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 * Some of this is documented in Documentation/x86/entry_64.txt
14 * NOTE: This code handles signal-recognition, which happens every time
15 * after an interrupt and after each system call.
17 * Normal syscalls and interrupts don't save a full stack frame, this is
18 * only done for syscall tracing, signals or fork/exec et.al.
20 * A note on terminology:
21 * - top of stack: Architecture defined interrupt frame from SS to RIP
22 * at the top of the kernel process stack.
23 * - partial stack frame: partially saved registers up to R11.
24 * - full stack frame: Like partial stack frame, but all register saved.
27 * - CFI macros are used to generate dwarf2 unwind information for better
28 * backtraces. They don't change any code.
29 * - SAVE_ALL/RESTORE_ALL - Save/restore all registers
30 * - SAVE_ARGS/RESTORE_ARGS - Save/restore registers that C functions modify.
31 * There are unfortunately lots of special cases where some registers
32 * not touched. The macro is a big mess that should be cleaned up.
33 * - SAVE_REST/RESTORE_REST - Handle the registers not saved by SAVE_ARGS.
34 * Gives a full stack frame.
35 * - ENTRY/END Define functions in the symbol table.
36 * - FIXUP_TOP_OF_STACK/RESTORE_TOP_OF_STACK - Fix up the hardware stack
37 * frame that is otherwise undefined after a SYSCALL
38 * - TRACE_IRQ_* - Trace hard interrupt state for lock debugging.
39 * - errorentry/paranoidentry/zeroentry - Define exception entry points.
42 #include <linux/linkage.h>
43 #include <asm/segment.h>
44 #include <asm/cache.h>
45 #include <asm/errno.h>
46 #include <asm/dwarf2.h>
47 #include <asm/calling.h>
48 #include <asm/asm-offsets.h>
50 #include <asm/unistd.h>
51 #include <asm/thread_info.h>
52 #include <asm/hw_irq.h>
53 #include <asm/page_types.h>
54 #include <asm/irqflags.h>
55 #include <asm/paravirt.h>
56 #include <asm/ftrace.h>
57 #include <asm/percpu.h>
58 #include <linux/err.h>
60 /* Avoid __ASSEMBLER__'ifying <linux/audit.h> just for this. */
61 #include <linux/elf-em.h>
62 #define AUDIT_ARCH_X86_64 (EM_X86_64|__AUDIT_ARCH_64BIT|__AUDIT_ARCH_LE)
63 #define __AUDIT_ARCH_64BIT 0x80000000
64 #define __AUDIT_ARCH_LE 0x40000000
67 .section .entry.text, "ax"
69 #ifdef CONFIG_FUNCTION_TRACER
70 #ifdef CONFIG_DYNAMIC_FTRACE
76 cmpl $0, function_trace_stop
83 subq $MCOUNT_INSN_SIZE, %rdi
90 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
91 GLOBAL(ftrace_graph_call)
99 #else /* ! CONFIG_DYNAMIC_FTRACE */
101 cmpl $0, function_trace_stop
104 cmpq $ftrace_stub, ftrace_trace_function
107 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
108 cmpq $ftrace_stub, ftrace_graph_return
109 jnz ftrace_graph_caller
111 cmpq $ftrace_graph_entry_stub, ftrace_graph_entry
112 jnz ftrace_graph_caller
121 movq 0x38(%rsp), %rdi
123 subq $MCOUNT_INSN_SIZE, %rdi
125 call *ftrace_trace_function
131 #endif /* CONFIG_DYNAMIC_FTRACE */
132 #endif /* CONFIG_FUNCTION_TRACER */
134 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
135 ENTRY(ftrace_graph_caller)
136 cmpl $0, function_trace_stop
142 movq 0x38(%rsp), %rsi
144 subq $MCOUNT_INSN_SIZE, %rsi
146 call prepare_ftrace_return
151 END(ftrace_graph_caller)
153 GLOBAL(return_to_handler)
156 /* Save the return values */
161 call ftrace_return_to_handler
171 #ifndef CONFIG_PREEMPT
172 #define retint_kernel retint_restore_args
175 #ifdef CONFIG_PARAVIRT
176 ENTRY(native_usergs_sysret64)
179 ENDPROC(native_usergs_sysret64)
180 #endif /* CONFIG_PARAVIRT */
183 .macro TRACE_IRQS_IRETQ offset=ARGOFFSET
184 #ifdef CONFIG_TRACE_IRQFLAGS
185 bt $9,EFLAGS-\offset(%rsp) /* interrupts off? */
193 * C code is not supposed to know about undefined top of stack. Every time
194 * a C function with an pt_regs argument is called from the SYSCALL based
195 * fast path FIXUP_TOP_OF_STACK is needed.
