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
72 #ifdef CC_USING_FENTRY
73 # define function_hook __fentry__
75 # define function_hook mcount
78 #ifdef CONFIG_DYNAMIC_FTRACE
84 /* skip is set if stack has been adjusted */
85 .macro ftrace_caller_setup skip=0
86 MCOUNT_SAVE_FRAME \skip
88 /* Load the ftrace_ops into the 3rd parameter */
89 leaq function_trace_op, %rdx
91 /* Load ip into the first parameter */
93 subq $MCOUNT_INSN_SIZE, %rdi
94 /* Load the parent_ip into the second parameter */
95 #ifdef CC_USING_FENTRY
96 movq SS+16(%rsp), %rsi
103 /* Check if tracing was disabled (quick check) */
104 cmpl $0, function_trace_stop
108 /* regs go into 4th parameter (but make it NULL) */
117 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
118 GLOBAL(ftrace_graph_call)
126 ENTRY(ftrace_regs_caller)
127 /* Save the current flags before compare (in SS location)*/
130 /* Check if tracing was disabled (quick check) */
131 cmpl $0, function_trace_stop
132 jne ftrace_restore_flags
134 /* skip=8 to skip flags saved in SS */
135 ftrace_caller_setup 8
137 /* Save the rest of pt_regs */
146 /* Copy saved flags */
148 movq %rcx, EFLAGS(%rsp)
149 /* Kernel segments */
150 movq $__KERNEL_DS, %rcx
152 movq $__KERNEL_CS, %rcx
154 /* Stack - skipping return address */
155 leaq SS+16(%rsp), %rcx
158 /* regs go into 4th parameter */
161 GLOBAL(ftrace_regs_call)
164 /* Copy flags back to SS, to restore them */
165 movq EFLAGS(%rsp), %rax
168 /* restore the rest of pt_regs */
177 /* skip=8 to skip flags saved in SS */
178 MCOUNT_RESTORE_FRAME 8
184 ftrace_restore_flags:
188 END(ftrace_regs_caller)
191 #else /* ! CONFIG_DYNAMIC_FTRACE */
194 cmpl $0, function_trace_stop
197 cmpq $ftrace_stub, ftrace_trace_function
200 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
201 cmpq $ftrace_stub, ftrace_graph_return
202 jnz ftrace_graph_caller
204 cmpq $ftrace_graph_entry_stub, ftrace_graph_entry
205 jnz ftrace_graph_caller
215 #ifdef CC_USING_FENTRY
216 movq SS+16(%rsp), %rsi
220 subq $MCOUNT_INSN_SIZE, %rdi
222 call *ftrace_trace_function
228 #endif /* CONFIG_DYNAMIC_FTRACE */
229 #endif /* CONFIG_FUNCTION_TRACER */
231 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
232 ENTRY(ftrace_graph_caller)
235 #ifdef CC_USING_FENTRY
236 leaq SS+16(%rsp), %rdi
237 movq $0, %rdx /* No framepointers needed */
243 subq $MCOUNT_INSN_SIZE, %rsi
245 call prepare_ftrace_return
250 END(ftrace_graph_caller)
252 GLOBAL(return_to_handler)
255 /* Save the return values */
260 call ftrace_return_to_handler
270 #ifndef CONFIG_PREEMPT
271 #define retint_kernel retint_restore_args
274 #ifdef CONFIG_PARAVIRT
275 ENTRY(native_usergs_sysret64)
278 ENDPROC(native_usergs_sysret64)
279 #endif /* CONFIG_PARAVIRT */
282 .macro TRACE_IRQS_IRETQ offset=ARGOFFSET
283 #ifdef CONFIG_TRACE_IRQFLAGS
284 bt $9,EFLAGS-\offset(%rsp) /* interrupts off? */
292 * When dynamic function tracer is enabled it will add a breakpoint
293 * to all locations that it is about to modify, sync CPUs, update
294 * all the code, sync CPUs, then remove the breakpoints. In this time
295 * if lockdep is enabled, it might jump back into the debug handler
296 * outside the updating of the IST protection. (TRACE_IRQS_ON/OFF).
298 * We need to change the IDT table before calling TRACE_IRQS_ON/OFF to
299 * make sure the stack pointer does not get reset back to the top
300 * of the debug stack, and instead just reuses the current stack.
302 #if defined(CONFIG_DYNAMIC_FTRACE) && defined(CONFIG_TRACE_IRQFLAGS)
304 .macro TRACE_IRQS_OFF_DEBUG
305 call debug_stack_set_zero
307 call debug_stack_reset
310 .macro TRACE_IRQS_ON_DEBUG
311 call debug_stack_set_zero
313 call debug_stack_reset
316 .macro TRACE_IRQS_IRETQ_DEBUG offset=ARGOFFSET
317 bt $9,EFLAGS-\offset(%rsp) /* interrupts off? */
324 # define TRACE_IRQS_OFF_DEBUG TRACE_IRQS_OFF
325 # define TRACE_IRQS_ON_DEBUG TRACE_IRQS_ON
326 # define TRACE_IRQS_IRETQ_DEBUG TRACE_IRQS_IRETQ
330 * C code is not supposed to know about undefined top of stack. Every time
331 * a C function with an pt_regs argument is called from the SYSCALL based
332 * fast path FIXUP_TOP_OF_STACK is needed.
