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)
221 subq $MCOUNT_INSN_SIZE, %rsi
223 call prepare_ftrace_return
228 END(ftrace_graph_caller)
230 GLOBAL(return_to_handler)
233 /* Save the return values */
238 call ftrace_return_to_handler
248 #ifndef CONFIG_PREEMPT
249 #define retint_kernel retint_restore_args
252 #ifdef CONFIG_PARAVIRT
253 ENTRY(native_usergs_sysret64)
256 ENDPROC(native_usergs_sysret64)
257 #endif /* CONFIG_PARAVIRT */
260 .macro TRACE_IRQS_IRETQ offset=ARGOFFSET
261 #ifdef CONFIG_TRACE_IRQFLAGS
262 bt $9,EFLAGS-\offset(%rsp) /* interrupts off? */
270 * When dynamic function tracer is enabled it will add a breakpoint
271 * to all locations that it is about to modify, sync CPUs, update
272 * all the code, sync CPUs, then remove the breakpoints. In this time
273 * if lockdep is enabled, it might jump back into the debug handler
274 * outside the updating of the IST protection. (TRACE_IRQS_ON/OFF).
276 * We need to change the IDT table before calling TRACE_IRQS_ON/OFF to
277 * make sure the stack pointer does not get reset back to the top
278 * of the debug stack, and instead just reuses the current stack.
280 #if defined(CONFIG_DYNAMIC_FTRACE) && defined(CONFIG_TRACE_IRQFLAGS)
282 .macro TRACE_IRQS_OFF_DEBUG
283 call debug_stack_set_zero
285 call debug_stack_reset
288 .macro TRACE_IRQS_ON_DEBUG
289 call debug_stack_set_zero
291 call debug_stack_reset
294 .macro TRACE_IRQS_IRETQ_DEBUG offset=ARGOFFSET
295 bt $9,EFLAGS-\offset(%rsp) /* interrupts off? */
302 # define TRACE_IRQS_OFF_DEBUG TRACE_IRQS_OFF
303 # define TRACE_IRQS_ON_DEBUG TRACE_IRQS_ON
304 # define TRACE_IRQS_IRETQ_DEBUG TRACE_IRQS_IRETQ
308 * C code is not supposed to know about undefined top of stack. Every time
309 * a C function with an pt_regs argument is called from the SYSCALL based
310 * fast path FIXUP_TOP_OF_STACK is needed.
311 * RESTORE_TOP_OF_STACK syncs the syscall state after any possible ptregs
315 /* %rsp:at FRAMEEND */
316 .macro FIXUP_TOP_OF_STACK tmp offset=0
317 movq PER_CPU_VAR(old_rsp),\tmp
318 movq \tmp,RSP+\offset(%rsp)
319 movq $__USER_DS,SS+\offset(%rsp)
320 movq $__USER_CS,CS+\offset(%rsp)
321 movq $-1,RCX+\offset(%rsp)
322 movq R11+\offset(%rsp),\tmp /* get eflags */
323 movq \tmp,EFLAGS+\offset(%rsp)
326 .macro RESTORE_TOP_OF_STACK tmp offset=0
327 movq RSP+\offset(%rsp),\tmp
328 movq \tmp,PER_CPU_VAR(old_rsp)
329 movq EFLAGS+\offset(%rsp),\tmp
330 movq \tmp,R11+\offset(%rsp)
333 .macro FAKE_STACK_FRAME child_rip
334 /* push in order ss, rsp, eflags, cs, rip */
336 pushq_cfi $__KERNEL_DS /* ss */
337 /*CFI_REL_OFFSET ss,0*/
338 pushq_cfi %rax /* rsp */
340 pushq_cfi $(X86_EFLAGS_IF|X86_EFLAGS_BIT1) /* eflags - interrupts on */
341 /*CFI_REL_OFFSET rflags,0*/
342 pushq_cfi $__KERNEL_CS /* cs */
343 /*CFI_REL_OFFSET cs,0*/
344 pushq_cfi \child_rip /* rip */
346 pushq_cfi %rax /* orig rax */
349 .macro UNFAKE_STACK_FRAME
351 CFI_ADJUST_CFA_OFFSET -(6*8)
355 * initial frame state for interrupts (and exceptions without error code)
357 .macro EMPTY_FRAME start=1 offset=0
361 CFI_DEF_CFA rsp,8+\offset
363 CFI_DEF_CFA_OFFSET 8+\offset
368 * initial frame state for interrupts (and exceptions without error code)
370 .macro INTR_FRAME start=1 offset=0
371 EMPTY_FRAME \start, SS+8+\offset-RIP
372 /*CFI_REL_OFFSET ss, SS+\offset-RIP*/
373 CFI_REL_OFFSET rsp, RSP+\offset-RIP
374 /*CFI_REL_OFFSET rflags, EFLAGS+\offset-RIP*/
375 /*CFI_REL_OFFSET cs, CS+\offset-RIP*/
376 CFI_REL_OFFSET rip, RIP+\offset-RIP
380 * initial frame state for exceptions with error code (and interrupts
381 * with vector already pushed)
383 .macro XCPT_FRAME start=1 offset=0
384 INTR_FRAME \start, RIP+\offset-ORIG_RAX
385 /*CFI_REL_OFFSET orig_rax, ORIG_RAX-ORIG_RAX*/
389 * frame that enables calling into C.