196 * RESTORE_TOP_OF_STACK syncs the syscall state after any possible ptregs
200 /* %rsp:at FRAMEEND */
201 .macro FIXUP_TOP_OF_STACK tmp offset=0
202 movq PER_CPU_VAR(old_rsp),\tmp
203 movq \tmp,RSP+\offset(%rsp)
204 movq $__USER_DS,SS+\offset(%rsp)
205 movq $__USER_CS,CS+\offset(%rsp)
206 movq $-1,RCX+\offset(%rsp)
207 movq R11+\offset(%rsp),\tmp /* get eflags */
208 movq \tmp,EFLAGS+\offset(%rsp)
211 .macro RESTORE_TOP_OF_STACK tmp offset=0
212 movq RSP+\offset(%rsp),\tmp
213 movq \tmp,PER_CPU_VAR(old_rsp)
214 movq EFLAGS+\offset(%rsp),\tmp
215 movq \tmp,R11+\offset(%rsp)
218 .macro FAKE_STACK_FRAME child_rip
219 /* push in order ss, rsp, eflags, cs, rip */
221 pushq_cfi $__KERNEL_DS /* ss */
222 /*CFI_REL_OFFSET ss,0*/
223 pushq_cfi %rax /* rsp */
225 pushq_cfi $(X86_EFLAGS_IF|X86_EFLAGS_BIT1) /* eflags - interrupts on */
226 /*CFI_REL_OFFSET rflags,0*/
227 pushq_cfi $__KERNEL_CS /* cs */
228 /*CFI_REL_OFFSET cs,0*/
229 pushq_cfi \child_rip /* rip */
231 pushq_cfi %rax /* orig rax */
234 .macro UNFAKE_STACK_FRAME
236 CFI_ADJUST_CFA_OFFSET -(6*8)
240 * initial frame state for interrupts (and exceptions without error code)
242 .macro EMPTY_FRAME offset=0
245 CFI_DEF_CFA rsp,\offset
249 * initial frame state for interrupts (and exceptions without error code)
251 .macro INTR_FRAME start=1 offset=0
253 EMPTY_FRAME SS+8+\offset-RIP
255 CFI_DEF_CFA_OFFSET SS+8+\offset-RIP
257 /*CFI_REL_OFFSET ss, SS+\offset-RIP*/
258 CFI_REL_OFFSET rsp, RSP+\offset-RIP
259 /*CFI_REL_OFFSET rflags, EFLAGS+\offset-RIP*/
260 /*CFI_REL_OFFSET cs, CS+\offset-RIP*/
261 CFI_REL_OFFSET rip, RIP+\offset-RIP
265 * initial frame state for exceptions with error code (and interrupts
266 * with vector already pushed)
268 .macro XCPT_FRAME start=1 offset=0
269 INTR_FRAME \start, RIP+\offset-ORIG_RAX
273 * frame that enables calling into C.
275 .macro PARTIAL_FRAME start=1 offset=0
277 XCPT_FRAME \start, ORIG_RAX+\offset-ARGOFFSET
279 CFI_REL_OFFSET rdi, RDI+\offset-ARGOFFSET
280 CFI_REL_OFFSET rsi, RSI+\offset-ARGOFFSET
281 CFI_REL_OFFSET rdx, RDX+\offset-ARGOFFSET
282 CFI_REL_OFFSET rcx, RCX+\offset-ARGOFFSET
283 CFI_REL_OFFSET rax, RAX+\offset-ARGOFFSET
284 CFI_REL_OFFSET r8, R8+\offset-ARGOFFSET
285 CFI_REL_OFFSET r9, R9+\offset-ARGOFFSET
286 CFI_REL_OFFSET r10, R10+\offset-ARGOFFSET
287 CFI_REL_OFFSET r11, R11+\offset-ARGOFFSET
291 * frame that enables passing a complete pt_regs to a C function.
293 .macro DEFAULT_FRAME start=1 offset=0
295 PARTIAL_FRAME \start, R11+\offset-R15
297 CFI_REL_OFFSET rbx, RBX+\offset
298 CFI_REL_OFFSET rbp, RBP+\offset
299 CFI_REL_OFFSET r12, R12+\offset
300 CFI_REL_OFFSET r13, R13+\offset
301 CFI_REL_OFFSET r14, R14+\offset
302 CFI_REL_OFFSET r15, R15+\offset
305 /* save partial stack frame */
308 /* start from rbp in pt_regs and jump over */
309 movq_cfi rdi, RDI-RBP
310 movq_cfi rsi, RSI-RBP
311 movq_cfi rdx, RDX-RBP
312 movq_cfi rcx, RCX-RBP
313 movq_cfi rax, RAX-RBP
316 movq_cfi r10, R10-RBP
317 movq_cfi r11, R11-RBP
319 /* Save rbp so that we can unwind from get_irq_regs() */
322 /* Save previous stack value */
325 leaq -RBP(%rsp),%rdi /* arg1 for handler */
330 * irq_count is used to check if a CPU is already on an interrupt stack
331 * or not. While this is essentially redundant with preempt_count it is
332 * a little cheaper to use a separate counter in the PDA (short of
333 * moving irq_enter into assembly, which would be too much work)
335 1: incl PER_CPU_VAR(irq_count)
337 mov PER_CPU_VAR(irq_stack_ptr),%rsp
338 CFI_DEF_CFA_REGISTER rsi
340 2: /* Store previous stack value */
342 CFI_ESCAPE 0x0f /* DW_CFA_def_cfa_expression */, 6, \
343 0x77 /* DW_OP_breg7 */, 0, \
344 0x06 /* DW_OP_deref */, \
345 0x08 /* DW_OP_const1u */, SS+8-RBP, \
346 0x22 /* DW_OP_plus */
347 /* We entered an interrupt context - irqs are off: */
353 movq 5*8+16(%rsp), %r11 /* save return address */
354 movq %rbx, RBX+16(%rsp)
355 movq %rbp, RBP+16(%rsp)
356 movq %r12, R12+16(%rsp)
357 movq %r13, R13+16(%rsp)
358 movq %r14, R14+16(%rsp)
359 movq %r15, R15+16(%rsp)
360 movq %r11, 8(%rsp) /* return address */
361 FIXUP_TOP_OF_STACK %r11, 16
366 /* save complete stack frame */
367 .pushsection .kprobes.text, "ax"
369 XCPT_FRAME offset=ORIG_RAX-R15+8
371 movq %rdi, RDI+8(%rsp)
372 movq %rsi, RSI+8(%rsp)
378 movq %r10, R10+8(%rsp)
379 movq %r11, R11+8(%rsp)
381 movq %rbp, RBP+8(%rsp)
382 movq %r12, R12+8(%rsp)
383 movq %r13, R13+8(%rsp)
384 movq %r14, R14+8(%rsp)
385 movq %r15, R15+8(%rsp)
387 movl $MSR_GS_BASE,%ecx
390 js 1f /* negative -> in kernel */
399 * A newly forked process directly context switches into this address.