333 * RESTORE_TOP_OF_STACK syncs the syscall state after any possible ptregs
337 /* %rsp:at FRAMEEND */
338 .macro FIXUP_TOP_OF_STACK tmp offset=0
339 movq PER_CPU_VAR(old_rsp),\tmp
340 movq \tmp,RSP+\offset(%rsp)
341 movq $__USER_DS,SS+\offset(%rsp)
342 movq $__USER_CS,CS+\offset(%rsp)
343 movq $-1,RCX+\offset(%rsp)
344 movq R11+\offset(%rsp),\tmp /* get eflags */
345 movq \tmp,EFLAGS+\offset(%rsp)
348 .macro RESTORE_TOP_OF_STACK tmp offset=0
349 movq RSP+\offset(%rsp),\tmp
350 movq \tmp,PER_CPU_VAR(old_rsp)
351 movq EFLAGS+\offset(%rsp),\tmp
352 movq \tmp,R11+\offset(%rsp)
355 .macro FAKE_STACK_FRAME child_rip
356 /* push in order ss, rsp, eflags, cs, rip */
358 pushq_cfi $__KERNEL_DS /* ss */
359 /*CFI_REL_OFFSET ss,0*/
360 pushq_cfi %rax /* rsp */
362 pushq_cfi $(X86_EFLAGS_IF|X86_EFLAGS_BIT1) /* eflags - interrupts on */
363 /*CFI_REL_OFFSET rflags,0*/
364 pushq_cfi $__KERNEL_CS /* cs */
365 /*CFI_REL_OFFSET cs,0*/
366 pushq_cfi \child_rip /* rip */
368 pushq_cfi %rax /* orig rax */
371 .macro UNFAKE_STACK_FRAME
373 CFI_ADJUST_CFA_OFFSET -(6*8)
377 * initial frame state for interrupts (and exceptions without error code)
379 .macro EMPTY_FRAME start=1 offset=0
383 CFI_DEF_CFA rsp,8+\offset
385 CFI_DEF_CFA_OFFSET 8+\offset
390 * initial frame state for interrupts (and exceptions without error code)
392 .macro INTR_FRAME start=1 offset=0
393 EMPTY_FRAME \start, SS+8+\offset-RIP
394 /*CFI_REL_OFFSET ss, SS+\offset-RIP*/
395 CFI_REL_OFFSET rsp, RSP+\offset-RIP
396 /*CFI_REL_OFFSET rflags, EFLAGS+\offset-RIP*/
397 /*CFI_REL_OFFSET cs, CS+\offset-RIP*/
398 CFI_REL_OFFSET rip, RIP+\offset-RIP
402 * initial frame state for exceptions with error code (and interrupts
403 * with vector already pushed)
405 .macro XCPT_FRAME start=1 offset=0
406 INTR_FRAME \start, RIP+\offset-ORIG_RAX
407 /*CFI_REL_OFFSET orig_rax, ORIG_RAX-ORIG_RAX*/
411 * frame that enables calling into C.
413 .macro PARTIAL_FRAME start=1 offset=0
414 XCPT_FRAME \start, ORIG_RAX+\offset-ARGOFFSET
415 CFI_REL_OFFSET rdi, RDI+\offset-ARGOFFSET
416 CFI_REL_OFFSET rsi, RSI+\offset-ARGOFFSET
417 CFI_REL_OFFSET rdx, RDX+\offset-ARGOFFSET
418 CFI_REL_OFFSET rcx, RCX+\offset-ARGOFFSET
419 CFI_REL_OFFSET rax, RAX+\offset-ARGOFFSET
420 CFI_REL_OFFSET r8, R8+\offset-ARGOFFSET
421 CFI_REL_OFFSET r9, R9+\offset-ARGOFFSET
422 CFI_REL_OFFSET r10, R10+\offset-ARGOFFSET
423 CFI_REL_OFFSET r11, R11+\offset-ARGOFFSET
427 * frame that enables passing a complete pt_regs to a C function.
429 .macro DEFAULT_FRAME start=1 offset=0
430 PARTIAL_FRAME \start, R11+\offset-R15
431 CFI_REL_OFFSET rbx, RBX+\offset
432 CFI_REL_OFFSET rbp, RBP+\offset
433 CFI_REL_OFFSET r12, R12+\offset
434 CFI_REL_OFFSET r13, R13+\offset
435 CFI_REL_OFFSET r14, R14+\offset
436 CFI_REL_OFFSET r15, R15+\offset
439 /* save partial stack frame */
442 /* start from rbp in pt_regs and jump over */
443 movq_cfi rdi, RDI-RBP
444 movq_cfi rsi, RSI-RBP
445 movq_cfi rdx, RDX-RBP
446 movq_cfi rcx, RCX-RBP
447 movq_cfi rax, RAX-RBP
450 movq_cfi r10, R10-RBP
451 movq_cfi r11, R11-RBP
453 /* Save rbp so that we can unwind from get_irq_regs() */
456 /* Save previous stack value */
459 leaq -RBP(%rsp),%rdi /* arg1 for handler */
460 testl $3, CS-RBP(%rsi)
464 * irq_count is used to check if a CPU is already on an interrupt stack
465 * or not. While this is essentially redundant with preempt_count it is
466 * a little cheaper to use a separate counter in the PDA (short of
467 * moving irq_enter into assembly, which would be too much work)
469 1: incl PER_CPU_VAR(irq_count)
470 cmovzq PER_CPU_VAR(irq_stack_ptr),%rsp
471 CFI_DEF_CFA_REGISTER rsi
473 /* Store previous stack value */
475 CFI_ESCAPE 0x0f /* DW_CFA_def_cfa_expression */, 6, \
476 0x77 /* DW_OP_breg7 */, 0, \
477 0x06 /* DW_OP_deref */, \
478 0x08 /* DW_OP_const1u */, SS+8-RBP, \
479 0x22 /* DW_OP_plus */
480 /* We entered an interrupt context - irqs are off: */
485 PARTIAL_FRAME 1 REST_SKIP+8
486 movq 5*8+16(%rsp), %r11 /* save return address */
493 movq %r11, 8(%rsp) /* return address */
494 FIXUP_TOP_OF_STACK %r11, 16
499 /* save complete stack frame */
500 .pushsection .kprobes.text, "ax"
520 movl $MSR_GS_BASE,%ecx
523 js 1f /* negative -> in kernel */
532 * A newly forked process directly context switches into this address.
534 * rdi: prev task we switched from
539 LOCK ; btr $TIF_FORK,TI_flags(%r8)
541 pushq_cfi kernel_eflags(%rip)
542 popfq_cfi # reset kernel eflags
544 call schedule_tail # rdi: 'prev' task parameter
546 GET_THREAD_INFO(%rcx)
550 testl $3, CS-ARGOFFSET(%rsp) # from kernel_thread?
