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 * NOTE: This code handles signal-recognition, which happens every time
13 * after an interrupt and after each system call.
15 * Normal syscalls and interrupts don't save a full stack frame, this is
16 * only done for syscall tracing, signals or fork/exec et.al.
18 * A note on terminology:
19 * - top of stack: Architecture defined interrupt frame from SS to RIP
20 * at the top of the kernel process stack.
21 * - partial stack frame: partially saved registers upto R11.
22 * - full stack frame: Like partial stack frame, but all register saved.
25 * - CFI macros are used to generate dwarf2 unwind information for better
26 * backtraces. They don't change any code.
27 * - SAVE_ALL/RESTORE_ALL - Save/restore all registers
28 * - SAVE_ARGS/RESTORE_ARGS - Save/restore registers that C functions modify.
29 * There are unfortunately lots of special cases where some registers
30 * not touched. The macro is a big mess that should be cleaned up.
31 * - SAVE_REST/RESTORE_REST - Handle the registers not saved by SAVE_ARGS.
32 * Gives a full stack frame.
33 * - ENTRY/END Define functions in the symbol table.
34 * - FIXUP_TOP_OF_STACK/RESTORE_TOP_OF_STACK - Fix up the hardware stack
35 * frame that is otherwise undefined after a SYSCALL
36 * - TRACE_IRQ_* - Trace hard interrupt state for lock debugging.
37 * - errorentry/paranoidentry/zeroentry - Define exception entry points.
40 #include <linux/linkage.h>
41 #include <asm/segment.h>
42 #include <asm/cache.h>
43 #include <asm/errno.h>
44 #include <asm/dwarf2.h>
45 #include <asm/calling.h>
46 #include <asm/asm-offsets.h>
48 #include <asm/unistd.h>
49 #include <asm/thread_info.h>
50 #include <asm/hw_irq.h>
51 #include <asm/page_types.h>
52 #include <asm/irqflags.h>
53 #include <asm/paravirt.h>
54 #include <asm/ftrace.h>
55 #include <asm/percpu.h>
57 /* Avoid __ASSEMBLER__'ifying <linux/audit.h> just for this. */
58 #include <linux/elf-em.h>
59 #define AUDIT_ARCH_X86_64 (EM_X86_64|__AUDIT_ARCH_64BIT|__AUDIT_ARCH_LE)
60 #define __AUDIT_ARCH_64BIT 0x80000000
61 #define __AUDIT_ARCH_LE 0x40000000
64 #ifdef CONFIG_FUNCTION_TRACER
65 #ifdef CONFIG_DYNAMIC_FTRACE
71 cmpl $0, function_trace_stop
78 subq $MCOUNT_INSN_SIZE, %rdi
85 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
86 GLOBAL(ftrace_graph_call)
94 #else /* ! CONFIG_DYNAMIC_FTRACE */
96 cmpl $0, function_trace_stop
99 cmpq $ftrace_stub, ftrace_trace_function
102 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
103 cmpq $ftrace_stub, ftrace_graph_return
104 jnz ftrace_graph_caller
106 cmpq $ftrace_graph_entry_stub, ftrace_graph_entry
107 jnz ftrace_graph_caller
116 movq 0x38(%rsp), %rdi
118 subq $MCOUNT_INSN_SIZE, %rdi
120 call *ftrace_trace_function
126 #endif /* CONFIG_DYNAMIC_FTRACE */
127 #endif /* CONFIG_FUNCTION_TRACER */
129 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
130 ENTRY(ftrace_graph_caller)
131 cmpl $0, function_trace_stop
137 movq 0x38(%rsp), %rsi
139 subq $MCOUNT_INSN_SIZE, %rsi
141 call prepare_ftrace_return
146 END(ftrace_graph_caller)
148 GLOBAL(return_to_handler)
151 /* Save the return values */
156 call ftrace_return_to_handler
166 #ifndef CONFIG_PREEMPT
167 #define retint_kernel retint_restore_args
170 #ifdef CONFIG_PARAVIRT
171 ENTRY(native_usergs_sysret64)
174 ENDPROC(native_usergs_sysret64)
175 #endif /* CONFIG_PARAVIRT */
178 .macro TRACE_IRQS_IRETQ offset=ARGOFFSET
179 #ifdef CONFIG_TRACE_IRQFLAGS
180 bt $9,EFLAGS-\offset(%rsp) /* interrupts off? */
188 * C code is not supposed to know about undefined top of stack. Every time
189 * a C function with an pt_regs argument is called from the SYSCALL based
190 * fast path FIXUP_TOP_OF_STACK is needed.