401 * rdi: prev task we switched from
406 LOCK ; btr $TIF_FORK,TI_flags(%r8)
408 pushq_cfi kernel_eflags(%rip)
409 popfq_cfi # reset kernel eflags
411 call schedule_tail # rdi: 'prev' task parameter
413 GET_THREAD_INFO(%rcx)
417 testl $3, CS-ARGOFFSET(%rsp) # from kernel_thread?
418 jz retint_restore_args
420 testl $_TIF_IA32, TI_flags(%rcx) # 32-bit compat task needs IRET
421 jnz int_ret_from_sys_call
423 RESTORE_TOP_OF_STACK %rdi, -ARGOFFSET
424 jmp ret_from_sys_call # go to the SYSRET fastpath
430 * System call entry. Up to 6 arguments in registers are supported.
432 * SYSCALL does not save anything on the stack and does not change the
438 * rax system call number
440 * rcx return address for syscall/sysret, C arg3
443 * r10 arg3 (--> moved to rcx for C)
446 * r11 eflags for syscall/sysret, temporary for C
447 * r12-r15,rbp,rbx saved by C code, not touched.
449 * Interrupts are off on entry.
450 * Only called from user space.
452 * XXX if we had a free scratch register we could save the RSP into the stack frame
453 * and report it properly in ps. Unfortunately we haven't.
455 * When user can change the frames always force IRET. That is because
456 * it deals with uncanonical addresses better. SYSRET has trouble
457 * with them due to bugs in both AMD and Intel CPUs.
463 CFI_DEF_CFA rsp,KERNEL_STACK_OFFSET
465 /*CFI_REGISTER rflags,r11*/
468 * A hypervisor implementation might want to use a label
469 * after the swapgs, so that it can do the swapgs
470 * for the guest and jump here on syscall.
472 GLOBAL(system_call_after_swapgs)
474 movq %rsp,PER_CPU_VAR(old_rsp)
475 movq PER_CPU_VAR(kernel_stack),%rsp
477 * No need to follow this irqs off/on section - it's straight
480 ENABLE_INTERRUPTS(CLBR_NONE)
482 movq %rax,ORIG_RAX-ARGOFFSET(%rsp)
483 movq %rcx,RIP-ARGOFFSET(%rsp)
484 CFI_REL_OFFSET rip,RIP-ARGOFFSET
485 testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET)
487 system_call_fastpath:
488 cmpq $__NR_syscall_max,%rax
491 call *sys_call_table(,%rax,8) # XXX: rip relative
492 movq %rax,RAX-ARGOFFSET(%rsp)
494 * Syscall return path ending with SYSRET (fast path)
495 * Has incomplete stack frame and undefined top of stack.
498 movl $_TIF_ALLWORK_MASK,%edi
502 DISABLE_INTERRUPTS(CLBR_NONE)
504 movl TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET),%edx
509 * sysretq will re-enable interrupts:
512 movq RIP-ARGOFFSET(%rsp),%rcx
514 RESTORE_ARGS 1,-ARG_SKIP,0
515 /*CFI_REGISTER rflags,r11*/
516 movq PER_CPU_VAR(old_rsp), %rsp
520 /* Handle reschedules */
521 /* edx: work, edi: workmask */
523 bt $TIF_NEED_RESCHED,%edx
526 ENABLE_INTERRUPTS(CLBR_NONE)
532 /* Handle a signal */
535 ENABLE_INTERRUPTS(CLBR_NONE)
536 #ifdef CONFIG_AUDITSYSCALL
537 bt $TIF_SYSCALL_AUDIT,%edx
541 * We have a signal, or exit tracing or single-step.
542 * These all wind up with the iret return path anyway,
543 * so just join that path right now.
545 FIXUP_TOP_OF_STACK %r11, -ARGOFFSET
546 jmp int_check_syscall_exit_work
549 movq $-ENOSYS,RAX-ARGOFFSET(%rsp)
550 jmp ret_from_sys_call
552 #ifdef CONFIG_AUDITSYSCALL
554 * Fast path for syscall audit without full syscall trace.
555 * We just call __audit_syscall_entry() directly, and then
556 * jump back to the normal fast path.
559 movq %r10,%r9 /* 6th arg: 4th syscall arg */
560 movq %rdx,%r8 /* 5th arg: 3rd syscall arg */
561 movq %rsi,%rcx /* 4th arg: 2nd syscall arg */
562 movq %rdi,%rdx /* 3rd arg: 1st syscall arg */
563 movq %rax,%rsi /* 2nd arg: syscall number */
564 movl $AUDIT_ARCH_X86_64,%edi /* 1st arg: audit arch */
565 call __audit_syscall_entry
566 LOAD_ARGS 0 /* reload call-clobbered registers */
567 jmp system_call_fastpath
570 * Return fast path for syscall audit. Call __audit_syscall_exit()
571 * directly and then jump back to the fast path with TIF_SYSCALL_AUDIT
575 movq RAX-ARGOFFSET(%rsp),%rsi /* second arg, syscall return value */
576 cmpq $-MAX_ERRNO,%rsi /* is it < -MAX_ERRNO? */
577 setbe %al /* 1 if so, 0 if not */
578 movzbl %al,%edi /* zero-extend that into %edi */
579 call __audit_syscall_exit
580 movl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT),%edi
582 #endif /* CONFIG_AUDITSYSCALL */
584 /* Do syscall tracing */
586 #ifdef CONFIG_AUDITSYSCALL
587 testl $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET)
591 movq $-ENOSYS,RAX(%rsp) /* ptrace can change this for a bad syscall */
592 FIXUP_TOP_OF_STACK %rdi
594 call syscall_trace_enter
596 * Reload arg registers from stack in case ptrace changed them.