391 .macro PARTIAL_FRAME start=1 offset=0
392 XCPT_FRAME \start, ORIG_RAX+\offset-ARGOFFSET
393 CFI_REL_OFFSET rdi, RDI+\offset-ARGOFFSET
394 CFI_REL_OFFSET rsi, RSI+\offset-ARGOFFSET
395 CFI_REL_OFFSET rdx, RDX+\offset-ARGOFFSET
396 CFI_REL_OFFSET rcx, RCX+\offset-ARGOFFSET
397 CFI_REL_OFFSET rax, RAX+\offset-ARGOFFSET
398 CFI_REL_OFFSET r8, R8+\offset-ARGOFFSET
399 CFI_REL_OFFSET r9, R9+\offset-ARGOFFSET
400 CFI_REL_OFFSET r10, R10+\offset-ARGOFFSET
401 CFI_REL_OFFSET r11, R11+\offset-ARGOFFSET
405 * frame that enables passing a complete pt_regs to a C function.
407 .macro DEFAULT_FRAME start=1 offset=0
408 PARTIAL_FRAME \start, R11+\offset-R15
409 CFI_REL_OFFSET rbx, RBX+\offset
410 CFI_REL_OFFSET rbp, RBP+\offset
411 CFI_REL_OFFSET r12, R12+\offset
412 CFI_REL_OFFSET r13, R13+\offset
413 CFI_REL_OFFSET r14, R14+\offset
414 CFI_REL_OFFSET r15, R15+\offset
417 /* save partial stack frame */
420 /* start from rbp in pt_regs and jump over */
421 movq_cfi rdi, RDI-RBP
422 movq_cfi rsi, RSI-RBP
423 movq_cfi rdx, RDX-RBP
424 movq_cfi rcx, RCX-RBP
425 movq_cfi rax, RAX-RBP
428 movq_cfi r10, R10-RBP
429 movq_cfi r11, R11-RBP
431 /* Save rbp so that we can unwind from get_irq_regs() */
434 /* Save previous stack value */
437 leaq -RBP(%rsp),%rdi /* arg1 for handler */
438 testl $3, CS-RBP(%rsi)
442 * irq_count is used to check if a CPU is already on an interrupt stack
443 * or not. While this is essentially redundant with preempt_count it is
444 * a little cheaper to use a separate counter in the PDA (short of
445 * moving irq_enter into assembly, which would be too much work)
447 1: incl PER_CPU_VAR(irq_count)
448 cmovzq PER_CPU_VAR(irq_stack_ptr),%rsp
449 CFI_DEF_CFA_REGISTER rsi
451 /* Store previous stack value */
453 CFI_ESCAPE 0x0f /* DW_CFA_def_cfa_expression */, 6, \
454 0x77 /* DW_OP_breg7 */, 0, \
455 0x06 /* DW_OP_deref */, \
456 0x08 /* DW_OP_const1u */, SS+8-RBP, \
457 0x22 /* DW_OP_plus */
458 /* We entered an interrupt context - irqs are off: */
463 PARTIAL_FRAME 1 REST_SKIP+8
464 movq 5*8+16(%rsp), %r11 /* save return address */
471 movq %r11, 8(%rsp) /* return address */
472 FIXUP_TOP_OF_STACK %r11, 16
477 /* save complete stack frame */
478 .pushsection .kprobes.text, "ax"
498 movl $MSR_GS_BASE,%ecx
501 js 1f /* negative -> in kernel */
510 * A newly forked process directly context switches into this address.
512 * rdi: prev task we switched from
517 LOCK ; btr $TIF_FORK,TI_flags(%r8)
519 pushq_cfi kernel_eflags(%rip)
520 popfq_cfi # reset kernel eflags
522 call schedule_tail # rdi: 'prev' task parameter
524 GET_THREAD_INFO(%rcx)
528 testl $3, CS-ARGOFFSET(%rsp) # from kernel_thread?
529 jz retint_restore_args
531 testl $_TIF_IA32, TI_flags(%rcx) # 32-bit compat task needs IRET
532 jnz int_ret_from_sys_call
534 RESTORE_TOP_OF_STACK %rdi, -ARGOFFSET
535 jmp ret_from_sys_call # go to the SYSRET fastpath
541 * System call entry. Up to 6 arguments in registers are supported.
543 * SYSCALL does not save anything on the stack and does not change the
549 * rax system call number
551 * rcx return address for syscall/sysret, C arg3
554 * r10 arg3 (--> moved to rcx for C)
557 * r11 eflags for syscall/sysret, temporary for C
558 * r12-r15,rbp,rbx saved by C code, not touched.
560 * Interrupts are off on entry.
561 * Only called from user space.
563 * XXX if we had a free scratch register we could save the RSP into the stack frame
564 * and report it properly in ps. Unfortunately we haven't.
566 * When user can change the frames always force IRET. That is because
567 * it deals with uncanonical addresses better. SYSRET has trouble
568 * with them due to bugs in both AMD and Intel CPUs.
574 CFI_DEF_CFA rsp,KERNEL_STACK_OFFSET
576 /*CFI_REGISTER rflags,r11*/
579 * A hypervisor implementation might want to use a label
580 * after the swapgs, so that it can do the swapgs
581 * for the guest and jump here on syscall.
583 GLOBAL(system_call_after_swapgs)
585 movq %rsp,PER_CPU_VAR(old_rsp)
586 movq PER_CPU_VAR(kernel_stack),%rsp
588 * No need to follow this irqs off/on section - it's straight
591 ENABLE_INTERRUPTS(CLBR_NONE)
593 movq %rax,ORIG_RAX-ARGOFFSET(%rsp)
594 movq %rcx,RIP-ARGOFFSET(%rsp)
595 CFI_REL_OFFSET rip,RIP-ARGOFFSET
596 testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET)
598 system_call_fastpath:
599 #if __SYSCALL_MASK == ~0
600 cmpq $__NR_syscall_max,%rax
602 andl $__SYSCALL_MASK,%eax
603 cmpl $__NR_syscall_max,%eax
607 call *sys_call_table(,%rax,8) # XXX: rip relative
608 movq %rax,RAX-ARGOFFSET(%rsp)
610 * Syscall return path ending with SYSRET (fast path)
611 * Has incomplete stack frame and undefined top of stack.
