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_cfi $__KERNEL_DS /* ss */
217 /*CFI_REL_OFFSET ss,0*/
218 pushq_cfi %rax /* rsp */
220 pushq_cfi $X86_EFLAGS_IF /* eflags - interrupts on */
221 /*CFI_REL_OFFSET rflags,0*/
222 pushq_cfi $__KERNEL_CS /* cs */
223 /*CFI_REL_OFFSET cs,0*/
224 pushq_cfi \child_rip /* rip */
226 pushq_cfi %rax /* orig rax */
229 .macro UNFAKE_STACK_FRAME
231 CFI_ADJUST_CFA_OFFSET -(6*8)
235 * initial frame state for interrupts (and exceptions without error code)
237 .macro EMPTY_FRAME start=1 offset=0
241 CFI_DEF_CFA rsp,8+\offset
243 CFI_DEF_CFA_OFFSET 8+\offset
248 * initial frame state for interrupts (and exceptions without error code)
250 .macro INTR_FRAME start=1 offset=0
251 EMPTY_FRAME \start, SS+8+\offset-RIP
252 /*CFI_REL_OFFSET ss, SS+\offset-RIP*/
253 CFI_REL_OFFSET rsp, RSP+\offset-RIP
254 /*CFI_REL_OFFSET rflags, EFLAGS+\offset-RIP*/
255 /*CFI_REL_OFFSET cs, CS+\offset-RIP*/
256 CFI_REL_OFFSET rip, RIP+\offset-RIP
260 * initial frame state for exceptions with error code (and interrupts
261 * with vector already pushed)
263 .macro XCPT_FRAME start=1 offset=0
264 INTR_FRAME \start, RIP+\offset-ORIG_RAX
265 /*CFI_REL_OFFSET orig_rax, ORIG_RAX-ORIG_RAX*/
269 * frame that enables calling into C.
271 .macro PARTIAL_FRAME start=1 offset=0
272 XCPT_FRAME \start, ORIG_RAX+\offset-ARGOFFSET
273 CFI_REL_OFFSET rdi, RDI+\offset-ARGOFFSET
274 CFI_REL_OFFSET rsi, RSI+\offset-ARGOFFSET
275 CFI_REL_OFFSET rdx, RDX+\offset-ARGOFFSET
276 CFI_REL_OFFSET rcx, RCX+\offset-ARGOFFSET
277 CFI_REL_OFFSET rax, RAX+\offset-ARGOFFSET
278 CFI_REL_OFFSET r8, R8+\offset-ARGOFFSET
279 CFI_REL_OFFSET r9, R9+\offset-ARGOFFSET
280 CFI_REL_OFFSET r10, R10+\offset-ARGOFFSET
281 CFI_REL_OFFSET r11, R11+\offset-ARGOFFSET
285 * frame that enables passing a complete pt_regs to a C function.
287 .macro DEFAULT_FRAME start=1 offset=0
288 PARTIAL_FRAME \start, R11+\offset-R15
289 CFI_REL_OFFSET rbx, RBX+\offset
290 CFI_REL_OFFSET rbp, RBP+\offset
291 CFI_REL_OFFSET r12, R12+\offset
292 CFI_REL_OFFSET r13, R13+\offset
293 CFI_REL_OFFSET r14, R14+\offset
294 CFI_REL_OFFSET r15, R15+\offset
297 /* save partial stack frame */
298 .pushsection .kprobes.text, "ax"
303 * start from rbp in pt_regs and jump over
306 movq_cfi rdi, RDI+8-RBP
307 movq_cfi rsi, RSI+8-RBP
308 movq_cfi rdx, RDX+8-RBP
309 movq_cfi rcx, RCX+8-RBP
310 movq_cfi rax, RAX+8-RBP
311 movq_cfi r8, R8+8-RBP
312 movq_cfi r9, R9+8-RBP
313 movq_cfi r10, R10+8-RBP
314 movq_cfi r11, R11+8-RBP
316 leaq -RBP+8(%rsp),%rdi /* arg1 for handler */
317 movq_cfi rbp, 8 /* push %rbp */
318 leaq 8(%rsp), %rbp /* mov %rsp, %ebp */
323 * irq_count is used to check if a CPU is already on an interrupt stack
324 * or not. While this is essentially redundant with preempt_count it is
325 * a little cheaper to use a separate counter in the PDA (short of
326 * moving irq_enter into assembly, which would be too much work)
328 1: incl PER_CPU_VAR(irq_count)
330 popq_cfi %rax /* move return address... */
331 mov PER_CPU_VAR(irq_stack_ptr),%rsp
333 pushq_cfi %rbp /* backlink for unwinder */
334 pushq_cfi %rax /* ... to the new stack */
336 * 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 pushq_cfi kernel_eflags(%rip)
402 popfq_cfi # reset kernel eflags
404 call schedule_tail # rdi: 'prev' task parameter
406 GET_THREAD_INFO(%rcx)
410 testl $3, CS-ARGOFFSET(%rsp) # from kernel_thread?