597 * We don't reload %rax because syscall_trace_enter() returned
598 * the value it wants us to use in the table lookup.
600 LOAD_ARGS ARGOFFSET, 1
602 cmpq $__NR_syscall_max,%rax
603 ja int_ret_from_sys_call /* RAX(%rsp) set to -ENOSYS above */
604 movq %r10,%rcx /* fixup for C */
605 call *sys_call_table(,%rax,8)
606 movq %rax,RAX-ARGOFFSET(%rsp)
607 /* Use IRET because user could have changed frame */
610 * Syscall return path ending with IRET.
611 * Has correct top of stack, but partial stack frame.
613 GLOBAL(int_ret_from_sys_call)
614 DISABLE_INTERRUPTS(CLBR_NONE)
616 movl $_TIF_ALLWORK_MASK,%edi
617 /* edi: mask to check */
618 GLOBAL(int_with_check)
620 GET_THREAD_INFO(%rcx)
621 movl TI_flags(%rcx),%edx
624 andl $~TS_COMPAT,TI_status(%rcx)
627 /* Either reschedule or signal or syscall exit tracking needed. */
628 /* First do a reschedule test. */
629 /* edx: work, edi: workmask */
631 bt $TIF_NEED_RESCHED,%edx
634 ENABLE_INTERRUPTS(CLBR_NONE)
638 DISABLE_INTERRUPTS(CLBR_NONE)
642 /* handle signals and tracing -- both require a full stack frame */
645 ENABLE_INTERRUPTS(CLBR_NONE)
646 int_check_syscall_exit_work:
648 /* Check for syscall exit trace */
649 testl $_TIF_WORK_SYSCALL_EXIT,%edx
652 leaq 8(%rsp),%rdi # &ptregs -> arg1
653 call syscall_trace_leave
655 andl $~(_TIF_WORK_SYSCALL_EXIT|_TIF_SYSCALL_EMU),%edi
659 testl $_TIF_DO_NOTIFY_MASK,%edx
661 movq %rsp,%rdi # &ptregs -> arg1
662 xorl %esi,%esi # oldset -> arg2
663 call do_notify_resume
664 1: movl $_TIF_WORK_MASK,%edi
667 DISABLE_INTERRUPTS(CLBR_NONE)
674 * Certain special system calls that need to save a complete full stack frame.
676 .macro PTREGSCALL label,func,arg
678 PARTIAL_FRAME 1 8 /* offset 8: return address */
679 subq $REST_SKIP, %rsp
680 CFI_ADJUST_CFA_OFFSET REST_SKIP
682 DEFAULT_FRAME -2 8 /* offset 8: return address */
683 leaq 8(%rsp), \arg /* pt_regs pointer */
685 jmp ptregscall_common
690 PTREGSCALL stub_clone, sys_clone, %r8
691 PTREGSCALL stub_fork, sys_fork, %rdi
692 PTREGSCALL stub_vfork, sys_vfork, %rdi
693 PTREGSCALL stub_sigaltstack, sys_sigaltstack, %rdx
694 PTREGSCALL stub_iopl, sys_iopl, %rsi
696 ENTRY(ptregscall_common)
697 DEFAULT_FRAME 1 8 /* offset 8: return address */
698 RESTORE_TOP_OF_STACK %r11, 8
699 movq_cfi_restore R15+8, r15
700 movq_cfi_restore R14+8, r14
701 movq_cfi_restore R13+8, r13
702 movq_cfi_restore R12+8, r12
703 movq_cfi_restore RBP+8, rbp
704 movq_cfi_restore RBX+8, rbx
705 ret $REST_SKIP /* pop extended registers */
707 END(ptregscall_common)
714 FIXUP_TOP_OF_STACK %r11
717 RESTORE_TOP_OF_STACK %r11
720 jmp int_ret_from_sys_call
725 * sigreturn is special because it needs to restore all registers on return.
726 * This cannot be done with SYSRET, so use the IRET return path instead.
728 ENTRY(stub_rt_sigreturn)
734 FIXUP_TOP_OF_STACK %r11
735 call sys_rt_sigreturn
736 movq %rax,RAX(%rsp) # fixme, this could be done at the higher layer
738 jmp int_ret_from_sys_call
740 END(stub_rt_sigreturn)
743 * Build the entry stubs and pointer table with some assembler magic.
744 * We pack 7 stubs into a single 32-byte chunk, which will fit in a
745 * single cache line on all modern x86 implementations.
747 .section .init.rodata,"a"
751 .p2align CONFIG_X86_L1_CACHE_SHIFT
752 ENTRY(irq_entries_start)
754 vector=FIRST_EXTERNAL_VECTOR
755 .rept (NR_VECTORS-FIRST_EXTERNAL_VECTOR+6)/7
758 .if vector < NR_VECTORS
759 .if vector <> FIRST_EXTERNAL_VECTOR
760 CFI_ADJUST_CFA_OFFSET -8
762 1: pushq_cfi $(~vector+0x80) /* Note: always in signed byte range */
763 .if ((vector-FIRST_EXTERNAL_VECTOR)%7) <> 6
772 2: jmp common_interrupt
775 END(irq_entries_start)
782 * Interrupt entry/exit.
784 * Interrupt entry points save only callee clobbered registers in fast path.