551 jz retint_restore_args
553 testl $_TIF_IA32, TI_flags(%rcx) # 32-bit compat task needs IRET
554 jnz int_ret_from_sys_call
556 RESTORE_TOP_OF_STACK %rdi, -ARGOFFSET
557 jmp ret_from_sys_call # go to the SYSRET fastpath
563 * System call entry. Up to 6 arguments in registers are supported.
565 * SYSCALL does not save anything on the stack and does not change the
571 * rax system call number
573 * rcx return address for syscall/sysret, C arg3
576 * r10 arg3 (--> moved to rcx for C)
579 * r11 eflags for syscall/sysret, temporary for C
580 * r12-r15,rbp,rbx saved by C code, not touched.
582 * Interrupts are off on entry.
583 * Only called from user space.
585 * XXX if we had a free scratch register we could save the RSP into the stack frame
586 * and report it properly in ps. Unfortunately we haven't.
588 * When user can change the frames always force IRET. That is because
589 * it deals with uncanonical addresses better. SYSRET has trouble
590 * with them due to bugs in both AMD and Intel CPUs.
596 CFI_DEF_CFA rsp,KERNEL_STACK_OFFSET
598 /*CFI_REGISTER rflags,r11*/
601 * A hypervisor implementation might want to use a label
602 * after the swapgs, so that it can do the swapgs
603 * for the guest and jump here on syscall.
605 GLOBAL(system_call_after_swapgs)
607 movq %rsp,PER_CPU_VAR(old_rsp)
608 movq PER_CPU_VAR(kernel_stack),%rsp
610 * No need to follow this irqs off/on section - it's straight
613 ENABLE_INTERRUPTS(CLBR_NONE)
615 movq %rax,ORIG_RAX-ARGOFFSET(%rsp)
616 movq %rcx,RIP-ARGOFFSET(%rsp)
617 CFI_REL_OFFSET rip,RIP-ARGOFFSET
618 testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET)
620 system_call_fastpath:
621 #if __SYSCALL_MASK == ~0
622 cmpq $__NR_syscall_max,%rax
624 andl $__SYSCALL_MASK,%eax
625 cmpl $__NR_syscall_max,%eax
629 call *sys_call_table(,%rax,8) # XXX: rip relative
630 movq %rax,RAX-ARGOFFSET(%rsp)
632 * Syscall return path ending with SYSRET (fast path)
633 * Has incomplete stack frame and undefined top of stack.
636 movl $_TIF_ALLWORK_MASK,%edi
640 DISABLE_INTERRUPTS(CLBR_NONE)
642 movl TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET),%edx
647 * sysretq will re-enable interrupts:
650 movq RIP-ARGOFFSET(%rsp),%rcx
652 RESTORE_ARGS 1,-ARG_SKIP,0
653 /*CFI_REGISTER rflags,r11*/
654 movq PER_CPU_VAR(old_rsp), %rsp
658 /* Handle reschedules */
659 /* edx: work, edi: workmask */
661 bt $TIF_NEED_RESCHED,%edx
664 ENABLE_INTERRUPTS(CLBR_NONE)
670 /* Handle a signal */
673 ENABLE_INTERRUPTS(CLBR_NONE)
674 #ifdef CONFIG_AUDITSYSCALL
675 bt $TIF_SYSCALL_AUDIT,%edx
679 * We have a signal, or exit tracing or single-step.
680 * These all wind up with the iret return path anyway,
681 * so just join that path right now.
683 FIXUP_TOP_OF_STACK %r11, -ARGOFFSET
684 jmp int_check_syscall_exit_work
687 movq $-ENOSYS,RAX-ARGOFFSET(%rsp)
688 jmp ret_from_sys_call
690 #ifdef CONFIG_AUDITSYSCALL
692 * Fast path for syscall audit without full syscall trace.
693 * We just call __audit_syscall_entry() directly, and then
694 * jump back to the normal fast path.
697 movq %r10,%r9 /* 6th arg: 4th syscall arg */
698 movq %rdx,%r8 /* 5th arg: 3rd syscall arg */
699 movq %rsi,%rcx /* 4th arg: 2nd syscall arg */
700 movq %rdi,%rdx /* 3rd arg: 1st syscall arg */
701 movq %rax,%rsi /* 2nd arg: syscall number */
702 movl $AUDIT_ARCH_X86_64,%edi /* 1st arg: audit arch */
703 call __audit_syscall_entry
704 LOAD_ARGS 0 /* reload call-clobbered registers */
705 jmp system_call_fastpath
708 * Return fast path for syscall audit. Call __audit_syscall_exit()
709 * directly and then jump back to the fast path with TIF_SYSCALL_AUDIT
713 movq RAX-ARGOFFSET(%rsp),%rsi /* second arg, syscall return value */
714 cmpq $-MAX_ERRNO,%rsi /* is it < -MAX_ERRNO? */
715 setbe %al /* 1 if so, 0 if not */
716 movzbl %al,%edi /* zero-extend that into %edi */
717 call __audit_syscall_exit
718 movl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT),%edi
720 #endif /* CONFIG_AUDITSYSCALL */
722 /* Do syscall tracing */
724 #ifdef CONFIG_AUDITSYSCALL
725 testl $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET)
729 movq $-ENOSYS,RAX(%rsp) /* ptrace can change this for a bad syscall */
730 FIXUP_TOP_OF_STACK %rdi
732 call syscall_trace_enter
734 * Reload arg registers from stack in case ptrace changed them.
735 * We don't reload %rax because syscall_trace_enter() returned
736 * the value it wants us to use in the table lookup.
738 LOAD_ARGS ARGOFFSET, 1
740 #if __SYSCALL_MASK == ~0
741 cmpq $__NR_syscall_max,%rax
743 andl $__SYSCALL_MASK,%eax
744 cmpl $__NR_syscall_max,%eax
746 ja int_ret_from_sys_call /* RAX(%rsp) set to -ENOSYS above */
747 movq %r10,%rcx /* fixup for C */
748 call *sys_call_table(,%rax,8)
749 movq %rax,RAX-ARGOFFSET(%rsp)
750 /* Use IRET because user could have changed frame */
753 * Syscall return path ending with IRET.
