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"
302 movq_cfi rdi, RDI+16-ARGOFFSET
303 movq_cfi rsi, RSI+16-ARGOFFSET
304 movq_cfi rdx, RDX+16-ARGOFFSET
305 movq_cfi rcx, RCX+16-ARGOFFSET
306 movq_cfi rax, RAX+16-ARGOFFSET
307 movq_cfi r8, R8+16-ARGOFFSET
308 movq_cfi r9, R9+16-ARGOFFSET
309 movq_cfi r10, R10+16-ARGOFFSET
310 movq_cfi r11, R11+16-ARGOFFSET
312 leaq -ARGOFFSET+16(%rsp),%rdi /* arg1 for handler */
313 movq_cfi rbp, 8 /* push %rbp */
314 leaq 8(%rsp), %rbp /* mov %rsp, %ebp */
319 * irq_count is used to check if a CPU is already on an interrupt stack
320 * or not. While this is essentially redundant with preempt_count it is
321 * a little cheaper to use a separate counter in the PDA (short of
322 * moving irq_enter into assembly, which would be too much work)
324 1: incl PER_CPU_VAR(irq_count)
326 popq_cfi %rax /* move return address... */
327 mov PER_CPU_VAR(irq_stack_ptr),%rsp
329 pushq_cfi %rbp /* backlink for unwinder */
330 pushq_cfi %rax /* ... to the new stack */
332 * We entered an interrupt context - irqs are off:
341 PARTIAL_FRAME 1 REST_SKIP+8
342 movq 5*8+16(%rsp), %r11 /* save return address */
349 movq %r11, 8(%rsp) /* return address */
350 FIXUP_TOP_OF_STACK %r11, 16
355 /* save complete stack frame */
356 .pushsection .kprobes.text, "ax"
376 movl $MSR_GS_BASE,%ecx
379 js 1f /* negative -> in kernel */
388 * A newly forked process directly context switches into this address.
390 * rdi: prev task we switched from
395 LOCK ; btr $TIF_FORK,TI_flags(%r8)
397 pushq_cfi kernel_eflags(%rip)
398 popfq_cfi # reset kernel eflags
400 call schedule_tail # rdi: 'prev' task parameter
402 GET_THREAD_INFO(%rcx)
406 testl $3, CS-ARGOFFSET(%rsp) # from kernel_thread?
407 je int_ret_from_sys_call
409 testl $_TIF_IA32, TI_flags(%rcx) # 32-bit compat task needs IRET
410 jnz int_ret_from_sys_call
412 RESTORE_TOP_OF_STACK %rdi, -ARGOFFSET
413 jmp ret_from_sys_call # go to the SYSRET fastpath
419 * System call entry. Upto 6 arguments in registers are supported.
421 * SYSCALL does not save anything on the stack and does not change the
427 * rax system call number
429 * rcx return address for syscall/sysret, C arg3
432 * r10 arg3 (--> moved to rcx for C)
435 * r11 eflags for syscall/sysret, temporary for C
436 * r12-r15,rbp,rbx saved by C code, not touched.
438 * Interrupts are off on entry.
439 * Only called from user space.
441 * XXX if we had a free scratch register we could save the RSP into the stack frame
442 * and report it properly in ps. Unfortunately we haven't.
444 * When user can change the frames always force IRET. That is because
445 * it deals with uncanonical addresses better. SYSRET has trouble
446 * with them due to bugs in both AMD and Intel CPUs.
452 CFI_DEF_CFA rsp,KERNEL_STACK_OFFSET
454 /*CFI_REGISTER rflags,r11*/
457 * A hypervisor implementation might want to use a label
458 * after the swapgs, so that it can do the swapgs
459 * for the guest and jump here on syscall.