411 je int_ret_from_sys_call
413 testl $_TIF_IA32, TI_flags(%rcx) # 32-bit compat task needs IRET
414 jnz int_ret_from_sys_call
416 RESTORE_TOP_OF_STACK %rdi, -ARGOFFSET
417 jmp ret_from_sys_call # go to the SYSRET fastpath
423 * System call entry. Upto 6 arguments in registers are supported.
425 * SYSCALL does not save anything on the stack and does not change the
431 * rax system call number
433 * rcx return address for syscall/sysret, C arg3
436 * r10 arg3 (--> moved to rcx for C)
439 * r11 eflags for syscall/sysret, temporary for C
440 * r12-r15,rbp,rbx saved by C code, not touched.
442 * Interrupts are off on entry.
443 * Only called from user space.
445 * XXX if we had a free scratch register we could save the RSP into the stack frame
446 * and report it properly in ps. Unfortunately we haven't.
448 * When user can change the frames always force IRET. That is because
449 * it deals with uncanonical addresses better. SYSRET has trouble
450 * with them due to bugs in both AMD and Intel CPUs.
456 CFI_DEF_CFA rsp,KERNEL_STACK_OFFSET
458 /*CFI_REGISTER rflags,r11*/
461 * A hypervisor implementation might want to use a label
462 * after the swapgs, so that it can do the swapgs
463 * for the guest and jump here on syscall.
465 ENTRY(system_call_after_swapgs)
467 movq %rsp,PER_CPU_VAR(old_rsp)
468 movq PER_CPU_VAR(kernel_stack),%rsp
470 * No need to follow this irqs off/on section - it's straight
473 ENABLE_INTERRUPTS(CLBR_NONE)
475 movq %rax,ORIG_RAX-ARGOFFSET(%rsp)
476 movq %rcx,RIP-ARGOFFSET(%rsp)
477 CFI_REL_OFFSET rip,RIP-ARGOFFSET
478 GET_THREAD_INFO(%rcx)
479 testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%rcx)
481 system_call_fastpath:
482 cmpq $__NR_syscall_max,%rax
485 call *sys_call_table(,%rax,8) # XXX: rip relative
486 movq %rax,RAX-ARGOFFSET(%rsp)
488 * Syscall return path ending with SYSRET (fast path)
489 * Has incomplete stack frame and undefined top of stack.
492 movl $_TIF_ALLWORK_MASK,%edi
496 GET_THREAD_INFO(%rcx)
497 DISABLE_INTERRUPTS(CLBR_NONE)
499 movl TI_flags(%rcx),%edx
504 * sysretq will re-enable interrupts:
507 movq RIP-ARGOFFSET(%rsp),%rcx
509 RESTORE_ARGS 0,-ARG_SKIP,1
510 /*CFI_REGISTER rflags,r11*/
511 movq PER_CPU_VAR(old_rsp), %rsp
515 /* Handle reschedules */
516 /* edx: work, edi: workmask */
518 bt $TIF_NEED_RESCHED,%edx
521 ENABLE_INTERRUPTS(CLBR_NONE)
527 /* Handle a signal */
530 ENABLE_INTERRUPTS(CLBR_NONE)
531 #ifdef CONFIG_AUDITSYSCALL
532 bt $TIF_SYSCALL_AUDIT,%edx
536 * We have a signal, or exit tracing or single-step.
537 * These all wind up with the iret return path anyway,
538 * so just join that path right now.
540 FIXUP_TOP_OF_STACK %r11, -ARGOFFSET
541 jmp int_check_syscall_exit_work
544 movq $-ENOSYS,RAX-ARGOFFSET(%rsp)
545 jmp ret_from_sys_call
547 #ifdef CONFIG_AUDITSYSCALL
549 * Fast path for syscall audit without full syscall trace.
550 * We just call audit_syscall_entry() directly, and then
551 * jump back to the normal fast path.