786 * Entry runs with interrupts off.
789 /* 0(%rsp): ~(interrupt number) */
790 .macro interrupt func
791 /* reserve pt_regs for scratch regs and rbp */
792 subq $ORIG_RAX-RBP, %rsp
793 CFI_ADJUST_CFA_OFFSET ORIG_RAX-RBP
799 * Interrupt entry/exit should be protected against kprobes
801 .pushsection .kprobes.text, "ax"
803 * The interrupt stubs push (~vector+0x80) onto the stack and
804 * then jump to common_interrupt.
806 .p2align CONFIG_X86_L1_CACHE_SHIFT
809 addq $-0x80,(%rsp) /* Adjust vector to [-256,-1] range */
811 /* 0(%rsp): old_rsp-ARGOFFSET */
813 DISABLE_INTERRUPTS(CLBR_NONE)
815 decl PER_CPU_VAR(irq_count)
817 /* Restore saved previous stack */
819 CFI_DEF_CFA_REGISTER rsi
820 leaq ARGOFFSET-RBP(%rsi), %rsp
821 CFI_DEF_CFA_REGISTER rsp
822 CFI_ADJUST_CFA_OFFSET RBP-ARGOFFSET
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
963 apicinterrupt IRQ_MOVE_CLEANUP_VECTOR \
964 irq_move_cleanup_interrupt smp_irq_move_cleanup_interrupt
965 apicinterrupt REBOOT_VECTOR \
966 reboot_interrupt smp_reboot_interrupt
970 apicinterrupt UV_BAU_MESSAGE \
971 uv_bau_message_intr1 uv_bau_message_interrupt
973 apicinterrupt LOCAL_TIMER_VECTOR \
974 apic_timer_interrupt smp_apic_timer_interrupt
975 apicinterrupt X86_PLATFORM_IPI_VECTOR \
976 x86_platform_ipi smp_x86_platform_ipi
981 .irp idx,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15, \
982 16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31
983 .if NUM_INVALIDATE_TLB_VECTORS > \idx
984 ENTRY(invalidate_interrupt\idx)
985 pushq_cfi $~(INVALIDATE_TLB_VECTOR_START+\idx)
986 jmp .Lcommon_invalidate_interrupt0
987 CFI_ADJUST_CFA_OFFSET -8
988 END(invalidate_interrupt\idx)
992 apicinterrupt INVALIDATE_TLB_VECTOR_START, \
993 invalidate_interrupt0, smp_invalidate_interrupt
996 apicinterrupt THRESHOLD_APIC_VECTOR \
997 threshold_interrupt smp_threshold_interrupt
998 apicinterrupt THERMAL_APIC_VECTOR \
999 thermal_interrupt smp_thermal_interrupt
1002 apicinterrupt CALL_FUNCTION_SINGLE_VECTOR \
1003 call_function_single_interrupt smp_call_function_single_interrupt
1004 apicinterrupt CALL_FUNCTION_VECTOR \
1005 call_function_interrupt smp_call_function_interrupt
1006 apicinterrupt RESCHEDULE_VECTOR \
1007 reschedule_interrupt smp_reschedule_interrupt
1010 apicinterrupt ERROR_APIC_VECTOR \
1011 error_interrupt smp_error_interrupt
1012 apicinterrupt SPURIOUS_APIC_VECTOR \
1013 spurious_interrupt smp_spurious_interrupt
1015 #ifdef CONFIG_IRQ_WORK
1016 apicinterrupt IRQ_WORK_VECTOR \
1017 irq_work_interrupt smp_irq_work_interrupt
1021 * Exception entry points.
1023 .macro zeroentry 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 error_exit /* %ebx: no swapgs flag */
1040 .macro paranoidzeroentry sym do_sym
1043 PARAVIRT_ADJUST_EXCEPTION_FRAME
1044 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1045 subq $ORIG_RAX-R15, %rsp
1046 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1050 movq %rsp,%rdi /* pt_regs pointer */
1051 xorl %esi,%esi /* no error code */
1053 jmp paranoid_exit /* %ebx: no swapgs flag */
1058 #define INIT_TSS_IST(x) PER_CPU_VAR(init_tss) + (TSS_ist + ((x) - 1) * 8)
1059 .macro paranoidzeroentry_ist sym do_sym ist
1062 PARAVIRT_ADJUST_EXCEPTION_FRAME
1063 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1064 subq $ORIG_RAX-R15, %rsp
1065 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1069 movq %rsp,%rdi /* pt_regs pointer */
1070 xorl %esi,%esi /* no error code */
1071 subq $EXCEPTION_STKSZ, INIT_TSS_IST(\ist)
1073 addq $EXCEPTION_STKSZ, INIT_TSS_IST(\ist)
1074 jmp paranoid_exit /* %ebx: no swapgs flag */
1079 .macro errorentry sym do_sym
1082 PARAVIRT_ADJUST_EXCEPTION_FRAME
1083 subq $ORIG_RAX-R15, %rsp
1084 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 error_exit /* %ebx: no swapgs flag */
1096 /* error code is on the stack already */
1097 .macro paranoiderrorentry sym do_sym
1100 PARAVIRT_ADJUST_EXCEPTION_FRAME
1101 subq $ORIG_RAX-R15, %rsp
1102 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1106 movq %rsp,%rdi /* pt_regs pointer */
1107 movq ORIG_RAX(%rsp),%rsi /* get error code */
1108 movq $-1,ORIG_RAX(%rsp) /* no syscall to restart */
1110 jmp paranoid_exit /* %ebx: no swapgs flag */
1115 zeroentry divide_error do_divide_error
1116 zeroentry overflow do_overflow
1117 zeroentry bounds do_bounds
1118 zeroentry invalid_op do_invalid_op
1119 zeroentry device_not_available do_device_not_available
1120 paranoiderrorentry double_fault do_double_fault
1121 zeroentry coprocessor_segment_overrun do_coprocessor_segment_overrun
1122 errorentry invalid_TSS do_invalid_TSS
1123 errorentry segment_not_present do_segment_not_present
1124 zeroentry spurious_interrupt_bug do_spurious_interrupt_bug
1125 zeroentry coprocessor_error do_coprocessor_error
1126 errorentry alignment_check do_alignment_check
1127 zeroentry simd_coprocessor_error do_simd_coprocessor_error
1130 /* Reload gs selector with exception handling */
1131 /* edi: new selector */
1132 ENTRY(native_load_gs_index)
1135 DISABLE_INTERRUPTS(CLBR_ANY & ~CLBR_RDI)
1139 2: mfence /* workaround */
1144 END(native_load_gs_index)
1146 .section __ex_table,"a"
1148 .quad gs_change,bad_gs
1150 .section .fixup,"ax"
1151 /* running with kernelgs */
1153 SWAPGS /* switch back to user gs */
1159 ENTRY(kernel_thread_helper)
1160 pushq $0 # fake return address
1163 * Here we are in the child and the registers are set as they were
1164 * at kernel_thread() invocation in the parent.