191 * RESTORE_TOP_OF_STACK syncs the syscall state after any possible ptregs
195 /* %rsp:at FRAMEEND */
196 .macro FIXUP_TOP_OF_STACK tmp offset=0
197 movq PER_CPU_VAR(old_rsp),\tmp
198 movq \tmp,RSP+\offset(%rsp)
199 movq $__USER_DS,SS+\offset(%rsp)
200 movq $__USER_CS,CS+\offset(%rsp)
201 movq $-1,RCX+\offset(%rsp)
202 movq R11+\offset(%rsp),\tmp /* get eflags */
203 movq \tmp,EFLAGS+\offset(%rsp)
206 .macro RESTORE_TOP_OF_STACK tmp offset=0
207 movq RSP+\offset(%rsp),\tmp
208 movq \tmp,PER_CPU_VAR(old_rsp)
209 movq EFLAGS+\offset(%rsp),\tmp
210 movq \tmp,R11+\offset(%rsp)
213 .macro FAKE_STACK_FRAME child_rip
214 /* push in order ss, rsp, eflags, cs, rip */
216 pushq $__KERNEL_DS /* ss */
217 CFI_ADJUST_CFA_OFFSET 8
218 /*CFI_REL_OFFSET ss,0*/
220 CFI_ADJUST_CFA_OFFSET 8
222 pushq $X86_EFLAGS_IF /* eflags - interrupts on */
223 CFI_ADJUST_CFA_OFFSET 8
224 /*CFI_REL_OFFSET rflags,0*/
225 pushq $__KERNEL_CS /* cs */
226 CFI_ADJUST_CFA_OFFSET 8
227 /*CFI_REL_OFFSET cs,0*/
228 pushq \child_rip /* rip */
229 CFI_ADJUST_CFA_OFFSET 8
231 pushq %rax /* orig rax */
232 CFI_ADJUST_CFA_OFFSET 8
235 .macro UNFAKE_STACK_FRAME
237 CFI_ADJUST_CFA_OFFSET -(6*8)
241 * initial frame state for interrupts (and exceptions without error code)
243 .macro EMPTY_FRAME start=1 offset=0
247 CFI_DEF_CFA rsp,8+\offset
249 CFI_DEF_CFA_OFFSET 8+\offset
254 * initial frame state for interrupts (and exceptions without error code)
256 .macro INTR_FRAME start=1 offset=0
257 EMPTY_FRAME \start, SS+8+\offset-RIP
258 /*CFI_REL_OFFSET ss, SS+\offset-RIP*/
259 CFI_REL_OFFSET rsp, RSP+\offset-RIP
260 /*CFI_REL_OFFSET rflags, EFLAGS+\offset-RIP*/
261 /*CFI_REL_OFFSET cs, CS+\offset-RIP*/
262 CFI_REL_OFFSET rip, RIP+\offset-RIP
266 * initial frame state for exceptions with error code (and interrupts
267 * with vector already pushed)
269 .macro XCPT_FRAME start=1 offset=0
270 INTR_FRAME \start, RIP+\offset-ORIG_RAX
271 /*CFI_REL_OFFSET orig_rax, ORIG_RAX-ORIG_RAX*/
275 * frame that enables calling into C.
277 .macro PARTIAL_FRAME start=1 offset=0
278 XCPT_FRAME \start, ORIG_RAX+\offset-ARGOFFSET
279 CFI_REL_OFFSET rdi, RDI+\offset-ARGOFFSET
280 CFI_REL_OFFSET rsi, RSI+\offset-ARGOFFSET
281 CFI_REL_OFFSET rdx, RDX+\offset-ARGOFFSET
282 CFI_REL_OFFSET rcx, RCX+\offset-ARGOFFSET
283 CFI_REL_OFFSET rax, RAX+\offset-ARGOFFSET
284 CFI_REL_OFFSET r8, R8+\offset-ARGOFFSET
285 CFI_REL_OFFSET r9, R9+\offset-ARGOFFSET
286 CFI_REL_OFFSET r10, R10+\offset-ARGOFFSET
287 CFI_REL_OFFSET r11, R11+\offset-ARGOFFSET
291 * frame that enables passing a complete pt_regs to a C function.
293 .macro DEFAULT_FRAME start=1 offset=0
294 PARTIAL_FRAME \start, R11+\offset-R15
295 CFI_REL_OFFSET rbx, RBX+\offset
296 CFI_REL_OFFSET rbp, RBP+\offset
297 CFI_REL_OFFSET r12, R12+\offset
298 CFI_REL_OFFSET r13, R13+\offset
299 CFI_REL_OFFSET r14, R14+\offset
300 CFI_REL_OFFSET r15, R15+\offset
303 /* save partial stack frame */
307 movq_cfi rdi, RDI+16-ARGOFFSET
308 movq_cfi rsi, RSI+16-ARGOFFSET
309 movq_cfi rdx, RDX+16-ARGOFFSET
310 movq_cfi rcx, RCX+16-ARGOFFSET
311 movq_cfi rax, RAX+16-ARGOFFSET
312 movq_cfi r8, R8+16-ARGOFFSET
313 movq_cfi r9, R9+16-ARGOFFSET
314 movq_cfi r10, R10+16-ARGOFFSET
315 movq_cfi r11, R11+16-ARGOFFSET
317 leaq -ARGOFFSET+16(%rsp),%rdi /* arg1 for handler */
318 movq_cfi rbp, 8 /* push %rbp */
319 leaq 8(%rsp), %rbp /* mov %rsp, %ebp */
324 * irq_count is used to check if a CPU is already on an interrupt stack
325 * or not. While this is essentially redundant with preempt_count it is
326 * a little cheaper to use a separate counter in the PDA (short of
327 * moving irq_enter into assembly, which would be too much work)
329 1: incl PER_CPU_VAR(irq_count)
331 popq_cfi %rax /* move return address... */
332 mov PER_CPU_VAR(irq_stack_ptr),%rsp
334 pushq_cfi %rbp /* backlink for unwinder */
335 pushq_cfi %rax /* ... to the new stack */
337 * We entered an interrupt context - irqs are off:
345 PARTIAL_FRAME 1 REST_SKIP+8
346 movq 5*8+16(%rsp), %r11 /* save return address */
353 movq %r11, 8(%rsp) /* return address */
354 FIXUP_TOP_OF_STACK %r11, 16
359 /* save complete stack frame */
360 .pushsection .kprobes.text, "ax"
380 movl $MSR_GS_BASE,%ecx
383 js 1f /* negative -> in kernel */
392 * A newly forked process directly context switches into this address.
