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 <asm/context_tracking.h>
61 #include <asm/pgtable_types.h>
62 #include <linux/err.h>
64 /* Avoid __ASSEMBLER__'ifying <linux/audit.h> just for this. */
65 #include <linux/elf-em.h>
66 #define AUDIT_ARCH_X86_64 (EM_X86_64|__AUDIT_ARCH_64BIT|__AUDIT_ARCH_LE)
67 #define __AUDIT_ARCH_64BIT 0x80000000
68 #define __AUDIT_ARCH_LE 0x40000000
71 .section .entry.text, "ax"
73 #ifdef CONFIG_FUNCTION_TRACER
75 #ifdef CC_USING_FENTRY
76 # define function_hook __fentry__
78 # define function_hook mcount
81 #ifdef CONFIG_DYNAMIC_FTRACE
87 /* skip is set if stack has been adjusted */
88 .macro ftrace_caller_setup skip=0
89 MCOUNT_SAVE_FRAME \skip
91 /* Load the ftrace_ops into the 3rd parameter */
92 movq function_trace_op(%rip), %rdx
94 /* Load ip into the first parameter */
96 subq $MCOUNT_INSN_SIZE, %rdi
97 /* Load the parent_ip into the second parameter */
98 #ifdef CC_USING_FENTRY
99 movq SS+16(%rsp), %rsi
106 /* Check if tracing was disabled (quick check) */
107 cmpl $0, function_trace_stop
111 /* regs go into 4th parameter (but make it NULL) */
120 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
121 GLOBAL(ftrace_graph_call)
129 ENTRY(ftrace_regs_caller)
130 /* Save the current flags before compare (in SS location)*/
133 /* Check if tracing was disabled (quick check) */
134 cmpl $0, function_trace_stop
135 jne ftrace_restore_flags
137 /* skip=8 to skip flags saved in SS */
138 ftrace_caller_setup 8
140 /* Save the rest of pt_regs */
149 /* Copy saved flags */
151 movq %rcx, EFLAGS(%rsp)
152 /* Kernel segments */
153 movq $__KERNEL_DS, %rcx
155 movq $__KERNEL_CS, %rcx
157 /* Stack - skipping return address */
158 leaq SS+16(%rsp), %rcx
161 /* regs go into 4th parameter */
164 GLOBAL(ftrace_regs_call)
167 /* Copy flags back to SS, to restore them */
168 movq EFLAGS(%rsp), %rax
171 /* Handlers can change the RIP */
173 movq %rax, SS+8(%rsp)
175 /* restore the rest of pt_regs */
184 /* skip=8 to skip flags saved in SS */
185 MCOUNT_RESTORE_FRAME 8
191 ftrace_restore_flags:
195 END(ftrace_regs_caller)
198 #else /* ! CONFIG_DYNAMIC_FTRACE */
201 cmpl $0, function_trace_stop
204 cmpq $ftrace_stub, ftrace_trace_function
207 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
208 cmpq $ftrace_stub, ftrace_graph_return
209 jnz ftrace_graph_caller
211 cmpq $ftrace_graph_entry_stub, ftrace_graph_entry
212 jnz ftrace_graph_caller
222 #ifdef CC_USING_FENTRY
223 movq SS+16(%rsp), %rsi
227 subq $MCOUNT_INSN_SIZE, %rdi
229 call *ftrace_trace_function
235 #endif /* CONFIG_DYNAMIC_FTRACE */
236 #endif /* CONFIG_FUNCTION_TRACER */
238 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
239 ENTRY(ftrace_graph_caller)
242 #ifdef CC_USING_FENTRY
243 leaq SS+16(%rsp), %rdi
244 movq $0, %rdx /* No framepointers needed */
250 subq $MCOUNT_INSN_SIZE, %rsi
252 call prepare_ftrace_return
257 END(ftrace_graph_caller)
259 GLOBAL(return_to_handler)
262 /* Save the return values */
267 call ftrace_return_to_handler
277 #ifndef CONFIG_PREEMPT
278 #define retint_kernel retint_restore_args
281 #ifdef CONFIG_PARAVIRT
282 ENTRY(native_usergs_sysret64)
285 ENDPROC(native_usergs_sysret64)
286 #endif /* CONFIG_PARAVIRT */
289 .macro TRACE_IRQS_IRETQ offset=ARGOFFSET
290 #ifdef CONFIG_TRACE_IRQFLAGS
291 bt $9,EFLAGS-\offset(%rsp) /* interrupts off? */
299 * When dynamic function tracer is enabled it will add a breakpoint
300 * to all locations that it is about to modify, sync CPUs, update
301 * all the code, sync CPUs, then remove the breakpoints. In this time
302 * if lockdep is enabled, it might jump back into the debug handler
303 * outside the updating of the IST protection. (TRACE_IRQS_ON/OFF).
305 * We need to change the IDT table before calling TRACE_IRQS_ON/OFF to
306 * make sure the stack pointer does not get reset back to the top
307 * of the debug stack, and instead just reuses the current stack.
309 #if defined(CONFIG_DYNAMIC_FTRACE) && defined(CONFIG_TRACE_IRQFLAGS)
311 .macro TRACE_IRQS_OFF_DEBUG
312 call debug_stack_set_zero
314 call debug_stack_reset
317 .macro TRACE_IRQS_ON_DEBUG
318 call debug_stack_set_zero
320 call debug_stack_reset
323 .macro TRACE_IRQS_IRETQ_DEBUG offset=ARGOFFSET
324 bt $9,EFLAGS-\offset(%rsp) /* interrupts off? */
331 # define TRACE_IRQS_OFF_DEBUG TRACE_IRQS_OFF
332 # define TRACE_IRQS_ON_DEBUG TRACE_IRQS_ON
333 # define TRACE_IRQS_IRETQ_DEBUG TRACE_IRQS_IRETQ
337 * C code is not supposed to know about undefined top of stack. Every time
338 * a C function with an pt_regs argument is called from the SYSCALL based
339 * fast path FIXUP_TOP_OF_STACK is needed.
