o Modify the "copy" location to jump to the repeate_nmi
1688 * o return back to the first NMI
1690 * Now on exit of the first NMI, we first clear the stack variable
1691 * The NMI stack will tell any nested NMIs at that point that it is
1692 * nested. Then we pop the stack normally with iret, and if there was
1693 * a nested NMI that updated the copy interrupt stack frame, a
1694 * jump will be made to the repeat_nmi code that will handle the second
1698 /* Use %rdx as out temp variable throughout */
1700 CFI_REL_OFFSET rdx, 0
1703 * If %cs was not the kernel segment, then the NMI triggered in user
1704 * space, which means it is definitely not nested.
1706 cmpl $__KERNEL_CS, 16(%rsp)
1710 * Check the special variable on the stack to see if NMIs are
1717 * Now test if the previous stack was an NMI stack.
1718 * We need the double check. We check the NMI stack to satisfy the
1719 * race when the first NMI clears the variable before returning.
1720 * We check the variable because the first NMI could be in a
1721 * breakpoint routine using a breakpoint stack.
1724 test_in_nmi rdx, 4*8(%rsp), nested_nmi, first_nmi
1729 * Do nothing if we interrupted the fixup in repeat_nmi.
1730 * It's about to repeat the NMI handler, so we are fine
1731 * with ignoring this one.
1733 movq $repeat_nmi, %rdx
1736 movq $end_repeat_nmi, %rdx
1741 /* Set up the interrupted NMIs stack to jump to repeat_nmi */
1742 leaq -6*8(%rsp), %rdx
1744 CFI_ADJUST_CFA_OFFSET 6*8
1745 pushq_cfi $__KERNEL_DS
1748 pushq_cfi $__KERNEL_CS
1749 pushq_cfi $repeat_nmi
1751 /* Put stack back */
1753 CFI_ADJUST_CFA_OFFSET -11*8
1759 /* No need to check faults here */
1765 * Because nested NMIs will use the pushed location that we
1766 * stored in rdx, we must keep that space available.
1767 * Here's what our stack frame will look like:
1768 * +-------------------------+
1770 * | original Return RSP |
1771 * | original RFLAGS |
1774 * +-------------------------+
1775 * | temp storage for rdx |
1776 * +-------------------------+
1777 * | NMI executing variable |
1778 * +-------------------------+
1780 * | Saved Return RSP |
1784 * +-------------------------+
1786 * | copied Return RSP |
1790 * +-------------------------+
1792 * +-------------------------+
1794 * The saved stack frame is used to fix up the copied stack frame
1795 * that a nested NMI may change to make the interrupted NMI iret jump
1796 * to the repeat_nmi. The original stack frame and the temp storage
1797 * is also used by nested NMIs and can not be trusted on exit.
1799 /* Do not pop rdx, nested NMIs will corrupt that part of the stack */
1803 /* Set the NMI executing variable on the stack. */
1806 /* Copy the stack frame to the Saved frame */
1810 CFI_DEF_CFA_OFFSET SS+8-RIP
1812 /* Everything up to here is safe from nested NMIs */
1815 * If there was a nested NMI, the first NMI's iret will return
1816 * here. But NMIs are still enabled and we can take another
1817 * nested NMI. The nested NMI checks the interrupted RIP to see
1818 * if it is between repeat_nmi and end_repeat_nmi, and if so
1819 * it will just return, as we are about to repeat an NMI anyway.
1820 * This makes it safe to copy to the stack frame that a nested
1825 * Update the stack variable to say we are still in NMI (the update
1826 * is benign for the non-repeat case, where 1 was pushed just above
1827 * to this very stack slot).
1831 /* Make another copy, this one may be modified by nested NMIs */
1835 CFI_DEF_CFA_OFFSET SS+8-RIP
1839 * Everything below this point can be preempted by a nested
1840 * NMI if the first NMI took an exception and reset our iret stack
1841 * so that we repeat another NMI.
1843 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1844 subq $ORIG_RAX-R15, %rsp
1845 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1847 * Use save_paranoid to handle SWAPGS, but no need to use paranoid_exit
1848 * as we should not be calling schedule in NMI context.
1849 * Even with normal interrupts enabled. An NMI should not be
1850 * setting NEED_RESCHED or anything that normal interrupts and
1851 * exceptions might do.
1857 * Save off the CR2 register. If we take a page fault in the NMI then
1858 * it could corrupt the CR2 value. If the NMI preempts a page fault
1859 * handler before it was able to read the CR2 register, and then the
1860 * NMI itself takes a page fault, the page fault that was preempted
1861 * will read the information from the NMI page fault and not the
1862 * origin fault. Save it off and restore it if it changes.
1863 * Use the r12 callee-saved register.
1867 /* paranoidentry do_nmi, 0; without TRACE_IRQS_OFF */
1872 /* Did the NMI take a page fault? Restore cr2 if it did */
1879 testl %ebx,%ebx /* swapgs needed? */
1885 /* Clear the NMI executing stack variable */
1891 ENTRY(ignore_sysret)
1899 * End of kprobes section