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
77 cmpl $0, function_trace_stop
82 leaq function_trace_op, %rdx
85 subq $MCOUNT_INSN_SIZE, %rdi
92 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
93 GLOBAL(ftrace_graph_call)
101 #else /* ! CONFIG_DYNAMIC_FTRACE */
103 cmpl $0, function_trace_stop
106 cmpq $ftrace_stub, ftrace_trace_function
109 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
110 cmpq $ftrace_stub, ftrace_graph_return
111 jnz ftrace_graph_caller
113 cmpq $ftrace_graph_entry_stub, ftrace_graph_entry
114 jnz ftrace_graph_caller
123 movq 0x38(%rsp), %rdi
125 subq $MCOUNT_INSN_SIZE, %rdi
127 call *ftrace_trace_function
133 #endif /* CONFIG_DYNAMIC_FTRACE */
134 #endif /* CONFIG_FUNCTION_TRACER */
136 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
137 ENTRY(ftrace_graph_caller)
138 cmpl $0, function_trace_stop
144 movq 0x38(%rsp), %rsi
146 subq $MCOUNT_INSN_SIZE, %rsi
148 call prepare_ftrace_return
153 END(ftrace_graph_caller)
155 GLOBAL(return_to_handler)
158 /* Save the return values */
163 call ftrace_return_to_handler
173 #ifndef CONFIG_PREEMPT
174 #define retint_kernel retint_restore_args
177 #ifdef CONFIG_PARAVIRT
178 ENTRY(native_usergs_sysret64)
181 ENDPROC(native_usergs_sysret64)
182 #endif /* CONFIG_PARAVIRT */
185 .macro TRACE_IRQS_IRETQ offset=ARGOFFSET
186 #ifdef CONFIG_TRACE_IRQFLAGS
187 bt $9,EFLAGS-\offset(%rsp) /* interrupts off? */
195 * When dynamic function tracer is enabled it will add a breakpoint
196 * to all locations that it is about to modify, sync CPUs, update
197 * all the code, sync CPUs, then remove the breakpoints. In this time
198 * if lockdep is enabled, it might jump back into the debug handler
199 * outside the updating of the IST protection. (TRACE_IRQS_ON/OFF).
201 * We need to change the IDT table before calling TRACE_IRQS_ON/OFF to
202 * make sure the stack pointer does not get reset back to the top
203 * of the debug stack, and instead just reuses the current stack.
205 #if defined(CONFIG_DYNAMIC_FTRACE) && defined(CONFIG_TRACE_IRQFLAGS)
207 .macro TRACE_IRQS_OFF_DEBUG
208 call debug_stack_set_zero
210 call debug_stack_reset
213 .macro TRACE_IRQS_ON_DEBUG
214 call debug_stack_set_zero
216 call debug_stack_reset
219 .macro TRACE_IRQS_IRETQ_DEBUG offset=ARGOFFSET
220 bt $9,EFLAGS-\offset(%rsp) /* interrupts off? */
227 # define TRACE_IRQS_OFF_DEBUG TRACE_IRQS_OFF
228 # define TRACE_IRQS_ON_DEBUG TRACE_IRQS_ON
229 # define TRACE_IRQS_IRETQ_DEBUG TRACE_IRQS_IRETQ
233 * C code is not supposed to know about undefined top of stack. Every time
234 * a C function with an pt_regs argument is called from the SYSCALL based
235 * fast path FIXUP_TOP_OF_STACK is needed.
236 * RESTORE_TOP_OF_STACK syncs the syscall state after any possible ptregs
240 /* %rsp:at FRAMEEND */
241 .macro FIXUP_TOP_OF_STACK tmp offset=0
242 movq PER_CPU_VAR(old_rsp),\tmp
243 movq \tmp,RSP+\offset(%rsp)
244 movq $__USER_DS,SS+\offset(%rsp)
245 movq $__USER_CS,CS+\offset(%rsp)
246 movq $-1,RCX+\offset(%rsp)
247 movq R11+\offset(%rsp),\tmp /* get eflags */
248 movq \tmp,EFLAGS+\offset(%rsp)
251 .macro RESTORE_TOP_OF_STACK tmp offset=0
252 movq RSP+\offset(%rsp),\tmp
253 movq \tmp,PER_CPU_VAR(old_rsp)
254 movq EFLAGS+\offset(%rsp),\tmp
255 movq \tmp,R11+\offset(%rsp)
258 .macro FAKE_STACK_FRAME child_rip
259 /* push in order ss, rsp, eflags, cs, rip */
261 pushq_cfi $__KERNEL_DS /* ss */
262 /*CFI_REL_OFFSET ss,0*/
263 pushq_cfi %rax /* rsp */
265 pushq_cfi $(X86_EFLAGS_IF|X86_EFLAGS_BIT1) /* eflags - interrupts on */
266 /*CFI_REL_OFFSET rflags,0*/
267 pushq_cfi $__KERNEL_CS /* cs */
268 /*CFI_REL_OFFSET cs,0*/
269 pushq_cfi \child_rip /* rip */
271 pushq_cfi %rax /* orig rax */
274 .macro UNFAKE_STACK_FRAME
276 CFI_ADJUST_CFA_OFFSET -(6*8)
280 * initial frame state for interrupts (and exceptions without error code)
282 .macro EMPTY_FRAME start=1 offset=0
286 CFI_DEF_CFA rsp,8+\offset
288 CFI_DEF_CFA_OFFSET 8+\offset
293 * initial frame state for interrupts (and exceptions without error code)
295 .macro INTR_FRAME start=1 offset=0
296 EMPTY_FRAME \start, SS+8+\offset-RIP
297 /*CFI_REL_OFFSET ss, SS+\offset-RIP*/
298 CFI_REL_OFFSET rsp, RSP+\offset-RIP
299 /*CFI_REL_OFFSET rflags, EFLAGS+\offset-RIP*/
300 /*CFI_REL_OFFSET cs, CS+\offset-RIP*/
301 CFI_REL_OFFSET rip, RIP+\offset-RIP
305 * initial frame state for exceptions with error code (and interrupts
306 * with vector already pushed)
308 .macro XCPT_FRAME start=1 offset=0
309 INTR_FRAME \start, RIP+\offset-ORIG_RAX
310 /*CFI_REL_OFFSET orig_rax, ORIG_RAX-ORIG_RAX*/
314 * frame that enables calling into C.