554 movq %r10,%r9 /* 6th arg: 4th syscall arg */
555 movq %rdx,%r8 /* 5th arg: 3rd syscall arg */
556 movq %rsi,%rcx /* 4th arg: 2nd syscall arg */
557 movq %rdi,%rdx /* 3rd arg: 1st syscall arg */
558 movq %rax,%rsi /* 2nd arg: syscall number */
559 movl $AUDIT_ARCH_X86_64,%edi /* 1st arg: audit arch */
560 call audit_syscall_entry
561 LOAD_ARGS 0 /* reload call-clobbered registers */
562 jmp system_call_fastpath
565 * Return fast path for syscall audit. Call audit_syscall_exit()
566 * directly and then jump back to the fast path with TIF_SYSCALL_AUDIT
570 movq RAX-ARGOFFSET(%rsp),%rsi /* second arg, syscall return value */
571 cmpq $0,%rsi /* is it < 0? */
572 setl %al /* 1 if so, 0 if not */
573 movzbl %al,%edi /* zero-extend that into %edi */
574 inc %edi /* first arg, 0->1(AUDITSC_SUCCESS), 1->2(AUDITSC_FAILURE) */
575 call audit_syscall_exit
576 movl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT),%edi
578 #endif /* CONFIG_AUDITSYSCALL */
580 /* Do syscall tracing */
582 #ifdef CONFIG_AUDITSYSCALL
583 testl $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags(%rcx)
587 movq $-ENOSYS,RAX(%rsp) /* ptrace can change this for a bad syscall */
588 FIXUP_TOP_OF_STACK %rdi
590 call syscall_trace_enter
592 * Reload arg registers from stack in case ptrace changed them.
593 * We don't reload %rax because syscall_trace_enter() returned
594 * the value it wants us to use in the table lookup.
596 LOAD_ARGS ARGOFFSET, 1
598 cmpq $__NR_syscall_max,%rax
599 ja int_ret_from_sys_call /* RAX(%rsp) set to -ENOSYS above */
600 movq %r10,%rcx /* fixup for C */
601 call *sys_call_table(,%rax,8)
602 movq %rax,RAX-ARGOFFSET(%rsp)
603 /* Use IRET because user could have changed frame */
606 * Syscall return path ending with IRET.
607 * Has correct top of stack, but partial stack frame.
609 GLOBAL(int_ret_from_sys_call)
610 DISABLE_INTERRUPTS(CLBR_NONE)
612 testl $3,CS-ARGOFFSET(%rsp)
613 je retint_restore_args
614 movl $_TIF_ALLWORK_MASK,%edi
615 /* edi: mask to check */
616 GLOBAL(int_with_check)
618 GET_THREAD_INFO(%rcx)
619 movl TI_flags(%rcx),%edx
622 andl $~TS_COMPAT,TI_status(%rcx)
625 /* Either reschedule or signal or syscall exit tracking needed. */
626 /* First do a reschedule test. */
627 /* edx: work, edi: workmask */
629 bt $TIF_NEED_RESCHED,%edx
632 ENABLE_INTERRUPTS(CLBR_NONE)
636 DISABLE_INTERRUPTS(CLBR_NONE)
640 /* handle signals and tracing -- both require a full stack frame */
643 ENABLE_INTERRUPTS(CLBR_NONE)
644 int_check_syscall_exit_work:
646 /* Check for syscall exit trace */
647 testl $_TIF_WORK_SYSCALL_EXIT,%edx
650 leaq 8(%rsp),%rdi # &ptregs -> arg1
651 call syscall_trace_leave
653 andl $~(_TIF_WORK_SYSCALL_EXIT|_TIF_SYSCALL_EMU),%edi
657 testl $_TIF_DO_NOTIFY_MASK,%edx
659 movq %rsp,%rdi # &ptregs -> arg1
660 xorl %esi,%esi # oldset -> arg2
661 call do_notify_resume
662 1: movl $_TIF_WORK_MASK,%edi
665 DISABLE_INTERRUPTS(CLBR_NONE)
672 * Certain special system calls that need to save a complete full stack frame.