461 ENTRY(system_call_after_swapgs)
463 movq %rsp,PER_CPU_VAR(old_rsp)
464 movq PER_CPU_VAR(kernel_stack),%rsp
466 * No need to follow this irqs off/on section - it's straight
469 ENABLE_INTERRUPTS(CLBR_NONE)
471 movq %rax,ORIG_RAX-ARGOFFSET(%rsp)
472 movq %rcx,RIP-ARGOFFSET(%rsp)
473 CFI_REL_OFFSET rip,RIP-ARGOFFSET
474 GET_THREAD_INFO(%rcx)
475 testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%rcx)
477 system_call_fastpath:
478 cmpq $__NR_syscall_max,%rax
481 call *sys_call_table(,%rax,8) # XXX: rip relative
482 movq %rax,RAX-ARGOFFSET(%rsp)
484 * Syscall return path ending with SYSRET (fast path)
485 * Has incomplete stack frame and undefined top of stack.
488 movl $_TIF_ALLWORK_MASK,%edi
492 GET_THREAD_INFO(%rcx)
493 DISABLE_INTERRUPTS(CLBR_NONE)
495 movl TI_flags(%rcx),%edx
500 * sysretq will re-enable interrupts:
503 movq RIP-ARGOFFSET(%rsp),%rcx
505 RESTORE_ARGS 0,-ARG_SKIP,1
506 /*CFI_REGISTER rflags,r11*/
507 movq PER_CPU_VAR(old_rsp), %rsp
511 /* Handle reschedules */
512 /* edx: work, edi: workmask */
514 bt $TIF_NEED_RESCHED,%edx
517 ENABLE_INTERRUPTS(CLBR_NONE)
523 /* Handle a signal */
526 ENABLE_INTERRUPTS(CLBR_NONE)
527 #ifdef CONFIG_AUDITSYSCALL
528 bt $TIF_SYSCALL_AUDIT,%edx
532 * We have a signal, or exit tracing or single-step.
533 * These all wind up with the iret return path anyway,
534 * so just join that path right now.
536 FIXUP_TOP_OF_STACK %r11, -ARGOFFSET
537 jmp int_check_syscall_exit_work
540 movq $-ENOSYS,RAX-ARGOFFSET(%rsp)
541 jmp ret_from_sys_call
543 #ifdef CONFIG_AUDITSYSCALL
545 * Fast path for syscall audit without full syscall trace.
546 * We just call audit_syscall_entry() directly, and then
547 * jump back to the normal fast path.
550 movq %r10,%r9 /* 6th arg: 4th syscall arg */
551 movq %rdx,%r8 /* 5th arg: 3rd syscall arg */
552 movq %rsi,%rcx /* 4th arg: 2nd syscall arg */
553 movq %rdi,%rdx /* 3rd arg: 1st syscall arg */
554 movq %rax,%rsi /* 2nd arg: syscall number */
555 movl $AUDIT_ARCH_X86_64,%edi /* 1st arg: audit arch */
556 call audit_syscall_entry
557 LOAD_ARGS 0 /* reload call-clobbered registers */
558 jmp system_call_fastpath
561 * Return fast path for syscall audit. Call audit_syscall_exit()
562 * directly and then jump back to the fast path with TIF_SYSCALL_AUDIT
566 movq RAX-ARGOFFSET(%rsp),%rsi /* second arg, syscall return value */
567 cmpq $0,%rsi /* is it < 0? */
568 setl %al /* 1 if so, 0 if not */
569 movzbl %al,%edi /* zero-extend that into %edi */
570 inc %edi /* first arg, 0->1(AUDITSC_SUCCESS), 1->2(AUDITSC_FAILURE) */
571 call audit_syscall_exit
572 movl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT),%edi
574 #endif /* CONFIG_AUDITSYSCALL */
576 /* Do syscall tracing */
578 #ifdef CONFIG_AUDITSYSCALL
579 testl $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags(%rcx)
583 movq $-ENOSYS,RAX(%rsp) /* ptrace can change this for a bad syscall */
584 FIXUP_TOP_OF_STACK %rdi
586 call syscall_trace_enter
588 * Reload arg registers from stack in case ptrace changed them.
589 * We don't reload %rax because syscall_trace_enter() returned
590 * the value it wants us to use in the table lookup.
592 LOAD_ARGS ARGOFFSET, 1
594 cmpq $__NR_syscall_max,%rax
595 ja int_ret_from_sys_call /* RAX(%rsp) set to -ENOSYS above */
596 movq %r10,%rcx /* fixup for C */
597 call *sys_call_table(,%rax,8)
598 movq %rax,RAX-ARGOFFSET(%rsp)
599 /* Use IRET because user could have changed frame */
602 * Syscall return path ending with IRET.