394 * rdi: prev task we switched from
399 LOCK ; btr $TIF_FORK,TI_flags(%r8)
401 push kernel_eflags(%rip)
402 CFI_ADJUST_CFA_OFFSET 8
403 popf # reset kernel eflags
404 CFI_ADJUST_CFA_OFFSET -8
406 call schedule_tail # rdi: 'prev' task parameter
408 GET_THREAD_INFO(%rcx)
412 testl $3, CS-ARGOFFSET(%rsp) # from kernel_thread?
413 je int_ret_from_sys_call
415 testl $_TIF_IA32, TI_flags(%rcx) # 32-bit compat task needs IRET
416 jnz int_ret_from_sys_call
418 RESTORE_TOP_OF_STACK %rdi, -ARGOFFSET
419 jmp ret_from_sys_call # go to the SYSRET fastpath
425 * System call entry. Upto 6 arguments in registers are supported.
427 * SYSCALL does not save anything on the stack and does not change the
435 CFI_DEF_CFA rsp,KERNEL_STACK_OFFSET
437 /*CFI_REGISTER rflags,r11*/
440 * A hypervisor implementation might want to use a label
441 * after the swapgs, so that it can do the swapgs
442 * for the guest and jump here on syscall.
444 ENTRY(system_call_after_swapgs)
446 movq %rsp,PER_CPU_VAR(old_rsp)
447 movq PER_CPU_VAR(kernel_stack),%rsp
449 * No need to follow this irqs off/on section - it's straight
452 ENABLE_INTERRUPTS(CLBR_NONE)
454 movq %rax,ORIG_RAX-ARGOFFSET(%rsp)
455 movq %rcx,RIP-ARGOFFSET(%rsp)
456 CFI_REL_OFFSET rip,RIP-ARGOFFSET
457 GET_THREAD_INFO(%rcx)
458 testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%rcx)
460 system_call_fastpath:
461 cmpq $__NR_syscall_max,%rax
464 call *sys_call_table(,%rax,8) # XXX: rip relative
465 movq %rax,RAX-ARGOFFSET(%rsp)
467 * Syscall return path ending with SYSRET (fast path)
468 * Has incomplete stack frame and undefined top of stack.
471 movl $_TIF_ALLWORK_MASK,%edi
475 GET_THREAD_INFO(%rcx)
476 DISABLE_INTERRUPTS(CLBR_NONE)
478 movl TI_flags(%rcx),%edx
483 * sysretq will re-enable interrupts:
486 movq RIP-ARGOFFSET(%rsp),%rcx
488 RESTORE_ARGS 0,-ARG_SKIP,1
489 /*CFI_REGISTER rflags,r11*/
490 movq PER_CPU_VAR(old_rsp), %rsp
494 /* Handle reschedules */
495 /* edx: work, edi: workmask */
497 bt $TIF_NEED_RESCHED,%edx
500 ENABLE_INTERRUPTS(CLBR_NONE)
502 CFI_ADJUST_CFA_OFFSET 8
505 CFI_ADJUST_CFA_OFFSET -8
508 /* Handle a signal */
511 ENABLE_INTERRUPTS(CLBR_NONE)
512 #ifdef CONFIG_AUDITSYSCALL
513 bt $TIF_SYSCALL_AUDIT,%edx
517 * We have a signal, or exit tracing or single-step.
518 * These all wind up with the iret return path anyway,
519 * so just join that path right now.
521 FIXUP_TOP_OF_STACK %r11, -ARGOFFSET
522 jmp int_check_syscall_exit_work
525 movq $-ENOSYS,RAX-ARGOFFSET(%rsp)
526 jmp ret_from_sys_call
528 #ifdef CONFIG_AUDITSYSCALL
530 * Fast path for syscall audit without full syscall trace.
531 * We just call audit_syscall_entry() directly, and then
532 * jump back to the normal fast path.
535 movq %r10,%r9 /* 6th arg: 4th syscall arg */
536 movq %rdx,%r8 /* 5th arg: 3rd syscall arg */
537 movq %rsi,%rcx /* 4th arg: 2nd syscall arg */
538 movq %rdi,%rdx /* 3rd arg: 1st syscall arg */
539 movq %rax,%rsi /* 2nd arg: syscall number */
540 movl $AUDIT_ARCH_X86_64,%edi /* 1st arg: audit arch */
541 call audit_syscall_entry
542 LOAD_ARGS 0 /* reload call-clobbered registers */
543 jmp system_call_fastpath
546 * Return fast path for syscall audit. Call audit_syscall_exit()
547 * directly and then jump back to the fast path with TIF_SYSCALL_AUDIT
551 movq RAX-ARGOFFSET(%rsp),%rsi /* second arg, syscall return value */
552 cmpq $0,%rsi /* is it < 0? */
553 setl %al /* 1 if so, 0 if not */
554 movzbl %al,%edi /* zero-extend that into %edi */
555 inc %edi /* first arg, 0->1(AUDITSC_SUCCESS), 1->2(AUDITSC_FAILURE) */
556 call audit_syscall_exit
557 movl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT),%edi
559 #endif /* CONFIG_AUDITSYSCALL */
561 /* Do syscall tracing */
563 #ifdef CONFIG_AUDITSYSCALL
564 testl $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags(%rcx)
568 movq $-ENOSYS,RAX(%rsp) /* ptrace can change this for a bad syscall */
569 FIXUP_TOP_OF_STACK %rdi
571 call syscall_trace_enter
573 * Reload arg registers from stack in case ptrace changed them.
