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 <linux/err.h>
63 /* Avoid __ASSEMBLER__'ifying <linux/audit.h> just for this. */
64 #include <linux/elf-em.h>
65 #define AUDIT_ARCH_X86_64 (EM_X86_64|__AUDIT_ARCH_64BIT|__AUDIT_ARCH_LE)
66 #define __AUDIT_ARCH_64BIT 0x80000000
67 #define __AUDIT_ARCH_LE 0x40000000
70 .section .entry.text, "ax"
72 #ifdef CONFIG_FUNCTION_TRACER
74 #ifdef CC_USING_FENTRY
75 # define function_hook __fentry__
77 # define function_hook mcount
80 #ifdef CONFIG_DYNAMIC_FTRACE
86 /* skip is set if stack has been adjusted */
87 .macro ftrace_caller_setup skip=0
88 MCOUNT_SAVE_FRAME \skip
90 /* Load the ftrace_ops into the 3rd parameter */
91 leaq function_trace_op, %rdx
93 /* Load ip into the first parameter */
95 subq $MCOUNT_INSN_SIZE, %rdi
96 /* Load the parent_ip into the second parameter */
97 #ifdef CC_USING_FENTRY
98 movq SS+16(%rsp), %rsi
105 /* Check if tracing was disabled (quick check) */
106 cmpl $0, function_trace_stop
110 /* regs go into 4th parameter (but make it NULL) */
119 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
120 GLOBAL(ftrace_graph_call)
128 ENTRY(ftrace_regs_caller)
129 /* Save the current flags before compare (in SS location)*/
132 /* Check if tracing was disabled (quick check) */
133 cmpl $0, function_trace_stop
134 jne ftrace_restore_flags
136 /* skip=8 to skip flags saved in SS */
137 ftrace_caller_setup 8
139 /* Save the rest of pt_regs */
148 /* Copy saved flags */
150 movq %rcx, EFLAGS(%rsp)
151 /* Kernel segments */
152 movq $__KERNEL_DS, %rcx
154 movq $__KERNEL_CS, %rcx
156 /* Stack - skipping return address */
157 leaq SS+16(%rsp), %rcx
160 /* regs go into 4th parameter */
163 GLOBAL(ftrace_regs_call)
166 /* Copy flags back to SS, to restore them */
167 movq EFLAGS(%rsp), %rax
170 /* Handlers can change the RIP */
172 movq %rax, SS+8(%rsp)
174 /* restore the rest of pt_regs */
183 /* skip=8 to skip flags saved in SS */
184 MCOUNT_RESTORE_FRAME 8
190 ftrace_restore_flags:
194 END(ftrace_regs_caller)
197 #else /* ! CONFIG_DYNAMIC_FTRACE */
200 cmpl $0, function_trace_stop
203 cmpq $ftrace_stub, ftrace_trace_function
206 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
207 cmpq $ftrace_stub, ftrace_graph_return
208 jnz ftrace_graph_caller
210 cmpq $ftrace_graph_entry_stub, ftrace_graph_entry
211 jnz ftrace_graph_caller
221 #ifdef CC_USING_FENTRY
222 movq SS+16(%rsp), %rsi
226 subq $MCOUNT_INSN_SIZE, %rdi
228 call *ftrace_trace_function
234 #endif /* CONFIG_DYNAMIC_FTRACE */
235 #endif /* CONFIG_FUNCTION_TRACER */
237 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
238 ENTRY(ftrace_graph_caller)
241 #ifdef CC_USING_FENTRY
242 leaq SS+16(%rsp), %rdi
243 movq $0, %rdx /* No framepointers needed */
249 subq $MCOUNT_INSN_SIZE, %rsi
251 call prepare_ftrace_return
256 END(ftrace_graph_caller)
258 GLOBAL(return_to_handler)
261 /* Save the return values */
266 call ftrace_return_to_handler
276 #ifndef CONFIG_PREEMPT
277 #define retint_kernel retint_restore_args
280 #ifdef CONFIG_PARAVIRT
281 ENTRY(native_usergs_sysret64)
284 ENDPROC(native_usergs_sysret64)
285 #endif /* CONFIG_PARAVIRT */
288 .macro TRACE_IRQS_IRETQ offset=ARGOFFSET
289 #ifdef CONFIG_TRACE_IRQFLAGS
290 bt $9,EFLAGS-\offset(%rsp) /* interrupts off? */
298 * When dynamic function tracer is enabled it will add a breakpoint
299 * to all locations that it is about to modify, sync CPUs, update
300 * all the code, sync CPUs, then remove the breakpoints. In this time
301 * if lockdep is enabled, it might jump back into the debug handler
302 * outside the updating of the IST protection. (TRACE_IRQS_ON/OFF).
304 * We need to change the IDT table before calling TRACE_IRQS_ON/OFF to
305 * make sure the stack pointer does not get reset back to the top
306 * of the debug stack, and instead just reuses the current stack.
308 #if defined(CONFIG_DYNAMIC_FTRACE) && defined(CONFIG_TRACE_IRQFLAGS)
310 .macro TRACE_IRQS_OFF_DEBUG
311 call debug_stack_set_zero
313 call debug_stack_reset
316 .macro TRACE_IRQS_ON_DEBUG
317 call debug_stack_set_zero
319 call debug_stack_reset
322 .macro TRACE_IRQS_IRETQ_DEBUG offset=ARGOFFSET
323 bt $9,EFLAGS-\offset(%rsp) /* interrupts off? */
330 # define TRACE_IRQS_OFF_DEBUG TRACE_IRQS_OFF
331 # define TRACE_IRQS_ON_DEBUG TRACE_IRQS_ON
332 # define TRACE_IRQS_IRETQ_DEBUG TRACE_IRQS_IRETQ
336 * C code is not supposed to know about undefined top of stack. Every time
337 * a C function with an pt_regs argument is called from the SYSCALL based
338 * fast path FIXUP_TOP_OF_STACK is needed.