316 .macro PARTIAL_FRAME start=1 offset=0
317 XCPT_FRAME \start, ORIG_RAX+\offset-ARGOFFSET
318 CFI_REL_OFFSET rdi, RDI+\offset-ARGOFFSET
319 CFI_REL_OFFSET rsi, RSI+\offset-ARGOFFSET
320 CFI_REL_OFFSET rdx, RDX+\offset-ARGOFFSET
321 CFI_REL_OFFSET rcx, RCX+\offset-ARGOFFSET
322 CFI_REL_OFFSET rax, RAX+\offset-ARGOFFSET
323 CFI_REL_OFFSET r8, R8+\offset-ARGOFFSET
324 CFI_REL_OFFSET r9, R9+\offset-ARGOFFSET
325 CFI_REL_OFFSET r10, R10+\offset-ARGOFFSET
326 CFI_REL_OFFSET r11, R11+\offset-ARGOFFSET
330 * frame that enables passing a complete pt_regs to a C function.
332 .macro DEFAULT_FRAME start=1 offset=0
333 PARTIAL_FRAME \start, R11+\offset-R15
334 CFI_REL_OFFSET rbx, RBX+\offset
335 CFI_REL_OFFSET rbp, RBP+\offset
336 CFI_REL_OFFSET r12, R12+\offset
337 CFI_REL_OFFSET r13, R13+\offset
338 CFI_REL_OFFSET r14, R14+\offset
339 CFI_REL_OFFSET r15, R15+\offset
342 /* save partial stack frame */
345 /* start from rbp in pt_regs and jump over */
346 movq_cfi rdi, RDI-RBP
347 movq_cfi rsi, RSI-RBP
348 movq_cfi rdx, RDX-RBP
349 movq_cfi rcx, RCX-RBP
350 movq_cfi rax, RAX-RBP
353 movq_cfi r10, R10-RBP
354 movq_cfi r11, R11-RBP
356 /* Save rbp so that we can unwind from get_irq_regs() */
359 /* Save previous stack value */
362 leaq -RBP(%rsp),%rdi /* arg1 for handler */
363 testl $3, CS-RBP(%rsi)
367 * irq_count is used to check if a CPU is already on an interrupt stack
368 * or not. While this is essentially redundant with preempt_count it is
369 * a little cheaper to use a separate counter in the PDA (short of
370 * moving irq_enter into assembly, which would be too much work)
372 1: incl PER_CPU_VAR(irq_count)
373 cmovzq PER_CPU_VAR(irq_stack_ptr),%rsp
374 CFI_DEF_CFA_REGISTER rsi
376 /* Store previous stack value */
378 CFI_ESCAPE 0x0f /* DW_CFA_def_cfa_expression */, 6, \
379 0x77 /* DW_OP_breg7 */, 0, \
380 0x06 /* DW_OP_deref */, \
381 0x08 /* DW_OP_const1u */, SS+8-RBP, \
382 0x22 /* DW_OP_plus */
383 /* We entered an interrupt context - irqs are off: */
388 PARTIAL_FRAME 1 REST_SKIP+8
389 movq 5*8+16(%rsp), %r11 /* save return address */
396 movq %r11, 8(%rsp) /* return address */
397 FIXUP_TOP_OF_STACK %r11, 16
402 /* save complete stack frame */
403 .pushsection .kprobes.text, "ax"
423 movl $MSR_GS_BASE,%ecx
426 js 1f /* negative -> in kernel */
435 * A newly forked process directly context switches into this address.
437 * rdi: prev task we switched from
442 LOCK ; btr $TIF_FORK,TI_flags(%r8)
444 pushq_cfi kernel_eflags(%rip)
445 popfq_cfi # reset kernel eflags
447 call schedule_tail # rdi: 'prev' task parameter
449 GET_THREAD_INFO(%rcx)
453 testl $3, CS-ARGOFFSET(%rsp) # from kernel_thread?
454 jz retint_restore_args
456 testl $_TIF_IA32, TI_flags(%rcx) # 32-bit compat task needs IRET
457 jnz int_ret_from_sys_call
459 RESTORE_TOP_OF_STACK %rdi, -ARGOFFSET
460 jmp ret_from_sys_call # go to the SYSRET fastpath
466 * System call entry. Up to 6 arguments in registers are supported.
468 * SYSCALL does not save anything on the stack and does not change the
474 * rax system call number
476 * rcx return address for syscall/sysret, C arg3
479 * r10 arg3 (--> moved to rcx for C)
482 * r11 eflags for syscall/sysret, temporary for C
483 * r12-r15,rbp,rbx saved by C code, not touched.
485 * Interrupts are off on entry.
486 * Only called from user space.
488 * XXX if we had a free scratch register we could save the RSP into the stack frame
489 * and report it properly in ps. Unfortunately we haven't.
491 * When user can change the frames always force IRET. That is because
492 * it deals with uncanonical addresses better. SYSRET has trouble
493 * with them due to bugs in both AMD and Intel CPUs.
499 CFI_DEF_CFA rsp,KERNEL_STACK_OFFSET
501 /*CFI_REGISTER rflags,r11*/
504 * A hypervisor implementation might want to use a label
505 * after the swapgs, so that it can do the swapgs
506 * for the guest and jump here on syscall.
