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 */
301 movq_cfi rdi, RDI+16-ARGOFFSET
302 movq_cfi rsi, RSI+16-ARGOFFSET
303 movq_cfi rdx, RDX+16-ARGOFFSET
304 movq_cfi rcx, RCX+16-ARGOFFSET
305 movq_cfi rax, RAX+16-ARGOFFSET
306 movq_cfi r8, R8+16-ARGOFFSET
307 movq_cfi r9, R9+16-ARGOFFSET
308 movq_cfi r10, R10+16-ARGOFFSET
309 movq_cfi r11, R11+16-ARGOFFSET
311 leaq -ARGOFFSET+16(%rsp),%rdi /* arg1 for handler */
312 movq_cfi rbp, 8 /* push %rbp */
313 leaq 8(%rsp), %rbp /* mov %rsp, %ebp */
318 * irq_count is used to check if a CPU is already on an interrupt stack
319 * or not. While this is essentially redundant with preempt_count it is
320 * a little cheaper to use a separate counter in the PDA (short of
321 * moving irq_enter into assembly, which would be too much work)
323 1: incl PER_CPU_VAR(irq_count)
325 popq_cfi %rax /* move return address... */
326 mov PER_CPU_VAR(irq_stack_ptr),%rsp
328 pushq_cfi %rbp /* backlink for unwinder */
329 pushq_cfi %rax /* ... to the new stack */
331 * We entered an interrupt context - irqs are off:
339 PARTIAL_FRAME 1 REST_SKIP+8
340 movq 5*8+16(%rsp), %r11 /* save return address */
347 movq %r11, 8(%rsp) /* return address */
348 FIXUP_TOP_OF_STACK %r11, 16
353 /* save complete stack frame */
354 .pushsection .kprobes.text, "ax"
374 movl $MSR_GS_BASE,%ecx
377 js 1f /* negative -> in kernel */
386 * A newly forked process directly context switches into this address.
388 * rdi: prev task we switched from
393 LOCK ; btr $TIF_FORK,TI_flags(%r8)
395 pushq_cfi kernel_eflags(%rip)
396 popfq_cfi # reset kernel eflags
398 call schedule_tail # rdi: 'prev' task parameter
400 GET_THREAD_INFO(%rcx)
404 testl $3, CS-ARGOFFSET(%rsp) # from kernel_thread?
405 je int_ret_from_sys_call
407 testl $_TIF_IA32, TI_flags(%rcx) # 32-bit compat task needs IRET
408 jnz int_ret_from_sys_call
410 RESTORE_TOP_OF_STACK %rdi, -ARGOFFSET
411 jmp ret_from_sys_call # go to the SYSRET fastpath
417 * System call entry. Upto 6 arguments in registers are supported.
419 * SYSCALL does not save anything on the stack and does not change the
425 * rax system call number
427 * rcx return address for syscall/sysret, C arg3
430 * r10 arg3 (--> moved to rcx for C)
433 * r11 eflags for syscall/sysret, temporary for C
434 * r12-r15,rbp,rbx saved by C code, not touched.
436 * Interrupts are off on entry.
437 * Only called from user space.
439 * XXX if we had a free scratch register we could save the RSP into the stack frame
440 * and report it properly in ps. Unfortunately we haven't.
442 * When user can change the frames always force IRET. That is because
443 * it deals with uncanonical addresses better. SYSRET has trouble
444 * with them due to bugs in both AMD and Intel CPUs.
450 CFI_DEF_CFA rsp,KERNEL_STACK_OFFSET
452 /*CFI_REGISTER rflags,r11*/
455 * A hypervisor implementation might want to use a label
456 * after the swapgs, so that it can do the swapgs
457 * for the guest and jump here on syscall.