674 .macro PTREGSCALL label,func,arg
676 PARTIAL_FRAME 1 8 /* offset 8: return address */
677 subq $REST_SKIP, %rsp
678 CFI_ADJUST_CFA_OFFSET REST_SKIP
680 DEFAULT_FRAME 0 8 /* offset 8: return address */
681 leaq 8(%rsp), \arg /* pt_regs pointer */
683 jmp ptregscall_common
688 PTREGSCALL stub_clone, sys_clone, %r8
689 PTREGSCALL stub_fork, sys_fork, %rdi
690 PTREGSCALL stub_vfork, sys_vfork, %rdi
691 PTREGSCALL stub_sigaltstack, sys_sigaltstack, %rdx
692 PTREGSCALL stub_iopl, sys_iopl, %rsi
694 ENTRY(ptregscall_common)
695 DEFAULT_FRAME 1 8 /* offset 8: return address */
696 RESTORE_TOP_OF_STACK %r11, 8
697 movq_cfi_restore R15+8, r15
698 movq_cfi_restore R14+8, r14
699 movq_cfi_restore R13+8, r13
700 movq_cfi_restore R12+8, r12
701 movq_cfi_restore RBP+8, rbp
702 movq_cfi_restore RBX+8, rbx
703 ret $REST_SKIP /* pop extended registers */
705 END(ptregscall_common)
712 FIXUP_TOP_OF_STACK %r11
715 RESTORE_TOP_OF_STACK %r11
718 jmp int_ret_from_sys_call
723 * sigreturn is special because it needs to restore all registers on return.
724 * This cannot be done with SYSRET, so use the IRET return path instead.
726 ENTRY(stub_rt_sigreturn)
732 FIXUP_TOP_OF_STACK %r11
733 call sys_rt_sigreturn
734 movq %rax,RAX(%rsp) # fixme, this could be done at the higher layer
736 jmp int_ret_from_sys_call
738 END(stub_rt_sigreturn)
741 * Build the entry stubs and pointer table with some assembler magic.
742 * We pack 7 stubs into a single 32-byte chunk, which will fit in a
743 * single cache line on all modern x86 implementations.
745 .section .init.rodata,"a"
749 .p2align CONFIG_X86_L1_CACHE_SHIFT
750 ENTRY(irq_entries_start)
752 vector=FIRST_EXTERNAL_VECTOR
753 .rept (NR_VECTORS-FIRST_EXTERNAL_VECTOR+6)/7
756 .if vector < NR_VECTORS
757 .if vector <> FIRST_EXTERNAL_VECTOR
758 CFI_ADJUST_CFA_OFFSET -8
760 1: pushq_cfi $(~vector+0x80) /* Note: always in signed byte range */
761 .if ((vector-FIRST_EXTERNAL_VECTOR)%7) <> 6
770 2: jmp common_interrupt
773 END(irq_entries_start)
780 * Interrupt entry/exit.
782 * Interrupt entry points save only callee clobbered registers in fast path.
784 * Entry runs with interrupts off.
787 /* 0(%rsp): ~(interrupt number) */
788 .macro interrupt func
789 /* reserve pt_regs for scratch regs and rbp */
790 subq $ORIG_RAX-RBP, %rsp
791 CFI_ADJUST_CFA_OFFSET ORIG_RAX-RBP
798 * Interrupt entry/exit should be protected against kprobes
800 .pushsection .kprobes.text, "ax"
802 * The interrupt stubs push (~vector+0x80) onto the stack and
803 * then jump to common_interrupt.
805 .p2align CONFIG_X86_L1_CACHE_SHIFT
808 addq $-0x80,(%rsp) /* Adjust vector to [-256,-1] range */
810 /* 0(%rsp): old_rsp-ARGOFFSET */
812 DISABLE_INTERRUPTS(CLBR_NONE)
814 decl PER_CPU_VAR(irq_count)
818 CFI_DEF_CFA_REGISTER rsp
819 CFI_ADJUST_CFA_OFFSET -8
821 /* we did not save rbx, restore only from ARGOFFSET */
823 CFI_ADJUST_CFA_OFFSET -8
825 GET_THREAD_INFO(%rcx)
826 testl $3,CS-ARGOFFSET(%rsp)
829 /* Interrupt came from user space */
831 * Has a correct top of stack, but a partial stack frame
832 * %rcx: thread info. Interrupts off.
834 retint_with_reschedule:
835 movl $_TIF_WORK_MASK,%edi
838 movl TI_flags(%rcx),%edx
843 retint_swapgs: /* return to user-space */
845 * The iretq could re-enable interrupts:
847 DISABLE_INTERRUPTS(CLBR_ANY)
852 retint_restore_args: /* return to kernel space */
853 DISABLE_INTERRUPTS(CLBR_ANY)
855 * The iretq could re-enable interrupts:
864 .section __ex_table, "a"
865 .quad irq_return, bad_iret
868 #ifdef CONFIG_PARAVIRT
872 .section __ex_table,"a"
873 .quad native_iret, bad_iret
880 * The iret traps when the %cs or %ss being restored is bogus.
