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 <asm/pgtable_types.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 /* 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 start=1 offset=0
246 CFI_DEF_CFA rsp,8+\offset
248 CFI_DEF_CFA_OFFSET 8+\offset
253 * initial frame state for interrupts (and exceptions without error code)
255 .macro INTR_FRAME start=1 offset=0
256 EMPTY_FRAME \start, 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
270 /*CFI_REL_OFFSET orig_rax, ORIG_RAX-ORIG_RAX*/
274 * frame that enables calling into C.
276 .macro PARTIAL_FRAME start=1 offset=0
277 XCPT_FRAME \start, ORIG_RAX+\offset-ARGOFFSET
278 CFI_REL_OFFSET rdi, RDI+\offset-ARGOFFSET
279 CFI_REL_OFFSET rsi, RSI+\offset-ARGOFFSET
280 CFI_REL_OFFSET rdx, RDX+\offset-ARGOFFSET
281 CFI_REL_OFFSET rcx, RCX+\offset-ARGOFFSET
282 CFI_REL_OFFSET rax, RAX+\offset-ARGOFFSET
283 CFI_REL_OFFSET r8, R8+\offset-ARGOFFSET
284 CFI_REL_OFFSET r9, R9+\offset-ARGOFFSET
285 CFI_REL_OFFSET r10, R10+\offset-ARGOFFSET
286 CFI_REL_OFFSET r11, R11+\offset-ARGOFFSET
290 * frame that enables passing a complete pt_regs to a C function.
292 .macro DEFAULT_FRAME start=1 offset=0
293 PARTIAL_FRAME \start, R11+\offset-R15
294 CFI_REL_OFFSET rbx, RBX+\offset
295 CFI_REL_OFFSET rbp, RBP+\offset
296 CFI_REL_OFFSET r12, R12+\offset
297 CFI_REL_OFFSET r13, R13+\offset
298 CFI_REL_OFFSET r14, R14+\offset
299 CFI_REL_OFFSET r15, R15+\offset
302 /* save partial stack frame */
305 /* start from rbp in pt_regs and jump over */
306 movq_cfi rdi, RDI-RBP
307 movq_cfi rsi, RSI-RBP
308 movq_cfi rdx, RDX-RBP
309 movq_cfi rcx, RCX-RBP
310 movq_cfi rax, RAX-RBP
313 movq_cfi r10, R10-RBP
314 movq_cfi r11, R11-RBP
316 /* Save rbp so that we can unwind from get_irq_regs() */
319 /* Save previous stack value */
322 leaq -RBP(%rsp),%rdi /* arg1 for handler */
327 * irq_count is used to check if a CPU is already on an interrupt stack
328 * or not. While this is essentially redundant with preempt_count it is
329 * a little cheaper to use a separate counter in the PDA (short of
330 * moving irq_enter into assembly, which would be too much work)
332 1: incl PER_CPU_VAR(irq_count)
334 mov PER_CPU_VAR(irq_stack_ptr),%rsp
335 CFI_DEF_CFA_REGISTER rsi
337 2: /* Store previous stack value */
339 CFI_ESCAPE 0x0f /* DW_CFA_def_cfa_expression */, 6, \
340 0x77 /* DW_OP_breg7 */, 0, \
341 0x06 /* DW_OP_deref */, \
342 0x08 /* DW_OP_const1u */, SS+8-RBP, \
343 0x22 /* DW_OP_plus */
344 /* We entered an interrupt context - irqs are off: */
349 PARTIAL_FRAME 1 REST_SKIP+8
350 movq 5*8+16(%rsp), %r11 /* save return address */
357 movq %r11, 8(%rsp) /* return address */
358 FIXUP_TOP_OF_STACK %r11, 16
363 /* save complete stack frame */
364 .pushsection .kprobes.text, "ax"
384 movl $MSR_GS_BASE,%ecx
387 js 1f /* negative -> in kernel */
396 * A newly forked process directly context switches into this address.
398 * rdi: prev task we switched from
403 LOCK ; btr $TIF_FORK,TI_flags(%r8)
405 pushq_cfi kernel_eflags(%rip)
406 popfq_cfi # reset kernel eflags
408 call schedule_tail # rdi: 'prev' task parameter
410 GET_THREAD_INFO(%rcx)
414 testl $3, CS-ARGOFFSET(%rsp) # from kernel_thread?