508 GLOBAL(system_call_after_swapgs)
510 movq %rsp,PER_CPU_VAR(old_rsp)
511 movq PER_CPU_VAR(kernel_stack),%rsp
513 * No need to follow this irqs off/on section - it's straight
516 ENABLE_INTERRUPTS(CLBR_NONE)
518 movq %rax,ORIG_RAX-ARGOFFSET(%rsp)
519 movq %rcx,RIP-ARGOFFSET(%rsp)
520 CFI_REL_OFFSET rip,RIP-ARGOFFSET
521 testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET)
523 system_call_fastpath:
524 #if __SYSCALL_MASK == ~0
525 cmpq $__NR_syscall_max,%rax
527 andl $__SYSCALL_MASK,%eax
528 cmpl $__NR_syscall_max,%eax
532 call *sys_call_table(,%rax,8) # XXX: rip relative
533 movq %rax,RAX-ARGOFFSET(%rsp)
535 * Syscall return path ending with SYSRET (fast path)
536 * Has incomplete stack frame and undefined top of stack.
539 movl $_TIF_ALLWORK_MASK,%edi
543 DISABLE_INTERRUPTS(CLBR_NONE)
545 movl TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET),%edx
550 * sysretq will re-enable interrupts:
553 movq RIP-ARGOFFSET(%rsp),%rcx
555 RESTORE_ARGS 1,-ARG_SKIP,0
556 /*CFI_REGISTER rflags,r11*/
557 movq PER_CPU_VAR(old_rsp), %rsp
561 /* Handle reschedules */
562 /* edx: work, edi: workmask */
564 bt $TIF_NEED_RESCHED,%edx
567 ENABLE_INTERRUPTS(CLBR_NONE)
573 /* Handle a signal */
576 ENABLE_INTERRUPTS(CLBR_NONE)
577 #ifdef CONFIG_AUDITSYSCALL
578 bt $TIF_SYSCALL_AUDIT,%edx
582 * We have a signal, or exit tracing or single-step.
583 * These all wind up with the iret return path anyway,
584 * so just join that path right now.
586 FIXUP_TOP_OF_STACK %r11, -ARGOFFSET
587 jmp int_check_syscall_exit_work
590 movq $-ENOSYS,RAX-ARGOFFSET(%rsp)
591 jmp ret_from_sys_call
593 #ifdef CONFIG_AUDITSYSCALL
595 * Fast path for syscall audit without full syscall trace.
596 * We just call __audit_syscall_entry() directly, and then
597 * jump back to the normal fast path.
600 movq %r10,%r9 /* 6th arg: 4th syscall arg */
601 movq %rdx,%r8 /* 5th arg: 3rd syscall arg */
602 movq %rsi,%rcx /* 4th arg: 2nd syscall arg */
603 movq %rdi,%rdx /* 3rd arg: 1st syscall arg */
604 movq %rax,%rsi /* 2nd arg: syscall number */
605 movl $AUDIT_ARCH_X86_64,%edi /* 1st arg: audit arch */
606 call __audit_syscall_entry
607 LOAD_ARGS 0 /* reload call-clobbered registers */
608 jmp system_call_fastpath
611 * Return fast path for syscall audit. Call __audit_syscall_exit()
612 * directly and then jump back to the fast path with TIF_SYSCALL_AUDIT
616 movq RAX-ARGOFFSET(%rsp),%rsi /* second arg, syscall return value */
617 cmpq $-MAX_ERRNO,%rsi /* is it < -MAX_ERRNO? */
618 setbe %al /* 1 if so, 0 if not */
619 movzbl %al,%edi /* zero-extend that into %edi */
620 call __audit_syscall_exit
621 movl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT),%edi
623 #endif /* CONFIG_AUDITSYSCALL */
625 /* Do syscall tracing */
627 #ifdef CONFIG_AUDITSYSCALL
628 testl $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET)
632 movq $-ENOSYS,RAX(%rsp) /* ptrace can change this for a bad syscall */
633 FIXUP_TOP_OF_STACK %rdi
635 call syscall_trace_enter
637 * Reload arg registers from stack in case ptrace changed them.
638 * We don't reload %rax because syscall_trace_enter() returned
639 * the value it wants us to use in the table lookup.
641 LOAD_ARGS ARGOFFSET, 1
643 #if __SYSCALL_MASK == ~0
644 cmpq $__NR_syscall_max,%rax
646 andl $__SYSCALL_MASK,%eax
647 cmpl $__NR_syscall_max,%eax
649 ja int_ret_from_sys_call /* RAX(%rsp) set to -ENOSYS above */
650 movq %r10,%rcx /* fixup for C */
651 call *sys_call_table(,%rax,8)
652 movq %rax,RAX-ARGOFFSET(%rsp)
653 /* Use IRET because user could have changed frame */
656 * Syscall return path ending with IRET.
657 * Has correct top of stack, but partial stack frame.
659 GLOBAL(int_ret_from_sys_call)
660 DISABLE_INTERRUPTS(CLBR_NONE)
662 movl $_TIF_ALLWORK_MASK,%edi
663 /* edi: mask to check */
664 GLOBAL(int_with_check)
666 GET_THREAD_INFO(%rcx)
667 movl TI_flags(%rcx),%edx
670 andl $~TS_COMPAT,TI_status(%rcx)
673 /* Either reschedule or signal or syscall exit tracking needed. */
674 /* First do a reschedule test. */
675 /* edx: work, edi: workmask */
677 bt $TIF_NEED_RESCHED,%edx
680 ENABLE_INTERRUPTS(CLBR_NONE)
684 DISABLE_INTERRUPTS(CLBR_NONE)
688 /* handle signals and tracing -- both require a full stack frame */
691 ENABLE_INTERRUPTS(CLBR_NONE)
692 int_check_syscall_exit_work:
694 /* Check for syscall exit trace */
695 testl $_TIF_WORK_SYSCALL_EXIT,%edx
698 leaq 8(%rsp),%rdi # &ptregs -> arg1
699 call syscall_trace_leave
701 andl $~(_TIF_WORK_SYSCALL_EXIT|_TIF_SYSCALL_EMU),%edi
705 testl $_TIF_DO_NOTIFY_MASK,%edx
707 movq %rsp,%rdi # &ptregs -> arg1
708 xorl %esi,%esi # oldset -> arg2
709 call do_notify_resume
710 1: movl $_TIF_WORK_MASK,%edi
713 DISABLE_INTERRUPTS(CLBR_NONE)
720 * Certain special system calls that need to save a complete full stack frame.
