2 * linux/arch/x86_64/entry.S
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 * Copyright (C) 2000, 2001, 2002 Andi Kleen SuSE Labs
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
84 subq $MCOUNT_INSN_SIZE, %rdi
91 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
92 GLOBAL(ftrace_graph_call)
100 #else /* ! CONFIG_DYNAMIC_FTRACE */
102 cmpl $0, function_trace_stop
105 cmpq $ftrace_stub, ftrace_trace_function
108 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
109 cmpq $ftrace_stub, ftrace_graph_return
110 jnz ftrace_graph_caller
112 cmpq $ftrace_graph_entry_stub, ftrace_graph_entry
113 jnz ftrace_graph_caller
122 movq 0x38(%rsp), %rdi
124 subq $MCOUNT_INSN_SIZE, %rdi
126 call *ftrace_trace_function
132 #endif /* CONFIG_DYNAMIC_FTRACE */
133 #endif /* CONFIG_FUNCTION_TRACER */
135 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
136 ENTRY(ftrace_graph_caller)
137 cmpl $0, function_trace_stop
143 movq 0x38(%rsp), %rsi
145 subq $MCOUNT_INSN_SIZE, %rsi
147 call prepare_ftrace_return
152 END(ftrace_graph_caller)
154 GLOBAL(return_to_handler)
157 /* Save the return values */
162 call ftrace_return_to_handler
172 #ifndef CONFIG_PREEMPT
173 #define retint_kernel retint_restore_args
176 #ifdef CONFIG_PARAVIRT
177 ENTRY(native_usergs_sysret64)
180 ENDPROC(native_usergs_sysret64)
181 #endif /* CONFIG_PARAVIRT */
184 .macro TRACE_IRQS_IRETQ offset=ARGOFFSET
185 #ifdef CONFIG_TRACE_IRQFLAGS
186 bt $9,EFLAGS-\offset(%rsp) /* interrupts off? */
194 * When dynamic function tracer is enabled it will add a breakpoint
195 * to all locations that it is about to modify, sync CPUs, update
196 * all the code, sync CPUs, then remove the breakpoints. In this time
197 * if lockdep is enabled, it might jump back into the debug handler
198 * outside the updating of the IST protection. (TRACE_IRQS_ON/OFF).
200 * We need to change the IDT table before calling TRACE_IRQS_ON/OFF to
201 * make sure the stack pointer does not get reset back to the top
202 * of the debug stack, and instead just reuses the current stack.
204 #if defined(CONFIG_DYNAMIC_FTRACE) && defined(CONFIG_TRACE_IRQFLAGS)
206 .macro TRACE_IRQS_OFF_DEBUG
207 call debug_stack_set_zero
209 call debug_stack_reset
212 .macro TRACE_IRQS_ON_DEBUG
213 call debug_stack_set_zero
215 call debug_stack_reset
218 .macro TRACE_IRQS_IRETQ_DEBUG offset=ARGOFFSET
219 bt $9,EFLAGS-\offset(%rsp) /* interrupts off? */
226 # define TRACE_IRQS_OFF_DEBUG TRACE_IRQS_OFF
227 # define TRACE_IRQS_ON_DEBUG TRACE_IRQS_ON
228 # define TRACE_IRQS_IRETQ_DEBUG TRACE_IRQS_IRETQ
232 * C code is not supposed to know about undefined top of stack. Every time
233 * a C function with an pt_regs argument is called from the SYSCALL based
234 * fast path FIXUP_TOP_OF_STACK is needed.
235 * RESTORE_TOP_OF_STACK syncs the syscall state after any possible ptregs
239 /* %rsp:at FRAMEEND */
240 .macro FIXUP_TOP_OF_STACK tmp offset=0
241 movq PER_CPU_VAR(old_rsp),\tmp
242 movq \tmp,RSP+\offset(%rsp)
243 movq $__USER_DS,SS+\offset(%rsp)
244 movq $__USER_CS,CS+\offset(%rsp)
245 movq $-1,RCX+\offset(%rsp)
246 movq R11+\offset(%rsp),\tmp /* get eflags */
247 movq \tmp,EFLAGS+\offset(%rsp)
250 .macro RESTORE_TOP_OF_STACK tmp offset=0
251 movq RSP+\offset(%rsp),\tmp
252 movq \tmp,PER_CPU_VAR(old_rsp)
253 movq EFLAGS+\offset(%rsp),\tmp
254 movq \tmp,R11+\offset(%rsp)
257 .macro FAKE_STACK_FRAME child_rip
258 /* push in order ss, rsp, eflags, cs, rip */
260 pushq_cfi $__KERNEL_DS /* ss */
261 /*CFI_REL_OFFSET ss,0*/
262 pushq_cfi %rax /* rsp */
264 pushq_cfi $(X86_EFLAGS_IF|X86_EFLAGS_BIT1) /* eflags - interrupts on */
265 /*CFI_REL_OFFSET rflags,0*/
266 pushq_cfi $__KERNEL_CS /* cs */
267 /*CFI_REL_OFFSET cs,0*/
268 pushq_cfi \child_rip /* rip */
270 pushq_cfi %rax /* orig rax */
273 .macro UNFAKE_STACK_FRAME
275 CFI_ADJUST_CFA_OFFSET -(6*8)
279 * initial frame state for interrupts (and exceptions without error code)
281 .macro EMPTY_FRAME start=1 offset=0
285 CFI_DEF_CFA rsp,8+\offset
287 CFI_DEF_CFA_OFFSET 8+\offset
292 * initial frame state for interrupts (and exceptions without error code)
294 .macro INTR_FRAME start=1 offset=0
295 EMPTY_FRAME \start, SS+8+\offset-RIP
296 /*CFI_REL_OFFSET ss, SS+\offset-RIP*/
297 CFI_REL_OFFSET rsp, RSP+\offset-RIP
298 /*CFI_REL_OFFSET rflags, EFLAGS+\offset-RIP*/
299 /*CFI_REL_OFFSET cs, CS+\offset-RIP*/
300 CFI_REL_OFFSET rip, RIP+\offset-RIP
304 * initial frame state for exceptions with error code (and interrupts
305 * with vector already pushed)
307 .macro XCPT_FRAME start=1 offset=0
308 INTR_FRAME \start, RIP+\offset-ORIG_RAX
309 /*CFI_REL_OFFSET orig_rax, ORIG_RAX-ORIG_RAX*/
313 * frame that enables calling into C.