1170 ud2 # padding for call trace
1172 END(kernel_thread_helper)
1175 * execve(). This function needs to use IRET, not SYSRET, to set up all state properly.
1177 * C extern interface:
1178 * extern long execve(const char *name, char **argv, char **envp)
1180 * asm input arguments:
1181 * rdi: name, rsi: argv, rdx: envp
1183 * We want to fallback into:
1184 * extern long sys_execve(const char *name, char **argv,char **envp, struct pt_regs *regs)
1186 * do_sys_execve asm fallback arguments:
1187 * rdi: name, rsi: argv, rdx: envp, rcx: fake frame on the stack
1189 ENTRY(kernel_execve)
1195 movq %rax, RAX(%rsp)
1198 je int_ret_from_sys_call
1205 /* Call softirq on interrupt stack. Interrupts are off. */
1209 CFI_REL_OFFSET rbp,0
1211 CFI_DEF_CFA_REGISTER rbp
1212 incl PER_CPU_VAR(irq_count)
1213 cmove PER_CPU_VAR(irq_stack_ptr),%rsp
1214 push %rbp # backlink for old unwinder
1218 CFI_DEF_CFA_REGISTER rsp
1219 CFI_ADJUST_CFA_OFFSET -8
1220 decl PER_CPU_VAR(irq_count)
1225 #ifdef CONFIG_STACK_UNWIND
1226 ENTRY(arch_unwind_init_running)
1228 movq %r15, R15(%rdi)
1229 movq %r14, R14(%rdi)
1231 movq %r13, R13(%rdi)
1232 movq %r12, R12(%rdi)
1234 movq %rbp, RBP(%rdi)
1235 movq %rbx, RBX(%rdi)
1238 movq %rax, R11(%rdi)
1239 movq %rax, R10(%rdi)
1242 movq %rax, RAX(%rdi)
1243 movq %rax, RCX(%rdi)
1244 movq %rax, RDX(%rdi)
1245 movq %rax, RSI(%rdi)
1246 movq %rax, RDI(%rdi)
1247 movq %rax, ORIG_RAX(%rdi)
1250 movq $__KERNEL_CS, CS(%rdi)
1251 movq %rax, EFLAGS(%rdi)
1253 movq $__KERNEL_DS, SS(%rdi)
1256 END(arch_unwind_init_running)
1259 #ifdef CONFIG_PARAVIRT_XEN
1260 zeroentry xen_hypervisor_callback xen_do_hypervisor_callback
1263 * A note on the "critical region" in our callback handler.
1264 * We want to avoid stacking callback handlers due to events occurring
1265 * during handling of the last event. To do this, we keep events disabled
1266 * until we've done all processing. HOWEVER, we must enable events before
1267 * popping the stack frame (can't be done atomically) and so it would still
1268 * be possible to get enough handler activations to overflow the stack.
1269 * Although unlikely, bugs of that kind are hard to track down, so we'd
1270 * like to avoid the possibility.
1271 * So, on entry to the handler we detect whether we interrupted an
1272 * existing activation in its critical region -- if so, we pop the current
1273 * activation and restart the handler using the previous one.
1275 ENTRY(xen_do_hypervisor_callback) # do_hypervisor_callback(struct *pt_regs)
1278 * Since we don't modify %rdi, evtchn_do_upall(struct *pt_regs) will
1279 * see the correct pointer to the pt_regs
1281 movq %rdi, %rsp # we don't return, adjust the stack frame
1284 11: incl PER_CPU_VAR(irq_count)
1286 CFI_DEF_CFA_REGISTER rbp
1287 cmovzq PER_CPU_VAR(irq_stack_ptr),%rsp
1288 pushq %rbp # backlink for old unwinder
1289 call xen_evtchn_do_upcall
1291 CFI_DEF_CFA_REGISTER rsp
1292 decl PER_CPU_VAR(irq_count)
1295 END(xen_do_hypervisor_callback)
1298 * Hypervisor uses this for application faults while it executes.
1299 * We get here for two reasons:
1300 * 1. Fault while reloading DS, ES, FS or GS
1301 * 2. Fault while executing IRET
1302 * Category 1 we do not need to fix up as Xen has already reloaded all segment
1303 * registers that could be reloaded and zeroed the others.