603 * Has correct top of stack, but partial stack frame.
605 GLOBAL(int_ret_from_sys_call)
606 DISABLE_INTERRUPTS(CLBR_NONE)
608 testl $3,CS-ARGOFFSET(%rsp)
609 je retint_restore_args
610 movl $_TIF_ALLWORK_MASK,%edi
611 /* edi: mask to check */
612 GLOBAL(int_with_check)
614 GET_THREAD_INFO(%rcx)
615 movl TI_flags(%rcx),%edx
618 andl $~TS_COMPAT,TI_status(%rcx)
621 /* Either reschedule or signal or syscall exit tracking needed. */
622 /* First do a reschedule test. */
623 /* edx: work, edi: workmask */
625 bt $TIF_NEED_RESCHED,%edx
628 ENABLE_INTERRUPTS(CLBR_NONE)
632 DISABLE_INTERRUPTS(CLBR_NONE)
636 /* handle signals and tracing -- both require a full stack frame */
639 ENABLE_INTERRUPTS(CLBR_NONE)
640 int_check_syscall_exit_work:
642 /* Check for syscall exit trace */
643 testl $_TIF_WORK_SYSCALL_EXIT,%edx
646 leaq 8(%rsp),%rdi # &ptregs -> arg1
647 call syscall_trace_leave
649 andl $~(_TIF_WORK_SYSCALL_EXIT|_TIF_SYSCALL_EMU),%edi
653 testl $_TIF_DO_NOTIFY_MASK,%edx
655 movq %rsp,%rdi # &ptregs -> arg1
656 xorl %esi,%esi # oldset -> arg2
657 call do_notify_resume
658 1: movl $_TIF_WORK_MASK,%edi
661 DISABLE_INTERRUPTS(CLBR_NONE)
668 * Certain special system calls that need to save a complete full stack frame.
670 .macro PTREGSCALL label,func,arg
672 PARTIAL_FRAME 1 8 /* offset 8: return address */
673 subq $REST_SKIP, %rsp
674 CFI_ADJUST_CFA_OFFSET REST_SKIP
676 DEFAULT_FRAME 0 8 /* offset 8: return address */
677 leaq 8(%rsp), \arg /* pt_regs pointer */
679 jmp ptregscall_common
684 PTREGSCALL stub_clone, sys_clone, %r8
685 PTREGSCALL stub_fork, sys_fork, %rdi
686 PTREGSCALL stub_vfork, sys_vfork, %rdi
687 PTREGSCALL stub_sigaltstack, sys_sigaltstack, %rdx
688 PTREGSCALL stub_iopl, sys_iopl, %rsi
690 ENTRY(ptregscall_common)
691 DEFAULT_FRAME 1 8 /* offset 8: return address */
692 RESTORE_TOP_OF_STACK %r11, 8
693 movq_cfi_restore R15+8, r15
694 movq_cfi_restore R14+8, r14
695 movq_cfi_restore R13+8, r13
696 movq_cfi_restore R12+8, r12
697 movq_cfi_restore RBP+8, rbp
698 movq_cfi_restore RBX+8, rbx
699 ret $REST_SKIP /* pop extended registers */
701 END(ptregscall_common)
708 FIXUP_TOP_OF_STACK %r11
711 RESTORE_TOP_OF_STACK %r11
714 jmp int_ret_from_sys_call
719 * sigreturn is special because it needs to restore all registers on return.
720 * This cannot be done with SYSRET, so use the IRET return path instead.
722 ENTRY(stub_rt_sigreturn)
728 FIXUP_TOP_OF_STACK %r11
729 call sys_rt_sigreturn
730 movq %rax,RAX(%rsp) # fixme, this could be done at the higher layer
732 jmp int_ret_from_sys_call
734 END(stub_rt_sigreturn)
737 * Build the entry stubs and pointer table with some assembler magic.