614 movl $_TIF_ALLWORK_MASK,%edi
618 DISABLE_INTERRUPTS(CLBR_NONE)
620 movl TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET),%edx
625 * sysretq will re-enable interrupts:
628 movq RIP-ARGOFFSET(%rsp),%rcx
630 RESTORE_ARGS 1,-ARG_SKIP,0
631 /*CFI_REGISTER rflags,r11*/
632 movq PER_CPU_VAR(old_rsp), %rsp
636 /* Handle reschedules */
637 /* edx: work, edi: workmask */
639 bt $TIF_NEED_RESCHED,%edx
642 ENABLE_INTERRUPTS(CLBR_NONE)
648 /* Handle a signal */
651 ENABLE_INTERRUPTS(CLBR_NONE)
652 #ifdef CONFIG_AUDITSYSCALL
653 bt $TIF_SYSCALL_AUDIT,%edx
657 * We have a signal, or exit tracing or single-step.
658 * These all wind up with the iret return path anyway,
659 * so just join that path right now.
661 FIXUP_TOP_OF_STACK %r11, -ARGOFFSET
662 jmp int_check_syscall_exit_work
665 movq $-ENOSYS,RAX-ARGOFFSET(%rsp)
666 jmp ret_from_sys_call
668 #ifdef CONFIG_AUDITSYSCALL
670 * Fast path for syscall audit without full syscall trace.
671 * We just call __audit_syscall_entry() directly, and then
672 * jump back to the normal fast path.
675 movq %r10,%r9 /* 6th arg: 4th syscall arg */
676 movq %rdx,%r8 /* 5th arg: 3rd syscall arg */
677 movq %rsi,%rcx /* 4th arg: 2nd syscall arg */
678 movq %rdi,%rdx /* 3rd arg: 1st syscall arg */
679 movq %rax,%rsi /* 2nd arg: syscall number */
680 movl $AUDIT_ARCH_X86_64,%edi /* 1st arg: audit arch */
681 call __audit_syscall_entry
682 LOAD_ARGS 0 /* reload call-clobbered registers */
683 jmp system_call_fastpath
686 * Return fast path for syscall audit. Call __audit_syscall_exit()
687 * directly and then jump back to the fast path with TIF_SYSCALL_AUDIT
691 movq RAX-ARGOFFSET(%rsp),%rsi /* second arg, syscall return value */
692 cmpq $-MAX_ERRNO,%rsi /* is it < -MAX_ERRNO? */
693 setbe %al /* 1 if so, 0 if not */
694 movzbl %al,%edi /* zero-extend that into %edi */
695 call __audit_syscall_exit
696 movl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT),%edi
698 #endif /* CONFIG_AUDITSYSCALL */
700 /* Do syscall tracing */
702 #ifdef CONFIG_AUDITSYSCALL
703 testl $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET)
707 movq $-ENOSYS,RAX(%rsp) /* ptrace can change this for a bad syscall */
708 FIXUP_TOP_OF_STACK %rdi
710 call syscall_trace_enter
712 * Reload arg registers from stack in case ptrace changed them.
713 * We don't reload %rax because syscall_trace_enter() returned
714 * the value it wants us to use in the table lookup.
716 LOAD_ARGS ARGOFFSET, 1
718 #if __SYSCALL_MASK == ~0
719 cmpq $__NR_syscall_max,%rax
721 andl $__SYSCALL_MASK,%eax
722 cmpl $__NR_syscall_max,%eax
724 ja int_ret_from_sys_call /* RAX(%rsp) set to -ENOSYS above */
725 movq %r10,%rcx /* fixup for C */
726 call *sys_call_table(,%rax,8)
727 movq %rax,RAX-ARGOFFSET(%rsp)
728 /* Use IRET because user could have changed frame */
731 * Syscall return path ending with IRET.
732 * Has correct top of stack, but partial stack frame.
734 GLOBAL(int_ret_from_sys_call)
735 DISABLE_INTERRUPTS(CLBR_NONE)
737 movl $_TIF_ALLWORK_MASK,%edi
738 /* edi: mask to check */
739 GLOBAL(int_with_check)
741 GET_THREAD_INFO(%rcx)
742 movl TI_flags(%rcx),%edx
745 andl $~TS_COMPAT,TI_status(%rcx)
748 /* Either reschedule or signal or syscall exit tracking needed. */
749 /* First do a reschedule test. */
750 /* edx: work, edi: workmask */
752 bt $TIF_NEED_RESCHED,%edx
755 ENABLE_INTERRUPTS(CLBR_NONE)
759 DISABLE_INTERRUPTS(CLBR_NONE)
763 /* handle signals and tracing -- both require a full stack frame */
766 ENABLE_INTERRUPTS(CLBR_NONE)
767 int_check_syscall_exit_work:
769 /* Check for syscall exit trace */
770 testl $_TIF_WORK_SYSCALL_EXIT,%edx
773 leaq 8(%rsp),%rdi # &ptregs -> arg1
774 call syscall_trace_leave
776 andl $~(_TIF_WORK_SYSCALL_EXIT|_TIF_SYSCALL_EMU),%edi
780 testl $_TIF_DO_NOTIFY_MASK,%edx
782 movq %rsp,%rdi # &ptregs -> arg1
783 xorl %esi,%esi # oldset -> arg2
784 call do_notify_resume
785 1: movl $_TIF_WORK_MASK,%edi
788 DISABLE_INTERRUPTS(CLBR_NONE)
795 * Certain special system calls that need to save a complete full stack frame.