722 .macro PTREGSCALL label,func,arg
724 PARTIAL_FRAME 1 8 /* offset 8: return address */
725 subq $REST_SKIP, %rsp
726 CFI_ADJUST_CFA_OFFSET REST_SKIP
728 DEFAULT_FRAME 0 8 /* offset 8: return address */
729 leaq 8(%rsp), \arg /* pt_regs pointer */
731 jmp ptregscall_common
736 PTREGSCALL stub_clone, sys_clone, %r8
737 PTREGSCALL stub_fork, sys_fork, %rdi
738 PTREGSCALL stub_vfork, sys_vfork, %rdi
739 PTREGSCALL stub_sigaltstack, sys_sigaltstack, %rdx
740 PTREGSCALL stub_iopl, sys_iopl, %rsi
742 ENTRY(ptregscall_common)
743 DEFAULT_FRAME 1 8 /* offset 8: return address */
744 RESTORE_TOP_OF_STACK %r11, 8
745 movq_cfi_restore R15+8, r15
746 movq_cfi_restore R14+8, r14
747 movq_cfi_restore R13+8, r13
748 movq_cfi_restore R12+8, r12
749 movq_cfi_restore RBP+8, rbp
750 movq_cfi_restore RBX+8, rbx
751 ret $REST_SKIP /* pop extended registers */
753 END(ptregscall_common)
760 FIXUP_TOP_OF_STACK %r11
763 RESTORE_TOP_OF_STACK %r11
766 jmp int_ret_from_sys_call
771 * sigreturn is special because it needs to restore all registers on return.
772 * This cannot be done with SYSRET, so use the IRET return path instead.
774 ENTRY(stub_rt_sigreturn)
780 FIXUP_TOP_OF_STACK %r11
781 call sys_rt_sigreturn
782 movq %rax,RAX(%rsp) # fixme, this could be done at the higher layer
784 jmp int_ret_from_sys_call
786 END(stub_rt_sigreturn)
788 #ifdef CONFIG_X86_X32_ABI
789 PTREGSCALL stub_x32_sigaltstack, sys32_sigaltstack, %rdx
791 ENTRY(stub_x32_rt_sigreturn)
797 FIXUP_TOP_OF_STACK %r11
798 call sys32_x32_rt_sigreturn
799 movq %rax,RAX(%rsp) # fixme, this could be done at the higher layer
801 jmp int_ret_from_sys_call
803 END(stub_x32_rt_sigreturn)
805 ENTRY(stub_x32_execve)
810 FIXUP_TOP_OF_STACK %r11
813 RESTORE_TOP_OF_STACK %r11
816 jmp int_ret_from_sys_call
823 * Build the entry stubs and pointer table with some assembler magic.
824 * We pack 7 stubs into a single 32-byte chunk, which will fit in a
825 * single cache line on all modern x86 implementations.
827 .section .init.rodata,"a"
831 .p2align CONFIG_X86_L1_CACHE_SHIFT
832 ENTRY(irq_entries_start)
834 vector=FIRST_EXTERNAL_VECTOR
835 .rept (NR_VECTORS-FIRST_EXTERNAL_VECTOR+6)/7
838 .if vector < NR_VECTORS
839 .if vector <> FIRST_EXTERNAL_VECTOR
840 CFI_ADJUST_CFA_OFFSET -8
842 1: pushq_cfi $(~vector+0x80) /* Note: always in signed byte range */
843 .if ((vector-FIRST_EXTERNAL_VECTOR)%7) <> 6
852 2: jmp common_interrupt
855 END(irq_entries_start)
862 * Interrupt entry/exit.
864 * Interrupt entry points save only callee clobbered registers in fast path.
866 * Entry runs with interrupts off.
869 /* 0(%rsp): ~(interrupt number) */
870 .macro interrupt func
871 /* reserve pt_regs for scratch regs and rbp */
872 subq $ORIG_RAX-RBP, %rsp
873 CFI_ADJUST_CFA_OFFSET ORIG_RAX-RBP
879 * Interrupt entry/exit should be protected against kprobes
881 .pushsection .kprobes.text, "ax"
883 * The interrupt stubs push (~vector+0x80) onto the stack and
884 * then jump to common_interrupt.
886 .p2align CONFIG_X86_L1_CACHE_SHIFT
889 addq $-0x80,(%rsp) /* Adjust vector to [-256,-1] range */
891 /* 0(%rsp): old_rsp-ARGOFFSET */
893 DISABLE_INTERRUPTS(CLBR_NONE)
895 decl PER_CPU_VAR(irq_count)
897 /* Restore saved previous stack */
899 CFI_DEF_CFA rsi,SS+8-RBP /* reg/off reset after def_cfa_expr */
900 leaq ARGOFFSET-RBP(%rsi), %rsp
901 CFI_DEF_CFA_REGISTER rsp
902 CFI_ADJUST_CFA_OFFSET RBP-ARGOFFSET
905 GET_THREAD_INFO(%rcx)
906 testl $3,CS-ARGOFFSET(%rsp)
909 /* Interrupt came from user space */
911 * Has a correct top of stack, but a partial stack frame
912 * %rcx: thread info. Interrupts off.