340 * RESTORE_TOP_OF_STACK syncs the syscall state after any possible ptregs
344 /* %rsp:at FRAMEEND */
345 .macro FIXUP_TOP_OF_STACK tmp offset=0
346 movq PER_CPU_VAR(old_rsp),\tmp
347 movq \tmp,RSP+\offset(%rsp)
348 movq $__USER_DS,SS+\offset(%rsp)
349 movq $__USER_CS,CS+\offset(%rsp)
350 movq $-1,RCX+\offset(%rsp)
351 movq R11+\offset(%rsp),\tmp /* get eflags */
352 movq \tmp,EFLAGS+\offset(%rsp)
355 .macro RESTORE_TOP_OF_STACK tmp offset=0
356 movq RSP+\offset(%rsp),\tmp
357 movq \tmp,PER_CPU_VAR(old_rsp)
358 movq EFLAGS+\offset(%rsp),\tmp
359 movq \tmp,R11+\offset(%rsp)
362 .macro FAKE_STACK_FRAME child_rip
363 /* push in order ss, rsp, eflags, cs, rip */
365 pushq_cfi $__KERNEL_DS /* ss */
366 /*CFI_REL_OFFSET ss,0*/
367 pushq_cfi %rax /* rsp */
369 pushq_cfi $(X86_EFLAGS_IF|X86_EFLAGS_FIXED) /* eflags - interrupts on */
370 /*CFI_REL_OFFSET rflags,0*/
371 pushq_cfi $__KERNEL_CS /* cs */
372 /*CFI_REL_OFFSET cs,0*/
373 pushq_cfi \child_rip /* rip */
375 pushq_cfi %rax /* orig rax */
378 .macro UNFAKE_STACK_FRAME
380 CFI_ADJUST_CFA_OFFSET -(6*8)
384 * initial frame state for interrupts (and exceptions without error code)
386 .macro EMPTY_FRAME start=1 offset=0
390 CFI_DEF_CFA rsp,8+\offset
392 CFI_DEF_CFA_OFFSET 8+\offset
397 * initial frame state for interrupts (and exceptions without error code)
399 .macro INTR_FRAME start=1 offset=0
400 EMPTY_FRAME \start, SS+8+\offset-RIP
401 /*CFI_REL_OFFSET ss, SS+\offset-RIP*/
402 CFI_REL_OFFSET rsp, RSP+\offset-RIP
403 /*CFI_REL_OFFSET rflags, EFLAGS+\offset-RIP*/
404 /*CFI_REL_OFFSET cs, CS+\offset-RIP*/
405 CFI_REL_OFFSET rip, RIP+\offset-RIP
409 * initial frame state for exceptions with error code (and interrupts
410 * with vector already pushed)
412 .macro XCPT_FRAME start=1 offset=0
413 INTR_FRAME \start, RIP+\offset-ORIG_RAX
414 /*CFI_REL_OFFSET orig_rax, ORIG_RAX-ORIG_RAX*/
418 * frame that enables calling into C.
420 .macro PARTIAL_FRAME start=1 offset=0
421 XCPT_FRAME \start, ORIG_RAX+\offset-ARGOFFSET
422 CFI_REL_OFFSET rdi, RDI+\offset-ARGOFFSET
423 CFI_REL_OFFSET rsi, RSI+\offset-ARGOFFSET
424 CFI_REL_OFFSET rdx, RDX+\offset-ARGOFFSET
425 CFI_REL_OFFSET rcx, RCX+\offset-ARGOFFSET
426 CFI_REL_OFFSET rax, RAX+\offset-ARGOFFSET
427 CFI_REL_OFFSET r8, R8+\offset-ARGOFFSET
428 CFI_REL_OFFSET r9, R9+\offset-ARGOFFSET
429 CFI_REL_OFFSET r10, R10+\offset-ARGOFFSET
430 CFI_REL_OFFSET r11, R11+\offset-ARGOFFSET
434 * frame that enables passing a complete pt_regs to a C function.
436 .macro DEFAULT_FRAME start=1 offset=0
437 PARTIAL_FRAME \start, R11+\offset-R15
438 CFI_REL_OFFSET rbx, RBX+\offset
439 CFI_REL_OFFSET rbp, RBP+\offset
440 CFI_REL_OFFSET r12, R12+\offset
441 CFI_REL_OFFSET r13, R13+\offset
442 CFI_REL_OFFSET r14, R14+\offset
443 CFI_REL_OFFSET r15, R15+\offset
446 /* save partial stack frame */
449 /* start from rbp in pt_regs and jump over */
450 movq_cfi rdi, (RDI-RBP)
451 movq_cfi rsi, (RSI-RBP)
452 movq_cfi rdx, (RDX-RBP)
453 movq_cfi rcx, (RCX-RBP)
454 movq_cfi rax, (RAX-RBP)
455 movq_cfi r8, (R8-RBP)
456 movq_cfi r9, (R9-RBP)
457 movq_cfi r10, (R10-RBP)
458 movq_cfi r11, (R11-RBP)
460 /* Save rbp so that we can unwind from get_irq_regs() */
463 /* Save previous stack value */
466 leaq -RBP(%rsp),%rdi /* arg1 for handler */
467 testl $3, CS-RBP(%rsi)
471 * irq_count is used to check if a CPU is already on an interrupt stack
472 * or not. While this is essentially redundant with preempt_count it is
473 * a little cheaper to use a separate counter in the PDA (short of
474 * moving irq_enter into assembly, which would be too much work)
476 1: incl PER_CPU_VAR(irq_count)
477 cmovzq PER_CPU_VAR(irq_stack_ptr),%rsp
478 CFI_DEF_CFA_REGISTER rsi
480 /* Store previous stack value */
482 CFI_ESCAPE 0x0f /* DW_CFA_def_cfa_expression */, 6, \
483 0x77 /* DW_OP_breg7 */, 0, \
484 0x06 /* DW_OP_deref */, \
485 0x08 /* DW_OP_const1u */, SS+8-RBP, \
486 0x22 /* DW_OP_plus */
487 /* We entered an interrupt context - irqs are off: */
492 PARTIAL_FRAME 1 (REST_SKIP+8)
493 movq 5*8+16(%rsp), %r11 /* save return address */
500 movq %r11, 8(%rsp) /* return address */
501 FIXUP_TOP_OF_STACK %r11, 16
506 /* save complete stack frame */
507 .pushsection .kprobes.text, "ax"
527 movl $MSR_GS_BASE,%ecx
530 js 1f /* negative -> in kernel */
539 * A newly forked process directly context switches into this address.