881 * We've lost the original trap vector and error code.
882 * #GPF is the most likely one to get for an invalid selector.
883 * So pretend we completed the iret and took the #GPF in user mode.
885 * We are now running with the kernel GS after exception recovery.
886 * But error_entry expects us to have user GS to match the user %cs,
892 jmp general_protection
896 /* edi: workmask, edx: work */
899 bt $TIF_NEED_RESCHED,%edx
902 ENABLE_INTERRUPTS(CLBR_NONE)
906 GET_THREAD_INFO(%rcx)
907 DISABLE_INTERRUPTS(CLBR_NONE)
912 testl $_TIF_DO_NOTIFY_MASK,%edx
915 ENABLE_INTERRUPTS(CLBR_NONE)
917 movq $-1,ORIG_RAX(%rsp)
918 xorl %esi,%esi # oldset
919 movq %rsp,%rdi # &pt_regs
920 call do_notify_resume
922 DISABLE_INTERRUPTS(CLBR_NONE)
924 GET_THREAD_INFO(%rcx)
925 jmp retint_with_reschedule
927 #ifdef CONFIG_PREEMPT
928 /* Returning to kernel space. Check if we need preemption */
929 /* rcx: threadinfo. interrupts off. */
931 cmpl $0,TI_preempt_count(%rcx)
932 jnz retint_restore_args
933 bt $TIF_NEED_RESCHED,TI_flags(%rcx)
934 jnc retint_restore_args
935 bt $9,EFLAGS-ARGOFFSET(%rsp) /* interrupts off? */
936 jnc retint_restore_args
937 call preempt_schedule_irq
942 END(common_interrupt)
944 * End of kprobes section
951 .macro apicinterrupt num sym do_sym
962 apicinterrupt IRQ_MOVE_CLEANUP_VECTOR \
963 irq_move_cleanup_interrupt smp_irq_move_cleanup_interrupt
964 apicinterrupt REBOOT_VECTOR \
965 reboot_interrupt smp_reboot_interrupt
969 apicinterrupt UV_BAU_MESSAGE \
970 uv_bau_message_intr1 uv_bau_message_interrupt
972 apicinterrupt LOCAL_TIMER_VECTOR \
973 apic_timer_interrupt smp_apic_timer_interrupt
974 apicinterrupt X86_PLATFORM_IPI_VECTOR \
975 x86_platform_ipi smp_x86_platform_ipi
978 .irpc idx, "01234567"
979 apicinterrupt (INVALIDATE_TLB_VECTOR_START)+\idx \
980 invalidate_interrupt\idx smp_invalidate_interrupt
984 apicinterrupt THRESHOLD_APIC_VECTOR \
985 threshold_interrupt smp_threshold_interrupt
986 apicinterrupt THERMAL_APIC_VECTOR \
987 thermal_interrupt smp_thermal_interrupt
989 #ifdef CONFIG_X86_MCE
990 apicinterrupt MCE_SELF_VECTOR \
991 mce_self_interrupt smp_mce_self_interrupt
995 apicinterrupt CALL_FUNCTION_SINGLE_VECTOR \
996 call_function_single_interrupt smp_call_function_single_interrupt
997 apicinterrupt CALL_FUNCTION_VECTOR \
998 call_function_interrupt smp_call_function_interrupt
999 apicinterrupt RESCHEDULE_VECTOR \
1000 reschedule_interrupt smp_reschedule_interrupt
1003 apicinterrupt ERROR_APIC_VECTOR \
1004 error_interrupt smp_error_interrupt
1005 apicinterrupt SPURIOUS_APIC_VECTOR \
1006 spurious_interrupt smp_spurious_interrupt
1008 #ifdef CONFIG_IRQ_WORK
1009 apicinterrupt IRQ_WORK_VECTOR \
1010 irq_work_interrupt smp_irq_work_interrupt
1014 * Exception entry points.