339 * RESTORE_TOP_OF_STACK syncs the syscall state after any possible ptregs
343 /* %rsp:at FRAMEEND */
344 .macro FIXUP_TOP_OF_STACK tmp offset=0
345 movq PER_CPU_VAR(old_rsp),\tmp
346 movq \tmp,RSP+\offset(%rsp)
347 movq $__USER_DS,SS+\offset(%rsp)
348 movq $__USER_CS,CS+\offset(%rsp)
349 movq $-1,RCX+\offset(%rsp)
350 movq R11+\offset(%rsp),\tmp /* get eflags */
351 movq \tmp,EFLAGS+\offset(%rsp)
354 .macro RESTORE_TOP_OF_STACK tmp offset=0
355 movq RSP+\offset(%rsp),\tmp
356 movq \tmp,PER_CPU_VAR(old_rsp)
357 movq EFLAGS+\offset(%rsp),\tmp
358 movq \tmp,R11+\offset(%rsp)
361 .macro FAKE_STACK_FRAME child_rip
362 /* push in order ss, rsp, eflags, cs, rip */
364 pushq_cfi $__KERNEL_DS /* ss */
365 /*CFI_REL_OFFSET ss,0*/
366 pushq_cfi %rax /* rsp */
368 pushq_cfi $(X86_EFLAGS_IF|X86_EFLAGS_BIT1) /* eflags - interrupts on */
369 /*CFI_REL_OFFSET rflags,0*/
370 pushq_cfi $__KERNEL_CS /* cs */
371 /*CFI_REL_OFFSET cs,0*/
372 pushq_cfi \child_rip /* rip */
374 pushq_cfi %rax /* orig rax */
377 .macro UNFAKE_STACK_FRAME
379 CFI_ADJUST_CFA_OFFSET -(6*8)
383 * initial frame state for interrupts (and exceptions without error code)
385 .macro EMPTY_FRAME start=1 offset=0
389 CFI_DEF_CFA rsp,8+\offset
391 CFI_DEF_CFA_OFFSET 8+\offset
396 * initial frame state for interrupts (and exceptions without error code)
398 .macro INTR_FRAME start=1 offset=0
399 EMPTY_FRAME \start, SS+8+\offset-RIP
400 /*CFI_REL_OFFSET ss, SS+\offset-RIP*/
401 CFI_REL_OFFSET rsp, RSP+\offset-RIP
402 /*CFI_REL_OFFSET rflags, EFLAGS+\offset-RIP*/
403 /*CFI_REL_OFFSET cs, CS+\offset-RIP*/
404 CFI_REL_OFFSET rip, RIP+\offset-RIP
408 * initial frame state for exceptions with error code (and interrupts
409 * with vector already pushed)
411 .macro XCPT_FRAME start=1 offset=0
412 INTR_FRAME \start, RIP+\offset-ORIG_RAX
413 /*CFI_REL_OFFSET orig_rax, ORIG_RAX-ORIG_RAX*/
417 * frame that enables calling into C.
419 .macro PARTIAL_FRAME start=1 offset=0
420 XCPT_FRAME \start, ORIG_RAX+\offset-ARGOFFSET
421 CFI_REL_OFFSET rdi, RDI+\offset-ARGOFFSET
422 CFI_REL_OFFSET rsi, RSI+\offset-ARGOFFSET
423 CFI_REL_OFFSET rdx, RDX+\offset-ARGOFFSET
424 CFI_REL_OFFSET rcx, RCX+\offset-ARGOFFSET
425 CFI_REL_OFFSET rax, RAX+\offset-ARGOFFSET
426 CFI_REL_OFFSET r8, R8+\offset-ARGOFFSET
427 CFI_REL_OFFSET r9, R9+\offset-ARGOFFSET
428 CFI_REL_OFFSET r10, R10+\offset-ARGOFFSET
429 CFI_REL_OFFSET r11, R11+\offset-ARGOFFSET
433 * frame that enables passing a complete pt_regs to a C function.
435 .macro DEFAULT_FRAME start=1 offset=0
436 PARTIAL_FRAME \start, R11+\offset-R15
437 CFI_REL_OFFSET rbx, RBX+\offset
438 CFI_REL_OFFSET rbp, RBP+\offset
439 CFI_REL_OFFSET r12, R12+\offset
440 CFI_REL_OFFSET r13, R13+\offset
441 CFI_REL_OFFSET r14, R14+\offset
442 CFI_REL_OFFSET r15, R15+\offset
445 /* save partial stack frame */
448 /* start from rbp in pt_regs and jump over */
449 movq_cfi rdi, (RDI-RBP)
450 movq_cfi rsi, (RSI-RBP)
451 movq_cfi rdx, (RDX-RBP)
452 movq_cfi rcx, (RCX-RBP)
453 movq_cfi rax, (RAX-RBP)
454 movq_cfi r8, (R8-RBP)
455 movq_cfi r9, (R9-RBP)
456 movq_cfi r10, (R10-RBP)
457 movq_cfi r11, (R11-RBP)
459 /* Save rbp so that we can unwind from get_irq_regs() */
462 /* Save previous stack value */
465 leaq -RBP(%rsp),%rdi /* arg1 for handler */
466 testl $3, CS-RBP(%rsi)
470 * irq_count is used to check if a CPU is already on an interrupt stack
471 * or not. While this is essentially redundant with preempt_count it is
472 * a little cheaper to use a separate counter in the PDA (short of
473 * moving irq_enter into assembly, which would be too much work)
475 1: incl PER_CPU_VAR(irq_count)
476 cmovzq PER_CPU_VAR(irq_stack_ptr),%rsp
477 CFI_DEF_CFA_REGISTER rsi
479 /* Store previous stack value */
481 CFI_ESCAPE 0x0f /* DW_CFA_def_cfa_expression */, 6, \
482 0x77 /* DW_OP_breg7 */, 0, \
483 0x06 /* DW_OP_deref */, \
484 0x08 /* DW_OP_const1u */, SS+8-RBP, \
485 0x22 /* DW_OP_plus */
486 /* We entered an interrupt context - irqs are off: */
491 PARTIAL_FRAME 1 (REST_SKIP+8)
492 movq 5*8+16(%rsp), %r11 /* save return address */
499 movq %r11, 8(%rsp) /* return address */
500 FIXUP_TOP_OF_STACK %r11, 16
505 /* save complete stack frame */
506 .pushsection .kprobes.text, "ax"
526 movl $MSR_GS_BASE,%ecx
529 js 1f /* negative -> in kernel */
538 * A newly forked process directly context switches into this address.