541 * rdi: prev task we switched from
546 LOCK ; btr $TIF_FORK,TI_flags(%r8)
549 popfq_cfi # reset kernel eflags
551 call schedule_tail # rdi: 'prev' task parameter
553 GET_THREAD_INFO(%rcx)
557 testl $3, CS-ARGOFFSET(%rsp) # from kernel_thread?
560 testl $_TIF_IA32, TI_flags(%rcx) # 32-bit compat task needs IRET
561 jnz int_ret_from_sys_call
563 RESTORE_TOP_OF_STACK %rdi, -ARGOFFSET
564 jmp ret_from_sys_call # go to the SYSRET fastpath
567 subq $REST_SKIP, %rsp # leave space for volatiles
568 CFI_ADJUST_CFA_OFFSET REST_SKIP
573 jmp int_ret_from_sys_call
578 * System call entry. Up to 6 arguments in registers are supported.
580 * SYSCALL does not save anything on the stack and does not change the
581 * stack pointer. However, it does mask the flags register for us, so
582 * CLD and CLAC are not needed.
587 * rax system call number
589 * rcx return address for syscall/sysret, C arg3
592 * r10 arg3 (--> moved to rcx for C)
595 * r11 eflags for syscall/sysret, temporary for C
596 * r12-r15,rbp,rbx saved by C code, not touched.
598 * Interrupts are off on entry.
599 * Only called from user space.
601 * XXX if we had a free scratch register we could save the RSP into the stack frame
602 * and report it properly in ps. Unfortunately we haven't.
604 * When user can change the frames always force IRET. That is because
605 * it deals with uncanonical addresses better. SYSRET has trouble
606 * with them due to bugs in both AMD and Intel CPUs.
612 CFI_DEF_CFA rsp,KERNEL_STACK_OFFSET
614 /*CFI_REGISTER rflags,r11*/
617 * A hypervisor implementation might want to use a label
618 * after the swapgs, so that it can do the swapgs
619 * for the guest and jump here on syscall.
621 GLOBAL(system_call_after_swapgs)
623 movq %rsp,PER_CPU_VAR(old_rsp)
624 movq PER_CPU_VAR(kernel_stack),%rsp
626 * No need to follow this irqs off/on section - it's straight
629 ENABLE_INTERRUPTS(CLBR_NONE)
631 movq %rax,ORIG_RAX-ARGOFFSET(%rsp)
632 movq %rcx,RIP-ARGOFFSET(%rsp)
633 CFI_REL_OFFSET rip,RIP-ARGOFFSET
634 testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET)
636 system_call_fastpath:
637 #if __SYSCALL_MASK == ~0
638 cmpq $__NR_syscall_max,%rax
640 andl $__SYSCALL_MASK,%eax
641 cmpl $__NR_syscall_max,%eax
645 call *sys_call_table(,%rax,8) # XXX: rip relative
646 movq %rax,RAX-ARGOFFSET(%rsp)
648 * Syscall return path ending with SYSRET (fast path)
649 * Has incomplete stack frame and undefined top of stack.
652 movl $_TIF_ALLWORK_MASK,%edi
656 DISABLE_INTERRUPTS(CLBR_NONE)
658 movl TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET),%edx
663 * sysretq will re-enable interrupts:
666 movq RIP-ARGOFFSET(%rsp),%rcx
668 RESTORE_ARGS 1,-ARG_SKIP,0
669 /*CFI_REGISTER rflags,r11*/
670 movq PER_CPU_VAR(old_rsp), %rsp
674 /* Handle reschedules */
675 /* edx: work, edi: workmask */
677 bt $TIF_NEED_RESCHED,%edx
680 ENABLE_INTERRUPTS(CLBR_NONE)
686 /* Handle a signal */
689 ENABLE_INTERRUPTS(CLBR_NONE)
690 #ifdef CONFIG_AUDITSYSCALL
691 bt $TIF_SYSCALL_AUDIT,%edx
695 * We have a signal, or exit tracing or single-step.
696 * These all wind up with the iret return path anyway,
697 * so just join that path right now.
699 FIXUP_TOP_OF_STACK %r11, -ARGOFFSET
700 jmp int_check_syscall_exit_work
703 movq $-ENOSYS,RAX-ARGOFFSET(%rsp)
704 jmp ret_from_sys_call
706 #ifdef CONFIG_AUDITSYSCALL
708 * Fast path for syscall audit without full syscall trace.
709 * We just call __audit_syscall_entry() directly, and then
710 * jump back to the normal fast path.
713 movq %r10,%r9 /* 6th arg: 4th syscall arg */
714 movq %rdx,%r8 /* 5th arg: 3rd syscall arg */
715 movq %rsi,%rcx /* 4th arg: 2nd syscall arg */
716 movq %rdi,%rdx /* 3rd arg: 1st syscall arg */
717 movq %rax,%rsi /* 2nd arg: syscall number */
718 movl $AUDIT_ARCH_X86_64,%edi /* 1st arg: audit arch */
719 call __audit_syscall_entry
720 LOAD_ARGS 0 /* reload call-clobbered registers */
721 jmp system_call_fastpath
724 * Return fast path for syscall audit. Call __audit_syscall_exit()
725 * directly and then jump back to the fast path with TIF_SYSCALL_AUDIT
729 movq RAX-ARGOFFSET(%rsp),%rsi /* second arg, syscall return value */
730 cmpq $-MAX_ERRNO,%rsi /* is it < -MAX_ERRNO? */
731 setbe %al /* 1 if so, 0 if not */
732 movzbl %al,%edi /* zero-extend that into %edi */
733 call __audit_syscall_exit
734 movl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT),%edi
736 #endif /* CONFIG_AUDITSYSCALL */
738 /* Do syscall tracing */
740 #ifdef CONFIG_AUDITSYSCALL
741 testl $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET)
745 movq $-ENOSYS,RAX(%rsp) /* ptrace can change this for a bad syscall */
746 FIXUP_TOP_OF_STACK %rdi
748 call syscall_trace_enter
750 * Reload arg registers from stack in case ptrace changed them.
751 * We don't reload %rax because syscall_trace_enter() returned
752 * the value it wants us to use in the table lookup.