797 .macro PTREGSCALL label,func,arg
799 PARTIAL_FRAME 1 8 /* offset 8: return address */
800 subq $REST_SKIP, %rsp
801 CFI_ADJUST_CFA_OFFSET REST_SKIP
803 DEFAULT_FRAME 0 8 /* offset 8: return address */
804 leaq 8(%rsp), \arg /* pt_regs pointer */
806 jmp ptregscall_common
811 PTREGSCALL stub_clone, sys_clone, %r8
812 PTREGSCALL stub_fork, sys_fork, %rdi
813 PTREGSCALL stub_vfork, sys_vfork, %rdi
814 PTREGSCALL stub_sigaltstack, sys_sigaltstack, %rdx
815 PTREGSCALL stub_iopl, sys_iopl, %rsi
817 ENTRY(ptregscall_common)
818 DEFAULT_FRAME 1 8 /* offset 8: return address */
819 RESTORE_TOP_OF_STACK %r11, 8
820 movq_cfi_restore R15+8, r15
821 movq_cfi_restore R14+8, r14
822 movq_cfi_restore R13+8, r13
823 movq_cfi_restore R12+8, r12
824 movq_cfi_restore RBP+8, rbp
825 movq_cfi_restore RBX+8, rbx
826 ret $REST_SKIP /* pop extended registers */
828 END(ptregscall_common)
835 FIXUP_TOP_OF_STACK %r11
838 RESTORE_TOP_OF_STACK %r11
841 jmp int_ret_from_sys_call
846 * sigreturn is special because it needs to restore all registers on return.
847 * This cannot be done with SYSRET, so use the IRET return path instead.
849 ENTRY(stub_rt_sigreturn)
855 FIXUP_TOP_OF_STACK %r11
856 call sys_rt_sigreturn
857 movq %rax,RAX(%rsp) # fixme, this could be done at the higher layer
859 jmp int_ret_from_sys_call
861 END(stub_rt_sigreturn)
863 #ifdef CONFIG_X86_X32_ABI
864 PTREGSCALL stub_x32_sigaltstack, sys32_sigaltstack, %rdx
866 ENTRY(stub_x32_rt_sigreturn)
872 FIXUP_TOP_OF_STACK %r11
873 call sys32_x32_rt_sigreturn
874 movq %rax,RAX(%rsp) # fixme, this could be done at the higher layer
876 jmp int_ret_from_sys_call
878 END(stub_x32_rt_sigreturn)
880 ENTRY(stub_x32_execve)
885 FIXUP_TOP_OF_STACK %r11
888 RESTORE_TOP_OF_STACK %r11
891 jmp int_ret_from_sys_call
898 * Build the entry stubs and pointer table with some assembler magic.
899 * We pack 7 stubs into a single 32-byte chunk, which will fit in a
900 * single cache line on all modern x86 implementations.
902 .section .init.rodata,"a"
906 .p2align CONFIG_X86_L1_CACHE_SHIFT
907 ENTRY(irq_entries_start)
909 vector=FIRST_EXTERNAL_VECTOR
910 .rept (NR_VECTORS-FIRST_EXTERNAL_VECTOR+6)/7
913 .if vector < NR_VECTORS
914 .if vector <> FIRST_EXTERNAL_VECTOR
915 CFI_ADJUST_CFA_OFFSET -8
917 1: pushq_cfi $(~vector+0x80) /* Note: always in signed byte range */
918 .if ((vector-FIRST_EXTERNAL_VECTOR)%7) <> 6
927 2: jmp common_interrupt
930 END(irq_entries_start)
937 * Interrupt entry/exit.
939 * Interrupt entry points save only callee clobbered registers in fast path.
941 * Entry runs with interrupts off.
944 /* 0(%rsp): ~(interrupt number) */
945 .macro interrupt func
946 /* reserve pt_regs for scratch regs and rbp */
947 subq $ORIG_RAX-RBP, %rsp
948 CFI_ADJUST_CFA_OFFSET ORIG_RAX-RBP
954 * Interrupt entry/exit should be protected against kprobes
956 .pushsection .kprobes.text, "ax"
958 * The interrupt stubs push (~vector+0x80) onto the stack and
959 * then jump to common_interrupt.
961 .p2align CONFIG_X86_L1_CACHE_SHIFT
964 addq $-0x80,(%rsp) /* Adjust vector to [-256,-1] range */
966 /* 0(%rsp): old_rsp-ARGOFFSET */
968 DISABLE_INTERRUPTS(CLBR_NONE)
970 decl PER_CPU_VAR(irq_count)
972 /* Restore saved previous stack */
974 CFI_DEF_CFA rsi,SS+8-RBP /* reg/off reset after def_cfa_expr */
975 leaq ARGOFFSET-RBP(%rsi), %rsp
976 CFI_DEF_CFA_REGISTER rsp
977 CFI_ADJUST_CFA_OFFSET RBP-ARGOFFSET
980 GET_THREAD_INFO(%rcx)
981 testl $3,CS-ARGOFFSET(%rsp)
984 /* Interrupt came from user space */
986 * Has a correct top of stack, but a partial stack frame
987 * %rcx: thread info. Interrupts off.
989 retint_with_reschedule:
990 movl $_TIF_WORK_MASK,%edi
993 movl TI_flags(%rcx),%edx
998 retint_swapgs: /* return to user-space */
1000 * The iretq could re-enable interrupts:
1002 DISABLE_INTERRUPTS(CLBR_ANY)
1007 retint_restore_args: /* return to kernel space */
1008 DISABLE_INTERRUPTS(CLBR_ANY)
1010 * The iretq could re-enable interrupts:
1018 _ASM_EXTABLE(irq_return, bad_iret)
1020 #ifdef CONFIG_PARAVIRT
1023 _ASM_EXTABLE(native_iret, bad_iret)
1026 .section .fixup,"ax"
1029 * The iret traps when the %cs or %ss being restored is bogus.