754 * Has correct top of stack, but partial stack frame.
756 GLOBAL(int_ret_from_sys_call)
757 DISABLE_INTERRUPTS(CLBR_NONE)
759 movl $_TIF_ALLWORK_MASK,%edi
760 /* edi: mask to check */
761 GLOBAL(int_with_check)
763 GET_THREAD_INFO(%rcx)
764 movl TI_flags(%rcx),%edx
767 andl $~TS_COMPAT,TI_status(%rcx)
770 /* Either reschedule or signal or syscall exit tracking needed. */
771 /* First do a reschedule test. */
772 /* edx: work, edi: workmask */
774 bt $TIF_NEED_RESCHED,%edx
777 ENABLE_INTERRUPTS(CLBR_NONE)
781 DISABLE_INTERRUPTS(CLBR_NONE)
785 /* handle signals and tracing -- both require a full stack frame */
788 ENABLE_INTERRUPTS(CLBR_NONE)
789 int_check_syscall_exit_work:
791 /* Check for syscall exit trace */
792 testl $_TIF_WORK_SYSCALL_EXIT,%edx
795 leaq 8(%rsp),%rdi # &ptregs -> arg1
796 call syscall_trace_leave
798 andl $~(_TIF_WORK_SYSCALL_EXIT|_TIF_SYSCALL_EMU),%edi
802 testl $_TIF_DO_NOTIFY_MASK,%edx
804 movq %rsp,%rdi # &ptregs -> arg1
805 xorl %esi,%esi # oldset -> arg2
806 call do_notify_resume
807 1: movl $_TIF_WORK_MASK,%edi
810 DISABLE_INTERRUPTS(CLBR_NONE)
817 * Certain special system calls that need to save a complete full stack frame.
819 .macro PTREGSCALL label,func,arg
821 PARTIAL_FRAME 1 8 /* offset 8: return address */
822 subq $REST_SKIP, %rsp
823 CFI_ADJUST_CFA_OFFSET REST_SKIP
825 DEFAULT_FRAME 0 8 /* offset 8: return address */
826 leaq 8(%rsp), \arg /* pt_regs pointer */
828 jmp ptregscall_common
833 PTREGSCALL stub_clone, sys_clone, %r8
834 PTREGSCALL stub_fork, sys_fork, %rdi
835 PTREGSCALL stub_vfork, sys_vfork, %rdi
836 PTREGSCALL stub_sigaltstack, sys_sigaltstack, %rdx
837 PTREGSCALL stub_iopl, sys_iopl, %rsi
839 ENTRY(ptregscall_common)
840 DEFAULT_FRAME 1 8 /* offset 8: return address */
841 RESTORE_TOP_OF_STACK %r11, 8
842 movq_cfi_restore R15+8, r15
843 movq_cfi_restore R14+8, r14
844 movq_cfi_restore R13+8, r13
845 movq_cfi_restore R12+8, r12
846 movq_cfi_restore RBP+8, rbp
847 movq_cfi_restore RBX+8, rbx
848 ret $REST_SKIP /* pop extended registers */
850 END(ptregscall_common)
857 FIXUP_TOP_OF_STACK %r11
860 RESTORE_TOP_OF_STACK %r11
863 jmp int_ret_from_sys_call
868 * sigreturn is special because it needs to restore all registers on return.
869 * This cannot be done with SYSRET, so use the IRET return path instead.
871 ENTRY(stub_rt_sigreturn)
877 FIXUP_TOP_OF_STACK %r11
878 call sys_rt_sigreturn
879 movq %rax,RAX(%rsp) # fixme, this could be done at the higher layer
881 jmp int_ret_from_sys_call
883 END(stub_rt_sigreturn)
885 #ifdef CONFIG_X86_X32_ABI
886 PTREGSCALL stub_x32_sigaltstack, sys32_sigaltstack, %rdx
888 ENTRY(stub_x32_rt_sigreturn)
894 FIXUP_TOP_OF_STACK %r11
895 call sys32_x32_rt_sigreturn
896 movq %rax,RAX(%rsp) # fixme, this could be done at the higher layer
898 jmp int_ret_from_sys_call
900 END(stub_x32_rt_sigreturn)
902 ENTRY(stub_x32_execve)
907 FIXUP_TOP_OF_STACK %r11
910 RESTORE_TOP_OF_STACK %r11
913 jmp int_ret_from_sys_call
920 * Build the entry stubs and pointer table with some assembler magic.
921 * We pack 7 stubs into a single 32-byte chunk, which will fit in a
922 * single cache line on all modern x86 implementations.
924 .section .init.rodata,"a"
928 .p2align CONFIG_X86_L1_CACHE_SHIFT
929 ENTRY(irq_entries_start)
931 vector=FIRST_EXTERNAL_VECTOR
932 .rept (NR_VECTORS-FIRST_EXTERNAL_VECTOR+6)/7
935 .if vector < NR_VECTORS
936 .if vector <> FIRST_EXTERNAL_VECTOR
937 CFI_ADJUST_CFA_OFFSET -8
939 1: pushq_cfi $(~vector+0x80) /* Note: always in signed byte range */
940 .if ((vector-FIRST_EXTERNAL_VECTOR)%7) <> 6
949 2: jmp common_interrupt
952 END(irq_entries_start)
959 * Interrupt entry/exit.
961 * Interrupt entry points save only callee clobbered registers in fast path.
963 * Entry runs with interrupts off.
966 /* 0(%rsp): ~(interrupt number) */
967 .macro interrupt func
968 /* reserve pt_regs for scratch regs and rbp */
969 subq $ORIG_RAX-RBP, %rsp
970 CFI_ADJUST_CFA_OFFSET ORIG_RAX-RBP
976 * Interrupt entry/exit should be protected against kprobes
978 .pushsection .kprobes.text, "ax"
980 * The interrupt stubs push (~vector+0x80) onto the stack and
981 * then jump to common_interrupt.