914 retint_with_reschedule:
915 movl $_TIF_WORK_MASK,%edi
918 movl TI_flags(%rcx),%edx
923 retint_swapgs: /* return to user-space */
925 * The iretq could re-enable interrupts:
927 DISABLE_INTERRUPTS(CLBR_ANY)
932 retint_restore_args: /* return to kernel space */
933 DISABLE_INTERRUPTS(CLBR_ANY)
935 * The iretq could re-enable interrupts:
943 _ASM_EXTABLE(irq_return, bad_iret)
945 #ifdef CONFIG_PARAVIRT
948 _ASM_EXTABLE(native_iret, bad_iret)
954 * The iret traps when the %cs or %ss being restored is bogus.
955 * We've lost the original trap vector and error code.
956 * #GPF is the most likely one to get for an invalid selector.
957 * So pretend we completed the iret and took the #GPF in user mode.
959 * We are now running with the kernel GS after exception recovery.
960 * But error_entry expects us to have user GS to match the user %cs,
966 jmp general_protection
970 /* edi: workmask, edx: work */
973 bt $TIF_NEED_RESCHED,%edx
976 ENABLE_INTERRUPTS(CLBR_NONE)
980 GET_THREAD_INFO(%rcx)
981 DISABLE_INTERRUPTS(CLBR_NONE)
986 testl $_TIF_DO_NOTIFY_MASK,%edx
989 ENABLE_INTERRUPTS(CLBR_NONE)
991 movq $-1,ORIG_RAX(%rsp)
992 xorl %esi,%esi # oldset
993 movq %rsp,%rdi # &pt_regs
994 call do_notify_resume
996 DISABLE_INTERRUPTS(CLBR_NONE)
998 GET_THREAD_INFO(%rcx)
999 jmp retint_with_reschedule
1001 #ifdef CONFIG_PREEMPT
1002 /* Returning to kernel space. Check if we need preemption */
1003 /* rcx: threadinfo. interrupts off. */
1004 ENTRY(retint_kernel)
1005 cmpl $0,TI_preempt_count(%rcx)
1006 jnz retint_restore_args
1007 bt $TIF_NEED_RESCHED,TI_flags(%rcx)
1008 jnc retint_restore_args
1009 bt $9,EFLAGS-ARGOFFSET(%rsp) /* interrupts off? */
1010 jnc retint_restore_args
1011 call preempt_schedule_irq
1016 END(common_interrupt)
1018 * End of kprobes section
1025 .macro apicinterrupt num sym do_sym
1037 apicinterrupt IRQ_MOVE_CLEANUP_VECTOR \
1038 irq_move_cleanup_interrupt smp_irq_move_cleanup_interrupt
1039 apicinterrupt REBOOT_VECTOR \
1040 reboot_interrupt smp_reboot_interrupt
1043 #ifdef CONFIG_X86_UV
1044 apicinterrupt UV_BAU_MESSAGE \
1045 uv_bau_message_intr1 uv_bau_message_interrupt
1047 apicinterrupt LOCAL_TIMER_VECTOR \
1048 apic_timer_interrupt smp_apic_timer_interrupt
1049 apicinterrupt X86_PLATFORM_IPI_VECTOR \
1050 x86_platform_ipi smp_x86_platform_ipi
1055 .irp idx,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15, \
1056 16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31
1057 .if NUM_INVALIDATE_TLB_VECTORS > \idx
1058 ENTRY(invalidate_interrupt\idx)
1059 pushq_cfi $~(INVALIDATE_TLB_VECTOR_START+\idx)
1060 jmp .Lcommon_invalidate_interrupt0
1061 CFI_ADJUST_CFA_OFFSET -8
1062 END(invalidate_interrupt\idx)
1066 apicinterrupt INVALIDATE_TLB_VECTOR_START, \
1067 invalidate_interrupt0, smp_invalidate_interrupt
1070 apicinterrupt THRESHOLD_APIC_VECTOR \
1071 threshold_interrupt smp_threshold_interrupt
1072 apicinterrupt THERMAL_APIC_VECTOR \
1073 thermal_interrupt smp_thermal_interrupt
1076 apicinterrupt CALL_FUNCTION_SINGLE_VECTOR \
1077 call_function_single_interrupt smp_call_function_single_interrupt
1078 apicinterrupt CALL_FUNCTION_VECTOR \
1079 call_function_interrupt smp_call_function_interrupt
1080 apicinterrupt RESCHEDULE_VECTOR \
1081 reschedule_interrupt smp_reschedule_interrupt
1084 apicinterrupt ERROR_APIC_VECTOR \
1085 error_interrupt smp_error_interrupt
1086 apicinterrupt SPURIOUS_APIC_VECTOR \
1087 spurious_interrupt smp_spurious_interrupt
1089 #ifdef CONFIG_IRQ_WORK
1090 apicinterrupt IRQ_WORK_VECTOR \
1091 irq_work_interrupt smp_irq_work_interrupt
1095 * Exception entry points.