315 .macro PARTIAL_FRAME start=1 offset=0
316 XCPT_FRAME \start, ORIG_RAX+\offset-ARGOFFSET
317 CFI_REL_OFFSET rdi, RDI+\offset-ARGOFFSET
318 CFI_REL_OFFSET rsi, RSI+\offset-ARGOFFSET
319 CFI_REL_OFFSET rdx, RDX+\offset-ARGOFFSET
320 CFI_REL_OFFSET rcx, RCX+\offset-ARGOFFSET
321 CFI_REL_OFFSET rax, RAX+\offset-ARGOFFSET
322 CFI_REL_OFFSET r8, R8+\offset-ARGOFFSET
323 CFI_REL_OFFSET r9, R9+\offset-ARGOFFSET
324 CFI_REL_OFFSET r10, R10+\offset-ARGOFFSET
325 CFI_REL_OFFSET r11, R11+\offset-ARGOFFSET
329 * frame that enables passing a complete pt_regs to a C function.
331 .macro DEFAULT_FRAME start=1 offset=0
332 PARTIAL_FRAME \start, R11+\offset-R15
333 CFI_REL_OFFSET rbx, RBX+\offset
334 CFI_REL_OFFSET rbp, RBP+\offset
335 CFI_REL_OFFSET r12, R12+\offset
336 CFI_REL_OFFSET r13, R13+\offset
337 CFI_REL_OFFSET r14, R14+\offset
338 CFI_REL_OFFSET r15, R15+\offset
341 /* save partial stack frame */
344 /* start from rbp in pt_regs and jump over */
345 movq_cfi rdi, RDI-RBP
346 movq_cfi rsi, RSI-RBP
347 movq_cfi rdx, RDX-RBP
348 movq_cfi rcx, RCX-RBP
349 movq_cfi rax, RAX-RBP
352 movq_cfi r10, R10-RBP
353 movq_cfi r11, R11-RBP
355 /* Save rbp so that we can unwind from get_irq_regs() */
358 /* Save previous stack value */
361 leaq -RBP(%rsp),%rdi /* arg1 for handler */
362 testl $3, CS-RBP(%rsi)
366 * irq_count is used to check if a CPU is already on an interrupt stack
367 * or not. While this is essentially redundant with preempt_count it is
368 * a little cheaper to use a separate counter in the PDA (short of
369 * moving irq_enter into assembly, which would be too much work)
371 1: incl PER_CPU_VAR(irq_count)
372 cmovzq PER_CPU_VAR(irq_stack_ptr),%rsp
373 CFI_DEF_CFA_REGISTER rsi
375 /* Store previous stack value */
377 CFI_ESCAPE 0x0f /* DW_CFA_def_cfa_expression */, 6, \
378 0x77 /* DW_OP_breg7 */, 0, \
379 0x06 /* DW_OP_deref */, \
380 0x08 /* DW_OP_const1u */, SS+8-RBP, \
381 0x22 /* DW_OP_plus */
382 /* We entered an interrupt context - irqs are off: */
387 PARTIAL_FRAME 1 REST_SKIP+8
388 movq 5*8+16(%rsp), %r11 /* save return address */
395 movq %r11, 8(%rsp) /* return address */
396 FIXUP_TOP_OF_STACK %r11, 16
401 /* save complete stack frame */
402 .pushsection .kprobes.text, "ax"
422 movl $MSR_GS_BASE,%ecx
425 js 1f /* negative -> in kernel */
434 * A newly forked process directly context switches into this address.
436 * rdi: prev task we switched from
441 LOCK ; btr $TIF_FORK,TI_flags(%r8)
443 pushq_cfi kernel_eflags(%rip)
444 popfq_cfi # reset kernel eflags
446 call schedule_tail # rdi: 'prev' task parameter
448 GET_THREAD_INFO(%rcx)
452 testl $3, CS-ARGOFFSET(%rsp) # from kernel_thread?
455 testl $_TIF_IA32, TI_flags(%rcx) # 32-bit compat task needs IRET
456 jnz int_ret_from_sys_call
458 RESTORE_TOP_OF_STACK %rdi, -ARGOFFSET
459 jmp ret_from_sys_call # go to the SYSRET fastpath
462 subq $REST_SKIP, %rsp # leave space for volatiles
463 CFI_ADJUST_CFA_OFFSET REST_SKIP
468 jmp int_ret_from_sys_call
473 * System call entry. Up to 6 arguments in registers are supported.
475 * SYSCALL does not save anything on the stack and does not change the
481 * rax system call number
483 * rcx return address for syscall/sysret, C arg3
486 * r10 arg3 (--> moved to rcx for C)
489 * r11 eflags for syscall/sysret, temporary for C
490 * r12-r15,rbp,rbx saved by C code, not touched.
492 * Interrupts are off on entry.
493 * Only called from user space.
495 * XXX if we had a free scratch register we could save the RSP into the stack frame
496 * and report it properly in ps. Unfortunately we haven't.