983 .p2align CONFIG_X86_L1_CACHE_SHIFT
986 addq $-0x80,(%rsp) /* Adjust vector to [-256,-1] range */
988 /* 0(%rsp): old_rsp-ARGOFFSET */
990 DISABLE_INTERRUPTS(CLBR_NONE)
992 decl PER_CPU_VAR(irq_count)
994 /* Restore saved previous stack */
996 CFI_DEF_CFA rsi,SS+8-RBP /* reg/off reset after def_cfa_expr */
997 leaq ARGOFFSET-RBP(%rsi), %rsp
998 CFI_DEF_CFA_REGISTER rsp
999 CFI_ADJUST_CFA_OFFSET RBP-ARGOFFSET
1002 GET_THREAD_INFO(%rcx)
1003 testl $3,CS-ARGOFFSET(%rsp)
1006 /* Interrupt came from user space */
1008 * Has a correct top of stack, but a partial stack frame
1009 * %rcx: thread info. Interrupts off.
1011 retint_with_reschedule:
1012 movl $_TIF_WORK_MASK,%edi
1014 LOCKDEP_SYS_EXIT_IRQ
1015 movl TI_flags(%rcx),%edx
1020 retint_swapgs: /* return to user-space */
1022 * The iretq could re-enable interrupts:
1024 DISABLE_INTERRUPTS(CLBR_ANY)
1029 retint_restore_args: /* return to kernel space */
1030 DISABLE_INTERRUPTS(CLBR_ANY)
1032 * The iretq could re-enable interrupts:
1040 _ASM_EXTABLE(irq_return, bad_iret)
1042 #ifdef CONFIG_PARAVIRT
1045 _ASM_EXTABLE(native_iret, bad_iret)
1048 .section .fixup,"ax"
1051 * The iret traps when the %cs or %ss being restored is bogus.
1052 * We've lost the original trap vector and error code.
1053 * #GPF is the most likely one to get for an invalid selector.
1054 * So pretend we completed the iret and took the #GPF in user mode.
1056 * We are now running with the kernel GS after exception recovery.
1057 * But error_entry expects us to have user GS to match the user %cs,
1063 jmp general_protection
1067 /* edi: workmask, edx: work */
1070 bt $TIF_NEED_RESCHED,%edx
1073 ENABLE_INTERRUPTS(CLBR_NONE)
1077 GET_THREAD_INFO(%rcx)
1078 DISABLE_INTERRUPTS(CLBR_NONE)
1083 testl $_TIF_DO_NOTIFY_MASK,%edx
1086 ENABLE_INTERRUPTS(CLBR_NONE)
1088 movq $-1,ORIG_RAX(%rsp)
1089 xorl %esi,%esi # oldset
1090 movq %rsp,%rdi # &pt_regs
1091 call do_notify_resume
1093 DISABLE_INTERRUPTS(CLBR_NONE)
1095 GET_THREAD_INFO(%rcx)
1096 jmp retint_with_reschedule
1098 #ifdef CONFIG_PREEMPT
1099 /* Returning to kernel space. Check if we need preemption */
1100 /* rcx: threadinfo. interrupts off. */
1101 ENTRY(retint_kernel)
1102 cmpl $0,TI_preempt_count(%rcx)
1103 jnz retint_restore_args
1104 bt $TIF_NEED_RESCHED,TI_flags(%rcx)
1105 jnc retint_restore_args
1106 bt $9,EFLAGS-ARGOFFSET(%rsp) /* interrupts off? */
1107 jnc retint_restore_args
1108 call preempt_schedule_irq
1113 END(common_interrupt)
1115 * End of kprobes section
1122 .macro apicinterrupt num sym do_sym
1134 apicinterrupt IRQ_MOVE_CLEANUP_VECTOR \
1135 irq_move_cleanup_interrupt smp_irq_move_cleanup_interrupt
1136 apicinterrupt REBOOT_VECTOR \
1137 reboot_interrupt smp_reboot_interrupt
1140 #ifdef CONFIG_X86_UV
1141 apicinterrupt UV_BAU_MESSAGE \
1142 uv_bau_message_intr1 uv_bau_message_interrupt
1144 apicinterrupt LOCAL_TIMER_VECTOR \
1145 apic_timer_interrupt smp_apic_timer_interrupt
1146 apicinterrupt X86_PLATFORM_IPI_VECTOR \
1147 x86_platform_ipi smp_x86_platform_ipi
1149 apicinterrupt THRESHOLD_APIC_VECTOR \
1150 threshold_interrupt smp_threshold_interrupt
1151 apicinterrupt THERMAL_APIC_VECTOR \
1152 thermal_interrupt smp_thermal_interrupt
1155 apicinterrupt CALL_FUNCTION_SINGLE_VECTOR \
1156 call_function_single_interrupt smp_call_function_single_interrupt
1157 apicinterrupt CALL_FUNCTION_VECTOR \
1158 call_function_interrupt smp_call_function_interrupt
1159 apicinterrupt RESCHEDULE_VECTOR \
1160 reschedule_interrupt smp_reschedule_interrupt
1163 apicinterrupt ERROR_APIC_VECTOR \
1164 error_interrupt smp_error_interrupt
1165 apicinterrupt SPURIOUS_APIC_VECTOR \
1166 spurious_interrupt smp_spurious_interrupt
1168 #ifdef CONFIG_IRQ_WORK
1169 apicinterrupt IRQ_WORK_VECTOR \
1170 irq_work_interrupt smp_irq_work_interrupt
1174 * Exception entry points.