1030 * We've lost the original trap vector and error code.
1031 * #GPF is the most likely one to get for an invalid selector.
1032 * So pretend we completed the iret and took the #GPF in user mode.
1034 * We are now running with the kernel GS after exception recovery.
1035 * But error_entry expects us to have user GS to match the user %cs,
1041 jmp general_protection
1045 /* edi: workmask, edx: work */
1048 bt $TIF_NEED_RESCHED,%edx
1051 ENABLE_INTERRUPTS(CLBR_NONE)
1055 GET_THREAD_INFO(%rcx)
1056 DISABLE_INTERRUPTS(CLBR_NONE)
1061 testl $_TIF_DO_NOTIFY_MASK,%edx
1064 ENABLE_INTERRUPTS(CLBR_NONE)
1066 movq $-1,ORIG_RAX(%rsp)
1067 xorl %esi,%esi # oldset
1068 movq %rsp,%rdi # &pt_regs
1069 call do_notify_resume
1071 DISABLE_INTERRUPTS(CLBR_NONE)
1073 GET_THREAD_INFO(%rcx)
1074 jmp retint_with_reschedule
1076 #ifdef CONFIG_PREEMPT
1077 /* Returning to kernel space. Check if we need preemption */
1078 /* rcx: threadinfo. interrupts off. */
1079 ENTRY(retint_kernel)
1080 cmpl $0,TI_preempt_count(%rcx)
1081 jnz retint_restore_args
1082 bt $TIF_NEED_RESCHED,TI_flags(%rcx)
1083 jnc retint_restore_args
1084 bt $9,EFLAGS-ARGOFFSET(%rsp) /* interrupts off? */
1085 jnc retint_restore_args
1086 call preempt_schedule_irq
1091 END(common_interrupt)
1093 * End of kprobes section
1100 .macro apicinterrupt num sym do_sym
1112 apicinterrupt IRQ_MOVE_CLEANUP_VECTOR \
1113 irq_move_cleanup_interrupt smp_irq_move_cleanup_interrupt
1114 apicinterrupt REBOOT_VECTOR \
1115 reboot_interrupt smp_reboot_interrupt
1118 #ifdef CONFIG_X86_UV
1119 apicinterrupt UV_BAU_MESSAGE \
1120 uv_bau_message_intr1 uv_bau_message_interrupt
1122 apicinterrupt LOCAL_TIMER_VECTOR \
1123 apic_timer_interrupt smp_apic_timer_interrupt
1124 apicinterrupt X86_PLATFORM_IPI_VECTOR \
1125 x86_platform_ipi smp_x86_platform_ipi
1130 .irp idx,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15, \
1131 16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31
1132 .if NUM_INVALIDATE_TLB_VECTORS > \idx
1133 ENTRY(invalidate_interrupt\idx)
1134 pushq_cfi $~(INVALIDATE_TLB_VECTOR_START+\idx)
1135 jmp .Lcommon_invalidate_interrupt0
1136 CFI_ADJUST_CFA_OFFSET -8
1137 END(invalidate_interrupt\idx)
1141 apicinterrupt INVALIDATE_TLB_VECTOR_START, \
1142 invalidate_interrupt0, smp_invalidate_interrupt
1145 apicinterrupt THRESHOLD_APIC_VECTOR \
1146 threshold_interrupt smp_threshold_interrupt
1147 apicinterrupt THERMAL_APIC_VECTOR \
1148 thermal_interrupt smp_thermal_interrupt
1151 apicinterrupt CALL_FUNCTION_SINGLE_VECTOR \
1152 call_function_single_interrupt smp_call_function_single_interrupt
1153 apicinterrupt CALL_FUNCTION_VECTOR \
1154 call_function_interrupt smp_call_function_interrupt
1155 apicinterrupt RESCHEDULE_VECTOR \
1156 reschedule_interrupt smp_reschedule_interrupt
1159 apicinterrupt ERROR_APIC_VECTOR \
1160 error_interrupt smp_error_interrupt
1161 apicinterrupt SPURIOUS_APIC_VECTOR \
1162 spurious_interrupt smp_spurious_interrupt
1164 #ifdef CONFIG_IRQ_WORK
1165 apicinterrupt IRQ_WORK_VECTOR \
1166 irq_work_interrupt smp_irq_work_interrupt
1170 * Exception entry points.