415 je int_ret_from_sys_call
417 testl $_TIF_IA32, TI_flags(%rcx) # 32-bit compat task needs IRET
418 jnz int_ret_from_sys_call
420 RESTORE_TOP_OF_STACK %rdi, -ARGOFFSET
421 jmp ret_from_sys_call # go to the SYSRET fastpath
427 * System call entry. Up to 6 arguments in registers are supported.
429 * SYSCALL does not save anything on the stack and does not change the
435 * rax system call number
437 * rcx return address for syscall/sysret, C arg3
440 * r10 arg3 (--> moved to rcx for C)
443 * r11 eflags for syscall/sysret, temporary for C
444 * r12-r15,rbp,rbx saved by C code, not touched.
446 * Interrupts are off on entry.
447 * Only called from user space.
449 * XXX if we had a free scratch register we could save the RSP into the stack frame
450 * and report it properly in ps. Unfortunately we haven't.
452 * When user can change the frames always force IRET. That is because
453 * it deals with uncanonical addresses better. SYSRET has trouble
454 * with them due to bugs in both AMD and Intel CPUs.
460 CFI_DEF_CFA rsp,KERNEL_STACK_OFFSET
462 /*CFI_REGISTER rflags,r11*/
465 * A hypervisor implementation might want to use a label
466 * after the swapgs, so that it can do the swapgs
467 * for the guest and jump here on syscall.
469 ENTRY(system_call_after_swapgs)
471 movq %rsp,PER_CPU_VAR(old_rsp)
472 movq PER_CPU_VAR(kernel_stack),%rsp
474 * No need to follow this irqs off/on section - it's straight
477 ENABLE_INTERRUPTS(CLBR_NONE)
479 movq %rax,ORIG_RAX-ARGOFFSET(%rsp)
480 movq %rcx,RIP-ARGOFFSET(%rsp)
481 CFI_REL_OFFSET rip,RIP-ARGOFFSET
482 GET_THREAD_INFO(%rcx)
483 testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%rcx)
485 system_call_fastpath:
486 cmpq $__NR_syscall_max,%rax
489 call *sys_call_table(,%rax,8) # XXX: rip relative
490 movq %rax,RAX-ARGOFFSET(%rsp)
492 * Syscall return path ending with SYSRET (fast path)
493 * Has incomplete stack frame and undefined top of stack.
496 movl $_TIF_ALLWORK_MASK,%edi
500 GET_THREAD_INFO(%rcx)
501 DISABLE_INTERRUPTS(CLBR_NONE)
503 movl TI_flags(%rcx),%edx
508 * sysretq will re-enable interrupts:
511 movq RIP-ARGOFFSET(%rsp),%rcx
513 RESTORE_ARGS 1,-ARG_SKIP,0
514 /*CFI_REGISTER rflags,r11*/
515 movq PER_CPU_VAR(old_rsp), %rsp
519 /* Handle reschedules */
520 /* edx: work, edi: workmask */
522 bt $TIF_NEED_RESCHED,%edx
525 ENABLE_INTERRUPTS(CLBR_NONE)
531 /* Handle a signal */
534 ENABLE_INTERRUPTS(CLBR_NONE)
535 #ifdef CONFIG_AUDITSYSCALL
536 bt $TIF_SYSCALL_AUDIT,%edx
540 * We have a signal, or exit tracing or single-step.
541 * These all wind up with the iret return path anyway,
542 * so just join that path right now.
544 FIXUP_TOP_OF_STACK %r11, -ARGOFFSET
545 jmp int_check_syscall_exit_work
548 movq $-ENOSYS,RAX-ARGOFFSET(%rsp)
549 jmp ret_from_sys_call
551 #ifdef CONFIG_AUDITSYSCALL
553 * Fast path for syscall audit without full syscall trace.
554 * We just call audit_syscall_entry() directly, and then
555 * jump back to the normal fast path.