1176 .macro zeroentry sym do_sym
1179 PARAVIRT_ADJUST_EXCEPTION_FRAME
1180 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1181 subq $ORIG_RAX-R15, %rsp
1182 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1185 movq %rsp,%rdi /* pt_regs pointer */
1186 xorl %esi,%esi /* no error code */
1188 jmp error_exit /* %ebx: no swapgs flag */
1193 .macro paranoidzeroentry sym do_sym
1196 PARAVIRT_ADJUST_EXCEPTION_FRAME
1197 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1198 subq $ORIG_RAX-R15, %rsp
1199 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1202 movq %rsp,%rdi /* pt_regs pointer */
1203 xorl %esi,%esi /* no error code */
1205 jmp paranoid_exit /* %ebx: no swapgs flag */
1210 #define INIT_TSS_IST(x) PER_CPU_VAR(init_tss) + (TSS_ist + ((x) - 1) * 8)
1211 .macro paranoidzeroentry_ist sym do_sym ist
1214 PARAVIRT_ADJUST_EXCEPTION_FRAME
1215 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1216 subq $ORIG_RAX-R15, %rsp
1217 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1219 TRACE_IRQS_OFF_DEBUG
1220 movq %rsp,%rdi /* pt_regs pointer */
1221 xorl %esi,%esi /* no error code */
1222 subq $EXCEPTION_STKSZ, INIT_TSS_IST(\ist)
1224 addq $EXCEPTION_STKSZ, INIT_TSS_IST(\ist)
1225 jmp paranoid_exit /* %ebx: no swapgs flag */
1230 .macro errorentry sym do_sym
1233 PARAVIRT_ADJUST_EXCEPTION_FRAME
1234 subq $ORIG_RAX-R15, %rsp
1235 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1238 movq %rsp,%rdi /* pt_regs pointer */
1239 movq ORIG_RAX(%rsp),%rsi /* get error code */
1240 movq $-1,ORIG_RAX(%rsp) /* no syscall to restart */
1242 jmp error_exit /* %ebx: no swapgs flag */
1247 /* error code is on the stack already */
1248 .macro paranoiderrorentry sym do_sym
1251 PARAVIRT_ADJUST_EXCEPTION_FRAME
1252 subq $ORIG_RAX-R15, %rsp
1253 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1257 movq %rsp,%rdi /* pt_regs pointer */
1258 movq ORIG_RAX(%rsp),%rsi /* get error code */
1259 movq $-1,ORIG_RAX(%rsp) /* no syscall to restart */
1261 jmp paranoid_exit /* %ebx: no swapgs flag */
1266 zeroentry divide_error do_divide_error
1267 zeroentry overflow do_overflow
1268 zeroentry bounds do_bounds
1269 zeroentry invalid_op do_invalid_op
1270 zeroentry device_not_available do_device_not_available
1271 paranoiderrorentry double_fault do_double_fault
1272 zeroentry coprocessor_segment_overrun do_coprocessor_segment_overrun
1273 errorentry invalid_TSS do_invalid_TSS
1274 errorentry segment_not_present do_segment_not_present
1275 zeroentry spurious_interrupt_bug do_spurious_interrupt_bug
1276 zeroentry coprocessor_error do_coprocessor_error
1277 errorentry alignment_check do_alignment_check
1278 zeroentry simd_coprocessor_error do_simd_coprocessor_error
1281 /* Reload gs selector with exception handling */
1282 /* edi: new selector */
1283 ENTRY(native_load_gs_index)
1286 DISABLE_INTERRUPTS(CLBR_ANY & ~CLBR_RDI)
1290 2: mfence /* workaround */
1295 END(native_load_gs_index)
1297 _ASM_EXTABLE(gs_change,bad_gs)
1298 .section .fixup,"ax"
1299 /* running with kernelgs */
1301 SWAPGS /* switch back to user gs */
1307 ENTRY(kernel_thread_helper)
1308 pushq $0 # fake return address
1311 * Here we are in the child and the registers are set as they were
1312 * at kernel_thread() invocation in the parent.
1318 ud2 # padding for call trace
1320 END(kernel_thread_helper)
1323 * execve(). This function needs to use IRET, not SYSRET, to set up all state properly.
1325 * C extern interface:
1326 * extern long execve(const char *name, char **argv, char **envp)
1328 * asm input arguments:
1329 * rdi: name, rsi: argv, rdx: envp
1331 * We want to fallback into:
1332 * extern long sys_execve(const char *name, char **argv,char **envp, struct pt_regs *regs)
1334 * do_sys_execve asm fallback arguments:
1335 * rdi: name, rsi: argv, rdx: envp, rcx: fake frame on the stack
1337 ENTRY(kernel_execve)
1343 movq %rax, RAX(%rsp)
1346 je int_ret_from_sys_call
1353 /* Call softirq on interrupt stack. Interrupts are off. */
1357 CFI_REL_OFFSET rbp,0
1359 CFI_DEF_CFA_REGISTER rbp
1360 incl PER_CPU_VAR(irq_count)
1361 cmove PER_CPU_VAR(irq_stack_ptr),%rsp
1362 push %rbp # backlink for old unwinder
1366 CFI_DEF_CFA_REGISTER rsp
1367 CFI_ADJUST_CFA_OFFSET -8
1368 decl PER_CPU_VAR(irq_count)
1374 zeroentry xen_hypervisor_callback xen_do_hypervisor_callback
1377 * A note on the "critical region" in our callback handler.
1378 * We want to avoid stacking callback handlers due to events occurring
1379 * during handling of the last event. To do this, we keep events disabled
1380 * until we've done all processing. HOWEVER, we must enable events before
1381 * popping the stack frame (can't be done atomically) and so it would still
1382 * be possible to get enough handler activations to overflow the stack.
1383 * Although unlikely, bugs of that kind are hard to track down, so we'd
1384 * like to avoid the possibility.
1385 * So, on entry to the handler we detect whether we interrupted an
1386 * existing activation in its critical region -- if so, we pop the current
1387 * activation and restart the handler using the previous one.
1389 ENTRY(xen_do_hypervisor_callback) # do_hypervisor_callback(struct *pt_regs)
1392 * Since we don't modify %rdi, evtchn_do_upall(struct *pt_regs) will
1393 * see the correct pointer to the pt_regs
1395 movq %rdi, %rsp # we don't return, adjust the stack frame
1398 11: incl PER_CPU_VAR(irq_count)
1400 CFI_DEF_CFA_REGISTER rbp
1401 cmovzq PER_CPU_VAR(irq_stack_ptr),%rsp
1402 pushq %rbp # backlink for old unwinder
1403 call xen_evtchn_do_upcall
1405 CFI_DEF_CFA_REGISTER rsp
1406 decl PER_CPU_VAR(irq_count)
1409 END(xen_do_hypervisor_callback)
1412 * Hypervisor uses this for application faults while it executes.