754 LOAD_ARGS ARGOFFSET, 1
756 #if __SYSCALL_MASK == ~0
757 cmpq $__NR_syscall_max,%rax
759 andl $__SYSCALL_MASK,%eax
760 cmpl $__NR_syscall_max,%eax
762 ja int_ret_from_sys_call /* RAX(%rsp) set to -ENOSYS above */
763 movq %r10,%rcx /* fixup for C */
764 call *sys_call_table(,%rax,8)
765 movq %rax,RAX-ARGOFFSET(%rsp)
766 /* Use IRET because user could have changed frame */
769 * Syscall return path ending with IRET.
770 * Has correct top of stack, but partial stack frame.
772 GLOBAL(int_ret_from_sys_call)
773 DISABLE_INTERRUPTS(CLBR_NONE)
775 movl $_TIF_ALLWORK_MASK,%edi
776 /* edi: mask to check */
777 GLOBAL(int_with_check)
779 GET_THREAD_INFO(%rcx)
780 movl TI_flags(%rcx),%edx
783 andl $~TS_COMPAT,TI_status(%rcx)
786 /* Either reschedule or signal or syscall exit tracking needed. */
787 /* First do a reschedule test. */
788 /* edx: work, edi: workmask */
790 bt $TIF_NEED_RESCHED,%edx
793 ENABLE_INTERRUPTS(CLBR_NONE)
797 DISABLE_INTERRUPTS(CLBR_NONE)
801 /* handle signals and tracing -- both require a full stack frame */
804 ENABLE_INTERRUPTS(CLBR_NONE)
805 int_check_syscall_exit_work:
807 /* Check for syscall exit trace */
808 testl $_TIF_WORK_SYSCALL_EXIT,%edx
811 leaq 8(%rsp),%rdi # &ptregs -> arg1
812 call syscall_trace_leave
814 andl $~(_TIF_WORK_SYSCALL_EXIT|_TIF_SYSCALL_EMU),%edi
818 testl $_TIF_DO_NOTIFY_MASK,%edx
820 movq %rsp,%rdi # &ptregs -> arg1
821 xorl %esi,%esi # oldset -> arg2
822 call do_notify_resume
823 1: movl $_TIF_WORK_MASK,%edi
826 DISABLE_INTERRUPTS(CLBR_NONE)
832 .macro FORK_LIKE func
835 popq %r11 /* save return address */
838 pushq %r11 /* put it back on stack */
839 FIXUP_TOP_OF_STACK %r11, 8
840 DEFAULT_FRAME 0 8 /* offset 8: return address */
842 RESTORE_TOP_OF_STACK %r11, 8
843 ret $REST_SKIP /* pop extended registers */
848 .macro FIXED_FRAME label,func
851 PARTIAL_FRAME 0 8 /* offset 8: return address */
852 FIXUP_TOP_OF_STACK %r11, 8-ARGOFFSET
854 RESTORE_TOP_OF_STACK %r11, 8-ARGOFFSET
863 FIXED_FRAME stub_iopl, sys_iopl
865 ENTRY(ptregscall_common)
866 DEFAULT_FRAME 1 8 /* offset 8: return address */
867 RESTORE_TOP_OF_STACK %r11, 8
868 movq_cfi_restore R15+8, r15
869 movq_cfi_restore R14+8, r14
870 movq_cfi_restore R13+8, r13
871 movq_cfi_restore R12+8, r12
872 movq_cfi_restore RBP+8, rbp
873 movq_cfi_restore RBX+8, rbx
874 ret $REST_SKIP /* pop extended registers */
876 END(ptregscall_common)
883 FIXUP_TOP_OF_STACK %r11
887 jmp int_ret_from_sys_call
892 * sigreturn is special because it needs to restore all registers on return.
893 * This cannot be done with SYSRET, so use the IRET return path instead.
895 ENTRY(stub_rt_sigreturn)
900 FIXUP_TOP_OF_STACK %r11
901 call sys_rt_sigreturn
902 movq %rax,RAX(%rsp) # fixme, this could be done at the higher layer
904 jmp int_ret_from_sys_call
906 END(stub_rt_sigreturn)
908 #ifdef CONFIG_X86_X32_ABI
909 ENTRY(stub_x32_rt_sigreturn)
914 FIXUP_TOP_OF_STACK %r11
915 call sys32_x32_rt_sigreturn
916 movq %rax,RAX(%rsp) # fixme, this could be done at the higher layer
918 jmp int_ret_from_sys_call
920 END(stub_x32_rt_sigreturn)
922 ENTRY(stub_x32_execve)
927 FIXUP_TOP_OF_STACK %r11
928 call compat_sys_execve
929 RESTORE_TOP_OF_STACK %r11
932 jmp int_ret_from_sys_call
939 * Build the entry stubs and pointer table with some assembler magic.
940 * We pack 7 stubs into a single 32-byte chunk, which will fit in a
941 * single cache line on all modern x86 implementations.
943 .section .init.rodata,"a"
947 .p2align CONFIG_X86_L1_CACHE_SHIFT
948 ENTRY(irq_entries_start)
950 vector=FIRST_EXTERNAL_VECTOR
951 .rept (NR_VECTORS-FIRST_EXTERNAL_VECTOR+6)/7
954 .if vector < NR_VECTORS
955 .if vector <> FIRST_EXTERNAL_VECTOR
956 CFI_ADJUST_CFA_OFFSET -8
958 1: pushq_cfi $(~vector+0x80) /* Note: always in signed byte range */
959 .if ((vector-FIRST_EXTERNAL_VECTOR)%7) <> 6
968 2: jmp common_interrupt
971 END(irq_entries_start)
978 * Interrupt entry/exit.
980 * Interrupt entry points save only callee clobbered registers in fast path.
982 * Entry runs with interrupts off.
985 /* 0(%rsp): ~(interrupt number) */
986 .macro interrupt func
987 /* reserve pt_regs for scratch regs and rbp */
988 subq $ORIG_RAX-RBP, %rsp
989 CFI_ADJUST_CFA_OFFSET ORIG_RAX-RBP
995 * Interrupt entry/exit should be protected against kprobes
997 .pushsection .kprobes.text, "ax"
999 * The interrupt stubs push (~vector+0x80) onto the stack and
1000 * then jump to common_interrupt.