498 * When user can change the frames always force IRET. That is because
499 * it deals with uncanonical addresses better. SYSRET has trouble
500 * with them due to bugs in both AMD and Intel CPUs.
506 CFI_DEF_CFA rsp,KERNEL_STACK_OFFSET
508 /*CFI_REGISTER rflags,r11*/
511 * A hypervisor implementation might want to use a label
512 * after the swapgs, so that it can do the swapgs
513 * for the guest and jump here on syscall.
515 GLOBAL(system_call_after_swapgs)
517 movq %rsp,PER_CPU_VAR(old_rsp)
518 movq PER_CPU_VAR(kernel_stack),%rsp
520 * No need to follow this irqs off/on section - it's straight
523 ENABLE_INTERRUPTS(CLBR_NONE)
525 movq %rax,ORIG_RAX-ARGOFFSET(%rsp)
526 movq %rcx,RIP-ARGOFFSET(%rsp)
527 CFI_REL_OFFSET rip,RIP-ARGOFFSET
528 testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET)
530 system_call_fastpath:
531 #if __SYSCALL_MASK == ~0
532 cmpq $__NR_syscall_max,%rax
534 andl $__SYSCALL_MASK,%eax
535 cmpl $__NR_syscall_max,%eax
539 call *sys_call_table(,%rax,8) # XXX: rip relative
540 movq %rax,RAX-ARGOFFSET(%rsp)
542 * Syscall return path ending with SYSRET (fast path)
543 * Has incomplete stack frame and undefined top of stack.
546 movl $_TIF_ALLWORK_MASK,%edi
550 DISABLE_INTERRUPTS(CLBR_NONE)
552 movl TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET),%edx
557 * sysretq will re-enable interrupts:
560 movq RIP-ARGOFFSET(%rsp),%rcx
562 RESTORE_ARGS 1,-ARG_SKIP,0
563 /*CFI_REGISTER rflags,r11*/
564 movq PER_CPU_VAR(old_rsp), %rsp
568 /* Handle reschedules */
569 /* edx: work, edi: workmask */
571 bt $TIF_NEED_RESCHED,%edx
574 ENABLE_INTERRUPTS(CLBR_NONE)
580 /* Handle a signal */
583 ENABLE_INTERRUPTS(CLBR_NONE)
584 #ifdef CONFIG_AUDITSYSCALL
585 bt $TIF_SYSCALL_AUDIT,%edx
589 * We have a signal, or exit tracing or single-step.
590 * These all wind up with the iret return path anyway,
591 * so just join that path right now.
593 FIXUP_TOP_OF_STACK %r11, -ARGOFFSET
594 jmp int_check_syscall_exit_work
597 movq $-ENOSYS,RAX-ARGOFFSET(%rsp)
598 jmp ret_from_sys_call
600 #ifdef CONFIG_AUDITSYSCALL
602 * Fast path for syscall audit without full syscall trace.
603 * We just call __audit_syscall_entry() directly, and then
604 * jump back to the normal fast path.
607 movq %r10,%r9 /* 6th arg: 4th syscall arg */
608 movq %rdx,%r8 /* 5th arg: 3rd syscall arg */
609 movq %rsi,%rcx /* 4th arg: 2nd syscall arg */
610 movq %rdi,%rdx /* 3rd arg: 1st syscall arg */
611 movq %rax,%rsi /* 2nd arg: syscall number */
612 movl $AUDIT_ARCH_X86_64,%edi /* 1st arg: audit arch */
613 call __audit_syscall_entry
614 LOAD_ARGS 0 /* reload call-clobbered registers */
615 jmp system_call_fastpath
618 * Return fast path for syscall audit. Call __audit_syscall_exit()
619 * directly and then jump back to the fast path with TIF_SYSCALL_AUDIT
623 movq RAX-ARGOFFSET(%rsp),%rsi /* second arg, syscall return value */
624 cmpq $-MAX_ERRNO,%rsi /* is it < -MAX_ERRNO? */
625 setbe %al /* 1 if so, 0 if not */
626 movzbl %al,%edi /* zero-extend that into %edi */
627 call __audit_syscall_exit
628 movl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT),%edi
630 #endif /* CONFIG_AUDITSYSCALL */
632 /* Do syscall tracing */
634 #ifdef CONFIG_AUDITSYSCALL
635 testl $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET)
639 movq $-ENOSYS,RAX(%rsp) /* ptrace can change this for a bad syscall */
640 FIXUP_TOP_OF_STACK %rdi
642 call syscall_trace_enter
644 * Reload arg registers from stack in case ptrace changed them.
645 * We don't reload %rax because syscall_trace_enter() returned
646 * the value it wants us to use in the table lookup.
648 LOAD_ARGS ARGOFFSET, 1
650 #if __SYSCALL_MASK == ~0
651 cmpq $__NR_syscall_max,%rax
653 andl $__SYSCALL_MASK,%eax
654 cmpl $__NR_syscall_max,%eax
656 ja int_ret_from_sys_call /* RAX(%rsp) set to -ENOSYS above */
657 movq %r10,%rcx /* fixup for C */
658 call *sys_call_table(,%rax,8)
659 movq %rax,RAX-ARGOFFSET(%rsp)
660 /* Use IRET because user could have changed frame */
663 * Syscall return path ending with IRET.
664 * Has correct top of stack, but partial stack frame.