738 * We pack 7 stubs into a single 32-byte chunk, which will fit in a
739 * single cache line on all modern x86 implementations.
741 .section .init.rodata,"a"
745 .p2align CONFIG_X86_L1_CACHE_SHIFT
746 ENTRY(irq_entries_start)
748 vector=FIRST_EXTERNAL_VECTOR
749 .rept (NR_VECTORS-FIRST_EXTERNAL_VECTOR+6)/7
752 .if vector < NR_VECTORS
753 .if vector <> FIRST_EXTERNAL_VECTOR
754 CFI_ADJUST_CFA_OFFSET -8
756 1: pushq_cfi $(~vector+0x80) /* Note: always in signed byte range */
757 .if ((vector-FIRST_EXTERNAL_VECTOR)%7) <> 6
766 2: jmp common_interrupt
769 END(irq_entries_start)
776 * Interrupt entry/exit.
778 * Interrupt entry points save only callee clobbered registers in fast path.
780 * Entry runs with interrupts off.
783 /* 0(%rsp): ~(interrupt number) */
784 .macro interrupt func
785 subq $ORIG_RAX-ARGOFFSET+8, %rsp
786 CFI_ADJUST_CFA_OFFSET ORIG_RAX-ARGOFFSET+8
793 * Interrupt entry/exit should be protected against kprobes
795 .pushsection .kprobes.text, "ax"
797 * The interrupt stubs push (~vector+0x80) onto the stack and
798 * then jump to common_interrupt.
800 .p2align CONFIG_X86_L1_CACHE_SHIFT
803 addq $-0x80,(%rsp) /* Adjust vector to [-256,-1] range */
805 /* 0(%rsp): old_rsp-ARGOFFSET */
807 DISABLE_INTERRUPTS(CLBR_NONE)
809 decl PER_CPU_VAR(irq_count)
812 CFI_DEF_CFA_REGISTER rsp
813 CFI_ADJUST_CFA_OFFSET -8
815 GET_THREAD_INFO(%rcx)
816 testl $3,CS-ARGOFFSET(%rsp)
819 /* Interrupt came from user space */
821 * Has a correct top of stack, but a partial stack frame
822 * %rcx: thread info. Interrupts off.
824 retint_with_reschedule:
825 movl $_TIF_WORK_MASK,%edi
828 movl TI_flags(%rcx),%edx
833 retint_swapgs: /* return to user-space */
835 * The iretq could re-enable interrupts:
837 DISABLE_INTERRUPTS(CLBR_ANY)
842 retint_restore_args: /* return to kernel space */
843 DISABLE_INTERRUPTS(CLBR_ANY)
845 * The iretq could re-enable interrupts:
854 .section __ex_table, "a"
855 .quad irq_return, bad_iret
858 #ifdef CONFIG_PARAVIRT
862 .section __ex_table,"a"
863 .quad native_iret, bad_iret
870 * The iret traps when the %cs or %ss being restored is bogus.
871 * We've lost the original trap vector and error code.
872 * #GPF is the most likely one to get for an invalid selector.
873 * So pretend we completed the iret and took the #GPF in user mode.
875 * We are now running with the kernel GS after exception recovery.