1413 * We get here for two reasons:
1414 * 1. Fault while reloading DS, ES, FS or GS
1415 * 2. Fault while executing IRET
1416 * Category 1 we do not need to fix up as Xen has already reloaded all segment
1417 * registers that could be reloaded and zeroed the others.
1418 * Category 2 we fix up by killing the current process. We cannot use the
1419 * normal Linux return path in this case because if we use the IRET hypercall
1420 * to pop the stack frame we end up in an infinite loop of failsafe callbacks.
1421 * We distinguish between categories by comparing each saved segment register
1422 * with its current contents: any discrepancy means we in category 1.
1424 ENTRY(xen_failsafe_callback)
1426 /*CFI_REL_OFFSET gs,GS*/
1427 /*CFI_REL_OFFSET fs,FS*/
1428 /*CFI_REL_OFFSET es,ES*/
1429 /*CFI_REL_OFFSET ds,DS*/
1430 CFI_REL_OFFSET r11,8
1431 CFI_REL_OFFSET rcx,0
1445 /* All segments match their saved values => Category 2 (Bad IRET). */
1451 CFI_ADJUST_CFA_OFFSET -0x30
1452 pushq_cfi $0 /* RIP */
1455 jmp general_protection
1457 1: /* Segment mismatch => Category 1 (Bad segment). Retry the IRET. */
1463 CFI_ADJUST_CFA_OFFSET -0x30
1468 END(xen_failsafe_callback)
1470 apicinterrupt XEN_HVM_EVTCHN_CALLBACK \
1471 xen_hvm_callback_vector xen_evtchn_do_upcall
1473 #endif /* CONFIG_XEN */
1476 * Some functions should be protected against kprobes
1478 .pushsection .kprobes.text, "ax"
1480 paranoidzeroentry_ist debug do_debug DEBUG_STACK
1481 paranoidzeroentry_ist int3 do_int3 DEBUG_STACK
1482 paranoiderrorentry stack_segment do_stack_segment
1484 zeroentry xen_debug do_debug
1485 zeroentry xen_int3 do_int3
1486 errorentry xen_stack_segment do_stack_segment
1488 errorentry general_protection do_general_protection
1489 errorentry page_fault do_page_fault
1490 #ifdef CONFIG_KVM_GUEST
1491 errorentry async_page_fault do_async_page_fault
1493 #ifdef CONFIG_X86_MCE
1494 paranoidzeroentry machine_check *machine_check_vector(%rip)
1498 * "Paranoid" exit path from exception stack.
1499 * Paranoid because this is used by NMIs and cannot take
1500 * any kernel state for granted.
1501 * We don't do kernel preemption checks here, because only
1502 * NMI should be common and it does not enable IRQs and
1503 * cannot get reschedule ticks.
1505 * "trace" is 0 for the NMI handler only, because irq-tracing
1506 * is fundamentally NMI-unsafe. (we cannot change the soft and
1507 * hard flags at once, atomically)
1510 /* ebx: no swapgs flag */
1511 ENTRY(paranoid_exit)
1513 DISABLE_INTERRUPTS(CLBR_NONE)
1514 TRACE_IRQS_OFF_DEBUG
1515 testl %ebx,%ebx /* swapgs needed? */
1516 jnz paranoid_restore
1518 jnz paranoid_userspace
1525 TRACE_IRQS_IRETQ_DEBUG 0
1529 GET_THREAD_INFO(%rcx)
1530 movl TI_flags(%rcx),%ebx
1531 andl $_TIF_WORK_MASK,%ebx
1533 movq %rsp,%rdi /* &pt_regs */
1535 movq %rax,%rsp /* switch stack for scheduling */
1536 testl $_TIF_NEED_RESCHED,%ebx
1537 jnz paranoid_schedule
1538 movl %ebx,%edx /* arg3: thread flags */
1540 ENABLE_INTERRUPTS(CLBR_NONE)
1541 xorl %esi,%esi /* arg2: oldset */
1542 movq %rsp,%rdi /* arg1: &pt_regs */
1543 call do_notify_resume
1544 DISABLE_INTERRUPTS(CLBR_NONE)
1546 jmp paranoid_userspace
1549 ENABLE_INTERRUPTS(CLBR_ANY)
1551 DISABLE_INTERRUPTS(CLBR_ANY)
1553 jmp paranoid_userspace
1558 * Exception entry point. This expects an error code/orig_rax on the stack.
1559 * returns in "no swapgs flag" in %ebx.
1563 CFI_ADJUST_CFA_OFFSET 15*8
1564 /* oldrax contains error code */
1583 je error_kernelspace
1591 * There are two places in the kernel that can potentially fault with
1592 * usergs. Handle them here. The exception handlers after iret run with
1593 * kernel gs again, so don't set the user space flag. B stepping K8s
1594 * sometimes report an truncated RIP for IRET exceptions returning to
1595 * compat mode. Check for these here too.
1599 leaq irq_return(%rip),%rcx
1600 cmpq %rcx,RIP+8(%rsp)
1602 movl %ecx,%eax /* zero extend */
1603 cmpq %rax,RIP+8(%rsp)
1605 cmpq $gs_change,RIP+8(%rsp)
1610 /* Fix truncated RIP */
1611 movq %rcx,RIP+8(%rsp)
1617 /* ebx: no swapgs flag (1: don't need swapgs, 0: need it) */
1622 DISABLE_INTERRUPTS(CLBR_NONE)
1624 GET_THREAD_INFO(%rcx)
1627 LOCKDEP_SYS_EXIT_IRQ
1628 movl TI_flags(%rcx),%edx
1629 movl $_TIF_WORK_MASK,%edi
1637 * Test if a given stack is an NMI stack or not.