666 GLOBAL(int_ret_from_sys_call)
667 DISABLE_INTERRUPTS(CLBR_NONE)
669 movl $_TIF_ALLWORK_MASK,%edi
670 /* edi: mask to check */
671 GLOBAL(int_with_check)
673 GET_THREAD_INFO(%rcx)
674 movl TI_flags(%rcx),%edx
677 andl $~TS_COMPAT,TI_status(%rcx)
680 /* Either reschedule or signal or syscall exit tracking needed. */
681 /* First do a reschedule test. */
682 /* edx: work, edi: workmask */
684 bt $TIF_NEED_RESCHED,%edx
687 ENABLE_INTERRUPTS(CLBR_NONE)
691 DISABLE_INTERRUPTS(CLBR_NONE)
695 /* handle signals and tracing -- both require a full stack frame */
698 ENABLE_INTERRUPTS(CLBR_NONE)
699 int_check_syscall_exit_work:
701 /* Check for syscall exit trace */
702 testl $_TIF_WORK_SYSCALL_EXIT,%edx
705 leaq 8(%rsp),%rdi # &ptregs -> arg1
706 call syscall_trace_leave
708 andl $~(_TIF_WORK_SYSCALL_EXIT|_TIF_SYSCALL_EMU),%edi
712 testl $_TIF_DO_NOTIFY_MASK,%edx
714 movq %rsp,%rdi # &ptregs -> arg1
715 xorl %esi,%esi # oldset -> arg2
716 call do_notify_resume
717 1: movl $_TIF_WORK_MASK,%edi
720 DISABLE_INTERRUPTS(CLBR_NONE)
727 * Certain special system calls that need to save a complete full stack frame.
729 .macro PTREGSCALL label,func,arg
731 PARTIAL_FRAME 1 8 /* offset 8: return address */
732 subq $REST_SKIP, %rsp
733 CFI_ADJUST_CFA_OFFSET REST_SKIP
735 DEFAULT_FRAME 0 8 /* offset 8: return address */
736 leaq 8(%rsp), \arg /* pt_regs pointer */
738 jmp ptregscall_common
743 PTREGSCALL stub_clone, sys_clone, %r8
744 PTREGSCALL stub_fork, sys_fork, %rdi
745 PTREGSCALL stub_vfork, sys_vfork, %rdi
746 PTREGSCALL stub_sigaltstack, sys_sigaltstack, %rdx
747 PTREGSCALL stub_iopl, sys_iopl, %rsi
749 ENTRY(ptregscall_common)
750 DEFAULT_FRAME 1 8 /* offset 8: return address */
751 RESTORE_TOP_OF_STACK %r11, 8
752 movq_cfi_restore R15+8, r15
753 movq_cfi_restore R14+8, r14
754 movq_cfi_restore R13+8, r13
755 movq_cfi_restore R12+8, r12
756 movq_cfi_restore RBP+8, rbp
757 movq_cfi_restore RBX+8, rbx
758 ret $REST_SKIP /* pop extended registers */
760 END(ptregscall_common)
767 FIXUP_TOP_OF_STACK %r11
769 RESTORE_TOP_OF_STACK %r11
772 jmp int_ret_from_sys_call
777 * sigreturn is special because it needs to restore all registers on return.
778 * This cannot be done with SYSRET, so use the IRET return path instead.
780 ENTRY(stub_rt_sigreturn)
786 FIXUP_TOP_OF_STACK %r11
787 call sys_rt_sigreturn
788 movq %rax,RAX(%rsp) # fixme, this could be done at the higher layer
790 jmp int_ret_from_sys_call
792 END(stub_rt_sigreturn)
794 #ifdef CONFIG_X86_X32_ABI
795 PTREGSCALL stub_x32_sigaltstack, sys32_sigaltstack, %rdx
797 ENTRY(stub_x32_rt_sigreturn)
803 FIXUP_TOP_OF_STACK %r11
804 call sys32_x32_rt_sigreturn
805 movq %rax,RAX(%rsp) # fixme, this could be done at the higher layer
807 jmp int_ret_from_sys_call
809 END(stub_x32_rt_sigreturn)
811 ENTRY(stub_x32_execve)
816 FIXUP_TOP_OF_STACK %r11
817 call compat_sys_execve
818 RESTORE_TOP_OF_STACK %r11
821 jmp int_ret_from_sys_call
828 * Build the entry stubs and pointer table with some assembler magic.
829 * We pack 7 stubs into a single 32-byte chunk, which will fit in a
830 * single cache line on all modern x86 implementations.
832 .section .init.rodata,"a"
836 .p2align CONFIG_X86_L1_CACHE_SHIFT
837 ENTRY(irq_entries_start)
839 vector=FIRST_EXTERNAL_VECTOR
840 .rept (NR_VECTORS-FIRST_EXTERNAL_VECTOR+6)/7
843 .if vector < NR_VECTORS
844 .if vector <> FIRST_EXTERNAL_VECTOR
845 CFI_ADJUST_CFA_OFFSET -8
847 1: pushq_cfi $(~vector+0x80) /* Note: always in signed byte range */
848 .if ((vector-FIRST_EXTERNAL_VECTOR)%7) <> 6
857 2: jmp common_interrupt
860 END(irq_entries_start)
867 * Interrupt entry/exit.
869 * Interrupt entry points save only callee clobbered registers in fast path.
871 * Entry runs with interrupts off.
874 /* 0(%rsp): ~(interrupt number) */
875 .macro interrupt func
876 /* reserve pt_regs for scratch regs and rbp */
877 subq $ORIG_RAX-RBP, %rsp
878 CFI_ADJUST_CFA_OFFSET ORIG_RAX-RBP
884 * Interrupt entry/exit should be protected against kprobes
886 .pushsection .kprobes.text, "ax"
888 * The interrupt stubs push (~vector+0x80) onto the stack and
889 * then jump to common_interrupt.