540 * rdi: prev task we switched from
545 LOCK ; btr $TIF_FORK,TI_flags(%r8)
548 popfq_cfi # reset kernel eflags
550 call schedule_tail # rdi: 'prev' task parameter
552 GET_THREAD_INFO(%rcx)
556 testl $3, CS-ARGOFFSET(%rsp) # from kernel_thread?
559 testl $_TIF_IA32, TI_flags(%rcx) # 32-bit compat task needs IRET
560 jnz int_ret_from_sys_call
562 RESTORE_TOP_OF_STACK %rdi, -ARGOFFSET
563 jmp ret_from_sys_call # go to the SYSRET fastpath
566 subq $REST_SKIP, %rsp # leave space for volatiles
567 CFI_ADJUST_CFA_OFFSET REST_SKIP
572 jmp int_ret_from_sys_call
577 * System call entry. Up to 6 arguments in registers are supported.
579 * SYSCALL does not save anything on the stack and does not change the
580 * stack pointer. However, it does mask the flags register for us, so
581 * CLD and CLAC are not needed.
586 * rax system call number
588 * rcx return address for syscall/sysret, C arg3
591 * r10 arg3 (--> moved to rcx for C)
594 * r11 eflags for syscall/sysret, temporary for C
595 * r12-r15,rbp,rbx saved by C code, not touched.
597 * Interrupts are off on entry.
598 * Only called from user space.
600 * XXX if we had a free scratch register we could save the RSP into the stack frame
601 * and report it properly in ps. Unfortunately we haven't.
603 * When user can change the frames always force IRET. That is because
604 * it deals with uncanonical addresses better. SYSRET has trouble
605 * with them due to bugs in both AMD and Intel CPUs.
611 CFI_DEF_CFA rsp,KERNEL_STACK_OFFSET
613 /*CFI_REGISTER rflags,r11*/
616 * A hypervisor implementation might want to use a label
617 * after the swapgs, so that it can do the swapgs
618 * for the guest and jump here on syscall.
620 GLOBAL(system_call_after_swapgs)
622 movq %rsp,PER_CPU_VAR(old_rsp)
623 movq PER_CPU_VAR(kernel_stack),%rsp
625 * No need to follow this irqs off/on section - it's straight
628 ENABLE_INTERRUPTS(CLBR_NONE)
630 movq %rax,ORIG_RAX-ARGOFFSET(%rsp)
631 movq %rcx,RIP-ARGOFFSET(%rsp)
632 CFI_REL_OFFSET rip,RIP-ARGOFFSET
633 testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET)
635 system_call_fastpath:
636 #if __SYSCALL_MASK == ~0
637 cmpq $__NR_syscall_max,%rax
639 andl $__SYSCALL_MASK,%eax
640 cmpl $__NR_syscall_max,%eax
644 call *sys_call_table(,%rax,8) # XXX: rip relative
645 movq %rax,RAX-ARGOFFSET(%rsp)
647 * Syscall return path ending with SYSRET (fast path)
648 * Has incomplete stack frame and undefined top of stack.
651 movl $_TIF_ALLWORK_MASK,%edi
655 DISABLE_INTERRUPTS(CLBR_NONE)
657 movl TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET),%edx
662 * sysretq will re-enable interrupts:
665 movq RIP-ARGOFFSET(%rsp),%rcx
667 RESTORE_ARGS 1,-ARG_SKIP,0
668 /*CFI_REGISTER rflags,r11*/
669 movq PER_CPU_VAR(old_rsp), %rsp
673 /* Handle reschedules */
674 /* edx: work, edi: workmask */
676 bt $TIF_NEED_RESCHED,%edx
679 ENABLE_INTERRUPTS(CLBR_NONE)
685 /* Handle a signal */
688 ENABLE_INTERRUPTS(CLBR_NONE)
689 #ifdef CONFIG_AUDITSYSCALL
690 bt $TIF_SYSCALL_AUDIT,%edx
694 * We have a signal, or exit tracing or single-step.
695 * These all wind up with the iret return path anyway,
696 * so just join that path right now.
698 FIXUP_TOP_OF_STACK %r11, -ARGOFFSET
699 jmp int_check_syscall_exit_work
702 movq $-ENOSYS,RAX-ARGOFFSET(%rsp)
703 jmp ret_from_sys_call
705 #ifdef CONFIG_AUDITSYSCALL
707 * Fast path for syscall audit without full syscall trace.
708 * We just call __audit_syscall_entry() directly, and then
709 * jump back to the normal fast path.
712 movq %r10,%r9 /* 6th arg: 4th syscall arg */
713 movq %rdx,%r8 /* 5th arg: 3rd syscall arg */
714 movq %rsi,%rcx /* 4th arg: 2nd syscall arg */
715 movq %rdi,%rdx /* 3rd arg: 1st syscall arg */
716 movq %rax,%rsi /* 2nd arg: syscall number */
717 movl $AUDIT_ARCH_X86_64,%edi /* 1st arg: audit arch */
718 call __audit_syscall_entry
719 LOAD_ARGS 0 /* reload call-clobbered registers */
720 jmp system_call_fastpath
723 * Return fast path for syscall audit. Call __audit_syscall_exit()
724 * directly and then jump back to the fast path with TIF_SYSCALL_AUDIT
728 movq RAX-ARGOFFSET(%rsp),%rsi /* second arg, syscall return value */
729 cmpq $-MAX_ERRNO,%rsi /* is it < -MAX_ERRNO? */
730 setbe %al /* 1 if so, 0 if not */
731 movzbl %al,%edi /* zero-extend that into %edi */
732 call __audit_syscall_exit
733 movl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT),%edi
735 #endif /* CONFIG_AUDITSYSCALL */
737 /* Do syscall tracing */
739 #ifdef CONFIG_AUDITSYSCALL
740 testl $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET)
744 movq $-ENOSYS,RAX(%rsp) /* ptrace can change this for a bad syscall */
745 FIXUP_TOP_OF_STACK %rdi
747 call syscall_trace_enter
749 * Reload arg registers from stack in case ptrace changed them.
750 * We don't reload %rax because syscall_trace_enter() returned
751 * the value it wants us to use in the table lookup.