1097 .macro zeroentry sym do_sym
1100 PARAVIRT_ADJUST_EXCEPTION_FRAME
1101 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1102 subq $ORIG_RAX-R15, %rsp
1103 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1106 movq %rsp,%rdi /* pt_regs pointer */
1107 xorl %esi,%esi /* no error code */
1109 jmp error_exit /* %ebx: no swapgs flag */
1114 .macro paranoidzeroentry sym do_sym
1117 PARAVIRT_ADJUST_EXCEPTION_FRAME
1118 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1119 subq $ORIG_RAX-R15, %rsp
1120 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1123 movq %rsp,%rdi /* pt_regs pointer */
1124 xorl %esi,%esi /* no error code */
1126 jmp paranoid_exit /* %ebx: no swapgs flag */
1131 #define INIT_TSS_IST(x) PER_CPU_VAR(init_tss) + (TSS_ist + ((x) - 1) * 8)
1132 .macro paranoidzeroentry_ist sym do_sym ist
1135 PARAVIRT_ADJUST_EXCEPTION_FRAME
1136 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1137 subq $ORIG_RAX-R15, %rsp
1138 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1140 TRACE_IRQS_OFF_DEBUG
1141 movq %rsp,%rdi /* pt_regs pointer */
1142 xorl %esi,%esi /* no error code */
1143 subq $EXCEPTION_STKSZ, INIT_TSS_IST(\ist)
1145 addq $EXCEPTION_STKSZ, INIT_TSS_IST(\ist)
1146 jmp paranoid_exit /* %ebx: no swapgs flag */
1151 .macro errorentry sym do_sym
1154 PARAVIRT_ADJUST_EXCEPTION_FRAME
1155 subq $ORIG_RAX-R15, %rsp
1156 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1159 movq %rsp,%rdi /* pt_regs pointer */
1160 movq ORIG_RAX(%rsp),%rsi /* get error code */
1161 movq $-1,ORIG_RAX(%rsp) /* no syscall to restart */
1163 jmp error_exit /* %ebx: no swapgs flag */
1168 /* error code is on the stack already */
1169 .macro paranoiderrorentry sym do_sym
1172 PARAVIRT_ADJUST_EXCEPTION_FRAME
1173 subq $ORIG_RAX-R15, %rsp
1174 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1178 movq %rsp,%rdi /* pt_regs pointer */
1179 movq ORIG_RAX(%rsp),%rsi /* get error code */
1180 movq $-1,ORIG_RAX(%rsp) /* no syscall to restart */
1182 jmp paranoid_exit /* %ebx: no swapgs flag */
1187 zeroentry divide_error do_divide_error
1188 zeroentry overflow do_overflow
1189 zeroentry bounds do_bounds
1190 zeroentry invalid_op do_invalid_op
1191 zeroentry device_not_available do_device_not_available
1192 paranoiderrorentry double_fault do_double_fault
1193 zeroentry coprocessor_segment_overrun do_coprocessor_segment_overrun
1194 errorentry invalid_TSS do_invalid_TSS
1195 errorentry segment_not_present do_segment_not_present
1196 zeroentry spurious_interrupt_bug do_spurious_interrupt_bug
1197 zeroentry coprocessor_error do_coprocessor_error
1198 errorentry alignment_check do_alignment_check
1199 zeroentry simd_coprocessor_error do_simd_coprocessor_error
1202 /* Reload gs selector with exception handling */
1203 /* edi: new selector */
1204 ENTRY(native_load_gs_index)
1207 DISABLE_INTERRUPTS(CLBR_ANY & ~CLBR_RDI)
1211 2: mfence /* workaround */
1216 END(native_load_gs_index)
1218 _ASM_EXTABLE(gs_change,bad_gs)
1219 .section .fixup,"ax"
1220 /* running with kernelgs */
1222 SWAPGS /* switch back to user gs */
1228 ENTRY(kernel_thread_helper)
1229 pushq $0 # fake return address
1232 * Here we are in the child and the registers are set as they were
1233 * at kernel_thread() invocation in the parent.
1239 ud2 # padding for call trace
1241 END(kernel_thread_helper)
1244 * execve(). This function needs to use IRET, not SYSRET, to set up all state properly.
1246 * C extern interface:
1247 * extern long execve(const char *name, char **argv, char **envp)
1249 * asm input arguments:
1250 * rdi: name, rsi: argv, rdx: envp
1252 * We want to fallback into:
1253 * extern long sys_execve(const char *name, char **argv,char **envp, struct pt_regs *regs)
1255 * do_sys_execve asm fallback arguments:
1256 * rdi: name, rsi: argv, rdx: envp, rcx: fake frame on the stack
1258 ENTRY(kernel_execve)
1264 movq %rax, RAX(%rsp)
1267 je int_ret_from_sys_call
1274 /* Call softirq on interrupt stack. Interrupts are off. */
1278 CFI_REL_OFFSET rbp,0
1280 CFI_DEF_CFA_REGISTER rbp
1281 incl PER_CPU_VAR(irq_count)
1282 cmove PER_CPU_VAR(irq_stack_ptr),%rsp
1283 push %rbp # backlink for old unwinder
1287 CFI_DEF_CFA_REGISTER rsp
1288 CFI_ADJUST_CFA_OFFSET -8
1289 decl PER_CPU_VAR(irq_count)
1295 zeroentry xen_hypervisor_callback xen_do_hypervisor_callback
1298 * A note on the "critical region" in our callback handler.
1299 * We want to avoid stacking callback handlers due to events occurring
1300 * during handling of the last event. To do this, we keep events disabled
1301 * until we've done all processing. HOWEVER, we must enable events before
1302 * popping the stack frame (can't be done atomically) and so it would still
1303 * be possible to get enough handler activations to overflow the stack.
1304 * Although unlikely, bugs of that kind are hard to track down, so we'd
1305 * like to avoid the possibility.
1306 * So, on entry to the handler we detect whether we interrupted an
1307 * existing activation in its critical region -- if so, we pop the current
1308 * activation and restart the handler using the previous one.
1310 ENTRY(xen_do_hypervisor_callback) # do_hypervisor_callback(struct *pt_regs)
1313 * Since we don't modify %rdi, evtchn_do_upall(struct *pt_regs) will
1314 * see the correct pointer to the pt_regs
1316 movq %rdi, %rsp # we don't return, adjust the stack frame
1319 11: incl PER_CPU_VAR(irq_count)
1321 CFI_DEF_CFA_REGISTER rbp
1322 cmovzq PER_CPU_VAR(irq_stack_ptr),%rsp
1323 pushq %rbp # backlink for old unwinder
1324 call xen_evtchn_do_upcall
1326 CFI_DEF_CFA_REGISTER rsp
1327 decl PER_CPU_VAR(irq_count)
1330 END(xen_do_hypervisor_callback)
1333 * Hypervisor uses this for application faults while it executes.