459 ENTRY(system_call_after_swapgs)
461 movq %rsp,PER_CPU_VAR(old_rsp)
462 movq PER_CPU_VAR(kernel_stack),%rsp
464 * No need to follow this irqs off/on section - it's straight
467 ENABLE_INTERRUPTS(CLBR_NONE)
469 movq %rax,ORIG_RAX-ARGOFFSET(%rsp)
470 movq %rcx,RIP-ARGOFFSET(%rsp)
471 CFI_REL_OFFSET rip,RIP-ARGOFFSET
472 GET_THREAD_INFO(%rcx)
473 testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%rcx)
475 system_call_fastpath:
476 cmpq $__NR_syscall_max,%rax
479 call *sys_call_table(,%rax,8) # XXX: rip relative
480 movq %rax,RAX-ARGOFFSET(%rsp)
482 * Syscall return path ending with SYSRET (fast path)
483 * Has incomplete stack frame and undefined top of stack.
486 movl $_TIF_ALLWORK_MASK,%edi
490 GET_THREAD_INFO(%rcx)
491 DISABLE_INTERRUPTS(CLBR_NONE)
493 movl TI_flags(%rcx),%edx
498 * sysretq will re-enable interrupts:
501 movq RIP-ARGOFFSET(%rsp),%rcx
503 RESTORE_ARGS 0,-ARG_SKIP,1
504 /*CFI_REGISTER rflags,r11*/
505 movq PER_CPU_VAR(old_rsp), %rsp
509 /* Handle reschedules */
510 /* edx: work, edi: workmask */
512 bt $TIF_NEED_RESCHED,%edx
515 ENABLE_INTERRUPTS(CLBR_NONE)
521 /* Handle a signal */
524 ENABLE_INTERRUPTS(CLBR_NONE)
525 #ifdef CONFIG_AUDITSYSCALL
526 bt $TIF_SYSCALL_AUDIT,%edx
530 * We have a signal, or exit tracing or single-step.
531 * These all wind up with the iret return path anyway,
532 * so just join that path right now.
534 FIXUP_TOP_OF_STACK %r11, -ARGOFFSET
535 jmp int_check_syscall_exit_work
538 movq $-ENOSYS,RAX-ARGOFFSET(%rsp)
539 jmp ret_from_sys_call
541 #ifdef CONFIG_AUDITSYSCALL
543 * Fast path for syscall audit without full syscall trace.
544 * We just call audit_syscall_entry() directly, and then
545 * jump back to the normal fast path.
548 movq %r10,%r9 /* 6th arg: 4th syscall arg */
549 movq %rdx,%r8 /* 5th arg: 3rd syscall arg */
550 movq %rsi,%rcx /* 4th arg: 2nd syscall arg */
551 movq %rdi,%rdx /* 3rd arg: 1st syscall arg */
552 movq %rax,%rsi /* 2nd arg: syscall number */
553 movl $AUDIT_ARCH_X86_64,%edi /* 1st arg: audit arch */
554 call audit_syscall_entry
555 LOAD_ARGS 0 /* reload call-clobbered registers */
556 jmp system_call_fastpath
559 * Return fast path for syscall audit. Call audit_syscall_exit()
560 * directly and then jump back to the fast path with TIF_SYSCALL_AUDIT
564 movq RAX-ARGOFFSET(%rsp),%rsi /* second arg, syscall return value */
565 cmpq $0,%rsi /* is it < 0? */
566 setl %al /* 1 if so, 0 if not */
567 movzbl %al,%edi /* zero-extend that into %edi */
568 inc %edi /* first arg, 0->1(AUDITSC_SUCCESS), 1->2(AUDITSC_FAILURE) */
569 call audit_syscall_exit
570 movl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT),%edi
572 #endif /* CONFIG_AUDITSYSCALL */
574 /* Do syscall tracing */
576 #ifdef CONFIG_AUDITSYSCALL
577 testl $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags(%rcx)
581 movq $-ENOSYS,RAX(%rsp) /* ptrace can change this for a bad syscall */
582 FIXUP_TOP_OF_STACK %rdi
584 call syscall_trace_enter
586 * Reload arg registers from stack in case ptrace changed them.