753 LOAD_ARGS ARGOFFSET, 1
755 #if __SYSCALL_MASK == ~0
756 cmpq $__NR_syscall_max,%rax
758 andl $__SYSCALL_MASK,%eax
759 cmpl $__NR_syscall_max,%eax
761 ja int_ret_from_sys_call /* RAX(%rsp) set to -ENOSYS above */
762 movq %r10,%rcx /* fixup for C */
763 call *sys_call_table(,%rax,8)
764 movq %rax,RAX-ARGOFFSET(%rsp)
765 /* Use IRET because user could have changed frame */
768 * Syscall return path ending with IRET.
769 * Has correct top of stack, but partial stack frame.
771 GLOBAL(int_ret_from_sys_call)
772 DISABLE_INTERRUPTS(CLBR_NONE)
774 movl $_TIF_ALLWORK_MASK,%edi
775 /* edi: mask to check */
776 GLOBAL(int_with_check)
778 GET_THREAD_INFO(%rcx)
779 movl TI_flags(%rcx),%edx
782 andl $~TS_COMPAT,TI_status(%rcx)
785 /* Either reschedule or signal or syscall exit tracking needed. */
786 /* First do a reschedule test. */
787 /* edx: work, edi: workmask */
789 bt $TIF_NEED_RESCHED,%edx
792 ENABLE_INTERRUPTS(CLBR_NONE)
796 DISABLE_INTERRUPTS(CLBR_NONE)
800 /* handle signals and tracing -- both require a full stack frame */
803 ENABLE_INTERRUPTS(CLBR_NONE)
804 int_check_syscall_exit_work:
806 /* Check for syscall exit trace */
807 testl $_TIF_WORK_SYSCALL_EXIT,%edx
810 leaq 8(%rsp),%rdi # &ptregs -> arg1
811 call syscall_trace_leave
813 andl $~(_TIF_WORK_SYSCALL_EXIT|_TIF_SYSCALL_EMU),%edi
817 testl $_TIF_DO_NOTIFY_MASK,%edx
819 movq %rsp,%rdi # &ptregs -> arg1
820 xorl %esi,%esi # oldset -> arg2
821 call do_notify_resume
822 1: movl $_TIF_WORK_MASK,%edi
825 DISABLE_INTERRUPTS(CLBR_NONE)
831 .macro FORK_LIKE func
834 popq %r11 /* save return address */
837 pushq %r11 /* put it back on stack */
838 FIXUP_TOP_OF_STACK %r11, 8
839 DEFAULT_FRAME 0 8 /* offset 8: return address */
841 RESTORE_TOP_OF_STACK %r11, 8
842 ret $REST_SKIP /* pop extended registers */
847 .macro FIXED_FRAME label,func
850 PARTIAL_FRAME 0 8 /* offset 8: return address */
851 FIXUP_TOP_OF_STACK %r11, 8-ARGOFFSET
853 RESTORE_TOP_OF_STACK %r11, 8-ARGOFFSET
862 FIXED_FRAME stub_iopl, sys_iopl
864 ENTRY(ptregscall_common)
865 DEFAULT_FRAME 1 8 /* offset 8: return address */
866 RESTORE_TOP_OF_STACK %r11, 8
867 movq_cfi_restore R15+8, r15
868 movq_cfi_restore R14+8, r14
869 movq_cfi_restore R13+8, r13
870 movq_cfi_restore R12+8, r12
871 movq_cfi_restore RBP+8, rbp
872 movq_cfi_restore RBX+8, rbx
873 ret $REST_SKIP /* pop extended registers */
875 END(ptregscall_common)
882 FIXUP_TOP_OF_STACK %r11
886 jmp int_ret_from_sys_call
891 * sigreturn is special because it needs to restore all registers on return.
892 * This cannot be done with SYSRET, so use the IRET return path instead.
894 ENTRY(stub_rt_sigreturn)
899 FIXUP_TOP_OF_STACK %r11
900 call sys_rt_sigreturn
901 movq %rax,RAX(%rsp) # fixme, this could be done at the higher layer
903 jmp int_ret_from_sys_call
905 END(stub_rt_sigreturn)
907 #ifdef CONFIG_X86_X32_ABI
908 ENTRY(stub_x32_rt_sigreturn)
913 FIXUP_TOP_OF_STACK %r11
914 call sys32_x32_rt_sigreturn
915 movq %rax,RAX(%rsp) # fixme, this could be done at the higher layer
917 jmp int_ret_from_sys_call
919 END(stub_x32_rt_sigreturn)
921 ENTRY(stub_x32_execve)
926 FIXUP_TOP_OF_STACK %r11
927 call compat_sys_execve
928 RESTORE_TOP_OF_STACK %r11
931 jmp int_ret_from_sys_call
938 * Build the entry stubs and pointer table with some assembler magic.
939 * We pack 7 stubs into a single 32-byte chunk, which will fit in a
940 * single cache line on all modern x86 implementations.
942 .section .init.rodata,"a"
946 .p2align CONFIG_X86_L1_CACHE_SHIFT
947 ENTRY(irq_entries_start)
949 vector=FIRST_EXTERNAL_VECTOR
950 .rept (NR_VECTORS-FIRST_EXTERNAL_VECTOR+6)/7
953 .if vector < NR_VECTORS
954 .if vector <> FIRST_EXTERNAL_VECTOR
955 CFI_ADJUST_CFA_OFFSET -8
957 1: pushq_cfi $(~vector+0x80) /* Note: always in signed byte range */
958 .if ((vector-FIRST_EXTERNAL_VECTOR)%7) <> 6
967 2: jmp common_interrupt
970 END(irq_entries_start)
977 * Interrupt entry/exit.
979 * Interrupt entry points save only callee clobbered registers in fast path.
981 * Entry runs with interrupts off.
984 /* 0(%rsp): ~(interrupt number) */
985 .macro interrupt func
986 /* reserve pt_regs for scratch regs and rbp */
987 subq $ORIG_RAX-RBP, %rsp
988 CFI_ADJUST_CFA_OFFSET ORIG_RAX-RBP
994 * Interrupt entry/exit should be protected against kprobes
996 .pushsection .kprobes.text, "ax"
998 * The interrupt stubs push (~vector+0x80) onto the stack and
999 * then jump to common_interrupt.