574 * We don't reload %rax because syscall_trace_enter() returned
575 * the value it wants us to use in the table lookup.
577 LOAD_ARGS ARGOFFSET, 1
579 cmpq $__NR_syscall_max,%rax
580 ja int_ret_from_sys_call /* RAX(%rsp) set to -ENOSYS above */
581 movq %r10,%rcx /* fixup for C */
582 call *sys_call_table(,%rax,8)
583 movq %rax,RAX-ARGOFFSET(%rsp)
584 /* Use IRET because user could have changed frame */
587 * Syscall return path ending with IRET.
588 * Has correct top of stack, but partial stack frame.
590 GLOBAL(int_ret_from_sys_call)
591 DISABLE_INTERRUPTS(CLBR_NONE)
593 testl $3,CS-ARGOFFSET(%rsp)
594 je retint_restore_args
595 movl $_TIF_ALLWORK_MASK,%edi
596 /* edi: mask to check */
597 GLOBAL(int_with_check)
599 GET_THREAD_INFO(%rcx)
600 movl TI_flags(%rcx),%edx
603 andl $~TS_COMPAT,TI_status(%rcx)
606 /* Either reschedule or signal or syscall exit tracking needed. */
607 /* First do a reschedule test. */
608 /* edx: work, edi: workmask */
610 bt $TIF_NEED_RESCHED,%edx
613 ENABLE_INTERRUPTS(CLBR_NONE)
615 CFI_ADJUST_CFA_OFFSET 8
618 CFI_ADJUST_CFA_OFFSET -8
619 DISABLE_INTERRUPTS(CLBR_NONE)
623 /* handle signals and tracing -- both require a full stack frame */
626 ENABLE_INTERRUPTS(CLBR_NONE)
627 int_check_syscall_exit_work:
629 /* Check for syscall exit trace */
630 testl $_TIF_WORK_SYSCALL_EXIT,%edx
633 CFI_ADJUST_CFA_OFFSET 8
634 leaq 8(%rsp),%rdi # &ptregs -> arg1
635 call syscall_trace_leave
637 CFI_ADJUST_CFA_OFFSET -8
638 andl $~(_TIF_WORK_SYSCALL_EXIT|_TIF_SYSCALL_EMU),%edi
642 testl $_TIF_DO_NOTIFY_MASK,%edx
644 movq %rsp,%rdi # &ptregs -> arg1
645 xorl %esi,%esi # oldset -> arg2
646 call do_notify_resume
647 1: movl $_TIF_WORK_MASK,%edi
650 DISABLE_INTERRUPTS(CLBR_NONE)
657 * Certain special system calls that need to save a complete full stack frame.
659 .macro PTREGSCALL label,func,arg
661 PARTIAL_FRAME 1 8 /* offset 8: return address */
662 subq $REST_SKIP, %rsp
663 CFI_ADJUST_CFA_OFFSET REST_SKIP
665 DEFAULT_FRAME 0 8 /* offset 8: return address */
666 leaq 8(%rsp), \arg /* pt_regs pointer */
668 jmp ptregscall_common
673 PTREGSCALL stub_clone, sys_clone, %r8
674 PTREGSCALL stub_fork, sys_fork, %rdi
675 PTREGSCALL stub_vfork, sys_vfork, %rdi
676 PTREGSCALL stub_sigaltstack, sys_sigaltstack, %rdx
677 PTREGSCALL stub_iopl, sys_iopl, %rsi
679 ENTRY(ptregscall_common)
680 DEFAULT_FRAME 1 8 /* offset 8: return address */
681 RESTORE_TOP_OF_STACK %r11, 8
682 movq_cfi_restore R15+8, r15
683 movq_cfi_restore R14+8, r14
684 movq_cfi_restore R13+8, r13
685 movq_cfi_restore R12+8, r12
686 movq_cfi_restore RBP+8, rbp
687 movq_cfi_restore RBX+8, rbx
688 ret $REST_SKIP /* pop extended registers */
690 END(ptregscall_common)
695 CFI_ADJUST_CFA_OFFSET -8
696 CFI_REGISTER rip, r11
698 FIXUP_TOP_OF_STACK %r11
701 RESTORE_TOP_OF_STACK %r11
704 jmp int_ret_from_sys_call
709 * sigreturn is special because it needs to restore all registers on return.
710 * This cannot be done with SYSRET, so use the IRET return path instead.
712 ENTRY(stub_rt_sigreturn)
715 CFI_ADJUST_CFA_OFFSET -8
718 FIXUP_TOP_OF_STACK %r11
719 call sys_rt_sigreturn
720 movq %rax,RAX(%rsp) # fixme, this could be done at the higher layer
722 jmp int_ret_from_sys_call
724 END(stub_rt_sigreturn)
727 * Build the entry stubs and pointer table with some assembler magic.
728 * We pack 7 stubs into a single 32-byte chunk, which will fit in a
729 * single cache line on all modern x86 implementations.