891 .p2align CONFIG_X86_L1_CACHE_SHIFT
894 addq $-0x80,(%rsp) /* Adjust vector to [-256,-1] range */
896 /* 0(%rsp): old_rsp-ARGOFFSET */
898 DISABLE_INTERRUPTS(CLBR_NONE)
900 decl PER_CPU_VAR(irq_count)
902 /* Restore saved previous stack */
904 CFI_DEF_CFA rsi,SS+8-RBP /* reg/off reset after def_cfa_expr */
905 leaq ARGOFFSET-RBP(%rsi), %rsp
906 CFI_DEF_CFA_REGISTER rsp
907 CFI_ADJUST_CFA_OFFSET RBP-ARGOFFSET
910 GET_THREAD_INFO(%rcx)
911 testl $3,CS-ARGOFFSET(%rsp)
914 /* Interrupt came from user space */
916 * Has a correct top of stack, but a partial stack frame
917 * %rcx: thread info. Interrupts off.
919 retint_with_reschedule:
920 movl $_TIF_WORK_MASK,%edi
923 movl TI_flags(%rcx),%edx
928 retint_swapgs: /* return to user-space */
930 * The iretq could re-enable interrupts:
932 DISABLE_INTERRUPTS(CLBR_ANY)
937 retint_restore_args: /* return to kernel space */
938 DISABLE_INTERRUPTS(CLBR_ANY)
940 * The iretq could re-enable interrupts:
948 _ASM_EXTABLE(irq_return, bad_iret)
950 #ifdef CONFIG_PARAVIRT
953 _ASM_EXTABLE(native_iret, bad_iret)
959 * The iret traps when the %cs or %ss being restored is bogus.
960 * We've lost the original trap vector and error code.
961 * #GPF is the most likely one to get for an invalid selector.
962 * So pretend we completed the iret and took the #GPF in user mode.
964 * We are now running with the kernel GS after exception recovery.
965 * But error_entry expects us to have user GS to match the user %cs,
971 jmp general_protection
975 /* edi: workmask, edx: work */
978 bt $TIF_NEED_RESCHED,%edx
981 ENABLE_INTERRUPTS(CLBR_NONE)
985 GET_THREAD_INFO(%rcx)
986 DISABLE_INTERRUPTS(CLBR_NONE)
991 testl $_TIF_DO_NOTIFY_MASK,%edx
994 ENABLE_INTERRUPTS(CLBR_NONE)
996 movq $-1,ORIG_RAX(%rsp)
997 xorl %esi,%esi # oldset
998 movq %rsp,%rdi # &pt_regs
999 call do_notify_resume
1001 DISABLE_INTERRUPTS(CLBR_NONE)
1003 GET_THREAD_INFO(%rcx)
1004 jmp retint_with_reschedule
1006 #ifdef CONFIG_PREEMPT
1007 /* Returning to kernel space. Check if we need preemption */
1008 /* rcx: threadinfo. interrupts off. */
1009 ENTRY(retint_kernel)
1010 cmpl $0,TI_preempt_count(%rcx)
1011 jnz retint_restore_args
1012 bt $TIF_NEED_RESCHED,TI_flags(%rcx)
1013 jnc retint_restore_args
1014 bt $9,EFLAGS-ARGOFFSET(%rsp) /* interrupts off? */
1015 jnc retint_restore_args
1016 call preempt_schedule_irq
1021 END(common_interrupt)
1023 * End of kprobes section
1030 .macro apicinterrupt num sym do_sym
1042 apicinterrupt IRQ_MOVE_CLEANUP_VECTOR \
1043 irq_move_cleanup_interrupt smp_irq_move_cleanup_interrupt
1044 apicinterrupt REBOOT_VECTOR \
1045 reboot_interrupt smp_reboot_interrupt
1048 #ifdef CONFIG_X86_UV
1049 apicinterrupt UV_BAU_MESSAGE \
1050 uv_bau_message_intr1 uv_bau_message_interrupt
1052 apicinterrupt LOCAL_TIMER_VECTOR \
1053 apic_timer_interrupt smp_apic_timer_interrupt
1054 apicinterrupt X86_PLATFORM_IPI_VECTOR \
1055 x86_platform_ipi smp_x86_platform_ipi
1057 apicinterrupt THRESHOLD_APIC_VECTOR \
1058 threshold_interrupt smp_threshold_interrupt
1059 apicinterrupt THERMAL_APIC_VECTOR \
1060 thermal_interrupt smp_thermal_interrupt
1063 apicinterrupt CALL_FUNCTION_SINGLE_VECTOR \
1064 call_function_single_interrupt smp_call_function_single_interrupt
1065 apicinterrupt CALL_FUNCTION_VECTOR \
1066 call_function_interrupt smp_call_function_interrupt
1067 apicinterrupt RESCHEDULE_VECTOR \
1068 reschedule_interrupt smp_reschedule_interrupt
1071 apicinterrupt ERROR_APIC_VECTOR \
1072 error_interrupt smp_error_interrupt
1073 apicinterrupt SPURIOUS_APIC_VECTOR \
1074 spurious_interrupt smp_spurious_interrupt
1076 #ifdef CONFIG_IRQ_WORK
1077 apicinterrupt IRQ_WORK_VECTOR \
1078 irq_work_interrupt smp_irq_work_interrupt
1082 * Exception entry points.