1002 .p2align CONFIG_X86_L1_CACHE_SHIFT
1006 addq $-0x80,(%rsp) /* Adjust vector to [-256,-1] range */
1008 /* 0(%rsp): old_rsp-ARGOFFSET */
1010 DISABLE_INTERRUPTS(CLBR_NONE)
1012 decl PER_CPU_VAR(irq_count)
1014 /* Restore saved previous stack */
1016 CFI_DEF_CFA rsi,SS+8-RBP /* reg/off reset after def_cfa_expr */
1017 leaq ARGOFFSET-RBP(%rsi), %rsp
1018 CFI_DEF_CFA_REGISTER rsp
1019 CFI_ADJUST_CFA_OFFSET RBP-ARGOFFSET
1022 GET_THREAD_INFO(%rcx)
1023 testl $3,CS-ARGOFFSET(%rsp)
1026 /* Interrupt came from user space */
1028 * Has a correct top of stack, but a partial stack frame
1029 * %rcx: thread info. Interrupts off.
1031 retint_with_reschedule:
1032 movl $_TIF_WORK_MASK,%edi
1034 LOCKDEP_SYS_EXIT_IRQ
1035 movl TI_flags(%rcx),%edx
1040 retint_swapgs: /* return to user-space */
1042 * The iretq could re-enable interrupts:
1044 DISABLE_INTERRUPTS(CLBR_ANY)
1049 retint_restore_args: /* return to kernel space */
1050 DISABLE_INTERRUPTS(CLBR_ANY)
1052 * The iretq could re-enable interrupts:
1063 * Are we returning to a stack segment from the LDT? Note: in
1064 * 64-bit mode SS:RSP on the exception stack is always valid.
1066 #ifdef CONFIG_X86_ESPFIX64
1067 testb $4,(SS-RIP)(%rsp)
1068 jnz native_irq_return_ldt
1071 .global native_irq_return_iret
1072 native_irq_return_iret:
1074 * This may fault. Non-paranoid faults on return to userspace are
1075 * handled by fixup_bad_iret. These include #SS, #GP, and #NP.
1076 * Double-faults due to espfix64 are handled in do_double_fault.
1077 * Other faults here are fatal.
1081 #ifdef CONFIG_X86_ESPFIX64
1082 native_irq_return_ldt:
1086 movq PER_CPU_VAR(espfix_waddr),%rdi
1087 movq %rax,(0*8)(%rdi) /* RAX */
1088 movq (2*8)(%rsp),%rax /* RIP */
1089 movq %rax,(1*8)(%rdi)
1090 movq (3*8)(%rsp),%rax /* CS */
1091 movq %rax,(2*8)(%rdi)
1092 movq (4*8)(%rsp),%rax /* RFLAGS */
1093 movq %rax,(3*8)(%rdi)
1094 movq (6*8)(%rsp),%rax /* SS */
1095 movq %rax,(5*8)(%rdi)
1096 movq (5*8)(%rsp),%rax /* RSP */
1097 movq %rax,(4*8)(%rdi)
1098 andl $0xffff0000,%eax
1100 orq PER_CPU_VAR(espfix_stack),%rax
1104 jmp native_irq_return_iret
1107 /* edi: workmask, edx: work */
1110 bt $TIF_NEED_RESCHED,%edx
1113 ENABLE_INTERRUPTS(CLBR_NONE)
1117 GET_THREAD_INFO(%rcx)
1118 DISABLE_INTERRUPTS(CLBR_NONE)
1123 testl $_TIF_DO_NOTIFY_MASK,%edx
1126 ENABLE_INTERRUPTS(CLBR_NONE)
1128 movq $-1,ORIG_RAX(%rsp)
1129 xorl %esi,%esi # oldset
1130 movq %rsp,%rdi # &pt_regs
1131 call do_notify_resume
1133 DISABLE_INTERRUPTS(CLBR_NONE)
1135 GET_THREAD_INFO(%rcx)
1136 jmp retint_with_reschedule
1138 #ifdef CONFIG_PREEMPT
1139 /* Returning to kernel space. Check if we need preemption */
1140 /* rcx: threadinfo. interrupts off. */
1141 ENTRY(retint_kernel)
1142 cmpl $0,TI_preempt_count(%rcx)
1143 jnz retint_restore_args
1144 bt $TIF_NEED_RESCHED,TI_flags(%rcx)
1145 jnc retint_restore_args
1146 bt $9,EFLAGS-ARGOFFSET(%rsp) /* interrupts off? */
1147 jnc retint_restore_args
1148 call preempt_schedule_irq
1152 END(common_interrupt)
1155 * End of kprobes section
1162 .macro apicinterrupt num sym do_sym
1175 apicinterrupt IRQ_MOVE_CLEANUP_VECTOR \
1176 irq_move_cleanup_interrupt smp_irq_move_cleanup_interrupt
1177 apicinterrupt REBOOT_VECTOR \
1178 reboot_interrupt smp_reboot_interrupt
1181 #ifdef CONFIG_X86_UV
1182 apicinterrupt UV_BAU_MESSAGE \
1183 uv_bau_message_intr1 uv_bau_message_interrupt
1185 apicinterrupt LOCAL_TIMER_VECTOR \
1186 apic_timer_interrupt smp_apic_timer_interrupt
1187 apicinterrupt X86_PLATFORM_IPI_VECTOR \
1188 x86_platform_ipi smp_x86_platform_ipi
1190 #ifdef CONFIG_HAVE_KVM
1191 apicinterrupt POSTED_INTR_VECTOR \
1192 kvm_posted_intr_ipi smp_kvm_posted_intr_ipi
1195 apicinterrupt THRESHOLD_APIC_VECTOR \
1196 threshold_interrupt smp_threshold_interrupt
1197 apicinterrupt THERMAL_APIC_VECTOR \
1198 thermal_interrupt smp_thermal_interrupt
1201 apicinterrupt CALL_FUNCTION_SINGLE_VECTOR \
1202 call_function_single_interrupt smp_call_function_single_interrupt
1203 apicinterrupt CALL_FUNCTION_VECTOR \
1204 call_function_interrupt smp_call_function_interrupt
1205 apicinterrupt RESCHEDULE_VECTOR \
1206 reschedule_interrupt smp_reschedule_interrupt
1209 apicinterrupt ERROR_APIC_VECTOR \
1210 error_interrupt smp_error_interrupt
1211 apicinterrupt SPURIOUS_APIC_VECTOR \
1212 spurious_interrupt smp_spurious_interrupt
1214 #ifdef CONFIG_IRQ_WORK
1215 apicinterrupt IRQ_WORK_VECTOR \
1216 irq_work_interrupt smp_irq_work_interrupt
1220 * Exception entry points.