876 * But error_entry expects us to have user GS to match the user %cs,
882 jmp general_protection
886 /* edi: workmask, edx: work */
889 bt $TIF_NEED_RESCHED,%edx
892 ENABLE_INTERRUPTS(CLBR_NONE)
896 GET_THREAD_INFO(%rcx)
897 DISABLE_INTERRUPTS(CLBR_NONE)
902 testl $_TIF_DO_NOTIFY_MASK,%edx
905 ENABLE_INTERRUPTS(CLBR_NONE)
907 movq $-1,ORIG_RAX(%rsp)
908 xorl %esi,%esi # oldset
909 movq %rsp,%rdi # &pt_regs
910 call do_notify_resume
912 DISABLE_INTERRUPTS(CLBR_NONE)
914 GET_THREAD_INFO(%rcx)
915 jmp retint_with_reschedule
917 #ifdef CONFIG_PREEMPT
918 /* Returning to kernel space. Check if we need preemption */
919 /* rcx: threadinfo. interrupts off. */
921 cmpl $0,TI_preempt_count(%rcx)
922 jnz retint_restore_args
923 bt $TIF_NEED_RESCHED,TI_flags(%rcx)
924 jnc retint_restore_args
925 bt $9,EFLAGS-ARGOFFSET(%rsp) /* interrupts off? */
926 jnc retint_restore_args
927 call preempt_schedule_irq
932 END(common_interrupt)
934 * End of kprobes section
941 .macro apicinterrupt num sym do_sym
952 apicinterrupt IRQ_MOVE_CLEANUP_VECTOR \
953 irq_move_cleanup_interrupt smp_irq_move_cleanup_interrupt
954 apicinterrupt REBOOT_VECTOR \
955 reboot_interrupt smp_reboot_interrupt
959 apicinterrupt UV_BAU_MESSAGE \
960 uv_bau_message_intr1 uv_bau_message_interrupt
962 apicinterrupt LOCAL_TIMER_VECTOR \
963 apic_timer_interrupt smp_apic_timer_interrupt
964 apicinterrupt X86_PLATFORM_IPI_VECTOR \
965 x86_platform_ipi smp_x86_platform_ipi
968 .irpc idx, "01234567"
969 apicinterrupt (INVALIDATE_TLB_VECTOR_START)+\idx \
970 invalidate_interrupt\idx smp_invalidate_interrupt
974 apicinterrupt THRESHOLD_APIC_VECTOR \
975 threshold_interrupt smp_threshold_interrupt
976 apicinterrupt THERMAL_APIC_VECTOR \
977 thermal_interrupt smp_thermal_interrupt
979 #ifdef CONFIG_X86_MCE
980 apicinterrupt MCE_SELF_VECTOR \
981 mce_self_interrupt smp_mce_self_interrupt
985 apicinterrupt CALL_FUNCTION_SINGLE_VECTOR \
986 call_function_single_interrupt smp_call_function_single_interrupt
987 apicinterrupt CALL_FUNCTION_VECTOR \
988 call_function_interrupt smp_call_function_interrupt
989 apicinterrupt RESCHEDULE_VECTOR \
990 reschedule_interrupt smp_reschedule_interrupt
993 apicinterrupt ERROR_APIC_VECTOR \
994 error_interrupt smp_error_interrupt
995 apicinterrupt SPURIOUS_APIC_VECTOR \
996 spurious_interrupt smp_spurious_interrupt
998 #ifdef CONFIG_IRQ_WORK
999 apicinterrupt IRQ_WORK_VECTOR \
1000 irq_work_interrupt smp_irq_work_interrupt
1004 * Exception entry points.
1006 .macro zeroentry sym do_sym
1009 PARAVIRT_ADJUST_EXCEPTION_FRAME
1010 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1011 subq $ORIG_RAX-R15, %rsp
1012 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1015 movq %rsp,%rdi /* pt_regs pointer */
1016 xorl %esi,%esi /* no error code */
1018 jmp error_exit /* %ebx: no swapgs flag */
1023 .macro paranoidzeroentry sym do_sym
1026 PARAVIRT_ADJUST_EXCEPTION_FRAME
1027 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1028 subq $ORIG_RAX-R15, %rsp
1029 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1032 movq %rsp,%rdi /* pt_regs pointer */
1033 xorl %esi,%esi /* no error code */
1035 jmp paranoid_exit /* %ebx: no swapgs flag */
1040 #define INIT_TSS_IST(x) PER_CPU_VAR(init_tss) + (TSS_ist + ((x) - 1) * 8)
1041 .macro paranoidzeroentry_ist sym do_sym ist
1044 PARAVIRT_ADJUST_EXCEPTION_FRAME
1045 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1046 subq $ORIG_RAX-R15, %rsp
1047 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1050 movq %rsp,%rdi /* pt_regs pointer */
1051 xorl %esi,%esi /* no error code */
1052 subq $EXCEPTION_STKSZ, INIT_TSS_IST(\ist)