1001 .p2align CONFIG_X86_L1_CACHE_SHIFT
1005 addq $-0x80,(%rsp) /* Adjust vector to [-256,-1] range */
1007 /* 0(%rsp): old_rsp-ARGOFFSET */
1009 DISABLE_INTERRUPTS(CLBR_NONE)
1011 decl PER_CPU_VAR(irq_count)
1013 /* Restore saved previous stack */
1015 CFI_DEF_CFA rsi,SS+8-RBP /* reg/off reset after def_cfa_expr */
1016 leaq ARGOFFSET-RBP(%rsi), %rsp
1017 CFI_DEF_CFA_REGISTER rsp
1018 CFI_ADJUST_CFA_OFFSET RBP-ARGOFFSET
1021 GET_THREAD_INFO(%rcx)
1022 testl $3,CS-ARGOFFSET(%rsp)
1025 /* Interrupt came from user space */
1027 * Has a correct top of stack, but a partial stack frame
1028 * %rcx: thread info. Interrupts off.
1030 retint_with_reschedule:
1031 movl $_TIF_WORK_MASK,%edi
1033 LOCKDEP_SYS_EXIT_IRQ
1034 movl TI_flags(%rcx),%edx
1039 retint_swapgs: /* return to user-space */
1041 * The iretq could re-enable interrupts:
1043 DISABLE_INTERRUPTS(CLBR_ANY)
1048 retint_restore_args: /* return to kernel space */
1049 DISABLE_INTERRUPTS(CLBR_ANY)
1051 * The iretq could re-enable interrupts:
1059 _ASM_EXTABLE(irq_return, bad_iret)
1061 #ifdef CONFIG_PARAVIRT
1064 _ASM_EXTABLE(native_iret, bad_iret)
1067 .section .fixup,"ax"
1070 * The iret traps when the %cs or %ss being restored is bogus.
1071 * We've lost the original trap vector and error code.
1072 * #GPF is the most likely one to get for an invalid selector.
1073 * So pretend we completed the iret and took the #GPF in user mode.
1075 * We are now running with the kernel GS after exception recovery.
1076 * But error_entry expects us to have user GS to match the user %cs,
1082 jmp general_protection
1086 /* edi: workmask, edx: work */
1089 bt $TIF_NEED_RESCHED,%edx
1092 ENABLE_INTERRUPTS(CLBR_NONE)
1096 GET_THREAD_INFO(%rcx)
1097 DISABLE_INTERRUPTS(CLBR_NONE)
1102 testl $_TIF_DO_NOTIFY_MASK,%edx
1105 ENABLE_INTERRUPTS(CLBR_NONE)
1107 movq $-1,ORIG_RAX(%rsp)
1108 xorl %esi,%esi # oldset
1109 movq %rsp,%rdi # &pt_regs
1110 call do_notify_resume
1112 DISABLE_INTERRUPTS(CLBR_NONE)
1114 GET_THREAD_INFO(%rcx)
1115 jmp retint_with_reschedule
1117 #ifdef CONFIG_PREEMPT
1118 /* Returning to kernel space. Check if we need preemption */
1119 /* rcx: threadinfo. interrupts off. */
1120 ENTRY(retint_kernel)
1121 cmpl $0,TI_preempt_count(%rcx)
1122 jnz retint_restore_args
1123 bt $TIF_NEED_RESCHED,TI_flags(%rcx)
1124 jnc retint_restore_args
1125 bt $9,EFLAGS-ARGOFFSET(%rsp) /* interrupts off? */
1126 jnc retint_restore_args
1127 call preempt_schedule_irq
1132 END(common_interrupt)
1134 * End of kprobes section
1141 .macro apicinterrupt3 num sym do_sym
1153 #ifdef CONFIG_TRACING
1154 #define trace(sym) trace_##sym
1155 #define smp_trace(sym) smp_trace_##sym
1157 .macro trace_apicinterrupt num sym
1158 apicinterrupt3 \num trace(\sym) smp_trace(\sym)
1161 .macro trace_apicinterrupt num sym do_sym
1165 .macro apicinterrupt num sym do_sym
1166 apicinterrupt3 \num \sym \do_sym
1167 trace_apicinterrupt \num \sym
1171 apicinterrupt3 IRQ_MOVE_CLEANUP_VECTOR \
1172 irq_move_cleanup_interrupt smp_irq_move_cleanup_interrupt
1173 apicinterrupt3 REBOOT_VECTOR \
1174 reboot_interrupt smp_reboot_interrupt
1177 #ifdef CONFIG_X86_UV
1178 apicinterrupt3 UV_BAU_MESSAGE \
1179 uv_bau_message_intr1 uv_bau_message_interrupt
1181 apicinterrupt LOCAL_TIMER_VECTOR \
1182 apic_timer_interrupt smp_apic_timer_interrupt
1183 apicinterrupt X86_PLATFORM_IPI_VECTOR \
1184 x86_platform_ipi smp_x86_platform_ipi
1186 #ifdef CONFIG_HAVE_KVM
1187 apicinterrupt3 POSTED_INTR_VECTOR \
1188 kvm_posted_intr_ipi smp_kvm_posted_intr_ipi
1191 apicinterrupt THRESHOLD_APIC_VECTOR \
1192 threshold_interrupt smp_threshold_interrupt
1193 apicinterrupt THERMAL_APIC_VECTOR \
1194 thermal_interrupt smp_thermal_interrupt
1197 apicinterrupt CALL_FUNCTION_SINGLE_VECTOR \
1198 call_function_single_interrupt smp_call_function_single_interrupt
1199 apicinterrupt CALL_FUNCTION_VECTOR \
1200 call_function_interrupt smp_call_function_interrupt
1201 apicinterrupt RESCHEDULE_VECTOR \
1202 reschedule_interrupt smp_reschedule_interrupt
1205 apicinterrupt ERROR_APIC_VECTOR \
1206 error_interrupt smp_error_interrupt
1207 apicinterrupt SPURIOUS_APIC_VECTOR \
1208 spurious_interrupt smp_spurious_interrupt
1210 #ifdef CONFIG_IRQ_WORK
1211 apicinterrupt IRQ_WORK_VECTOR \
1212 irq_work_interrupt smp_irq_work_interrupt
1216 * Exception entry points.