1334 * We get here for two reasons:
1335 * 1. Fault while reloading DS, ES, FS or GS
1336 * 2. Fault while executing IRET
1337 * Category 1 we do not need to fix up as Xen has already reloaded all segment
1338 * registers that could be reloaded and zeroed the others.
1339 * Category 2 we fix up by killing the current process. We cannot use the
1340 * normal Linux return path in this case because if we use the IRET hypercall
1341 * to pop the stack frame we end up in an infinite loop of failsafe callbacks.
1342 * We distinguish between categories by comparing each saved segment register
1343 * with its current contents: any discrepancy means we in category 1.
1345 ENTRY(xen_failsafe_callback)
1347 /*CFI_REL_OFFSET gs,GS*/
1348 /*CFI_REL_OFFSET fs,FS*/
1349 /*CFI_REL_OFFSET es,ES*/
1350 /*CFI_REL_OFFSET ds,DS*/
1351 CFI_REL_OFFSET r11,8
1352 CFI_REL_OFFSET rcx,0
1366 /* All segments match their saved values => Category 2 (Bad IRET). */
1372 CFI_ADJUST_CFA_OFFSET -0x30
1373 pushq_cfi $0 /* RIP */
1376 jmp general_protection
1378 1: /* Segment mismatch => Category 1 (Bad segment). Retry the IRET. */
1384 CFI_ADJUST_CFA_OFFSET -0x30
1389 END(xen_failsafe_callback)
1391 apicinterrupt XEN_HVM_EVTCHN_CALLBACK \
1392 xen_hvm_callback_vector xen_evtchn_do_upcall
1394 #endif /* CONFIG_XEN */
1397 * Some functions should be protected against kprobes
1399 .pushsection .kprobes.text, "ax"
1401 paranoidzeroentry_ist debug do_debug DEBUG_STACK
1402 paranoidzeroentry_ist int3 do_int3 DEBUG_STACK
1403 paranoiderrorentry stack_segment do_stack_segment
1405 zeroentry xen_debug do_debug
1406 zeroentry xen_int3 do_int3
1407 errorentry xen_stack_segment do_stack_segment
1409 errorentry general_protection do_general_protection
1410 errorentry page_fault do_page_fault
1411 #ifdef CONFIG_KVM_GUEST
1412 errorentry async_page_fault do_async_page_fault
1414 #ifdef CONFIG_X86_MCE
1415 paranoidzeroentry machine_check *machine_check_vector(%rip)
1419 * "Paranoid" exit path from exception stack.
1420 * Paranoid because this is used by NMIs and cannot take
1421 * any kernel state for granted.
1422 * We don't do kernel preemption checks here, because only
1423 * NMI should be common and it does not enable IRQs and
1424 * cannot get reschedule ticks.
1426 * "trace" is 0 for the NMI handler only, because irq-tracing
1427 * is fundamentally NMI-unsafe. (we cannot change the soft and
1428 * hard flags at once, atomically)
1431 /* ebx: no swapgs flag */
1432 ENTRY(paranoid_exit)
1434 DISABLE_INTERRUPTS(CLBR_NONE)
1435 TRACE_IRQS_OFF_DEBUG
1436 testl %ebx,%ebx /* swapgs needed? */
1437 jnz paranoid_restore
1439 jnz paranoid_userspace
1446 TRACE_IRQS_IRETQ_DEBUG 0
1450 GET_THREAD_INFO(%rcx)
1451 movl TI_flags(%rcx),%ebx
1452 andl $_TIF_WORK_MASK,%ebx
1454 movq %rsp,%rdi /* &pt_regs */
1456 movq %rax,%rsp /* switch stack for scheduling */
1457 testl $_TIF_NEED_RESCHED,%ebx
1458 jnz paranoid_schedule
1459 movl %ebx,%edx /* arg3: thread flags */
1461 ENABLE_INTERRUPTS(CLBR_NONE)
1462 xorl %esi,%esi /* arg2: oldset */
1463 movq %rsp,%rdi /* arg1: &pt_regs */
1464 call do_notify_resume
1465 DISABLE_INTERRUPTS(CLBR_NONE)
1467 jmp paranoid_userspace
1470 ENABLE_INTERRUPTS(CLBR_ANY)
1472 DISABLE_INTERRUPTS(CLBR_ANY)
1474 jmp paranoid_userspace
1479 * Exception entry point. This expects an error code/orig_rax on the stack.
1480 * returns in "no swapgs flag" in %ebx.
1484 CFI_ADJUST_CFA_OFFSET 15*8
1485 /* oldrax contains error code */
1504 je error_kernelspace
1512 * There are two places in the kernel that can potentially fault with
1513 * usergs. Handle them here. The exception handlers after iret run with
1514 * kernel gs again, so don't set the user space flag. B stepping K8s
1515 * sometimes report an truncated RIP for IRET exceptions returning to
1516 * compat mode. Check for these here too.
1520 leaq irq_return(%rip),%rcx
1521 cmpq %rcx,RIP+8(%rsp)
1523 movl %ecx,%eax /* zero extend */
1524 cmpq %rax,RIP+8(%rsp)
1526 cmpq $gs_change,RIP+8(%rsp)
1531 /* Fix truncated RIP */
1532 movq %rcx,RIP+8(%rsp)
1538 /* ebx: no swapgs flag (1: don't need swapgs, 0: need it) */
1543 DISABLE_INTERRUPTS(CLBR_NONE)
1545 GET_THREAD_INFO(%rcx)
1548 LOCKDEP_SYS_EXIT_IRQ
1549 movl TI_flags(%rcx),%edx
1550 movl $_TIF_WORK_MASK,%edi
1558 * Test if a given stack is an NMI stack or not.