1054 addq $EXCEPTION_STKSZ, INIT_TSS_IST(\ist)
1055 jmp paranoid_exit /* %ebx: no swapgs flag */
1060 .macro errorentry sym do_sym
1063 PARAVIRT_ADJUST_EXCEPTION_FRAME
1064 subq $ORIG_RAX-R15, %rsp
1065 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1068 movq %rsp,%rdi /* pt_regs pointer */
1069 movq ORIG_RAX(%rsp),%rsi /* get error code */
1070 movq $-1,ORIG_RAX(%rsp) /* no syscall to restart */
1072 jmp error_exit /* %ebx: no swapgs flag */
1077 /* error code is on the stack already */
1078 .macro paranoiderrorentry sym do_sym
1081 PARAVIRT_ADJUST_EXCEPTION_FRAME
1082 subq $ORIG_RAX-R15, %rsp
1083 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1087 movq %rsp,%rdi /* pt_regs pointer */
1088 movq ORIG_RAX(%rsp),%rsi /* get error code */
1089 movq $-1,ORIG_RAX(%rsp) /* no syscall to restart */
1091 jmp paranoid_exit /* %ebx: no swapgs flag */
1096 zeroentry divide_error do_divide_error
1097 zeroentry overflow do_overflow
1098 zeroentry bounds do_bounds
1099 zeroentry invalid_op do_invalid_op
1100 zeroentry device_not_available do_device_not_available
1101 paranoiderrorentry double_fault do_double_fault
1102 zeroentry coprocessor_segment_overrun do_coprocessor_segment_overrun
1103 errorentry invalid_TSS do_invalid_TSS
1104 errorentry segment_not_present do_segment_not_present
1105 zeroentry spurious_interrupt_bug do_spurious_interrupt_bug
1106 zeroentry coprocessor_error do_coprocessor_error
1107 errorentry alignment_check do_alignment_check
1108 zeroentry simd_coprocessor_error do_simd_coprocessor_error
1110 /* Reload gs selector with exception handling */
1111 /* edi: new selector */
1112 ENTRY(native_load_gs_index)
1115 DISABLE_INTERRUPTS(CLBR_ANY & ~CLBR_RDI)
1119 2: mfence /* workaround */
1124 END(native_load_gs_index)
1126 .section __ex_table,"a"
1128 .quad gs_change,bad_gs
1130 .section .fixup,"ax"
1131 /* running with kernelgs */
1133 SWAPGS /* switch back to user gs */
1139 ENTRY(kernel_thread_helper)
1140 pushq $0 # fake return address
1143 * Here we are in the child and the registers are set as they were
1144 * at kernel_thread() invocation in the parent.
1150 ud2 # padding for call trace
1152 END(kernel_thread_helper)
1155 * execve(). This function needs to use IRET, not SYSRET, to set up all state properly.
1157 * C extern interface:
1158 * extern long execve(const char *name, char **argv, char **envp)
1160 * asm input arguments:
1161 * rdi: name, rsi: argv, rdx: envp
1163 * We want to fallback into:
1164 * extern long sys_execve(const char *name, char **argv,char **envp, struct pt_regs *regs)
1166 * do_sys_execve asm fallback arguments:
1167 * rdi: name, rsi: argv, rdx: envp, rcx: fake frame on the stack
1169 ENTRY(kernel_execve)
1175 movq %rax, RAX(%rsp)
1178 je int_ret_from_sys_call
1185 /* Call softirq on interrupt stack. Interrupts are off. */
1189 CFI_REL_OFFSET rbp,0
1191 CFI_DEF_CFA_REGISTER rbp
1192 incl PER_CPU_VAR(irq_count)
1193 cmove PER_CPU_VAR(irq_stack_ptr),%rsp
1194 push %rbp # backlink for old unwinder
1198 CFI_DEF_CFA_REGISTER rsp
1199 CFI_ADJUST_CFA_OFFSET -8
1200 decl PER_CPU_VAR(irq_count)
1206 zeroentry xen_hypervisor_callback xen_do_hypervisor_callback
1209 * A note on the "critical region" in our callback handler.
1210 * We want to avoid stacking callback handlers due to events occurring
1211 * during handling of the last event. To do this, we keep events disabled
1212 * until we've done all processing. HOWEVER, we must enable events before
1213 * popping the stack frame (can't be done atomically) and so it would still
1214 * be possible to get enough handler activations to overflow the stack.