1218 .macro zeroentry sym do_sym
1222 PARAVIRT_ADJUST_EXCEPTION_FRAME
1223 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1224 subq $ORIG_RAX-R15, %rsp
1225 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1228 movq %rsp,%rdi /* pt_regs pointer */
1229 xorl %esi,%esi /* no error code */
1231 jmp error_exit /* %ebx: no swapgs flag */
1236 .macro paranoidzeroentry sym do_sym
1240 PARAVIRT_ADJUST_EXCEPTION_FRAME
1241 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1242 subq $ORIG_RAX-R15, %rsp
1243 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1246 movq %rsp,%rdi /* pt_regs pointer */
1247 xorl %esi,%esi /* no error code */
1249 jmp paranoid_exit /* %ebx: no swapgs flag */
1254 #define INIT_TSS_IST(x) PER_CPU_VAR(init_tss) + (TSS_ist + ((x) - 1) * 8)
1255 .macro paranoidzeroentry_ist sym do_sym ist
1259 PARAVIRT_ADJUST_EXCEPTION_FRAME
1260 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1261 subq $ORIG_RAX-R15, %rsp
1262 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1264 TRACE_IRQS_OFF_DEBUG
1265 movq %rsp,%rdi /* pt_regs pointer */
1266 xorl %esi,%esi /* no error code */
1267 subq $EXCEPTION_STKSZ, INIT_TSS_IST(\ist)
1269 addq $EXCEPTION_STKSZ, INIT_TSS_IST(\ist)
1270 jmp paranoid_exit /* %ebx: no swapgs flag */
1275 .macro errorentry sym do_sym
1279 PARAVIRT_ADJUST_EXCEPTION_FRAME
1280 subq $ORIG_RAX-R15, %rsp
1281 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1284 movq %rsp,%rdi /* pt_regs pointer */
1285 movq ORIG_RAX(%rsp),%rsi /* get error code */
1286 movq $-1,ORIG_RAX(%rsp) /* no syscall to restart */
1288 jmp error_exit /* %ebx: no swapgs flag */
1293 /* error code is on the stack already */
1294 .macro paranoiderrorentry sym do_sym
1298 PARAVIRT_ADJUST_EXCEPTION_FRAME
1299 subq $ORIG_RAX-R15, %rsp
1300 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1304 movq %rsp,%rdi /* pt_regs pointer */
1305 movq ORIG_RAX(%rsp),%rsi /* get error code */
1306 movq $-1,ORIG_RAX(%rsp) /* no syscall to restart */
1308 jmp paranoid_exit /* %ebx: no swapgs flag */
1313 zeroentry divide_error do_divide_error
1314 zeroentry overflow do_overflow
1315 zeroentry bounds do_bounds
1316 zeroentry invalid_op do_invalid_op
1317 zeroentry device_not_available do_device_not_available
1318 paranoiderrorentry double_fault do_double_fault
1319 zeroentry coprocessor_segment_overrun do_coprocessor_segment_overrun
1320 errorentry invalid_TSS do_invalid_TSS
1321 errorentry segment_not_present do_segment_not_present
1322 zeroentry spurious_interrupt_bug do_spurious_interrupt_bug
1323 zeroentry coprocessor_error do_coprocessor_error
1324 errorentry alignment_check do_alignment_check
1325 zeroentry simd_coprocessor_error do_simd_coprocessor_error
1328 /* Reload gs selector with exception handling */
1329 /* edi: new selector */
1330 ENTRY(native_load_gs_index)
1333 DISABLE_INTERRUPTS(CLBR_ANY & ~CLBR_RDI)
1337 2: mfence /* workaround */
1342 END(native_load_gs_index)
1344 _ASM_EXTABLE(gs_change,bad_gs)
1345 .section .fixup,"ax"
1346 /* running with kernelgs */
1348 SWAPGS /* switch back to user gs */
1354 /* Call softirq on interrupt stack. Interrupts are off. */
1358 CFI_REL_OFFSET rbp,0
1360 CFI_DEF_CFA_REGISTER rbp
1361 incl PER_CPU_VAR(irq_count)
1362 cmove PER_CPU_VAR(irq_stack_ptr),%rsp
1363 push %rbp # backlink for old unwinder
1367 CFI_DEF_CFA_REGISTER rsp
1368 CFI_ADJUST_CFA_OFFSET -8
1369 decl PER_CPU_VAR(irq_count)
1375 zeroentry xen_hypervisor_callback xen_do_hypervisor_callback
1378 * A note on the "critical region" in our callback handler.
1379 * We want to avoid stacking callback handlers due to events occurring
1380 * during handling of the last event. To do this, we keep events disabled
1381 * until we've done all processing. HOWEVER, we must enable events before
1382 * popping the stack frame (can't be done atomically) and so it would still
1383 * be possible to get enough handler activations to overflow the stack.
1384 * Although unlikely, bugs of that kind are hard to track down, so we'd
1385 * like to avoid the possibility.
1386 * So, on entry to the handler we detect whether we interrupted an
1387 * existing activation in its critical region -- if so, we pop the current
1388 * activation and restart the handler using the previous one.
1390 ENTRY(xen_do_hypervisor_callback) # do_hypervisor_callback(struct *pt_regs)
1393 * Since we don't modify %rdi, evtchn_do_upall(struct *pt_regs) will
1394 * see the correct pointer to the pt_regs
1396 movq %rdi, %rsp # we don't return, adjust the stack frame
1399 11: incl PER_CPU_VAR(irq_count)
1401 CFI_DEF_CFA_REGISTER rbp
1402 cmovzq PER_CPU_VAR(irq_stack_ptr),%rsp
1403 pushq %rbp # backlink for old unwinder
1404 call xen_evtchn_do_upcall
1406 CFI_DEF_CFA_REGISTER rsp
1407 decl PER_CPU_VAR(irq_count)
1410 END(xen_do_hypervisor_callback)
1413 * Hypervisor uses this for application faults while it executes.
1414 * We get here for two reasons:
1415 * 1. Fault while reloading DS, ES, FS or GS
1416 * 2. Fault while executing IRET
1417 * Category 1 we do not need to fix up as Xen has already reloaded all segment
1418 * registers that could be reloaded and zeroed the others.