1222 .macro zeroentry sym do_sym
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
1232 movq %rsp,%rdi /* pt_regs pointer */
1233 xorl %esi,%esi /* no error code */
1235 jmp error_exit /* %ebx: no swapgs flag */
1240 .macro paranoidzeroentry sym do_sym
1244 PARAVIRT_ADJUST_EXCEPTION_FRAME
1245 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1246 subq $ORIG_RAX-R15, %rsp
1247 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1250 movq %rsp,%rdi /* pt_regs pointer */
1251 xorl %esi,%esi /* no error code */
1253 jmp paranoid_exit /* %ebx: no swapgs flag */
1258 #define INIT_TSS_IST(x) PER_CPU_VAR(init_tss) + (TSS_ist + ((x) - 1) * 8)
1259 .macro paranoidzeroentry_ist sym do_sym ist
1263 PARAVIRT_ADJUST_EXCEPTION_FRAME
1264 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1265 subq $ORIG_RAX-R15, %rsp
1266 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1268 TRACE_IRQS_OFF_DEBUG
1269 movq %rsp,%rdi /* pt_regs pointer */
1270 xorl %esi,%esi /* no error code */
1271 subq $EXCEPTION_STKSZ, INIT_TSS_IST(\ist)
1273 addq $EXCEPTION_STKSZ, INIT_TSS_IST(\ist)
1274 jmp paranoid_exit /* %ebx: no swapgs flag */
1279 .macro errorentry sym do_sym
1283 PARAVIRT_ADJUST_EXCEPTION_FRAME
1284 subq $ORIG_RAX-R15, %rsp
1285 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1288 movq %rsp,%rdi /* pt_regs pointer */
1289 movq ORIG_RAX(%rsp),%rsi /* get error code */
1290 movq $-1,ORIG_RAX(%rsp) /* no syscall to restart */
1292 jmp error_exit /* %ebx: no swapgs flag */
1297 /* error code is on the stack already */
1298 .macro paranoiderrorentry sym do_sym
1302 PARAVIRT_ADJUST_EXCEPTION_FRAME
1303 subq $ORIG_RAX-R15, %rsp
1304 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1308 movq %rsp,%rdi /* pt_regs pointer */
1309 movq ORIG_RAX(%rsp),%rsi /* get error code */
1310 movq $-1,ORIG_RAX(%rsp) /* no syscall to restart */
1312 jmp paranoid_exit /* %ebx: no swapgs flag */
1317 zeroentry divide_error do_divide_error
1318 zeroentry overflow do_overflow
1319 zeroentry bounds do_bounds
1320 zeroentry invalid_op do_invalid_op
1321 zeroentry device_not_available do_device_not_available
1322 paranoiderrorentry double_fault do_double_fault
1323 zeroentry coprocessor_segment_overrun do_coprocessor_segment_overrun
1324 errorentry invalid_TSS do_invalid_TSS
1325 errorentry segment_not_present do_segment_not_present
1326 zeroentry spurious_interrupt_bug do_spurious_interrupt_bug
1327 zeroentry coprocessor_error do_coprocessor_error
1328 errorentry alignment_check do_alignment_check
1329 zeroentry simd_coprocessor_error do_simd_coprocessor_error
1332 /* Reload gs selector with exception handling */
1333 /* edi: new selector */
1334 ENTRY(native_load_gs_index)
1337 DISABLE_INTERRUPTS(CLBR_ANY & ~CLBR_RDI)
1341 2: mfence /* workaround */
1346 END(native_load_gs_index)
1348 _ASM_EXTABLE(gs_change,bad_gs)
1349 .section .fixup,"ax"
1350 /* running with kernelgs */
1352 SWAPGS /* switch back to user gs */
1358 /* Call softirq on interrupt stack. Interrupts are off. */
1362 CFI_REL_OFFSET rbp,0
1364 CFI_DEF_CFA_REGISTER rbp
1365 incl PER_CPU_VAR(irq_count)
1366 cmove PER_CPU_VAR(irq_stack_ptr),%rsp
1367 push %rbp # backlink for old unwinder
1371 CFI_DEF_CFA_REGISTER rsp
1372 CFI_ADJUST_CFA_OFFSET -8
1373 decl PER_CPU_VAR(irq_count)
1379 zeroentry xen_hypervisor_callback xen_do_hypervisor_callback
1382 * A note on the "critical region" in our callback handler.
1383 * We want to avoid stacking callback handlers due to events occurring
1384 * during handling of the last event. To do this, we keep events disabled
1385 * until we've done all processing. HOWEVER, we must enable events before
1386 * popping the stack frame (can't be done atomically) and so it would still
1387 * be possible to get enough handler activations to overflow the stack.
1388 * Although unlikely, bugs of that kind are hard to track down, so we'd
1389 * like to avoid the possibility.
1390 * So, on entry to the handler we detect whether we interrupted an
1391 * existing activation in its critical region -- if so, we pop the current
1392 * activation and restart the handler using the previous one.
1394 ENTRY(xen_do_hypervisor_callback) # do_hypervisor_callback(struct *pt_regs)
1397 * Since we don't modify %rdi, evtchn_do_upall(struct *pt_regs) will
1398 * see the correct pointer to the pt_regs
1400 movq %rdi, %rsp # we don't return, adjust the stack frame
1403 11: incl PER_CPU_VAR(irq_count)
1405 CFI_DEF_CFA_REGISTER rbp
1406 cmovzq PER_CPU_VAR(irq_stack_ptr),%rsp
1407 pushq %rbp # backlink for old unwinder
1408 call xen_evtchn_do_upcall
1410 CFI_DEF_CFA_REGISTER rsp
1411 decl PER_CPU_VAR(irq_count)
1414 END(xen_do_hypervisor_callback)
1417 * Hypervisor uses this for application faults while it executes.