587 * We don't reload %rax because syscall_trace_enter() returned
588 * the value it wants us to use in the table lookup.
590 LOAD_ARGS ARGOFFSET, 1
592 cmpq $__NR_syscall_max,%rax
593 ja int_ret_from_sys_call /* RAX(%rsp) set to -ENOSYS above */
594 movq %r10,%rcx /* fixup for C */
595 call *sys_call_table(,%rax,8)
596 movq %rax,RAX-ARGOFFSET(%rsp)
597 /* Use IRET because user could have changed frame */
600 * Syscall return path ending with IRET.
601 * Has correct top of stack, but partial stack frame.
603 GLOBAL(int_ret_from_sys_call)
604 DISABLE_INTERRUPTS(CLBR_NONE)
606 testl $3,CS-ARGOFFSET(%rsp)
607 je retint_restore_args
608 movl $_TIF_ALLWORK_MASK,%edi
609 /* edi: mask to check */
610 GLOBAL(int_with_check)
612 GET_THREAD_INFO(%rcx)
613 movl TI_flags(%rcx),%edx
616 andl $~TS_COMPAT,TI_status(%rcx)
619 /* Either reschedule or signal or syscall exit tracking needed. */
620 /* First do a reschedule test. */
621 /* edx: work, edi: workmask */
623 bt $TIF_NEED_RESCHED,%edx
626 ENABLE_INTERRUPTS(CLBR_NONE)
630 DISABLE_INTERRUPTS(CLBR_NONE)
634 /* handle signals and tracing -- both require a full stack frame */
637 ENABLE_INTERRUPTS(CLBR_NONE)
638 int_check_syscall_exit_work:
640 /* Check for syscall exit trace */
641 testl $_TIF_WORK_SYSCALL_EXIT,%edx
644 leaq 8(%rsp),%rdi # &ptregs -> arg1
645 call syscall_trace_leave
647 andl $~(_TIF_WORK_SYSCALL_EXIT|_TIF_SYSCALL_EMU),%edi
651 testl $_TIF_DO_NOTIFY_MASK,%edx
653 movq %rsp,%rdi # &ptregs -> arg1
654 xorl %esi,%esi # oldset -> arg2
655 call do_notify_resume
656 1: movl $_TIF_WORK_MASK,%edi
659 DISABLE_INTERRUPTS(CLBR_NONE)
666 * Certain special system calls that need to save a complete full stack frame.
668 .macro PTREGSCALL label,func,arg
670 PARTIAL_FRAME 1 8 /* offset 8: return address */
671 subq $REST_SKIP, %rsp
672 CFI_ADJUST_CFA_OFFSET REST_SKIP
674 DEFAULT_FRAME 0 8 /* offset 8: return address */
675 leaq 8(%rsp), \arg /* pt_regs pointer */
677 jmp ptregscall_common
682 PTREGSCALL stub_clone, sys_clone, %r8
683 PTREGSCALL stub_fork, sys_fork, %rdi
684 PTREGSCALL stub_vfork, sys_vfork, %rdi
685 PTREGSCALL stub_sigaltstack, sys_sigaltstack, %rdx
686 PTREGSCALL stub_iopl, sys_iopl, %rsi
688 ENTRY(ptregscall_common)
689 DEFAULT_FRAME 1 8 /* offset 8: return address */
690 RESTORE_TOP_OF_STACK %r11, 8
691 movq_cfi_restore R15+8, r15
692 movq_cfi_restore R14+8, r14
693 movq_cfi_restore R13+8, r13
694 movq_cfi_restore R12+8, r12
695 movq_cfi_restore RBP+8, rbp
696 movq_cfi_restore RBX+8, rbx
697 ret $REST_SKIP /* pop extended registers */
699 END(ptregscall_common)
706 FIXUP_TOP_OF_STACK %r11
709 RESTORE_TOP_OF_STACK %r11
712 jmp int_ret_from_sys_call
717 * sigreturn is special because it needs to restore all registers on return.