731 .section .init.rodata,"a"
735 .p2align CONFIG_X86_L1_CACHE_SHIFT
736 ENTRY(irq_entries_start)
738 vector=FIRST_EXTERNAL_VECTOR
739 .rept (NR_VECTORS-FIRST_EXTERNAL_VECTOR+6)/7
742 .if vector < NR_VECTORS
743 .if vector <> FIRST_EXTERNAL_VECTOR
744 CFI_ADJUST_CFA_OFFSET -8
746 1: pushq $(~vector+0x80) /* Note: always in signed byte range */
747 CFI_ADJUST_CFA_OFFSET 8
748 .if ((vector-FIRST_EXTERNAL_VECTOR)%7) <> 6
757 2: jmp common_interrupt
760 END(irq_entries_start)
767 * Interrupt entry/exit.
769 * Interrupt entry points save only callee clobbered registers in fast path.
771 * Entry runs with interrupts off.
774 /* 0(%rsp): ~(interrupt number) */
775 .macro interrupt func
777 CFI_ADJUST_CFA_OFFSET 10*8
784 * Interrupt entry/exit should be protected against kprobes
786 .pushsection .kprobes.text, "ax"
788 * The interrupt stubs push (~vector+0x80) onto the stack and
789 * then jump to common_interrupt.
791 .p2align CONFIG_X86_L1_CACHE_SHIFT
794 addq $-0x80,(%rsp) /* Adjust vector to [-256,-1] range */
796 /* 0(%rsp): old_rsp-ARGOFFSET */
798 DISABLE_INTERRUPTS(CLBR_NONE)
800 decl PER_CPU_VAR(irq_count)
802 CFI_DEF_CFA_REGISTER rsp
803 CFI_ADJUST_CFA_OFFSET -8
805 GET_THREAD_INFO(%rcx)
806 testl $3,CS-ARGOFFSET(%rsp)
809 /* Interrupt came from user space */
811 * Has a correct top of stack, but a partial stack frame
812 * %rcx: thread info. Interrupts off.
814 retint_with_reschedule:
815 movl $_TIF_WORK_MASK,%edi
818 movl TI_flags(%rcx),%edx
823 retint_swapgs: /* return to user-space */
825 * The iretq could re-enable interrupts:
827 DISABLE_INTERRUPTS(CLBR_ANY)
832 retint_restore_args: /* return to kernel space */
833 DISABLE_INTERRUPTS(CLBR_ANY)
835 * The iretq could re-enable interrupts:
844 .section __ex_table, "a"
845 .quad irq_return, bad_iret
848 #ifdef CONFIG_PARAVIRT
852 .section __ex_table,"a"
853 .quad native_iret, bad_iret
860 * The iret traps when the %cs or %ss being restored is bogus.
861 * We've lost the original trap vector and error code.
862 * #GPF is the most likely one to get for an invalid selector.
863 * So pretend we completed the iret and took the #GPF in user mode.
865 * We are now running with the kernel GS after exception recovery.
866 * But error_entry expects us to have user GS to match the user %cs,
872 jmp general_protection
876 /* edi: workmask, edx: work */
879 bt $TIF_NEED_RESCHED,%edx
882 ENABLE_INTERRUPTS(CLBR_NONE)
884 CFI_ADJUST_CFA_OFFSET 8
887 CFI_ADJUST_CFA_OFFSET -8
888 GET_THREAD_INFO(%rcx)
889 DISABLE_INTERRUPTS(CLBR_NONE)
894 testl $_TIF_DO_NOTIFY_MASK,%edx
897 ENABLE_INTERRUPTS(CLBR_NONE)
899 movq $-1,ORIG_RAX(%rsp)
900 xorl %esi,%esi # oldset
901 movq %rsp,%rdi # &pt_regs
902 call do_notify_resume
904 DISABLE_INTERRUPTS(CLBR_NONE)
906 GET_THREAD_INFO(%rcx)
907 jmp retint_with_reschedule
909 #ifdef CONFIG_PREEMPT
910 /* Returning to kernel space. Check if we need preemption */
911 /* rcx: threadinfo. interrupts off. */
913 cmpl $0,TI_preempt_count(%rcx)
914 jnz retint_restore_args
915 bt $TIF_NEED_RESCHED,TI_flags(%rcx)
916 jnc retint_restore_args
917 bt $9,EFLAGS-ARGOFFSET(%rsp) /* interrupts off? */
918 jnc retint_restore_args
919 call preempt_schedule_irq
924 END(common_interrupt)
926 * End of kprobes section
933 .macro apicinterrupt num sym do_sym
937 CFI_ADJUST_CFA_OFFSET 8
945 apicinterrupt IRQ_MOVE_CLEANUP_VECTOR \
946 irq_move_cleanup_interrupt smp_irq_move_cleanup_interrupt
947 apicinterrupt REBOOT_VECTOR \
948 reboot_interrupt smp_reboot_interrupt
952 apicinterrupt UV_BAU_MESSAGE \
953 uv_bau_message_intr1 uv_bau_message_interrupt
955 apicinterrupt LOCAL_TIMER_VECTOR \
956 apic_timer_interrupt smp_apic_timer_interrupt
957 apicinterrupt X86_PLATFORM_IPI_VECTOR \
958 x86_platform_ipi smp_x86_platform_ipi
961 apicinterrupt INVALIDATE_TLB_VECTOR_START+0 \
962 invalidate_interrupt0 smp_invalidate_interrupt
963 apicinterrupt INVALIDATE_TLB_VECTOR_START+1 \
964 invalidate_interrupt1 smp_invalidate_interrupt
965 apicinterrupt INVALIDATE_TLB_VECTOR_START+2 \
966 invalidate_interrupt2 smp_invalidate_interrupt
967 apicinterrupt INVALIDATE_TLB_VECTOR_START+3 \
968 invalidate_interrupt3 smp_invalidate_interrupt
969 apicinterrupt INVALIDATE_TLB_VECTOR_START+4 \
970 invalidate_interrupt4 smp_invalidate_interrupt