1418 * We get here for two reasons:
1419 * 1. Fault while reloading DS, ES, FS or GS
1420 * 2. Fault while executing IRET
1421 * Category 1 we do not need to fix up as Xen has already reloaded all segment
1422 * registers that could be reloaded and zeroed the others.
1423 * Category 2 we fix up by killing the current process. We cannot use the
1424 * normal Linux return path in this case because if we use the IRET hypercall
1425 * to pop the stack frame we end up in an infinite loop of failsafe callbacks.
1426 * We distinguish between categories by comparing each saved segment register
1427 * with its current contents: any discrepancy means we in category 1.
1429 ENTRY(xen_failsafe_callback)
1431 /*CFI_REL_OFFSET gs,GS*/
1432 /*CFI_REL_OFFSET fs,FS*/
1433 /*CFI_REL_OFFSET es,ES*/
1434 /*CFI_REL_OFFSET ds,DS*/
1435 CFI_REL_OFFSET r11,8
1436 CFI_REL_OFFSET rcx,0
1450 /* All segments match their saved values => Category 2 (Bad IRET). */
1456 CFI_ADJUST_CFA_OFFSET -0x30
1457 pushq_cfi $0 /* RIP */
1460 jmp general_protection
1462 1: /* Segment mismatch => Category 1 (Bad segment). Retry the IRET. */
1468 CFI_ADJUST_CFA_OFFSET -0x30
1469 pushq_cfi $-1 /* orig_ax = -1 => not a system call */
1473 END(xen_failsafe_callback)
1475 apicinterrupt HYPERVISOR_CALLBACK_VECTOR \
1476 xen_hvm_callback_vector xen_evtchn_do_upcall
1478 #endif /* CONFIG_XEN */
1480 #if IS_ENABLED(CONFIG_HYPERV)
1481 apicinterrupt HYPERVISOR_CALLBACK_VECTOR \
1482 hyperv_callback_vector hyperv_vector_handler
1483 #endif /* CONFIG_HYPERV */
1486 * Some functions should be protected against kprobes
1488 .pushsection .kprobes.text, "ax"
1490 paranoidzeroentry_ist debug do_debug DEBUG_STACK
1491 paranoidzeroentry_ist int3 do_int3 DEBUG_STACK
1492 errorentry stack_segment do_stack_segment
1494 zeroentry xen_debug do_debug
1495 zeroentry xen_int3 do_int3
1496 errorentry xen_stack_segment do_stack_segment
1498 errorentry general_protection do_general_protection
1499 errorentry page_fault do_page_fault
1500 #ifdef CONFIG_KVM_GUEST
1501 errorentry async_page_fault do_async_page_fault
1503 #ifdef CONFIG_X86_MCE
1504 paranoidzeroentry machine_check *machine_check_vector(%rip)
1508 * "Paranoid" exit path from exception stack.
1509 * Paranoid because this is used by NMIs and cannot take
1510 * any kernel state for granted.
1511 * We don't do kernel preemption checks here, because only
1512 * NMI should be common and it does not enable IRQs and
1513 * cannot get reschedule ticks.
1515 * "trace" is 0 for the NMI handler only, because irq-tracing
1516 * is fundamentally NMI-unsafe. (we cannot change the soft and
1517 * hard flags at once, atomically)
1520 /* ebx: no swapgs flag */
1521 ENTRY(paranoid_exit)
1523 DISABLE_INTERRUPTS(CLBR_NONE)
1524 TRACE_IRQS_OFF_DEBUG
1525 testl %ebx,%ebx /* swapgs needed? */
1526 jnz paranoid_restore
1528 jnz paranoid_userspace
1535 TRACE_IRQS_IRETQ_DEBUG 0
1539 GET_THREAD_INFO(%rcx)
1540 movl TI_flags(%rcx),%ebx
1541 andl $_TIF_WORK_MASK,%ebx
1543 movq %rsp,%rdi /* &pt_regs */
1545 movq %rax,%rsp /* switch stack for scheduling */
1546 testl $_TIF_NEED_RESCHED,%ebx
1547 jnz paranoid_schedule
1548 movl %ebx,%edx /* arg3: thread flags */
1550 ENABLE_INTERRUPTS(CLBR_NONE)
1551 xorl %esi,%esi /* arg2: oldset */
1552 movq %rsp,%rdi /* arg1: &pt_regs */
1553 call do_notify_resume
1554 DISABLE_INTERRUPTS(CLBR_NONE)
1556 jmp paranoid_userspace
1559 ENABLE_INTERRUPTS(CLBR_ANY)
1561 DISABLE_INTERRUPTS(CLBR_ANY)
1563 jmp paranoid_userspace
1568 * Exception entry point. This expects an error code/orig_rax on the stack.
1569 * returns in "no swapgs flag" in %ebx.
1573 CFI_ADJUST_CFA_OFFSET 15*8
1574 /* oldrax contains error code */
1593 je error_kernelspace
1601 * There are two places in the kernel that can potentially fault with
1602 * usergs. Handle them here. B stepping K8s sometimes report a
1603 * truncated RIP for IRET exceptions returning to compat mode. Check
1604 * for these here too.
1608 leaq native_irq_return_iret(%rip),%rcx
1609 cmpq %rcx,RIP+8(%rsp)
1611 movl %ecx,%eax /* zero extend */
1612 cmpq %rax,RIP+8(%rsp)
1614 cmpq $gs_change,RIP+8(%rsp)
1619 /* Fix truncated RIP */
1620 movq %rcx,RIP+8(%rsp)
1628 decl %ebx /* Return to usergs */
1634 /* ebx: no swapgs flag (1: don't need swapgs, 0: need it) */
1639 DISABLE_INTERRUPTS(CLBR_NONE)
1641 GET_THREAD_INFO(%rcx)
1644 LOCKDEP_SYS_EXIT_IRQ
1645 movl TI_flags(%rcx),%edx
1646 movl $_TIF_WORK_MASK,%edi
1654 * Test if a given stack is an NMI stack or not.
1656 .macro test_in_nmi reg stack nmi_ret normal_ret
1659 subq $EXCEPTION_STKSZ, %\reg
1665 /* runs on exception stack */
1668 PARAVIRT_ADJUST_EXCEPTION_FRAME
1670 * We allow breakpoints in NMIs. If a breakpoint occurs, then
1671 * the iretq it performs will take us out of NMI context.