718 * This cannot be done with SYSRET, so use the IRET return path instead.
720 ENTRY(stub_rt_sigreturn)
726 FIXUP_TOP_OF_STACK %r11
727 call sys_rt_sigreturn
728 movq %rax,RAX(%rsp) # fixme, this could be done at the higher layer
730 jmp int_ret_from_sys_call
732 END(stub_rt_sigreturn)
735 * Build the entry stubs and pointer table with some assembler magic.
736 * We pack 7 stubs into a single 32-byte chunk, which will fit in a
737 * single cache line on all modern x86 implementations.
739 .section .init.rodata,"a"
743 .p2align CONFIG_X86_L1_CACHE_SHIFT
744 ENTRY(irq_entries_start)
746 vector=FIRST_EXTERNAL_VECTOR
747 .rept (NR_VECTORS-FIRST_EXTERNAL_VECTOR+6)/7
750 .if vector < NR_VECTORS
751 .if vector <> FIRST_EXTERNAL_VECTOR
752 CFI_ADJUST_CFA_OFFSET -8
754 1: pushq_cfi $(~vector+0x80) /* Note: always in signed byte range */
755 .if ((vector-FIRST_EXTERNAL_VECTOR)%7) <> 6
764 2: jmp common_interrupt
767 END(irq_entries_start)
774 * Interrupt entry/exit.
776 * Interrupt entry points save only callee clobbered registers in fast path.
778 * Entry runs with interrupts off.
781 /* 0(%rsp): ~(interrupt number) */
782 .macro interrupt func
783 subq $ORIG_RAX-ARGOFFSET+8, %rsp
784 CFI_ADJUST_CFA_OFFSET ORIG_RAX-ARGOFFSET+8
791 * Interrupt entry/exit should be protected against kprobes
793 .pushsection .kprobes.text, "ax"
795 * The interrupt stubs push (~vector+0x80) onto the stack and
796 * then jump to common_interrupt.
798 .p2align CONFIG_X86_L1_CACHE_SHIFT
801 addq $-0x80,(%rsp) /* Adjust vector to [-256,-1] range */
803 /* 0(%rsp): old_rsp-ARGOFFSET */
805 DISABLE_INTERRUPTS(CLBR_NONE)
807 decl PER_CPU_VAR(irq_count)
810 CFI_DEF_CFA_REGISTER rsp
811 CFI_ADJUST_CFA_OFFSET -8
813 GET_THREAD_INFO(%rcx)
814 testl $3,CS-ARGOFFSET(%rsp)
817 /* Interrupt came from user space */
819 * Has a correct top of stack, but a partial stack frame
820 * %rcx: thread info. Interrupts off.
822 retint_with_reschedule:
823 movl $_TIF_WORK_MASK,%edi
826 movl TI_flags(%rcx),%edx
831 retint_swapgs: /* return to user-space */
833 * The iretq could re-enable interrupts:
835 DISABLE_INTERRUPTS(CLBR_ANY)
840 retint_restore_args: /* return to kernel space */
841 DISABLE_INTERRUPTS(CLBR_ANY)
843 * The iretq could re-enable interrupts:
852 .section __ex_table, "a"
853 .quad irq_return, bad_iret
856 #ifdef CONFIG_PARAVIRT
860 .section __ex_table,"a"
861 .quad native_iret, bad_iret
868 * The iret traps when the %cs or %ss being restored is bogus.
869 * We've lost the original trap vector and error code.
870 * #GPF is the most likely one to get for an invalid selector.
871 * So pretend we completed the iret and took the #GPF in user mode.
873 * We are now running with the kernel GS after exception recovery.