1016 .macro zeroentry sym do_sym
1019 PARAVIRT_ADJUST_EXCEPTION_FRAME
1020 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1021 subq $ORIG_RAX-R15, %rsp
1022 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1025 movq %rsp,%rdi /* pt_regs pointer */
1026 xorl %esi,%esi /* no error code */
1028 jmp error_exit /* %ebx: no swapgs flag */
1033 .macro paranoidzeroentry sym do_sym
1036 PARAVIRT_ADJUST_EXCEPTION_FRAME
1037 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1038 subq $ORIG_RAX-R15, %rsp
1039 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1042 movq %rsp,%rdi /* pt_regs pointer */
1043 xorl %esi,%esi /* no error code */
1045 jmp paranoid_exit /* %ebx: no swapgs flag */
1050 #define INIT_TSS_IST(x) PER_CPU_VAR(init_tss) + (TSS_ist + ((x) - 1) * 8)
1051 .macro paranoidzeroentry_ist sym do_sym ist
1054 PARAVIRT_ADJUST_EXCEPTION_FRAME
1055 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1056 subq $ORIG_RAX-R15, %rsp
1057 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1060 movq %rsp,%rdi /* pt_regs pointer */
1061 xorl %esi,%esi /* no error code */
1062 subq $EXCEPTION_STKSZ, INIT_TSS_IST(\ist)
1064 addq $EXCEPTION_STKSZ, INIT_TSS_IST(\ist)
1065 jmp paranoid_exit /* %ebx: no swapgs flag */
1070 .macro errorentry sym do_sym
1073 PARAVIRT_ADJUST_EXCEPTION_FRAME
1074 subq $ORIG_RAX-R15, %rsp
1075 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1078 movq %rsp,%rdi /* pt_regs pointer */
1079 movq ORIG_RAX(%rsp),%rsi /* get error code */
1080 movq $-1,ORIG_RAX(%rsp) /* no syscall to restart */
1082 jmp error_exit /* %ebx: no swapgs flag */
1087 /* error code is on the stack already */
1088 .macro paranoiderrorentry sym do_sym
1091 PARAVIRT_ADJUST_EXCEPTION_FRAME
1092 subq $ORIG_RAX-R15, %rsp
1093 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1097 movq %rsp,%rdi /* pt_regs pointer */
1098 movq ORIG_RAX(%rsp),%rsi /* get error code */
1099 movq $-1,ORIG_RAX(%rsp) /* no syscall to restart */
1101 jmp paranoid_exit /* %ebx: no swapgs flag */
1106 zeroentry divide_error do_divide_error
1107 zeroentry overflow do_overflow
1108 zeroentry bounds do_bounds
1109 zeroentry invalid_op do_invalid_op
1110 zeroentry device_not_available do_device_not_available
1111 paranoiderrorentry double_fault do_double_fault
1112 zeroentry coprocessor_segment_overrun do_coprocessor_segment_overrun
1113 errorentry invalid_TSS do_invalid_TSS
1114 errorentry segment_not_present do_segment_not_present
1115 zeroentry spurious_interrupt_bug do_spurious_interrupt_bug
1116 zeroentry coprocessor_error do_coprocessor_error
1117 errorentry alignment_check do_alignment_check
1118 zeroentry simd_coprocessor_error do_simd_coprocessor_error
1120 /* Reload gs selector with exception handling */
1121 /* edi: new selector */
1122 ENTRY(native_load_gs_index)
1125 DISABLE_INTERRUPTS(CLBR_ANY & ~CLBR_RDI)
1129 2: mfence /* workaround */
1134 END(native_load_gs_index)
1136 .section __ex_table,"a"
1138 .quad gs_change,bad_gs
1140 .section .fixup,"ax"
1141 /* running with kernelgs */
1143 SWAPGS /* switch back to user gs */
1149 ENTRY(kernel_thread_helper)
1150 pushq $0 # fake return address
1153 * Here we are in the child and the registers are set as they were
1154 * at kernel_thread() invocation in the parent.
1160 ud2 # padding for call trace
1162 END(kernel_thread_helper)
1165 * execve(). This function needs to use IRET, not SYSRET, to set up all state properly.
1167 * C extern interface:
1168 * extern long execve(const char *name, char **argv, char **envp)
1170 * asm input arguments:
1171 * rdi: name, rsi: argv, rdx: envp
1173 * We want to fallback into:
1174 * extern long sys_execve(const char *name, char **argv,char **envp, struct pt_regs *regs)
1176 * do_sys_execve asm fallback arguments:
1177 * rdi: name, rsi: argv, rdx: envp, rcx: fake frame on the stack
1179 ENTRY(kernel_execve)
1185 movq %rax, RAX(%rsp)
1188 je int_ret_from_sys_call
1195 /* Call softirq on interrupt stack. Interrupts are off. */
1199 CFI_REL_OFFSET rbp,0
1201 CFI_DEF_CFA_REGISTER rbp
1202 incl PER_CPU_VAR(irq_count)
1203 cmove PER_CPU_VAR(irq_stack_ptr),%rsp
1204 push %rbp # backlink for old unwinder
1208 CFI_DEF_CFA_REGISTER rsp
1209 CFI_ADJUST_CFA_OFFSET -8
1210 decl PER_CPU_VAR(irq_count)
1216 zeroentry xen_hypervisor_callback xen_do_hypervisor_callback
1219 * A note on the "critical region" in our callback handler.