1304 * Category 2 we fix up by killing the current process. We cannot use the
1305 * normal Linux return path in this case because if we use the IRET hypercall
1306 * to pop the stack frame we end up in an infinite loop of failsafe callbacks.
1307 * We distinguish between categories by comparing each saved segment register
1308 * with its current contents: any discrepancy means we in category 1.
1310 ENTRY(xen_failsafe_callback)
1312 /*CFI_REL_OFFSET gs,GS*/
1313 /*CFI_REL_OFFSET fs,FS*/
1314 /*CFI_REL_OFFSET es,ES*/
1315 /*CFI_REL_OFFSET ds,DS*/
1316 CFI_REL_OFFSET r11,8
1317 CFI_REL_OFFSET rcx,0
1331 /* All segments match their saved values => Category 2 (Bad IRET). */
1337 CFI_ADJUST_CFA_OFFSET -0x30
1338 pushq_cfi $0 /* RIP */
1341 jmp general_protection
1343 1: /* Segment mismatch => Category 1 (Bad segment). Retry the IRET. */
1349 CFI_ADJUST_CFA_OFFSET -0x30
1354 END(xen_failsafe_callback)
1356 apicinterrupt XEN_HVM_EVTCHN_CALLBACK \
1357 xen_hvm_callback_vector xen_evtchn_do_upcall
1359 #endif /* CONFIG_PARAVIRT_XEN */
1362 * Some functions should be protected against kprobes
1364 .pushsection .kprobes.text, "ax"
1366 paranoidzeroentry_ist debug do_debug DEBUG_STACK
1367 paranoidzeroentry_ist int3 do_int3 DEBUG_STACK
1368 paranoiderrorentry stack_segment do_stack_segment
1369 #ifdef CONFIG_PARAVIRT_XEN
1370 zeroentry xen_debug do_debug
1371 zeroentry xen_int3 do_int3
1372 errorentry xen_stack_segment do_stack_segment
1374 errorentry general_protection do_general_protection
1375 errorentry page_fault do_page_fault
1376 #ifdef CONFIG_KVM_GUEST
1377 errorentry async_page_fault do_async_page_fault
1379 #ifdef CONFIG_X86_MCE
1380 paranoidzeroentry machine_check *machine_check_vector(%rip)
1384 * "Paranoid" exit path from exception stack.
1385 * Paranoid because this is used by NMIs and cannot take
1386 * any kernel state for granted.
1387 * We don't do kernel preemption checks here, because only
1388 * NMI should be common and it does not enable IRQs and
1389 * cannot get reschedule ticks.
1391 * "trace" is 0 for the NMI handler only, because irq-tracing
1392 * is fundamentally NMI-unsafe. (we cannot change the soft and
1393 * hard flags at once, atomically)
1396 /* ebx: no swapgs flag */
1397 ENTRY(paranoid_exit)
1399 DISABLE_INTERRUPTS(CLBR_NONE)
1401 testl %ebx,%ebx /* swapgs needed? */
1402 jnz paranoid_restore
1404 jnz paranoid_userspace
1415 GET_THREAD_INFO(%rcx)
1416 movl TI_flags(%rcx),%ebx
1417 andl $_TIF_WORK_MASK,%ebx
1419 movq %rsp,%rdi /* &pt_regs */
1421 movq %rax,%rsp /* switch stack for scheduling */
1422 testl $_TIF_NEED_RESCHED,%ebx
1423 jnz paranoid_schedule
1424 movl %ebx,%edx /* arg3: thread flags */
1426 ENABLE_INTERRUPTS(CLBR_NONE)
1427 xorl %esi,%esi /* arg2: oldset */
1428 movq %rsp,%rdi /* arg1: &pt_regs */
1429 call do_notify_resume
1430 DISABLE_INTERRUPTS(CLBR_NONE)
1432 jmp paranoid_userspace
1435 ENABLE_INTERRUPTS(CLBR_ANY)
1437 DISABLE_INTERRUPTS(CLBR_ANY)
1439 jmp paranoid_userspace
1444 * Exception entry point. This expects an error code/orig_rax on the stack.
1445 * returns in "no swapgs flag" in %ebx.
1448 XCPT_FRAME offset=ORIG_RAX-R15+8
1449 /* oldrax contains error code */
1451 movq %rdi, RDI+8(%rsp)
1452 movq %rsi, RSI+8(%rsp)
1453 movq %rdx, RDX+8(%rsp)
1454 movq %rcx, RCX+8(%rsp)
1455 movq %rax, RAX+8(%rsp)
1456 movq %r8, R8+8(%rsp)
1457 movq %r9, R9+8(%rsp)
1458 movq %r10, R10+8(%rsp)
1459 movq %r11, R11+8(%rsp)
1461 movq %rbp, RBP+8(%rsp)
1462 movq %r12, R12+8(%rsp)
1463 movq %r13, R13+8(%rsp)
1464 movq %r14, R14+8(%rsp)
1465 movq %r15, R15+8(%rsp)
1468 je error_kernelspace
1476 * There are two places in the kernel that can potentially fault with
1477 * usergs. Handle them here. The exception handlers after iret run with
1478 * kernel gs again, so don't set the user space flag. B stepping K8s
1479 * sometimes report an truncated RIP for IRET exceptions returning to
1480 * compat mode. Check for these here too.
1483 CFI_REL_OFFSET rcx, RCX+8
1485 leaq irq_return(%rip),%rcx
1486 cmpq %rcx,RIP+8(%rsp)
1488 movl %ecx,%eax /* zero extend */
1489 cmpq %rax,RIP+8(%rsp)
1491 cmpq $gs_change,RIP+8(%rsp)
1496 /* Fix truncated RIP */
1497 movq %rcx,RIP+8(%rsp)
1503 /* ebx: no swapgs flag (1: don't need swapgs, 0: need it) */
1508 DISABLE_INTERRUPTS(CLBR_NONE)
1510 GET_THREAD_INFO(%rcx)
1513 LOCKDEP_SYS_EXIT_IRQ
1514 movl TI_flags(%rcx),%edx
1515 movl $_TIF_WORK_MASK,%edi
1523 * Test if a given stack is an NMI stack or not.