1215 * Although unlikely, bugs of that kind are hard to track down, so we'd
1216 * like to avoid the possibility.
1217 * So, on entry to the handler we detect whether we interrupted an
1218 * existing activation in its critical region -- if so, we pop the current
1219 * activation and restart the handler using the previous one.
1221 ENTRY(xen_do_hypervisor_callback) # do_hypervisor_callback(struct *pt_regs)
1224 * Since we don't modify %rdi, evtchn_do_upall(struct *pt_regs) will
1225 * see the correct pointer to the pt_regs
1227 movq %rdi, %rsp # we don't return, adjust the stack frame
1230 11: incl PER_CPU_VAR(irq_count)
1232 CFI_DEF_CFA_REGISTER rbp
1233 cmovzq PER_CPU_VAR(irq_stack_ptr),%rsp
1234 pushq %rbp # backlink for old unwinder
1235 call xen_evtchn_do_upcall
1237 CFI_DEF_CFA_REGISTER rsp
1238 decl PER_CPU_VAR(irq_count)
1241 END(do_hypervisor_callback)
1244 * Hypervisor uses this for application faults while it executes.
1245 * We get here for two reasons:
1246 * 1. Fault while reloading DS, ES, FS or GS
1247 * 2. Fault while executing IRET
1248 * Category 1 we do not need to fix up as Xen has already reloaded all segment
1249 * registers that could be reloaded and zeroed the others.
1250 * Category 2 we fix up by killing the current process. We cannot use the
1251 * normal Linux return path in this case because if we use the IRET hypercall
1252 * to pop the stack frame we end up in an infinite loop of failsafe callbacks.
1253 * We distinguish between categories by comparing each saved segment register
1254 * with its current contents: any discrepancy means we in category 1.
1256 ENTRY(xen_failsafe_callback)
1258 /*CFI_REL_OFFSET gs,GS*/
1259 /*CFI_REL_OFFSET fs,FS*/
1260 /*CFI_REL_OFFSET es,ES*/
1261 /*CFI_REL_OFFSET ds,DS*/
1262 CFI_REL_OFFSET r11,8
1263 CFI_REL_OFFSET rcx,0
1277 /* All segments match their saved values => Category 2 (Bad IRET). */
1283 CFI_ADJUST_CFA_OFFSET -0x30
1284 pushq_cfi $0 /* RIP */
1287 jmp general_protection
1289 1: /* Segment mismatch => Category 1 (Bad segment). Retry the IRET. */
1295 CFI_ADJUST_CFA_OFFSET -0x30
1300 END(xen_failsafe_callback)
1302 apicinterrupt XEN_HVM_EVTCHN_CALLBACK \
1303 xen_hvm_callback_vector xen_evtchn_do_upcall
1305 #endif /* CONFIG_XEN */
1308 * Some functions should be protected against kprobes
1310 .pushsection .kprobes.text, "ax"
1312 paranoidzeroentry_ist debug do_debug DEBUG_STACK
1313 paranoidzeroentry_ist int3 do_int3 DEBUG_STACK
1314 paranoiderrorentry stack_segment do_stack_segment
1316 zeroentry xen_debug do_debug
1317 zeroentry xen_int3 do_int3
1318 errorentry xen_stack_segment do_stack_segment
1320 errorentry general_protection do_general_protection
1321 errorentry page_fault do_page_fault
1322 #ifdef CONFIG_X86_MCE
1323 paranoidzeroentry machine_check *machine_check_vector(%rip)
1327 * "Paranoid" exit path from exception stack.
1328 * Paranoid because this is used by NMIs and cannot take
1329 * any kernel state for granted.
1330 * We don't do kernel preemption checks here, because only
1331 * NMI should be common and it does not enable IRQs and
1332 * cannot get reschedule ticks.