1172 .macro zeroentry sym do_sym
1175 PARAVIRT_ADJUST_EXCEPTION_FRAME
1176 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1177 subq $ORIG_RAX-R15, %rsp
1178 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1181 movq %rsp,%rdi /* pt_regs pointer */
1182 xorl %esi,%esi /* no error code */
1184 jmp error_exit /* %ebx: no swapgs flag */
1189 .macro paranoidzeroentry sym do_sym
1192 PARAVIRT_ADJUST_EXCEPTION_FRAME
1193 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1194 subq $ORIG_RAX-R15, %rsp
1195 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1198 movq %rsp,%rdi /* pt_regs pointer */
1199 xorl %esi,%esi /* no error code */
1201 jmp paranoid_exit /* %ebx: no swapgs flag */
1206 #define INIT_TSS_IST(x) PER_CPU_VAR(init_tss) + (TSS_ist + ((x) - 1) * 8)
1207 .macro paranoidzeroentry_ist sym do_sym ist
1210 PARAVIRT_ADJUST_EXCEPTION_FRAME
1211 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1212 subq $ORIG_RAX-R15, %rsp
1213 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1215 TRACE_IRQS_OFF_DEBUG
1216 movq %rsp,%rdi /* pt_regs pointer */
1217 xorl %esi,%esi /* no error code */
1218 subq $EXCEPTION_STKSZ, INIT_TSS_IST(\ist)
1220 addq $EXCEPTION_STKSZ, INIT_TSS_IST(\ist)
1221 jmp paranoid_exit /* %ebx: no swapgs flag */
1226 .macro errorentry sym do_sym
1229 PARAVIRT_ADJUST_EXCEPTION_FRAME
1230 subq $ORIG_RAX-R15, %rsp
1231 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1234 movq %rsp,%rdi /* pt_regs pointer */
1235 movq ORIG_RAX(%rsp),%rsi /* get error code */
1236 movq $-1,ORIG_RAX(%rsp) /* no syscall to restart */
1238 jmp error_exit /* %ebx: no swapgs flag */
1243 /* error code is on the stack already */
1244 .macro paranoiderrorentry sym do_sym
1247 PARAVIRT_ADJUST_EXCEPTION_FRAME
1248 subq $ORIG_RAX-R15, %rsp
1249 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1253 movq %rsp,%rdi /* pt_regs pointer */
1254 movq ORIG_RAX(%rsp),%rsi /* get error code */
1255 movq $-1,ORIG_RAX(%rsp) /* no syscall to restart */
1257 jmp paranoid_exit /* %ebx: no swapgs flag */
1262 zeroentry divide_error do_divide_error
1263 zeroentry overflow do_overflow
1264 zeroentry bounds do_bounds
1265 zeroentry invalid_op do_invalid_op
1266 zeroentry device_not_available do_device_not_available
1267 paranoiderrorentry double_fault do_double_fault
1268 zeroentry coprocessor_segment_overrun do_coprocessor_segment_overrun
1269 errorentry invalid_TSS do_invalid_TSS
1270 errorentry segment_not_present do_segment_not_present
1271 zeroentry spurious_interrupt_bug do_spurious_interrupt_bug
1272 zeroentry coprocessor_error do_coprocessor_error
1273 errorentry alignment_check do_alignment_check
1274 zeroentry simd_coprocessor_error do_simd_coprocessor_error
1277 /* Reload gs selector with exception handling */
1278 /* edi: new selector */
1279 ENTRY(native_load_gs_index)
1282 DISABLE_INTERRUPTS(CLBR_ANY & ~CLBR_RDI)
1286 2: mfence /* workaround */
1291 END(native_load_gs_index)
1293 _ASM_EXTABLE(gs_change,bad_gs)
1294 .section .fixup,"ax"
1295 /* running with kernelgs */
1297 SWAPGS /* switch back to user gs */
1303 ENTRY(kernel_thread_helper)
1304 pushq $0 # fake return address
1307 * Here we are in the child and the registers are set as they were
1308 * at kernel_thread() invocation in the parent.
1314 ud2 # padding for call trace
1316 END(kernel_thread_helper)
1319 * execve(). This function needs to use IRET, not SYSRET, to set up all state properly.
1321 * C extern interface:
1322 * extern long execve(const char *name, char **argv, char **envp)
1324 * asm input arguments:
1325 * rdi: name, rsi: argv, rdx: envp
1327 * We want to fallback into:
1328 * extern long sys_execve(const char *name, char **argv,char **envp, struct pt_regs *regs)
1330 * do_sys_execve asm fallback arguments:
1331 * rdi: name, rsi: argv, rdx: envp, rcx: fake frame on the stack
1333 ENTRY(kernel_execve)
1339 movq %rax, RAX(%rsp)
1342 je int_ret_from_sys_call
1349 /* Call softirq on interrupt stack. Interrupts are off. */
1353 CFI_REL_OFFSET rbp,0
1355 CFI_DEF_CFA_REGISTER rbp
1356 incl PER_CPU_VAR(irq_count)
1357 cmove PER_CPU_VAR(irq_stack_ptr),%rsp
1358 push %rbp # backlink for old unwinder
1362 CFI_DEF_CFA_REGISTER rsp
1363 CFI_ADJUST_CFA_OFFSET -8
1364 decl PER_CPU_VAR(irq_count)
1370 zeroentry xen_hypervisor_callback xen_do_hypervisor_callback
1373 * A note on the "critical region" in our callback handler.
1374 * We want to avoid stacking callback handlers due to events occurring
1375 * during handling of the last event. To do this, we keep events disabled
1376 * until we've done all processing. HOWEVER, we must enable events before
1377 * popping the stack frame (can't be done atomically) and so it would still
1378 * be possible to get enough handler activations to overflow the stack.
1379 * Although unlikely, bugs of that kind are hard to track down, so we'd
1380 * like to avoid the possibility.
1381 * So, on entry to the handler we detect whether we interrupted an
1382 * existing activation in its critical region -- if so, we pop the current
1383 * activation and restart the handler using the previous one.
1385 ENTRY(xen_do_hypervisor_callback) # do_hypervisor_callback(struct *pt_regs)
1388 * Since we don't modify %rdi, evtchn_do_upall(struct *pt_regs) will
1389 * see the correct pointer to the pt_regs
1391 movq %rdi, %rsp # we don't return, adjust the stack frame
1394 11: incl PER_CPU_VAR(irq_count)
1396 CFI_DEF_CFA_REGISTER rbp
1397 cmovzq PER_CPU_VAR(irq_stack_ptr),%rsp
1398 pushq %rbp # backlink for old unwinder
1399 call xen_evtchn_do_upcall
1401 CFI_DEF_CFA_REGISTER rsp
1402 decl PER_CPU_VAR(irq_count)
1405 END(xen_do_hypervisor_callback)
1408 * Hypervisor uses this for application faults while it executes.
1409 * We get here for two reasons:
1410 * 1. Fault while reloading DS, ES, FS or GS
1411 * 2. Fault while executing IRET
1412 * Category 1 we do not need to fix up as Xen has already reloaded all segment
1413 * registers that could be reloaded and zeroed the others.