1419 * Category 2 we fix up by killing the current process. We cannot use the
1420 * normal Linux return path in this case because if we use the IRET hypercall
1421 * to pop the stack frame we end up in an infinite loop of failsafe callbacks.
1422 * We distinguish between categories by comparing each saved segment register
1423 * with its current contents: any discrepancy means we in category 1.
1425 ENTRY(xen_failsafe_callback)
1427 /*CFI_REL_OFFSET gs,GS*/
1428 /*CFI_REL_OFFSET fs,FS*/
1429 /*CFI_REL_OFFSET es,ES*/
1430 /*CFI_REL_OFFSET ds,DS*/
1431 CFI_REL_OFFSET r11,8
1432 CFI_REL_OFFSET rcx,0
1446 /* All segments match their saved values => Category 2 (Bad IRET). */
1452 CFI_ADJUST_CFA_OFFSET -0x30
1453 pushq_cfi $0 /* RIP */
1456 jmp general_protection
1458 1: /* Segment mismatch => Category 1 (Bad segment). Retry the IRET. */
1464 CFI_ADJUST_CFA_OFFSET -0x30
1465 pushq_cfi $-1 /* orig_ax = -1 => not a system call */
1469 END(xen_failsafe_callback)
1471 apicinterrupt3 HYPERVISOR_CALLBACK_VECTOR \
1472 xen_hvm_callback_vector xen_evtchn_do_upcall
1474 #endif /* CONFIG_XEN */
1476 #if IS_ENABLED(CONFIG_HYPERV)
1477 apicinterrupt3 HYPERVISOR_CALLBACK_VECTOR \
1478 hyperv_callback_vector hyperv_vector_handler
1479 #endif /* CONFIG_HYPERV */
1482 * Some functions should be protected against kprobes
1484 .pushsection .kprobes.text, "ax"
1486 paranoidzeroentry_ist debug do_debug DEBUG_STACK
1487 paranoidzeroentry_ist int3 do_int3 DEBUG_STACK
1488 paranoiderrorentry stack_segment do_stack_segment
1490 zeroentry xen_debug do_debug
1491 zeroentry xen_int3 do_int3
1492 errorentry xen_stack_segment do_stack_segment
1494 errorentry general_protection do_general_protection
1495 errorentry page_fault do_page_fault
1496 #ifdef CONFIG_KVM_GUEST
1497 errorentry async_page_fault do_async_page_fault
1499 #ifdef CONFIG_X86_MCE
1500 paranoidzeroentry machine_check *machine_check_vector(%rip)
1504 * "Paranoid" exit path from exception stack.
1505 * Paranoid because this is used by NMIs and cannot take
1506 * any kernel state for granted.
1507 * We don't do kernel preemption checks here, because only
1508 * NMI should be common and it does not enable IRQs and
1509 * cannot get reschedule ticks.
1511 * "trace" is 0 for the NMI handler only, because irq-tracing
1512 * is fundamentally NMI-unsafe. (we cannot change the soft and
1513 * hard flags at once, atomically)
1516 /* ebx: no swapgs flag */
1517 ENTRY(paranoid_exit)
1519 DISABLE_INTERRUPTS(CLBR_NONE)
1520 TRACE_IRQS_OFF_DEBUG
1521 testl %ebx,%ebx /* swapgs needed? */
1522 jnz paranoid_restore
1524 jnz paranoid_userspace
1531 TRACE_IRQS_IRETQ_DEBUG 0
1535 GET_THREAD_INFO(%rcx)
1536 movl TI_flags(%rcx),%ebx
1537 andl $_TIF_WORK_MASK,%ebx
1539 movq %rsp,%rdi /* &pt_regs */
1541 movq %rax,%rsp /* switch stack for scheduling */
1542 testl $_TIF_NEED_RESCHED,%ebx
1543 jnz paranoid_schedule
1544 movl %ebx,%edx /* arg3: thread flags */
1546 ENABLE_INTERRUPTS(CLBR_NONE)
1547 xorl %esi,%esi /* arg2: oldset */
1548 movq %rsp,%rdi /* arg1: &pt_regs */
1549 call do_notify_resume
1550 DISABLE_INTERRUPTS(CLBR_NONE)
1552 jmp paranoid_userspace
1555 ENABLE_INTERRUPTS(CLBR_ANY)
1557 DISABLE_INTERRUPTS(CLBR_ANY)
1559 jmp paranoid_userspace
1564 * Exception entry point. This expects an error code/orig_rax on the stack.
1565 * returns in "no swapgs flag" in %ebx.
1569 CFI_ADJUST_CFA_OFFSET 15*8
1570 /* oldrax contains error code */
1589 je error_kernelspace
1597 * There are two places in the kernel that can potentially fault with
1598 * usergs. Handle them here. The exception handlers after iret run with
1599 * kernel gs again, so don't set the user space flag. B stepping K8s
1600 * sometimes report an truncated RIP for IRET exceptions returning to
1601 * compat mode. Check for these here too.
1605 leaq irq_return(%rip),%rcx
1606 cmpq %rcx,RIP+8(%rsp)
1608 movl %ecx,%eax /* zero extend */
1609 cmpq %rax,RIP+8(%rsp)
1611 cmpq $gs_change,RIP+8(%rsp)
1616 /* Fix truncated RIP */
1617 movq %rcx,RIP+8(%rsp)
1623 /* ebx: no swapgs flag (1: don't need swapgs, 0: need it) */
1628 DISABLE_INTERRUPTS(CLBR_NONE)
1630 GET_THREAD_INFO(%rcx)
1633 LOCKDEP_SYS_EXIT_IRQ
1634 movl TI_flags(%rcx),%edx
1635 movl $_TIF_WORK_MASK,%edi
1643 * Test if a given stack is an NMI stack or not.
1645 .macro test_in_nmi reg stack nmi_ret normal_ret
1648 subq $EXCEPTION_STKSZ, %\reg
1654 /* runs on exception stack */
1657 PARAVIRT_ADJUST_EXCEPTION_FRAME
1659 * We allow breakpoints in NMIs. If a breakpoint occurs, then
1660 * the iretq it performs will take us out of NMI context.