971 apicinterrupt INVALIDATE_TLB_VECTOR_START+5 \
972 invalidate_interrupt5 smp_invalidate_interrupt
973 apicinterrupt INVALIDATE_TLB_VECTOR_START+6 \
974 invalidate_interrupt6 smp_invalidate_interrupt
975 apicinterrupt INVALIDATE_TLB_VECTOR_START+7 \
976 invalidate_interrupt7 smp_invalidate_interrupt
979 apicinterrupt THRESHOLD_APIC_VECTOR \
980 threshold_interrupt smp_threshold_interrupt
981 apicinterrupt THERMAL_APIC_VECTOR \
982 thermal_interrupt smp_thermal_interrupt
984 #ifdef CONFIG_X86_MCE
985 apicinterrupt MCE_SELF_VECTOR \
986 mce_self_interrupt smp_mce_self_interrupt
990 apicinterrupt CALL_FUNCTION_SINGLE_VECTOR \
991 call_function_single_interrupt smp_call_function_single_interrupt
992 apicinterrupt CALL_FUNCTION_VECTOR \
993 call_function_interrupt smp_call_function_interrupt
994 apicinterrupt RESCHEDULE_VECTOR \
995 reschedule_interrupt smp_reschedule_interrupt
998 apicinterrupt ERROR_APIC_VECTOR \
999 error_interrupt smp_error_interrupt
1000 apicinterrupt SPURIOUS_APIC_VECTOR \
1001 spurious_interrupt smp_spurious_interrupt
1003 #ifdef CONFIG_PERF_EVENTS
1004 apicinterrupt LOCAL_PENDING_VECTOR \
1005 perf_pending_interrupt smp_perf_pending_interrupt
1009 * Exception entry points.
1011 .macro zeroentry sym do_sym
1014 PARAVIRT_ADJUST_EXCEPTION_FRAME
1015 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1017 CFI_ADJUST_CFA_OFFSET 15*8
1020 movq %rsp,%rdi /* pt_regs pointer */
1021 xorl %esi,%esi /* no error code */
1023 jmp error_exit /* %ebx: no swapgs flag */
1028 .macro paranoidzeroentry sym do_sym
1031 PARAVIRT_ADJUST_EXCEPTION_FRAME
1032 pushq $-1 /* ORIG_RAX: no syscall to restart */
1033 CFI_ADJUST_CFA_OFFSET 8
1037 movq %rsp,%rdi /* pt_regs pointer */
1038 xorl %esi,%esi /* no error code */
1040 jmp paranoid_exit /* %ebx: no swapgs flag */
1045 #define INIT_TSS_IST(x) PER_CPU_VAR(init_tss) + (TSS_ist + ((x) - 1) * 8)
1046 .macro paranoidzeroentry_ist sym do_sym ist
1049 PARAVIRT_ADJUST_EXCEPTION_FRAME
1050 pushq $-1 /* ORIG_RAX: no syscall to restart */
1051 CFI_ADJUST_CFA_OFFSET 8
1055 movq %rsp,%rdi /* pt_regs pointer */
1056 xorl %esi,%esi /* no error code */
1057 subq $EXCEPTION_STKSZ, INIT_TSS_IST(\ist)
1059 addq $EXCEPTION_STKSZ, INIT_TSS_IST(\ist)
1060 jmp paranoid_exit /* %ebx: no swapgs flag */
1065 .macro errorentry sym do_sym
1068 PARAVIRT_ADJUST_EXCEPTION_FRAME
1070 CFI_ADJUST_CFA_OFFSET 15*8
1073 movq %rsp,%rdi /* pt_regs pointer */
1074 movq ORIG_RAX(%rsp),%rsi /* get error code */
1075 movq $-1,ORIG_RAX(%rsp) /* no syscall to restart */
1077 jmp error_exit /* %ebx: no swapgs flag */
1082 /* error code is on the stack already */
1083 .macro paranoiderrorentry sym do_sym
1086 PARAVIRT_ADJUST_EXCEPTION_FRAME
1088 CFI_ADJUST_CFA_OFFSET 15*8
1092 movq %rsp,%rdi /* pt_regs pointer */
1093 movq ORIG_RAX(%rsp),%rsi /* get error code */
1094 movq $-1,ORIG_RAX(%rsp) /* no syscall to restart */
1096 jmp paranoid_exit /* %ebx: no swapgs flag */
1101 zeroentry divide_error do_divide_error
1102 zeroentry overflow do_overflow
1103 zeroentry bounds do_bounds
1104 zeroentry invalid_op do_invalid_op
1105 zeroentry device_not_available do_device_not_available
1106 paranoiderrorentry double_fault do_double_fault
1107 zeroentry coprocessor_segment_overrun do_coprocessor_segment_overrun
1108 errorentry invalid_TSS do_invalid_TSS
1109 errorentry segment_not_present do_segment_not_present
1110 zeroentry spurious_interrupt_bug do_spurious_interrupt_bug
1111 zeroentry coprocessor_error do_coprocessor_error
1112 errorentry alignment_check do_alignment_check
1113 zeroentry simd_coprocessor_error do_simd_coprocessor_error
1115 /* Reload gs selector with exception handling */
1116 /* edi: new selector */
1117 ENTRY(native_load_gs_index)
1120 CFI_ADJUST_CFA_OFFSET 8
1121 DISABLE_INTERRUPTS(CLBR_ANY & ~CLBR_RDI)
1128 CFI_ADJUST_CFA_OFFSET -8
1131 END(native_load_gs_index)
1133 .section __ex_table,"a"
1135 .quad gs_change,bad_gs
1137 .section .fixup,"ax"
1138 /* running with kernelgs */
1140 SWAPGS /* switch back to user gs */
1146 ENTRY(kernel_thread_helper)
1147 pushq $0 # fake return address
1150 * Here we are in the child and the registers are set as they were
1151 * at kernel_thread() invocation in the parent.