1084 .macro zeroentry sym do_sym
1087 PARAVIRT_ADJUST_EXCEPTION_FRAME
1088 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1089 subq $ORIG_RAX-R15, %rsp
1090 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1093 movq %rsp,%rdi /* pt_regs pointer */
1094 xorl %esi,%esi /* no error code */
1096 jmp error_exit /* %ebx: no swapgs flag */
1101 .macro paranoidzeroentry sym do_sym
1104 PARAVIRT_ADJUST_EXCEPTION_FRAME
1105 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1106 subq $ORIG_RAX-R15, %rsp
1107 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1110 movq %rsp,%rdi /* pt_regs pointer */
1111 xorl %esi,%esi /* no error code */
1113 jmp paranoid_exit /* %ebx: no swapgs flag */
1118 #define INIT_TSS_IST(x) PER_CPU_VAR(init_tss) + (TSS_ist + ((x) - 1) * 8)
1119 .macro paranoidzeroentry_ist sym do_sym ist
1122 PARAVIRT_ADJUST_EXCEPTION_FRAME
1123 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1124 subq $ORIG_RAX-R15, %rsp
1125 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1127 TRACE_IRQS_OFF_DEBUG
1128 movq %rsp,%rdi /* pt_regs pointer */
1129 xorl %esi,%esi /* no error code */
1130 subq $EXCEPTION_STKSZ, INIT_TSS_IST(\ist)
1132 addq $EXCEPTION_STKSZ, INIT_TSS_IST(\ist)
1133 jmp paranoid_exit /* %ebx: no swapgs flag */
1138 .macro errorentry sym do_sym
1141 PARAVIRT_ADJUST_EXCEPTION_FRAME
1142 subq $ORIG_RAX-R15, %rsp
1143 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1146 movq %rsp,%rdi /* pt_regs pointer */
1147 movq ORIG_RAX(%rsp),%rsi /* get error code */
1148 movq $-1,ORIG_RAX(%rsp) /* no syscall to restart */
1150 jmp error_exit /* %ebx: no swapgs flag */
1155 /* error code is on the stack already */
1156 .macro paranoiderrorentry sym do_sym
1159 PARAVIRT_ADJUST_EXCEPTION_FRAME
1160 subq $ORIG_RAX-R15, %rsp
1161 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1165 movq %rsp,%rdi /* pt_regs pointer */
1166 movq ORIG_RAX(%rsp),%rsi /* get error code */
1167 movq $-1,ORIG_RAX(%rsp) /* no syscall to restart */
1169 jmp paranoid_exit /* %ebx: no swapgs flag */
1174 zeroentry divide_error do_divide_error
1175 zeroentry overflow do_overflow
1176 zeroentry bounds do_bounds
1177 zeroentry invalid_op do_invalid_op
1178 zeroentry device_not_available do_device_not_available
1179 paranoiderrorentry double_fault do_double_fault
1180 zeroentry coprocessor_segment_overrun do_coprocessor_segment_overrun
1181 errorentry invalid_TSS do_invalid_TSS
1182 errorentry segment_not_present do_segment_not_present
1183 zeroentry spurious_interrupt_bug do_spurious_interrupt_bug
1184 zeroentry coprocessor_error do_coprocessor_error
1185 errorentry alignment_check do_alignment_check
1186 zeroentry simd_coprocessor_error do_simd_coprocessor_error
1189 /* Reload gs selector with exception handling */
1190 /* edi: new selector */
1191 ENTRY(native_load_gs_index)
1194 DISABLE_INTERRUPTS(CLBR_ANY & ~CLBR_RDI)
1198 2: mfence /* workaround */
1203 END(native_load_gs_index)
1205 _ASM_EXTABLE(gs_change,bad_gs)
1206 .section .fixup,"ax"
1207 /* running with kernelgs */
1209 SWAPGS /* switch back to user gs */
1215 /* Call softirq on interrupt stack. Interrupts are off. */
1219 CFI_REL_OFFSET rbp,0
1221 CFI_DEF_CFA_REGISTER rbp
1222 incl PER_CPU_VAR(irq_count)
1223 cmove PER_CPU_VAR(irq_stack_ptr),%rsp
1224 push %rbp # backlink for old unwinder
1228 CFI_DEF_CFA_REGISTER rsp
1229 CFI_ADJUST_CFA_OFFSET -8
1230 decl PER_CPU_VAR(irq_count)
1236 zeroentry xen_hypervisor_callback xen_do_hypervisor_callback
1239 * A note on the "critical region" in our callback handler.
1240 * We want to avoid stacking callback handlers due to events occurring
1241 * during handling of the last event. To do this, we keep events disabled
1242 * until we've done all processing. HOWEVER, we must enable events before
1243 * popping the stack frame (can't be done atomically) and so it would still
1244 * be possible to get enough handler activations to overflow the stack.
1245 * Although unlikely, bugs of that kind are hard to track down, so we'd
1246 * like to avoid the possibility.
1247 * So, on entry to the handler we detect whether we interrupted an
1248 * existing activation in its critical region -- if so, we pop the current
1249 * activation and restart the handler using the previous one.
1251 ENTRY(xen_do_hypervisor_callback) # do_hypervisor_callback(struct *pt_regs)
1254 * Since we don't modify %rdi, evtchn_do_upall(struct *pt_regs) will
1255 * see the correct pointer to the pt_regs
1257 movq %rdi, %rsp # we don't return, adjust the stack frame
1260 11: incl PER_CPU_VAR(irq_count)
1262 CFI_DEF_CFA_REGISTER rbp
1263 cmovzq PER_CPU_VAR(irq_stack_ptr),%rsp
1264 pushq %rbp # backlink for old unwinder
1265 call xen_evtchn_do_upcall
1267 CFI_DEF_CFA_REGISTER rsp
1268 decl PER_CPU_VAR(irq_count)
1271 END(xen_do_hypervisor_callback)
1274 * Hypervisor uses this for application faults while it executes.
1275 * We get here for two reasons:
1276 * 1. Fault while reloading DS, ES, FS or GS
1277 * 2. Fault while executing IRET
1278 * Category 1 we do not need to fix up as Xen has already reloaded all segment
1279 * registers that could be reloaded and zeroed the others.
1280 * Category 2 we fix up by killing the current process. We cannot use the
1281 * normal Linux return path in this case because if we use the IRET hypercall
1282 * to pop the stack frame we end up in an infinite loop of failsafe callbacks.