1525 .macro test_in_nmi reg stack nmi_ret normal_ret
1528 subq $EXCEPTION_STKSZ, %\reg
1534 /* runs on exception stack */
1537 PARAVIRT_ADJUST_EXCEPTION_FRAME
1539 * We allow breakpoints in NMIs. If a breakpoint occurs, then
1540 * the iretq it performs will take us out of NMI context.
1541 * This means that we can have nested NMIs where the next
1542 * NMI is using the top of the stack of the previous NMI. We
1543 * can't let it execute because the nested NMI will corrupt the
1544 * stack of the previous NMI. NMI handlers are not re-entrant
1547 * To handle this case we do the following:
1548 * Check the a special location on the stack that contains
1549 * a variable that is set when NMIs are executing.
1550 * The interrupted task's stack is also checked to see if it
1552 * If the variable is not set and the stack is not the NMI
1554 * o Set the special variable on the stack
1555 * o Copy the interrupt frame into a "saved" location on the stack
1556 * o Copy the interrupt frame into a "copy" location on the stack
1557 * o Continue processing the NMI
1558 * If the variable is set or the previous stack is the NMI stack:
1559 * o Modify the "copy" location to jump to the repeate_nmi
1560 * o return back to the first NMI
1562 * Now on exit of the first NMI, we first clear the stack variable
1563 * The NMI stack will tell any nested NMIs at that point that it is
1564 * nested. Then we pop the stack normally with iret, and if there was
1565 * a nested NMI that updated the copy interrupt stack frame, a
1566 * jump will be made to the repeat_nmi code that will handle the second
1570 /* Use %rdx as out temp variable throughout */
1574 * If %cs was not the kernel segment, then the NMI triggered in user
1575 * space, which means it is definitely not nested.
1577 cmpl $__KERNEL_CS, 16(%rsp)
1581 * Check the special variable on the stack to see if NMIs are
1588 * Now test if the previous stack was an NMI stack.
1589 * We need the double check. We check the NMI stack to satisfy the
1590 * race when the first NMI clears the variable before returning.
1591 * We check the variable because the first NMI could be in a
1592 * breakpoint routine using a breakpoint stack.
1595 test_in_nmi rdx, 4*8(%rsp), nested_nmi, first_nmi
1599 * Do nothing if we interrupted the fixup in repeat_nmi.
1600 * It's about to repeat the NMI handler, so we are fine
1601 * with ignoring this one.
1603 movq $repeat_nmi, %rdx
1606 movq $end_repeat_nmi, %rdx
1611 /* Set up the interrupted NMIs stack to jump to repeat_nmi */
1612 leaq -6*8(%rsp), %rdx
1614 CFI_ADJUST_CFA_OFFSET 6*8
1615 pushq_cfi $__KERNEL_DS
1618 pushq_cfi $__KERNEL_CS
1619 pushq_cfi $repeat_nmi
1621 /* Put stack back */
1623 CFI_ADJUST_CFA_OFFSET -11*8
1628 /* No need to check faults here */
1633 * Because nested NMIs will use the pushed location that we
1634 * stored in rdx, we must keep that space available.
1635 * Here's what our stack frame will look like:
1636 * +-------------------------+
1638 * | original Return RSP |
1639 * | original RFLAGS |
1642 * +-------------------------+
1643 * | temp storage for rdx |
1644 * +-------------------------+
1645 * | NMI executing variable |
1646 * +-------------------------+
1648 * | Saved Return RSP |
1652 * +-------------------------+
1654 * | copied Return RSP |
1658 * +-------------------------+
1660 * +-------------------------+
1662 * The saved RIP is used to fix up the copied RIP that a nested
1663 * NMI may zero out. The original stack frame and the temp storage
1664 * is also used by nested NMIs and can not be trusted on exit.
1666 /* Set the NMI executing variable on the stack. */
1669 /* Copy the stack frame to the Saved frame */
1674 /* Make another copy, this one may be modified by nested NMIs */
1679 /* Do not pop rdx, nested NMIs will corrupt it */
1680 movq 11*8(%rsp), %rdx
1683 * Everything below this point can be preempted by a nested
1684 * NMI if the first NMI took an exception. Repeated NMIs
1685 * caused by an exception and nested NMI will start here, and
1686 * can still be preempted by another NMI.
1689 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1690 subq $ORIG_RAX-R15, %rsp
1691 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1693 * Use save_paranoid to handle SWAPGS, but no need to use paranoid_exit
1694 * as we should not be calling schedule in NMI context.
1695 * Even with normal interrupts enabled. An NMI should not be
1696 * setting NEED_RESCHED or anything that normal interrupts and
1697 * exceptions might do.
1701 /* paranoidentry do_nmi, 0; without TRACE_IRQS_OFF */
1705 testl %ebx,%ebx /* swapgs needed? */
1711 /* Clear the NMI executing stack variable */
1718 * If an NMI hit an iret because of an exception or breakpoint,
1719 * it can lose its NMI context, and a nested NMI may come in.
1720 * In that case, the nested NMI will change the preempted NMI's
1721 * stack to jump to here when it does the final iret.
1725 /* Update the stack variable to say we are still in NMI */
1728 /* copy the saved stack back to copy stack */
1737 ENTRY(ignore_sysret)
1745 * End of kprobes section