1157 ud2 # padding for call trace
1159 END(kernel_thread_helper)
1162 * execve(). This function needs to use IRET, not SYSRET, to set up all state properly.
1164 * C extern interface:
1165 * extern long execve(const char *name, char **argv, char **envp)
1167 * asm input arguments:
1168 * rdi: name, rsi: argv, rdx: envp
1170 * We want to fallback into:
1171 * extern long sys_execve(const char *name, char **argv,char **envp, struct pt_regs *regs)
1173 * do_sys_execve asm fallback arguments:
1174 * rdi: name, rsi: argv, rdx: envp, rcx: fake frame on the stack
1176 ENTRY(kernel_execve)
1182 movq %rax, RAX(%rsp)
1185 je int_ret_from_sys_call
1192 /* Call softirq on interrupt stack. Interrupts are off. */
1196 CFI_ADJUST_CFA_OFFSET 8
1197 CFI_REL_OFFSET rbp,0
1199 CFI_DEF_CFA_REGISTER rbp
1200 incl PER_CPU_VAR(irq_count)
1201 cmove PER_CPU_VAR(irq_stack_ptr),%rsp
1202 push %rbp # backlink for old unwinder
1205 CFI_DEF_CFA_REGISTER rsp
1206 CFI_ADJUST_CFA_OFFSET -8
1207 decl PER_CPU_VAR(irq_count)
1213 zeroentry xen_hypervisor_callback xen_do_hypervisor_callback
1216 * A note on the "critical region" in our callback handler.
1217 * We want to avoid stacking callback handlers due to events occurring
1218 * during handling of the last event. To do this, we keep events disabled
1219 * until we've done all processing. HOWEVER, we must enable events before
1220 * popping the stack frame (can't be done atomically) and so it would still
1221 * be possible to get enough handler activations to overflow the stack.
1222 * Although unlikely, bugs of that kind are hard to track down, so we'd
1223 * like to avoid the possibility.
1224 * So, on entry to the handler we detect whether we interrupted an
1225 * existing activation in its critical region -- if so, we pop the current
1226 * activation and restart the handler using the previous one.
1228 ENTRY(xen_do_hypervisor_callback) # do_hypervisor_callback(struct *pt_regs)
1231 * Since we don't modify %rdi, evtchn_do_upall(struct *pt_regs) will
1232 * see the correct pointer to the pt_regs
1234 movq %rdi, %rsp # we don't return, adjust the stack frame
1237 11: incl PER_CPU_VAR(irq_count)
1239 CFI_DEF_CFA_REGISTER rbp
1240 cmovzq PER_CPU_VAR(irq_stack_ptr),%rsp
1241 pushq %rbp # backlink for old unwinder
1242 call xen_evtchn_do_upcall
1244 CFI_DEF_CFA_REGISTER rsp
1245 decl PER_CPU_VAR(irq_count)
1248 END(do_hypervisor_callback)
1251 * Hypervisor uses this for application faults while it executes.
1252 * We get here for two reasons:
1253 * 1. Fault while reloading DS, ES, FS or GS
1254 * 2. Fault while executing IRET
1255 * Category 1 we do not need to fix up as Xen has already reloaded all segment
1256 * registers that could be reloaded and zeroed the others.
1257 * Category 2 we fix up by killing the current process. We cannot use the
1258 * normal Linux return path in this case because if we use the IRET hypercall
1259 * to pop the stack frame we end up in an infinite loop of failsafe callbacks.
1260 * We distinguish between categories by comparing each saved segment register
1261 * with its current contents: any discrepancy means we in category 1.
1263 ENTRY(xen_failsafe_callback)
1265 /*CFI_REL_OFFSET gs,GS*/
1266 /*CFI_REL_OFFSET fs,FS*/
1267 /*CFI_REL_OFFSET es,ES*/
1268 /*CFI_REL_OFFSET ds,DS*/
1269 CFI_REL_OFFSET r11,8
1270 CFI_REL_OFFSET rcx,0
1284 /* All segments match their saved values => Category 2 (Bad IRET). */
1290 CFI_ADJUST_CFA_OFFSET -0x30
1291 pushq_cfi $0 /* RIP */
1294 jmp general_protection
1296 1: /* Segment mismatch => Category 1 (Bad segment). Retry the IRET. */
1302 CFI_ADJUST_CFA_OFFSET -0x30
1307 END(xen_failsafe_callback)
1309 apicinterrupt XEN_HVM_EVTCHN_CALLBACK \
1310 xen_hvm_callback_vector xen_evtchn_do_upcall
1312 #endif /* CONFIG_XEN */
1315 * Some functions should be protected against kprobes
1317 .pushsection .kprobes.text, "ax"
1319 paranoidzeroentry_ist debug do_debug DEBUG_STACK
1320 paranoidzeroentry_ist int3 do_int3 DEBUG_STACK
1321 paranoiderrorentry stack_segment do_stack_segment
1323 zeroentry xen_debug do_debug
1324 zeroentry xen_int3 do_int3
1325 errorentry xen_stack_segment do_stack_segment
1327 errorentry general_protection do_general_protection
1328 errorentry page_fault do_page_fault
1329 #ifdef CONFIG_X86_MCE
1330 paranoidzeroentry machine_check *machine_check_vector(%rip)
1334 * "Paranoid" exit path from exception stack.