1283 * We distinguish between categories by comparing each saved segment register
1284 * with its current contents: any discrepancy means we in category 1.
1286 ENTRY(xen_failsafe_callback)
1288 /*CFI_REL_OFFSET gs,GS*/
1289 /*CFI_REL_OFFSET fs,FS*/
1290 /*CFI_REL_OFFSET es,ES*/
1291 /*CFI_REL_OFFSET ds,DS*/
1292 CFI_REL_OFFSET r11,8
1293 CFI_REL_OFFSET rcx,0
1307 /* All segments match their saved values => Category 2 (Bad IRET). */
1313 CFI_ADJUST_CFA_OFFSET -0x30
1314 pushq_cfi $0 /* RIP */
1317 jmp general_protection
1319 1: /* Segment mismatch => Category 1 (Bad segment). Retry the IRET. */
1325 CFI_ADJUST_CFA_OFFSET -0x30
1330 END(xen_failsafe_callback)
1332 apicinterrupt XEN_HVM_EVTCHN_CALLBACK \
1333 xen_hvm_callback_vector xen_evtchn_do_upcall
1335 #endif /* CONFIG_XEN */
1338 * Some functions should be protected against kprobes
1340 .pushsection .kprobes.text, "ax"
1342 paranoidzeroentry_ist debug do_debug DEBUG_STACK
1343 paranoidzeroentry_ist int3 do_int3 DEBUG_STACK
1344 paranoiderrorentry stack_segment do_stack_segment
1346 zeroentry xen_debug do_debug
1347 zeroentry xen_int3 do_int3
1348 errorentry xen_stack_segment do_stack_segment
1350 errorentry general_protection do_general_protection
1351 errorentry page_fault do_page_fault
1352 #ifdef CONFIG_KVM_GUEST
1353 errorentry async_page_fault do_async_page_fault
1355 #ifdef CONFIG_X86_MCE
1356 paranoidzeroentry machine_check *machine_check_vector(%rip)
1360 * "Paranoid" exit path from exception stack.
1361 * Paranoid because this is used by NMIs and cannot take
1362 * any kernel state for granted.
1363 * We don't do kernel preemption checks here, because only
1364 * NMI should be common and it does not enable IRQs and
1365 * cannot get reschedule ticks.
1367 * "trace" is 0 for the NMI handler only, because irq-tracing
1368 * is fundamentally NMI-unsafe. (we cannot change the soft and
1369 * hard flags at once, atomically)
1372 /* ebx: no swapgs flag */
1373 ENTRY(paranoid_exit)
1375 DISABLE_INTERRUPTS(CLBR_NONE)
1376 TRACE_IRQS_OFF_DEBUG
1377 testl %ebx,%ebx /* swapgs needed? */
1378 jnz paranoid_restore
1380 jnz paranoid_userspace
1387 TRACE_IRQS_IRETQ_DEBUG 0
1391 GET_THREAD_INFO(%rcx)
1392 movl TI_flags(%rcx),%ebx
1393 andl $_TIF_WORK_MASK,%ebx
1395 movq %rsp,%rdi /* &pt_regs */
1397 movq %rax,%rsp /* switch stack for scheduling */
1398 testl $_TIF_NEED_RESCHED,%ebx
1399 jnz paranoid_schedule
1400 movl %ebx,%edx /* arg3: thread flags */
1402 ENABLE_INTERRUPTS(CLBR_NONE)
1403 xorl %esi,%esi /* arg2: oldset */
1404 movq %rsp,%rdi /* arg1: &pt_regs */
1405 call do_notify_resume
1406 DISABLE_INTERRUPTS(CLBR_NONE)
1408 jmp paranoid_userspace
1411 ENABLE_INTERRUPTS(CLBR_ANY)
1413 DISABLE_INTERRUPTS(CLBR_ANY)
1415 jmp paranoid_userspace
1420 * Exception entry point. This expects an error code/orig_rax on the stack.
1421 * returns in "no swapgs flag" in %ebx.
1425 CFI_ADJUST_CFA_OFFSET 15*8
1426 /* oldrax contains error code */
1445 je error_kernelspace
1453 * There are two places in the kernel that can potentially fault with
1454 * usergs. Handle them here. The exception handlers after iret run with
1455 * kernel gs again, so don't set the user space flag. B stepping K8s
1456 * sometimes report an truncated RIP for IRET exceptions returning to
1457 * compat mode. Check for these here too.
1461 leaq irq_return(%rip),%rcx
1462 cmpq %rcx,RIP+8(%rsp)
1464 movl %ecx,%eax /* zero extend */
1465 cmpq %rax,RIP+8(%rsp)
1467 cmpq $gs_change,RIP+8(%rsp)
1472 /* Fix truncated RIP */
1473 movq %rcx,RIP+8(%rsp)
1479 /* ebx: no swapgs flag (1: don't need swapgs, 0: need it) */
1484 DISABLE_INTERRUPTS(CLBR_NONE)
1486 GET_THREAD_INFO(%rcx)
1489 LOCKDEP_SYS_EXIT_IRQ
1490 movl TI_flags(%rcx),%edx
1491 movl $_TIF_WORK_MASK,%edi
1499 * Test if a given stack is an NMI stack or not.
1501 .macro test_in_nmi reg stack nmi_ret normal_ret
1504 subq $EXCEPTION_STKSZ, %\reg
1510 /* runs on exception stack */
1513 PARAVIRT_ADJUST_EXCEPTION_FRAME
1515 * We allow breakpoints in NMIs. If a breakpoint occurs, then
1516 * the iretq it performs will take us out of NMI context.