874 * But error_entry expects us to have user GS to match the user %cs,
880 jmp general_protection
884 /* edi: workmask, edx: work */
887 bt $TIF_NEED_RESCHED,%edx
890 ENABLE_INTERRUPTS(CLBR_NONE)
894 GET_THREAD_INFO(%rcx)
895 DISABLE_INTERRUPTS(CLBR_NONE)
900 testl $_TIF_DO_NOTIFY_MASK,%edx
903 ENABLE_INTERRUPTS(CLBR_NONE)
905 movq $-1,ORIG_RAX(%rsp)
906 xorl %esi,%esi # oldset
907 movq %rsp,%rdi # &pt_regs
908 call do_notify_resume
910 DISABLE_INTERRUPTS(CLBR_NONE)
912 GET_THREAD_INFO(%rcx)
913 jmp retint_with_reschedule
915 #ifdef CONFIG_PREEMPT
916 /* Returning to kernel space. Check if we need preemption */
917 /* rcx: threadinfo. interrupts off. */
919 cmpl $0,TI_preempt_count(%rcx)
920 jnz retint_restore_args
921 bt $TIF_NEED_RESCHED,TI_flags(%rcx)
922 jnc retint_restore_args
923 bt $9,EFLAGS-ARGOFFSET(%rsp) /* interrupts off? */
924 jnc retint_restore_args
925 call preempt_schedule_irq
930 END(common_interrupt)
932 * End of kprobes section
939 .macro apicinterrupt num sym do_sym
950 apicinterrupt IRQ_MOVE_CLEANUP_VECTOR \
951 irq_move_cleanup_interrupt smp_irq_move_cleanup_interrupt
952 apicinterrupt REBOOT_VECTOR \
953 reboot_interrupt smp_reboot_interrupt
957 apicinterrupt UV_BAU_MESSAGE \
958 uv_bau_message_intr1 uv_bau_message_interrupt
960 apicinterrupt LOCAL_TIMER_VECTOR \
961 apic_timer_interrupt smp_apic_timer_interrupt
962 apicinterrupt X86_PLATFORM_IPI_VECTOR \
963 x86_platform_ipi smp_x86_platform_ipi
966 apicinterrupt INVALIDATE_TLB_VECTOR_START+0 \
967 invalidate_interrupt0 smp_invalidate_interrupt
968 apicinterrupt INVALIDATE_TLB_VECTOR_START+1 \
969 invalidate_interrupt1 smp_invalidate_interrupt
970 apicinterrupt INVALIDATE_TLB_VECTOR_START+2 \
971 invalidate_interrupt2 smp_invalidate_interrupt
972 apicinterrupt INVALIDATE_TLB_VECTOR_START+3 \
973 invalidate_interrupt3 smp_invalidate_interrupt
974 apicinterrupt INVALIDATE_TLB_VECTOR_START+4 \
975 invalidate_interrupt4 smp_invalidate_interrupt
976 apicinterrupt INVALIDATE_TLB_VECTOR_START+5 \
977 invalidate_interrupt5 smp_invalidate_interrupt
978 apicinterrupt INVALIDATE_TLB_VECTOR_START+6 \
979 invalidate_interrupt6 smp_invalidate_interrupt
980 apicinterrupt INVALIDATE_TLB_VECTOR_START+7 \
981 invalidate_interrupt7 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_X86_MCE
1333 paranoidzeroentry machine_check *machine_check_vector(%rip)
1337 * "Paranoid" exit path from exception stack.
1338 * Paranoid because this is used by NMIs and cannot take
1339 * any kernel state for granted.
1340 * We don't do kernel preemption checks here, because only
1341 * NMI should be common and it does not enable IRQs and
1342 * cannot get reschedule ticks.