1335 * Paranoid because this is used by NMIs and cannot take
1336 * any kernel state for granted.
1337 * We don't do kernel preemption checks here, because only
1338 * NMI should be common and it does not enable IRQs and
1339 * cannot get reschedule ticks.
1341 * "trace" is 0 for the NMI handler only, because irq-tracing
1342 * is fundamentally NMI-unsafe. (we cannot change the soft and
1343 * hard flags at once, atomically)
1346 /* ebx: no swapgs flag */
1347 ENTRY(paranoid_exit)
1349 DISABLE_INTERRUPTS(CLBR_NONE)
1351 testl %ebx,%ebx /* swapgs needed? */
1352 jnz paranoid_restore
1354 jnz paranoid_userspace
1365 GET_THREAD_INFO(%rcx)
1366 movl TI_flags(%rcx),%ebx
1367 andl $_TIF_WORK_MASK,%ebx
1369 movq %rsp,%rdi /* &pt_regs */
1371 movq %rax,%rsp /* switch stack for scheduling */
1372 testl $_TIF_NEED_RESCHED,%ebx
1373 jnz paranoid_schedule
1374 movl %ebx,%edx /* arg3: thread flags */
1376 ENABLE_INTERRUPTS(CLBR_NONE)
1377 xorl %esi,%esi /* arg2: oldset */
1378 movq %rsp,%rdi /* arg1: &pt_regs */
1379 call do_notify_resume
1380 DISABLE_INTERRUPTS(CLBR_NONE)
1382 jmp paranoid_userspace
1385 ENABLE_INTERRUPTS(CLBR_ANY)
1387 DISABLE_INTERRUPTS(CLBR_ANY)
1389 jmp paranoid_userspace
1394 * Exception entry point. This expects an error code/orig_rax on the stack.
1395 * returns in "no swapgs flag" in %ebx.
1399 CFI_ADJUST_CFA_OFFSET 15*8
1400 /* oldrax contains error code */
1419 je error_kernelspace
1428 * There are two places in the kernel that can potentially fault with
1429 * usergs. Handle them here. The exception handlers after iret run with
1430 * kernel gs again, so don't set the user space flag. B stepping K8s
1431 * sometimes report an truncated RIP for IRET exceptions returning to
1432 * compat mode. Check for these here too.
1436 leaq irq_return(%rip),%rcx
1437 cmpq %rcx,RIP+8(%rsp)
1439 movl %ecx,%eax /* zero extend */
1440 cmpq %rax,RIP+8(%rsp)
1442 cmpq $gs_change,RIP+8(%rsp)
1447 /* Fix truncated RIP */
1448 movq %rcx,RIP+8(%rsp)
1453 /* ebx: no swapgs flag (1: don't need swapgs, 0: need it) */
1458 DISABLE_INTERRUPTS(CLBR_NONE)
1460 GET_THREAD_INFO(%rcx)
1463 LOCKDEP_SYS_EXIT_IRQ
1464 movl TI_flags(%rcx),%edx
1465 movl $_TIF_WORK_MASK,%edi
1473 /* runs on exception stack */
1476 PARAVIRT_ADJUST_EXCEPTION_FRAME
1479 CFI_ADJUST_CFA_OFFSET 15*8
1482 /* paranoidentry do_nmi, 0; without TRACE_IRQS_OFF */
1486 #ifdef CONFIG_TRACE_IRQFLAGS
1487 /* paranoidexit; without TRACE_IRQS_OFF */
1488 /* ebx: no swapgs flag */
1489 DISABLE_INTERRUPTS(CLBR_NONE)
1490 testl %ebx,%ebx /* swapgs needed? */
1500 GET_THREAD_INFO(%rcx)
1501 movl TI_flags(%rcx),%ebx
1502 andl $_TIF_WORK_MASK,%ebx
1504 movq %rsp,%rdi /* &pt_regs */
1506 movq %rax,%rsp /* switch stack for scheduling */
1507 testl $_TIF_NEED_RESCHED,%ebx
1509 movl %ebx,%edx /* arg3: thread flags */
1510 ENABLE_INTERRUPTS(CLBR_NONE)
1511 xorl %esi,%esi /* arg2: oldset */
1512 movq %rsp,%rdi /* arg1: &pt_regs */
1513 call do_notify_resume
1514 DISABLE_INTERRUPTS(CLBR_NONE)
1517 ENABLE_INTERRUPTS(CLBR_ANY)
1519 DISABLE_INTERRUPTS(CLBR_ANY)
1528 ENTRY(ignore_sysret)
1536 * End of kprobes section