1517 * This means that we can have nested NMIs where the next
1518 * NMI is using the top of the stack of the previous NMI. We
1519 * can't let it execute because the nested NMI will corrupt the
1520 * stack of the previous NMI. NMI handlers are not re-entrant
1523 * To handle this case we do the following:
1524 * Check the a special location on the stack that contains
1525 * a variable that is set when NMIs are executing.
1526 * The interrupted task's stack is also checked to see if it
1528 * If the variable is not set and the stack is not the NMI
1530 * o Set the special variable on the stack
1531 * o Copy the interrupt frame into a "saved" location on the stack
1532 * o Copy the interrupt frame into a "copy" location on the stack
1533 * o Continue processing the NMI
1534 * If the variable is set or the previous stack is the NMI stack:
1535 * o Modify the "copy" location to jump to the repeate_nmi
1536 * o return back to the first NMI
1538 * Now on exit of the first NMI, we first clear the stack variable
1539 * The NMI stack will tell any nested NMIs at that point that it is
1540 * nested. Then we pop the stack normally with iret, and if there was
1541 * a nested NMI that updated the copy interrupt stack frame, a
1542 * jump will be made to the repeat_nmi code that will handle the second
1546 /* Use %rdx as out temp variable throughout */
1548 CFI_REL_OFFSET rdx, 0
1551 * If %cs was not the kernel segment, then the NMI triggered in user
1552 * space, which means it is definitely not nested.
1554 cmpl $__KERNEL_CS, 16(%rsp)
1558 * Check the special variable on the stack to see if NMIs are
1565 * Now test if the previous stack was an NMI stack.
1566 * We need the double check. We check the NMI stack to satisfy the
1567 * race when the first NMI clears the variable before returning.
1568 * We check the variable because the first NMI could be in a
1569 * breakpoint routine using a breakpoint stack.
1572 test_in_nmi rdx, 4*8(%rsp), nested_nmi, first_nmi
1577 * Do nothing if we interrupted the fixup in repeat_nmi.
1578 * It's about to repeat the NMI handler, so we are fine
1579 * with ignoring this one.
1581 movq $repeat_nmi, %rdx
1584 movq $end_repeat_nmi, %rdx
1589 /* Set up the interrupted NMIs stack to jump to repeat_nmi */
1590 leaq -6*8(%rsp), %rdx
1592 CFI_ADJUST_CFA_OFFSET 6*8
1593 pushq_cfi $__KERNEL_DS
1596 pushq_cfi $__KERNEL_CS
1597 pushq_cfi $repeat_nmi
1599 /* Put stack back */
1601 CFI_ADJUST_CFA_OFFSET -11*8
1607 /* No need to check faults here */
1613 * Because nested NMIs will use the pushed location that we
1614 * stored in rdx, we must keep that space available.
1615 * Here's what our stack frame will look like:
1616 * +-------------------------+
1618 * | original Return RSP |
1619 * | original RFLAGS |
1622 * +-------------------------+
1623 * | temp storage for rdx |
1624 * +-------------------------+
1625 * | NMI executing variable |
1626 * +-------------------------+
1628 * | Saved Return RSP |
1632 * +-------------------------+
1634 * | copied Return RSP |
1638 * +-------------------------+
1640 * +-------------------------+
1642 * The saved stack frame is used to fix up the copied stack frame
1643 * that a nested NMI may change to make the interrupted NMI iret jump
1644 * to the repeat_nmi. The original stack frame and the temp storage
1645 * is also used by nested NMIs and can not be trusted on exit.
1647 /* Do not pop rdx, nested NMIs will corrupt that part of the stack */
1651 /* Set the NMI executing variable on the stack. */
1654 /* Copy the stack frame to the Saved frame */
1658 CFI_DEF_CFA_OFFSET SS+8-RIP
1660 /* Everything up to here is safe from nested NMIs */
1663 * If there was a nested NMI, the first NMI's iret will return
1664 * here. But NMIs are still enabled and we can take another
1665 * nested NMI. The nested NMI checks the interrupted RIP to see
1666 * if it is between repeat_nmi and end_repeat_nmi, and if so
1667 * it will just return, as we are about to repeat an NMI anyway.
1668 * This makes it safe to copy to the stack frame that a nested
1673 * Update the stack variable to say we are still in NMI (the update
1674 * is benign for the non-repeat case, where 1 was pushed just above
1675 * to this very stack slot).
1679 /* Make another copy, this one may be modified by nested NMIs */
1683 CFI_DEF_CFA_OFFSET SS+8-RIP
1687 * Everything below this point can be preempted by a nested
1688 * NMI if the first NMI took an exception and reset our iret stack
1689 * so that we repeat another NMI.
1691 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1692 subq $ORIG_RAX-R15, %rsp
1693 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1695 * Use save_paranoid to handle SWAPGS, but no need to use paranoid_exit
1696 * as we should not be calling schedule in NMI context.
1697 * Even with normal interrupts enabled. An NMI should not be
1698 * setting NEED_RESCHED or anything that normal interrupts and
1699 * exceptions might do.
1705 * Save off the CR2 register. If we take a page fault in the NMI then
1706 * it could corrupt the CR2 value. If the NMI preempts a page fault
1707 * handler before it was able to read the CR2 register, and then the
1708 * NMI itself takes a page fault, the page fault that was preempted
1709 * will read the information from the NMI page fault and not the
1710 * origin fault. Save it off and restore it if it changes.
1711 * Use the r12 callee-saved register.
1715 /* paranoidentry do_nmi, 0; without TRACE_IRQS_OFF */
1720 /* Did the NMI take a page fault? Restore cr2 if it did */
1727 testl %ebx,%ebx /* swapgs needed? */
1733 /* Clear the NMI executing stack variable */
1739 ENTRY(ignore_sysret)
1747 * End of kprobes section