1672 * This means that we can have nested NMIs where the next
1673 * NMI is using the top of the stack of the previous NMI. We
1674 * can't let it execute because the nested NMI will corrupt the
1675 * stack of the previous NMI. NMI handlers are not re-entrant
1678 * To handle this case we do the following:
1679 * Check the a special location on the stack that contains
1680 * a variable that is set when NMIs are executing.
1681 * The interrupted task's stack is also checked to see if it
1683 * If the variable is not set and the stack is not the NMI
1685 * o Set the special variable on the stack
1686 * o Copy the interrupt frame into a "saved" location on the stack
1687 * o Copy the interrupt frame into a "copy" location on the stack
1688 * o Continue processing the NMI
1689 * If the variable is set or the previous stack is the NMI stack:
1690 * o Modify the "copy" location to jump to the repeate_nmi
1691 * o return back to the first NMI
1693 * Now on exit of the first NMI, we first clear the stack variable
1694 * The NMI stack will tell any nested NMIs at that point that it is
1695 * nested. Then we pop the stack normally with iret, and if there was
1696 * a nested NMI that updated the copy interrupt stack frame, a
1697 * jump will be made to the repeat_nmi code that will handle the second
1701 /* Use %rdx as out temp variable throughout */
1703 CFI_REL_OFFSET rdx, 0
1706 * If %cs was not the kernel segment, then the NMI triggered in user
1707 * space, which means it is definitely not nested.
1709 cmpl $__KERNEL_CS, 16(%rsp)
1713 * Check the special variable on the stack to see if NMIs are
1720 * Now test if the previous stack was an NMI stack.
1721 * We need the double check. We check the NMI stack to satisfy the
1722 * race when the first NMI clears the variable before returning.
1723 * We check the variable because the first NMI could be in a
1724 * breakpoint routine using a breakpoint stack.
1727 test_in_nmi rdx, 4*8(%rsp), nested_nmi, first_nmi
1732 * Do nothing if we interrupted the fixup in repeat_nmi.
1733 * It's about to repeat the NMI handler, so we are fine
1734 * with ignoring this one.
1736 movq $repeat_nmi, %rdx
1739 movq $end_repeat_nmi, %rdx
1744 /* Set up the interrupted NMIs stack to jump to repeat_nmi */
1745 leaq -1*8(%rsp), %rdx
1747 CFI_ADJUST_CFA_OFFSET 1*8
1748 leaq -10*8(%rsp), %rdx
1749 pushq_cfi $__KERNEL_DS
1752 pushq_cfi $__KERNEL_CS
1753 pushq_cfi $repeat_nmi
1755 /* Put stack back */
1757 CFI_ADJUST_CFA_OFFSET -6*8
1763 /* No need to check faults here */
1769 * Because nested NMIs will use the pushed location that we
1770 * stored in rdx, we must keep that space available.
1771 * Here's what our stack frame will look like:
1772 * +-------------------------+
1774 * | original Return RSP |
1775 * | original RFLAGS |
1778 * +-------------------------+
1779 * | temp storage for rdx |
1780 * +-------------------------+
1781 * | NMI executing variable |
1782 * +-------------------------+
1784 * | copied Return RSP |
1788 * +-------------------------+
1790 * | Saved Return RSP |
1794 * +-------------------------+
1796 * +-------------------------+
1798 * The saved stack frame is used to fix up the copied stack frame
1799 * that a nested NMI may change to make the interrupted NMI iret jump
1800 * to the repeat_nmi. The original stack frame and the temp storage
1801 * is also used by nested NMIs and can not be trusted on exit.
1803 /* Do not pop rdx, nested NMIs will corrupt that part of the stack */
1807 /* Set the NMI executing variable on the stack. */
1811 * Leave room for the "copied" frame
1814 CFI_ADJUST_CFA_OFFSET 5*8
1816 /* Copy the stack frame to the Saved frame */
1818 pushq_cfi 11*8(%rsp)
1820 CFI_DEF_CFA_OFFSET SS+8-RIP
1822 /* Everything up to here is safe from nested NMIs */
1825 * If there was a nested NMI, the first NMI's iret will return
1826 * here. But NMIs are still enabled and we can take another
1827 * nested NMI. The nested NMI checks the interrupted RIP to see
1828 * if it is between repeat_nmi and end_repeat_nmi, and if so
1829 * it will just return, as we are about to repeat an NMI anyway.
1830 * This makes it safe to copy to the stack frame that a nested
1835 * Update the stack variable to say we are still in NMI (the update
1836 * is benign for the non-repeat case, where 1 was pushed just above
1837 * to this very stack slot).
1841 /* Make another copy, this one may be modified by nested NMIs */
1843 CFI_ADJUST_CFA_OFFSET -10*8
1845 pushq_cfi -6*8(%rsp)
1848 CFI_DEF_CFA_OFFSET SS+8-RIP
1852 * Everything below this point can be preempted by a nested
1853 * NMI if the first NMI took an exception and reset our iret stack
1854 * so that we repeat another NMI.
1856 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1857 subq $ORIG_RAX-R15, %rsp
1858 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1860 * Use save_paranoid to handle SWAPGS, but no need to use paranoid_exit
1861 * as we should not be calling schedule in NMI context.
1862 * Even with normal interrupts enabled. An NMI should not be
1863 * setting NEED_RESCHED or anything that normal interrupts and
1864 * exceptions might do.
1870 * Save off the CR2 register. If we take a page fault in the NMI then
1871 * it could corrupt the CR2 value. If the NMI preempts a page fault
1872 * handler before it was able to read the CR2 register, and then the
1873 * NMI itself takes a page fault, the page fault that was preempted
1874 * will read the information from the NMI page fault and not the
1875 * origin fault. Save it off and restore it if it changes.
1876 * Use the r12 callee-saved register.
1880 /* paranoidentry do_nmi, 0; without TRACE_IRQS_OFF */
1885 /* Did the NMI take a page fault? Restore cr2 if it did */
1892 testl %ebx,%ebx /* swapgs needed? */
1897 /* Pop the extra iret frame at once */
1900 /* Clear the NMI executing stack variable */
1906 ENTRY(ignore_sysret)
1914 * End of kprobes section