1661 * This means that we can have nested NMIs where the next
1662 * NMI is using the top of the stack of the previous NMI. We
1663 * can't let it execute because the nested NMI will corrupt the
1664 * stack of the previous NMI. NMI handlers are not re-entrant
1667 * To handle this case we do the following:
1668 * Check the a special location on the stack that contains
1669 * a variable that is set when NMIs are executing.
1670 * The interrupted task's stack is also checked to see if it
1672 * If the variable is not set and the stack is not the NMI
1674 * o Set the special variable on the stack
1675 * o Copy the interrupt frame into a "saved" location on the stack
1676 * o Copy the interrupt frame into a "copy" location on the stack
1677 * o Continue processing the NMI
1678 * If the variable is set or the previous stack is the NMI stack:
1679 * o Modify the "copy" location to jump to the repeate_nmi
1680 * o return back to the first NMI
1682 * Now on exit of the first NMI, we first clear the stack variable
1683 * The NMI stack will tell any nested NMIs at that point that it is
1684 * nested. Then we pop the stack normally with iret, and if there was
1685 * a nested NMI that updated the copy interrupt stack frame, a
1686 * jump will be made to the repeat_nmi code that will handle the second
1690 /* Use %rdx as out temp variable throughout */
1692 CFI_REL_OFFSET rdx, 0
1695 * If %cs was not the kernel segment, then the NMI triggered in user
1696 * space, which means it is definitely not nested.
1698 cmpl $__KERNEL_CS, 16(%rsp)
1702 * Check the special variable on the stack to see if NMIs are
1709 * Now test if the previous stack was an NMI stack.
1710 * We need the double check. We check the NMI stack to satisfy the
1711 * race when the first NMI clears the variable before returning.
1712 * We check the variable because the first NMI could be in a
1713 * breakpoint routine using a breakpoint stack.
1716 test_in_nmi rdx, 4*8(%rsp), nested_nmi, first_nmi
1721 * Do nothing if we interrupted the fixup in repeat_nmi.
1722 * It's about to repeat the NMI handler, so we are fine
1723 * with ignoring this one.
1725 movq $repeat_nmi, %rdx
1728 movq $end_repeat_nmi, %rdx
1733 /* Set up the interrupted NMIs stack to jump to repeat_nmi */
1734 leaq -1*8(%rsp), %rdx
1736 CFI_ADJUST_CFA_OFFSET 1*8
1737 leaq -10*8(%rsp), %rdx
1738 pushq_cfi $__KERNEL_DS
1741 pushq_cfi $__KERNEL_CS
1742 pushq_cfi $repeat_nmi
1744 /* Put stack back */
1746 CFI_ADJUST_CFA_OFFSET -6*8
1752 /* No need to check faults here */
1758 * Because nested NMIs will use the pushed location that we
1759 * stored in rdx, we must keep that space available.
1760 * Here's what our stack frame will look like:
1761 * +-------------------------+
1763 * | original Return RSP |
1764 * | original RFLAGS |
1767 * +-------------------------+
1768 * | temp storage for rdx |
1769 * +-------------------------+
1770 * | NMI executing variable |
1771 * +-------------------------+
1773 * | copied Return RSP |
1777 * +-------------------------+
1779 * | Saved Return RSP |
1783 * +-------------------------+
1785 * +-------------------------+
1787 * The saved stack frame is used to fix up the copied stack frame
1788 * that a nested NMI may change to make the interrupted NMI iret jump
1789 * to the repeat_nmi. The original stack frame and the temp storage
1790 * is also used by nested NMIs and can not be trusted on exit.
1792 /* Do not pop rdx, nested NMIs will corrupt that part of the stack */
1796 /* Set the NMI executing variable on the stack. */
1800 * Leave room for the "copied" frame
1803 CFI_ADJUST_CFA_OFFSET 5*8
1805 /* Copy the stack frame to the Saved frame */
1807 pushq_cfi 11*8(%rsp)
1809 CFI_DEF_CFA_OFFSET SS+8-RIP
1811 /* Everything up to here is safe from nested NMIs */
1814 * If there was a nested NMI, the first NMI's iret will return
1815 * here. But NMIs are still enabled and we can take another
1816 * nested NMI. The nested NMI checks the interrupted RIP to see
1817 * if it is between repeat_nmi and end_repeat_nmi, and if so
1818 * it will just return, as we are about to repeat an NMI anyway.
1819 * This makes it safe to copy to the stack frame that a nested
1824 * Update the stack variable to say we are still in NMI (the update
1825 * is benign for the non-repeat case, where 1 was pushed just above
1826 * to this very stack slot).
1830 /* Make another copy, this one may be modified by nested NMIs */
1832 CFI_ADJUST_CFA_OFFSET -10*8
1834 pushq_cfi -6*8(%rsp)
1837 CFI_DEF_CFA_OFFSET SS+8-RIP
1841 * Everything below this point can be preempted by a nested
1842 * NMI if the first NMI took an exception and reset our iret stack
1843 * so that we repeat another NMI.
1845 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1846 subq $ORIG_RAX-R15, %rsp
1847 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1849 * Use save_paranoid to handle SWAPGS, but no need to use paranoid_exit
1850 * as we should not be calling schedule in NMI context.
1851 * Even with normal interrupts enabled. An NMI should not be
1852 * setting NEED_RESCHED or anything that normal interrupts and
1853 * exceptions might do.
1859 * Save off the CR2 register. If we take a page fault in the NMI then
1860 * it could corrupt the CR2 value. If the NMI preempts a page fault
1861 * handler before it was able to read the CR2 register, and then the
1862 * NMI itself takes a page fault, the page fault that was preempted
1863 * will read the information from the NMI page fault and not the
1864 * origin fault. Save it off and restore it if it changes.
1865 * Use the r12 callee-saved register.
1869 /* paranoidentry do_nmi, 0; without TRACE_IRQS_OFF */
1874 /* Did the NMI take a page fault? Restore cr2 if it did */
1881 testl %ebx,%ebx /* swapgs needed? */
1886 /* Pop the extra iret frame at once */
1889 /* Clear the NMI executing stack variable */
1895 ENTRY(ignore_sysret)
1903 * End of kprobes section