1414 * Category 2 we fix up by killing the current process. We cannot use the
1415 * normal Linux return path in this case because if we use the IRET hypercall
1416 * to pop the stack frame we end up in an infinite loop of failsafe callbacks.
1417 * We distinguish between categories by comparing each saved segment register
1418 * with its current contents: any discrepancy means we in category 1.
1420 ENTRY(xen_failsafe_callback)
1422 /*CFI_REL_OFFSET gs,GS*/
1423 /*CFI_REL_OFFSET fs,FS*/
1424 /*CFI_REL_OFFSET es,ES*/
1425 /*CFI_REL_OFFSET ds,DS*/
1426 CFI_REL_OFFSET r11,8
1427 CFI_REL_OFFSET rcx,0
1441 /* All segments match their saved values => Category 2 (Bad IRET). */
1447 CFI_ADJUST_CFA_OFFSET -0x30
1448 pushq_cfi $0 /* RIP */
1451 jmp general_protection
1453 1: /* Segment mismatch => Category 1 (Bad segment). Retry the IRET. */
1459 CFI_ADJUST_CFA_OFFSET -0x30
1464 END(xen_failsafe_callback)
1466 apicinterrupt XEN_HVM_EVTCHN_CALLBACK \
1467 xen_hvm_callback_vector xen_evtchn_do_upcall
1469 #endif /* CONFIG_XEN */
1472 * Some functions should be protected against kprobes
1474 .pushsection .kprobes.text, "ax"
1476 paranoidzeroentry_ist debug do_debug DEBUG_STACK
1477 paranoidzeroentry_ist int3 do_int3 DEBUG_STACK
1478 paranoiderrorentry stack_segment do_stack_segment
1480 zeroentry xen_debug do_debug
1481 zeroentry xen_int3 do_int3
1482 errorentry xen_stack_segment do_stack_segment
1484 errorentry general_protection do_general_protection
1485 errorentry page_fault do_page_fault
1486 #ifdef CONFIG_KVM_GUEST
1487 errorentry async_page_fault do_async_page_fault
1489 #ifdef CONFIG_X86_MCE
1490 paranoidzeroentry machine_check *machine_check_vector(%rip)
1494 * "Paranoid" exit path from exception stack.
1495 * Paranoid because this is used by NMIs and cannot take
1496 * any kernel state for granted.
1497 * We don't do kernel preemption checks here, because only
1498 * NMI should be common and it does not enable IRQs and
1499 * cannot get reschedule ticks.
1501 * "trace" is 0 for the NMI handler only, because irq-tracing
1502 * is fundamentally NMI-unsafe. (we cannot change the soft and
1503 * hard flags at once, atomically)
1506 /* ebx: no swapgs flag */
1507 ENTRY(paranoid_exit)
1509 DISABLE_INTERRUPTS(CLBR_NONE)
1510 TRACE_IRQS_OFF_DEBUG
1511 testl %ebx,%ebx /* swapgs needed? */
1512 jnz paranoid_restore
1514 jnz paranoid_userspace
1521 TRACE_IRQS_IRETQ_DEBUG 0
1525 GET_THREAD_INFO(%rcx)
1526 movl TI_flags(%rcx),%ebx
1527 andl $_TIF_WORK_MASK,%ebx
1529 movq %rsp,%rdi /* &pt_regs */
1531 movq %rax,%rsp /* switch stack for scheduling */
1532 testl $_TIF_NEED_RESCHED,%ebx
1533 jnz paranoid_schedule
1534 movl %ebx,%edx /* arg3: thread flags */
1536 ENABLE_INTERRUPTS(CLBR_NONE)
1537 xorl %esi,%esi /* arg2: oldset */
1538 movq %rsp,%rdi /* arg1: &pt_regs */
1539 call do_notify_resume
1540 DISABLE_INTERRUPTS(CLBR_NONE)
1542 jmp paranoid_userspace
1545 ENABLE_INTERRUPTS(CLBR_ANY)
1547 DISABLE_INTERRUPTS(CLBR_ANY)
1549 jmp paranoid_userspace
1554 * Exception entry point. This expects an error code/orig_rax on the stack.
1555 * returns in "no swapgs flag" in %ebx.
1559 CFI_ADJUST_CFA_OFFSET 15*8
1560 /* oldrax contains error code */
1579 je error_kernelspace
1587 * There are two places in the kernel that can potentially fault with
1588 * usergs. Handle them here. The exception handlers after iret run with
1589 * kernel gs again, so don't set the user space flag. B stepping K8s
1590 * sometimes report an truncated RIP for IRET exceptions returning to
1591 * compat mode. Check for these here too.
1595 leaq irq_return(%rip),%rcx
1596 cmpq %rcx,RIP+8(%rsp)
1598 movl %ecx,%eax /* zero extend */
1599 cmpq %rax,RIP+8(%rsp)
1601 cmpq $gs_change,RIP+8(%rsp)
1606 /* Fix truncated RIP */
1607 movq %rcx,RIP+8(%rsp)
1613 /* ebx: no swapgs flag (1: don't need swapgs, 0: need it) */
1618 DISABLE_INTERRUPTS(CLBR_NONE)
1620 GET_THREAD_INFO(%rcx)
1623 LOCKDEP_SYS_EXIT_IRQ
1624 movl TI_flags(%rcx),%edx
1625 movl $_TIF_WORK_MASK,%edi
1633 * Test if a given stack is an NMI stack or not.
1635 .macro test_in_nmi reg stack nmi_ret normal_ret
1638 subq $EXCEPTION_STKSZ, %\reg
1644 /* runs on exception stack */
1647 PARAVIRT_ADJUST_EXCEPTION_FRAME
1649 * We allow breakpoints in NMIs. If a breakpoint occurs, then
1650 * the iretq it performs will take us out of NMI context.