1220 * We want to avoid stacking callback handlers due to events occurring
1221 * during handling of the last event. To do this, we keep events disabled
1222 * until we've done all processing. HOWEVER, we must enable events before
1223 * popping the stack frame (can't be done atomically) and so it would still
1224 * be possible to get enough handler activations to overflow the stack.
1225 * Although unlikely, bugs of that kind are hard to track down, so we'd
1226 * like to avoid the possibility.
1227 * So, on entry to the handler we detect whether we interrupted an
1228 * existing activation in its critical region -- if so, we pop the current
1229 * activation and restart the handler using the previous one.
1231 ENTRY(xen_do_hypervisor_callback) # do_hypervisor_callback(struct *pt_regs)
1234 * Since we don't modify %rdi, evtchn_do_upall(struct *pt_regs) will
1235 * see the correct pointer to the pt_regs
1237 movq %rdi, %rsp # we don't return, adjust the stack frame
1240 11: incl PER_CPU_VAR(irq_count)
1242 CFI_DEF_CFA_REGISTER rbp
1243 cmovzq PER_CPU_VAR(irq_stack_ptr),%rsp
1244 pushq %rbp # backlink for old unwinder
1245 call xen_evtchn_do_upcall
1247 CFI_DEF_CFA_REGISTER rsp
1248 decl PER_CPU_VAR(irq_count)
1251 END(do_hypervisor_callback)
1254 * Hypervisor uses this for application faults while it executes.
1255 * We get here for two reasons:
1256 * 1. Fault while reloading DS, ES, FS or GS
1257 * 2. Fault while executing IRET
1258 * Category 1 we do not need to fix up as Xen has already reloaded all segment
1259 * registers that could be reloaded and zeroed the others.
1260 * Category 2 we fix up by killing the current process. We cannot use the
1261 * normal Linux return path in this case because if we use the IRET hypercall
1262 * to pop the stack frame we end up in an infinite loop of failsafe callbacks.
1263 * We distinguish between categories by comparing each saved segment register
1264 * with its current contents: any discrepancy means we in category 1.
1266 ENTRY(xen_failsafe_callback)
1268 /*CFI_REL_OFFSET gs,GS*/
1269 /*CFI_REL_OFFSET fs,FS*/
1270 /*CFI_REL_OFFSET es,ES*/
1271 /*CFI_REL_OFFSET ds,DS*/
1272 CFI_REL_OFFSET r11,8
1273 CFI_REL_OFFSET rcx,0
1287 /* All segments match their saved values => Category 2 (Bad IRET). */
1293 CFI_ADJUST_CFA_OFFSET -0x30
1294 pushq_cfi $0 /* RIP */
1297 jmp general_protection
1299 1: /* Segment mismatch => Category 1 (Bad segment). Retry the IRET. */
1305 CFI_ADJUST_CFA_OFFSET -0x30
1310 END(xen_failsafe_callback)
1312 apicinterrupt XEN_HVM_EVTCHN_CALLBACK \
1313 xen_hvm_callback_vector xen_evtchn_do_upcall
1315 #endif /* CONFIG_XEN */
1318 * Some functions should be protected against kprobes
1320 .pushsection .kprobes.text, "ax"
1322 paranoidzeroentry_ist debug do_debug DEBUG_STACK
1323 paranoidzeroentry_ist int3 do_int3 DEBUG_STACK
1324 paranoiderrorentry stack_segment do_stack_segment
1326 zeroentry xen_debug do_debug
1327 zeroentry xen_int3 do_int3
1328 errorentry xen_stack_segment do_stack_segment
1330 errorentry general_protection do_general_protection
1331 errorentry page_fault do_page_fault
1332 #ifdef CONFIG_KVM_GUEST
1333 errorentry async_page_fault do_async_page_fault
1335 #ifdef CONFIG_X86_MCE
1336 paranoidzeroentry machine_check *machine_check_vector(%rip)
1340 * "Paranoid" exit path from exception stack.
1341 * Paranoid because this is used by NMIs and cannot take
1342 * any kernel state for granted.