1560 .macro test_in_nmi reg stack nmi_ret normal_ret
1563 subq $EXCEPTION_STKSZ, %\reg
1569 /* runs on exception stack */
1572 PARAVIRT_ADJUST_EXCEPTION_FRAME
1574 * We allow breakpoints in NMIs. If a breakpoint occurs, then
1575 * the iretq it performs will take us out of NMI context.
1576 * This means that we can have nested NMIs where the next
1577 * NMI is using the top of the stack of the previous NMI. We
1578 * can't let it execute because the nested NMI will corrupt the
1579 * stack of the previous NMI. NMI handlers are not re-entrant
1582 * To handle this case we do the following:
1583 * Check the a special location on the stack that contains
1584 * a variable that is set when NMIs are executing.
1585 * The interrupted task's stack is also checked to see if it
1587 * If the variable is not set and the stack is not the NMI
1589 * o Set the special variable on the stack
1590 * o Copy the interrupt frame into a "saved" location on the stack
1591 * o Copy the interrupt frame into a "copy" location on the stack
1592 * o Continue processing the NMI
1593 * If the variable is set or the previous stack is the NMI stack:
1594 * o Modify the "copy" location to jump to the repeate_nmi
1595 * o return back to the first NMI
1597 * Now on exit of the first NMI, we first clear the stack variable
1598 * The NMI stack will tell any nested NMIs at that point that it is
1599 * nested. Then we pop the stack normally with iret, and if there was
1600 * a nested NMI that updated the copy interrupt stack frame, a
1601 * jump will be made to the repeat_nmi code that will handle the second
1605 /* Use %rdx as out temp variable throughout */
1607 CFI_REL_OFFSET rdx, 0
1610 * If %cs was not the kernel segment, then the NMI triggered in user
1611 * space, which means it is definitely not nested.
1613 cmpl $__KERNEL_CS, 16(%rsp)
1617 * Check the special variable on the stack to see if NMIs are
1624 * Now test if the previous stack was an NMI stack.
1625 * We need the double check. We check the NMI stack to satisfy the
1626 * race when the first NMI clears the variable before returning.
1627 * We check the variable because the first NMI could be in a
1628 * breakpoint routine using a breakpoint stack.
1631 test_in_nmi rdx, 4*8(%rsp), nested_nmi, first_nmi
1636 * Do nothing if we interrupted the fixup in repeat_nmi.
1637 * It's about to repeat the NMI handler, so we are fine
1638 * with ignoring this one.
1640 movq $repeat_nmi, %rdx
1643 movq $end_repeat_nmi, %rdx
1648 /* Set up the interrupted NMIs stack to jump to repeat_nmi */
1649 leaq -6*8(%rsp), %rdx
1651 CFI_ADJUST_CFA_OFFSET 6*8
1652 pushq_cfi $__KERNEL_DS
1655 pushq_cfi $__KERNEL_CS
1656 pushq_cfi $repeat_nmi
1658 /* Put stack back */
1660 CFI_ADJUST_CFA_OFFSET -11*8
1666 /* No need to check faults here */
1672 * Because nested NMIs will use the pushed location that we
1673 * stored in rdx, we must keep that space available.
1674 * Here's what our stack frame will look like:
1675 * +-------------------------+
1677 * | original Return RSP |
1678 * | original RFLAGS |
1681 * +-------------------------+
1682 * | temp storage for rdx |
1683 * +-------------------------+
1684 * | NMI executing variable |
1685 * +-------------------------+
1687 * | Saved Return RSP |
1691 * +-------------------------+
1693 * | copied Return RSP |
1697 * +-------------------------+
1699 * +-------------------------+
1701 * The saved stack frame is used to fix up the copied stack frame
1702 * that a nested NMI may change to make the interrupted NMI iret jump
1703 * to the repeat_nmi. The original stack frame and the temp storage
1704 * is also used by nested NMIs and can not be trusted on exit.
1706 /* Do not pop rdx, nested NMIs will corrupt that part of the stack */
1710 /* Set the NMI executing variable on the stack. */
1713 /* Copy the stack frame to the Saved frame */
1717 CFI_DEF_CFA_OFFSET SS+8-RIP
1719 /* Everything up to here is safe from nested NMIs */
1722 * If there was a nested NMI, the first NMI's iret will return
1723 * here. But NMIs are still enabled and we can take another
1724 * nested NMI. The nested NMI checks the interrupted RIP to see
1725 * if it is between repeat_nmi and end_repeat_nmi, and if so
1726 * it will just return, as we are about to repeat an NMI anyway.
1727 * This makes it safe to copy to the stack frame that a nested
1732 * Update the stack variable to say we are still in NMI (the update
1733 * is benign for the non-repeat case, where 1 was pushed just above
1734 * to this very stack slot).
1738 /* Make another copy, this one may be modified by nested NMIs */
1742 CFI_DEF_CFA_OFFSET SS+8-RIP
1746 * Everything below this point can be preempted by a nested
1747 * NMI if the first NMI took an exception and reset our iret stack
1748 * so that we repeat another NMI.
1750 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1751 subq $ORIG_RAX-R15, %rsp
1752 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1754 * Use save_paranoid to handle SWAPGS, but no need to use paranoid_exit
1755 * as we should not be calling schedule in NMI context.
1756 * Even with normal interrupts enabled. An NMI should not be
1757 * setting NEED_RESCHED or anything that normal interrupts and
1758 * exceptions might do.
1762 /* paranoidentry do_nmi, 0; without TRACE_IRQS_OFF */
1766 testl %ebx,%ebx /* swapgs needed? */
1772 /* Clear the NMI executing stack variable */
1778 ENTRY(ignore_sysret)
1786 * End of kprobes section