1651 * This means that we can have nested NMIs where the next
1652 * NMI is using the top of the stack of the previous NMI. We
1653 * can't let it execute because the nested NMI will corrupt the
1654 * stack of the previous NMI. NMI handlers are not re-entrant
1657 * To handle this case we do the following:
1658 * Check the a special location on the stack that contains
1659 * a variable that is set when NMIs are executing.
1660 * The interrupted task's stack is also checked to see if it
1662 * If the variable is not set and the stack is not the NMI
1664 * o Set the special variable on the stack
1665 * o Copy the interrupt frame into a "saved" location on the stack
1666 * o Copy the interrupt frame into a "copy" location on the stack
1667 * o Continue processing the NMI
1668 * If the variable is set or the previous stack is the NMI stack:
1669 * o Modify the "copy" location to jump to the repeate_nmi
1670 * o return back to the first NMI
1672 * Now on exit of the first NMI, we first clear the stack variable
1673 * The NMI stack will tell any nested NMIs at that point that it is
1674 * nested. Then we pop the stack normally with iret, and if there was
1675 * a nested NMI that updated the copy interrupt stack frame, a
1676 * jump will be made to the repeat_nmi code that will handle the second
1680 /* Use %rdx as out temp variable throughout */
1682 CFI_REL_OFFSET rdx, 0
1685 * If %cs was not the kernel segment, then the NMI triggered in user
1686 * space, which means it is definitely not nested.
1688 cmpl $__KERNEL_CS, 16(%rsp)
1692 * Check the special variable on the stack to see if NMIs are
1699 * Now test if the previous stack was an NMI stack.
1700 * We need the double check. We check the NMI stack to satisfy the
1701 * race when the first NMI clears the variable before returning.
1702 * We check the variable because the first NMI could be in a
1703 * breakpoint routine using a breakpoint stack.
1706 test_in_nmi rdx, 4*8(%rsp), nested_nmi, first_nmi
1711 * Do nothing if we interrupted the fixup in repeat_nmi.
1712 * It's about to repeat the NMI handler, so we are fine
1713 * with ignoring this one.
1715 movq $repeat_nmi, %rdx
1718 movq $end_repeat_nmi, %rdx
1723 /* Set up the interrupted NMIs stack to jump to repeat_nmi */
1724 leaq -6*8(%rsp), %rdx
1726 CFI_ADJUST_CFA_OFFSET 6*8
1727 pushq_cfi $__KERNEL_DS
1730 pushq_cfi $__KERNEL_CS
1731 pushq_cfi $repeat_nmi
1733 /* Put stack back */
1735 CFI_ADJUST_CFA_OFFSET -11*8
1741 /* No need to check faults here */
1747 * Because nested NMIs will use the pushed location that we
1748 * stored in rdx, we must keep that space available.
1749 * Here's what our stack frame will look like:
1750 * +-------------------------+
1752 * | original Return RSP |
1753 * | original RFLAGS |
1756 * +-------------------------+
1757 * | temp storage for rdx |
1758 * +-------------------------+
1759 * | NMI executing variable |
1760 * +-------------------------+
1762 * | Saved Return RSP |
1766 * +-------------------------+
1768 * | copied Return RSP |
1772 * +-------------------------+
1774 * +-------------------------+
1776 * The saved stack frame is used to fix up the copied stack frame
1777 * that a nested NMI may change to make the interrupted NMI iret jump
1778 * to the repeat_nmi. The original stack frame and the temp storage
1779 * is also used by nested NMIs and can not be trusted on exit.
1781 /* Do not pop rdx, nested NMIs will corrupt that part of the stack */
1785 /* Set the NMI executing variable on the stack. */
1788 /* Copy the stack frame to the Saved frame */
1792 CFI_DEF_CFA_OFFSET SS+8-RIP
1794 /* Everything up to here is safe from nested NMIs */
1797 * If there was a nested NMI, the first NMI's iret will return
1798 * here. But NMIs are still enabled and we can take another
1799 * nested NMI. The nested NMI checks the interrupted RIP to see
1800 * if it is between repeat_nmi and end_repeat_nmi, and if so
1801 * it will just return, as we are about to repeat an NMI anyway.
1802 * This makes it safe to copy to the stack frame that a nested
1807 * Update the stack variable to say we are still in NMI (the update
1808 * is benign for the non-repeat case, where 1 was pushed just above
1809 * to this very stack slot).
1813 /* Make another copy, this one may be modified by nested NMIs */
1817 CFI_DEF_CFA_OFFSET SS+8-RIP
1821 * Everything below this point can be preempted by a nested
1822 * NMI if the first NMI took an exception and reset our iret stack
1823 * so that we repeat another NMI.
1825 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1826 subq $ORIG_RAX-R15, %rsp
1827 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1829 * Use save_paranoid to handle SWAPGS, but no need to use paranoid_exit
1830 * as we should not be calling schedule in NMI context.
1831 * Even with normal interrupts enabled. An NMI should not be
1832 * setting NEED_RESCHED or anything that normal interrupts and
1833 * exceptions might do.
1839 * Save off the CR2 register. If we take a page fault in the NMI then
1840 * it could corrupt the CR2 value. If the NMI preempts a page fault
1841 * handler before it was able to read the CR2 register, and then the
1842 * NMI itself takes a page fault, the page fault that was preempted
1843 * will read the information from the NMI page fault and not the
1844 * origin fault. Save it off and restore it if it changes.
1845 * Use the r12 callee-saved register.
1849 /* paranoidentry do_nmi, 0; without TRACE_IRQS_OFF */
1854 /* Did the NMI take a page fault? Restore cr2 if it did */
1861 testl %ebx,%ebx /* swapgs needed? */
1867 /* Clear the NMI executing stack variable */
1873 ENTRY(ignore_sysret)
1881 * End of kprobes section