1334 * "trace" is 0 for the NMI handler only, because irq-tracing
1335 * is fundamentally NMI-unsafe. (we cannot change the soft and
1336 * hard flags at once, atomically)
1339 /* ebx: no swapgs flag */
1340 ENTRY(paranoid_exit)
1342 DISABLE_INTERRUPTS(CLBR_NONE)
1344 testl %ebx,%ebx /* swapgs needed? */
1345 jnz paranoid_restore
1347 jnz paranoid_userspace
1358 GET_THREAD_INFO(%rcx)
1359 movl TI_flags(%rcx),%ebx
1360 andl $_TIF_WORK_MASK,%ebx
1362 movq %rsp,%rdi /* &pt_regs */
1364 movq %rax,%rsp /* switch stack for scheduling */
1365 testl $_TIF_NEED_RESCHED,%ebx
1366 jnz paranoid_schedule
1367 movl %ebx,%edx /* arg3: thread flags */
1369 ENABLE_INTERRUPTS(CLBR_NONE)
1370 xorl %esi,%esi /* arg2: oldset */
1371 movq %rsp,%rdi /* arg1: &pt_regs */
1372 call do_notify_resume
1373 DISABLE_INTERRUPTS(CLBR_NONE)
1375 jmp paranoid_userspace
1378 ENABLE_INTERRUPTS(CLBR_ANY)
1380 DISABLE_INTERRUPTS(CLBR_ANY)
1382 jmp paranoid_userspace
1387 * Exception entry point. This expects an error code/orig_rax on the stack.
1388 * returns in "no swapgs flag" in %ebx.
1392 CFI_ADJUST_CFA_OFFSET 15*8
1393 /* oldrax contains error code */
1412 je error_kernelspace
1420 * There are two places in the kernel that can potentially fault with
1421 * usergs. Handle them here. The exception handlers after iret run with
1422 * kernel gs again, so don't set the user space flag. B stepping K8s
1423 * sometimes report an truncated RIP for IRET exceptions returning to
1424 * compat mode. Check for these here too.
1428 leaq irq_return(%rip),%rcx
1429 cmpq %rcx,RIP+8(%rsp)
1431 movl %ecx,%eax /* zero extend */
1432 cmpq %rax,RIP+8(%rsp)
1434 cmpq $gs_change,RIP+8(%rsp)
1439 /* Fix truncated RIP */
1440 movq %rcx,RIP+8(%rsp)
1446 /* ebx: no swapgs flag (1: don't need swapgs, 0: need it) */
1451 DISABLE_INTERRUPTS(CLBR_NONE)
1453 GET_THREAD_INFO(%rcx)
1456 LOCKDEP_SYS_EXIT_IRQ
1457 movl TI_flags(%rcx),%edx
1458 movl $_TIF_WORK_MASK,%edi
1466 /* runs on exception stack */
1469 PARAVIRT_ADJUST_EXCEPTION_FRAME
1471 subq $ORIG_RAX-R15, %rsp
1472 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1475 /* paranoidentry do_nmi, 0; without TRACE_IRQS_OFF */
1479 #ifdef CONFIG_TRACE_IRQFLAGS
1480 /* paranoidexit; without TRACE_IRQS_OFF */
1481 /* ebx: no swapgs flag */
1482 DISABLE_INTERRUPTS(CLBR_NONE)
1483 testl %ebx,%ebx /* swapgs needed? */
1493 GET_THREAD_INFO(%rcx)
1494 movl TI_flags(%rcx),%ebx
1495 andl $_TIF_WORK_MASK,%ebx
1497 movq %rsp,%rdi /* &pt_regs */
1499 movq %rax,%rsp /* switch stack for scheduling */
1500 testl $_TIF_NEED_RESCHED,%ebx
1502 movl %ebx,%edx /* arg3: thread flags */
1503 ENABLE_INTERRUPTS(CLBR_NONE)
1504 xorl %esi,%esi /* arg2: oldset */
1505 movq %rsp,%rdi /* arg1: &pt_regs */
1506 call do_notify_resume
1507 DISABLE_INTERRUPTS(CLBR_NONE)
1510 ENABLE_INTERRUPTS(CLBR_ANY)
1512 DISABLE_INTERRUPTS(CLBR_ANY)
1521 ENTRY(ignore_sysret)
1529 * End of kprobes section