1343 * We don't do kernel preemption checks here, because only
1344 * NMI should be common and it does not enable IRQs and
1345 * cannot get reschedule ticks.
1347 * "trace" is 0 for the NMI handler only, because irq-tracing
1348 * is fundamentally NMI-unsafe. (we cannot change the soft and
1349 * hard flags at once, atomically)
1352 /* ebx: no swapgs flag */
1353 ENTRY(paranoid_exit)
1355 DISABLE_INTERRUPTS(CLBR_NONE)
1357 testl %ebx,%ebx /* swapgs needed? */
1358 jnz paranoid_restore
1360 jnz paranoid_userspace
1371 GET_THREAD_INFO(%rcx)
1372 movl TI_flags(%rcx),%ebx
1373 andl $_TIF_WORK_MASK,%ebx
1375 movq %rsp,%rdi /* &pt_regs */
1377 movq %rax,%rsp /* switch stack for scheduling */
1378 testl $_TIF_NEED_RESCHED,%ebx
1379 jnz paranoid_schedule
1380 movl %ebx,%edx /* arg3: thread flags */
1382 ENABLE_INTERRUPTS(CLBR_NONE)
1383 xorl %esi,%esi /* arg2: oldset */
1384 movq %rsp,%rdi /* arg1: &pt_regs */
1385 call do_notify_resume
1386 DISABLE_INTERRUPTS(CLBR_NONE)
1388 jmp paranoid_userspace
1391 ENABLE_INTERRUPTS(CLBR_ANY)
1393 DISABLE_INTERRUPTS(CLBR_ANY)
1395 jmp paranoid_userspace
1400 * Exception entry point. This expects an error code/orig_rax on the stack.
1401 * returns in "no swapgs flag" in %ebx.
1405 CFI_ADJUST_CFA_OFFSET 15*8
1406 /* oldrax contains error code */
1425 je error_kernelspace
1433 * There are two places in the kernel that can potentially fault with
1434 * usergs. Handle them here. The exception handlers after iret run with
1435 * kernel gs again, so don't set the user space flag. B stepping K8s
1436 * sometimes report an truncated RIP for IRET exceptions returning to
1437 * compat mode. Check for these here too.
1441 leaq irq_return(%rip),%rcx
1442 cmpq %rcx,RIP+8(%rsp)
1444 movl %ecx,%eax /* zero extend */
1445 cmpq %rax,RIP+8(%rsp)
1447 cmpq $gs_change,RIP+8(%rsp)
1452 /* Fix truncated RIP */
1453 movq %rcx,RIP+8(%rsp)
1459 /* ebx: no swapgs flag (1: don't need swapgs, 0: need it) */
1464 DISABLE_INTERRUPTS(CLBR_NONE)
1466 GET_THREAD_INFO(%rcx)
1469 LOCKDEP_SYS_EXIT_IRQ
1470 movl TI_flags(%rcx),%edx
1471 movl $_TIF_WORK_MASK,%edi
1479 /* runs on exception stack */
1482 PARAVIRT_ADJUST_EXCEPTION_FRAME
1484 subq $ORIG_RAX-R15, %rsp
1485 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1488 /* paranoidentry do_nmi, 0; without TRACE_IRQS_OFF */
1492 #ifdef CONFIG_TRACE_IRQFLAGS
1493 /* paranoidexit; without TRACE_IRQS_OFF */
1494 /* ebx: no swapgs flag */
1495 DISABLE_INTERRUPTS(CLBR_NONE)
1496 testl %ebx,%ebx /* swapgs needed? */
1506 GET_THREAD_INFO(%rcx)
1507 movl TI_flags(%rcx),%ebx
1508 andl $_TIF_WORK_MASK,%ebx
1510 movq %rsp,%rdi /* &pt_regs */
1512 movq %rax,%rsp /* switch stack for scheduling */
1513 testl $_TIF_NEED_RESCHED,%ebx
1515 movl %ebx,%edx /* arg3: thread flags */
1516 ENABLE_INTERRUPTS(CLBR_NONE)
1517 xorl %esi,%esi /* arg2: oldset */
1518 movq %rsp,%rdi /* arg1: &pt_regs */
1519 call do_notify_resume
1520 DISABLE_INTERRUPTS(CLBR_NONE)
1523 ENABLE_INTERRUPTS(CLBR_ANY)
1525 DISABLE_INTERRUPTS(CLBR_ANY)
1534 ENTRY(ignore_sysret)
1542 * End of kprobes section