1344 * "trace" is 0 for the NMI handler only, because irq-tracing
1345 * is fundamentally NMI-unsafe. (we cannot change the soft and
1346 * hard flags at once, atomically)
1349 /* ebx: no swapgs flag */
1350 ENTRY(paranoid_exit)
1352 DISABLE_INTERRUPTS(CLBR_NONE)
1354 testl %ebx,%ebx /* swapgs needed? */
1355 jnz paranoid_restore
1357 jnz paranoid_userspace
1368 GET_THREAD_INFO(%rcx)
1369 movl TI_flags(%rcx),%ebx
1370 andl $_TIF_WORK_MASK,%ebx
1372 movq %rsp,%rdi /* &pt_regs */
1374 movq %rax,%rsp /* switch stack for scheduling */
1375 testl $_TIF_NEED_RESCHED,%ebx
1376 jnz paranoid_schedule
1377 movl %ebx,%edx /* arg3: thread flags */
1379 ENABLE_INTERRUPTS(CLBR_NONE)
1380 xorl %esi,%esi /* arg2: oldset */
1381 movq %rsp,%rdi /* arg1: &pt_regs */
1382 call do_notify_resume
1383 DISABLE_INTERRUPTS(CLBR_NONE)
1385 jmp paranoid_userspace
1388 ENABLE_INTERRUPTS(CLBR_ANY)
1390 DISABLE_INTERRUPTS(CLBR_ANY)
1392 jmp paranoid_userspace
1397 * Exception entry point. This expects an error code/orig_rax on the stack.
1398 * returns in "no swapgs flag" in %ebx.
1402 CFI_ADJUST_CFA_OFFSET 15*8
1403 /* oldrax contains error code */
1422 je error_kernelspace
1430 * There are two places in the kernel that can potentially fault with
1431 * usergs. Handle them here. The exception handlers after iret run with
1432 * kernel gs again, so don't set the user space flag. B stepping K8s
1433 * sometimes report an truncated RIP for IRET exceptions returning to
1434 * compat mode. Check for these here too.
1438 leaq irq_return(%rip),%rcx
1439 cmpq %rcx,RIP+8(%rsp)
1441 movl %ecx,%eax /* zero extend */
1442 cmpq %rax,RIP+8(%rsp)
1444 cmpq $gs_change,RIP+8(%rsp)
1449 /* Fix truncated RIP */
1450 movq %rcx,RIP+8(%rsp)
1456 /* ebx: no swapgs flag (1: don't need swapgs, 0: need it) */
1461 DISABLE_INTERRUPTS(CLBR_NONE)
1463 GET_THREAD_INFO(%rcx)
1466 LOCKDEP_SYS_EXIT_IRQ
1467 movl TI_flags(%rcx),%edx
1468 movl $_TIF_WORK_MASK,%edi
1476 /* runs on exception stack */
1479 PARAVIRT_ADJUST_EXCEPTION_FRAME
1481 subq $ORIG_RAX-R15, %rsp
1482 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1485 /* paranoidentry do_nmi, 0; without TRACE_IRQS_OFF */
1489 #ifdef CONFIG_TRACE_IRQFLAGS
1490 /* paranoidexit; without TRACE_IRQS_OFF */
1491 /* ebx: no swapgs flag */
1492 DISABLE_INTERRUPTS(CLBR_NONE)
1493 testl %ebx,%ebx /* swapgs needed? */
1503 GET_THREAD_INFO(%rcx)
1504 movl TI_flags(%rcx),%ebx
1505 andl $_TIF_WORK_MASK,%ebx
1507 movq %rsp,%rdi /* &pt_regs */
1509 movq %rax,%rsp /* switch stack for scheduling */
1510 testl $_TIF_NEED_RESCHED,%ebx
1512 movl %ebx,%edx /* arg3: thread flags */
1513 ENABLE_INTERRUPTS(CLBR_NONE)
1514 xorl %esi,%esi /* arg2: oldset */
1515 movq %rsp,%rdi /* arg1: &pt_regs */
1516 call do_notify_resume
1517 DISABLE_INTERRUPTS(CLBR_NONE)
1520 ENABLE_INTERRUPTS(CLBR_ANY)
1522 DISABLE_INTERRUPTS(CLBR_ANY)
1531 ENTRY(ignore_sysret)
1539 * End of kprobes section