1639 .macro test_in_nmi reg stack nmi_ret normal_ret
1642 subq $EXCEPTION_STKSZ, %\reg
1648 /* runs on exception stack */
1651 PARAVIRT_ADJUST_EXCEPTION_FRAME
1653 * We allow breakpoints in NMIs. If a breakpoint occurs, then
1654 * the iretq it performs will take us out of NMI context.
1655 * This means that we can have nested NMIs where the next
1656 * NMI is using the top of the stack of the previous NMI. We
1657 * can't let it execute because the nested NMI will corrupt the
1658 * stack of the previous NMI. NMI handlers are not re-entrant
1661 * To handle this case we do the following:
1662 * Check the a special location on the stack that contains
1663 * a variable that is set when NMIs are executing.
1664 * The interrupted task's stack is also checked to see if it
1666 * If the variable is not set and the stack is not the NMI
1668 * o Set the special variable on the stack
1669 * o Copy the interrupt frame into a "saved" location on the stack
1670 * o Copy the interrupt frame into a "copy" location on the stack
1671 * o Continue processing the NMI
1672 * If the variable is set or the previous stack is the NMI stack:
1673 * o Modify the "copy" location to jump to the repeate_nmi
1674 * o return back to the first NMI
1676 * Now on exit of the first NMI, we first clear the stack variable
1677 * The NMI stack will tell any nested NMIs at that point that it is
1678 * nested. Then we pop the stack normally with iret, and if there was
1679 * a nested NMI that updated the copy interrupt stack frame, a
1680 * jump will be made to the repeat_nmi code that will handle the second
1684 /* Use %rdx as out temp variable throughout */
1686 CFI_REL_OFFSET rdx, 0
1689 * If %cs was not the kernel segment, then the NMI triggered in user
1690 * space, which means it is definitely not nested.
1692 cmpl $__KERNEL_CS, 16(%rsp)
1696 * Check the special variable on the stack to see if NMIs are
1703 * Now test if the previous stack was an NMI stack.
1704 * We need the double check. We check the NMI stack to satisfy the
1705 * race when the first NMI clears the variable before returning.
1706 * We check the variable because the first NMI could be in a
1707 * breakpoint routine using a breakpoint stack.
1710 test_in_nmi rdx, 4*8(%rsp), nested_nmi, first_nmi
1715 * Do nothing if we interrupted the fixup in repeat_nmi.
1716 * It's about to repeat the NMI handler, so we are fine
1717 * with ignoring this one.
1719 movq $repeat_nmi, %rdx
1722 movq $end_repeat_nmi, %rdx
1727 /* Set up the interrupted NMIs stack to jump to repeat_nmi */
1728 leaq -6*8(%rsp), %rdx
1730 CFI_ADJUST_CFA_OFFSET 6*8
1731 pushq_cfi $__KERNEL_DS
1734 pushq_cfi $__KERNEL_CS
1735 pushq_cfi $repeat_nmi
1737 /* Put stack back */
1739 CFI_ADJUST_CFA_OFFSET -11*8
1745 /* No need to check faults here */
1751 * Because nested NMIs will use the pushed location that we
1752 * stored in rdx, we must keep that space available.
1753 * Here's what our stack frame will look like:
1754 * +-------------------------+
1756 * | original Return RSP |
1757 * | original RFLAGS |
1760 * +-------------------------+
1761 * | temp storage for rdx |
1762 * +-------------------------+
1763 * | NMI executing variable |
1764 * +-------------------------+
1766 * | Saved Return RSP |
1770 * +-------------------------+
1772 * | copied Return RSP |
1776 * +-------------------------+
1778 * +-------------------------+
1780 * The saved stack frame is used to fix up the copied stack frame
1781 * that a nested NMI may change to make the interrupted NMI iret jump
1782 * to the repeat_nmi. The original stack frame and the temp storage
1783 * is also used by nested NMIs and can not be trusted on exit.
1785 /* Do not pop rdx, nested NMIs will corrupt that part of the stack */
1789 /* Set the NMI executing variable on the stack. */
1792 /* Copy the stack frame to the Saved frame */
1796 CFI_DEF_CFA_OFFSET SS+8-RIP
1798 /* Everything up to here is safe from nested NMIs */
1801 * If there was a nested NMI, the first NMI's iret will return
1802 * here. But NMIs are still enabled and we can take another
1803 * nested NMI. The nested NMI checks the interrupted RIP to see
1804 * if it is between repeat_nmi and end_repeat_nmi, and if so
1805 * it will just return, as we are about to repeat an NMI anyway.
1806 * This makes it safe to copy to the stack frame that a nested
1811 * Update the stack variable to say we are still in NMI (the update
1812 * is benign for the non-repeat case, where 1 was pushed just above
1813 * to this very stack slot).
1817 /* Make another copy, this one may be modified by nested NMIs */
1821 CFI_DEF_CFA_OFFSET SS+8-RIP
1825 * Everything below this point can be preempted by a nested
1826 * NMI if the first NMI took an exception and reset our iret stack
1827 * so that we repeat another NMI.
1829 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1830 subq $ORIG_RAX-R15, %rsp
1831 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1833 * Use save_paranoid to handle SWAPGS, but no need to use paranoid_exit
1834 * as we should not be calling schedule in NMI context.
1835 * Even with normal interrupts enabled. An NMI should not be
1836 * setting NEED_RESCHED or anything that normal interrupts and
1837 * exceptions might do.
1843 * Save off the CR2 register. If we take a page fault in the NMI then
1844 * it could corrupt the CR2 value. If the NMI preempts a page fault
1845 * handler before it was able to read the CR2 register, and then the
1846 * NMI itself takes a page fault, the page fault that was preempted
1847 * will read the information from the NMI page fault and not the
1848 * origin fault. Save it off and restore it if it changes.
1849 * Use the r12 callee-saved register.
1853 /* paranoidentry do_nmi, 0; without TRACE_IRQS_OFF */
1858 /* Did the NMI take a page fault? Restore cr2 if it did */
1865 testl %ebx,%ebx /* swapgs needed? */
1871 /* Clear the NMI executing stack variable */
1877 ENTRY(ignore_sysret)
1885 * End of kprobes section