558 movq %r10,%r9 /* 6th arg: 4th syscall arg */
559 movq %rdx,%r8 /* 5th arg: 3rd syscall arg */
560 movq %rsi,%rcx /* 4th arg: 2nd syscall arg */
561 movq %rdi,%rdx /* 3rd arg: 1st syscall arg */
562 movq %rax,%rsi /* 2nd arg: syscall number */
563 movl $AUDIT_ARCH_X86_64,%edi /* 1st arg: audit arch */
564 call audit_syscall_entry
565 LOAD_ARGS 0 /* reload call-clobbered registers */
566 jmp system_call_fastpath
569 * Return fast path for syscall audit. Call audit_syscall_exit()
570 * directly and then jump back to the fast path with TIF_SYSCALL_AUDIT
574 movq RAX-ARGOFFSET(%rsp),%rsi /* second arg, syscall return value */
575 cmpq $0,%rsi /* is it < 0? */
576 setl %al /* 1 if so, 0 if not */
577 movzbl %al,%edi /* zero-extend that into %edi */
578 inc %edi /* first arg, 0->1(AUDITSC_SUCCESS), 1->2(AUDITSC_FAILURE) */
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(%rcx)
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 testl $3,CS-ARGOFFSET(%rsp)
617 je retint_restore_args
618 movl $_TIF_ALLWORK_MASK,%edi
619 /* edi: mask to check */
620 GLOBAL(int_with_check)
622 GET_THREAD_INFO(%rcx)
623 movl TI_flags(%rcx),%edx
626 andl $~TS_COMPAT,TI_status(%rcx)
629 /* Either reschedule or signal or syscall exit tracking needed. */
630 /* First do a reschedule test. */
631 /* edx: work, edi: workmask */
633 bt $TIF_NEED_RESCHED,%edx
636 ENABLE_INTERRUPTS(CLBR_NONE)
640 DISABLE_INTERRUPTS(CLBR_NONE)
644 /* handle signals and tracing -- both require a full stack frame */
647 ENABLE_INTERRUPTS(CLBR_NONE)
648 int_check_syscall_exit_work:
650 /* Check for syscall exit trace */
651 testl $_TIF_WORK_SYSCALL_EXIT,%edx
654 leaq 8(%rsp),%rdi # &ptregs -> arg1
655 call syscall_trace_leave
657 andl $~(_TIF_WORK_SYSCALL_EXIT|_TIF_SYSCALL_EMU),%edi
661 testl $_TIF_DO_NOTIFY_MASK,%edx
663 movq %rsp,%rdi # &ptregs -> arg1
664 xorl %esi,%esi # oldset -> arg2
665 call do_notify_resume
666 1: movl $_TIF_WORK_MASK,%edi
669 DISABLE_INTERRUPTS(CLBR_NONE)
676 * Certain special system calls that need to save a complete full stack frame.
678 .macro PTREGSCALL label,func,arg
680 PARTIAL_FRAME 1 8 /* offset 8: return address */
681 subq $REST_SKIP, %rsp
682 CFI_ADJUST_CFA_OFFSET REST_SKIP
684 DEFAULT_FRAME 0 8 /* offset 8: return address */
685 leaq 8(%rsp), \arg /* pt_regs pointer */
687 jmp ptregscall_common
692 PTREGSCALL stub_clone, sys_clone, %r8
693 PTREGSCALL stub_fork, sys_fork, %rdi
694 PTREGSCALL stub_vfork, sys_vfork, %rdi
695 PTREGSCALL stub_sigaltstack, sys_sigaltstack, %rdx
696 PTREGSCALL stub_iopl, sys_iopl, %rsi
698 ENTRY(ptregscall_common)
699 DEFAULT_FRAME 1 8 /* offset 8: return address */
700 RESTORE_TOP_OF_STACK %r11, 8
701 movq_cfi_restore R15+8, r15
702 movq_cfi_restore R14+8, r14
703 movq_cfi_restore R13+8, r13
704 movq_cfi_restore R12+8, r12
705 movq_cfi_restore RBP+8, rbp
706 movq_cfi_restore RBX+8, rbx
707 ret $REST_SKIP /* pop extended registers */
709 END(ptregscall_common)
716 FIXUP_TOP_OF_STACK %r11
719 RESTORE_TOP_OF_STACK %r11
722 jmp int_ret_from_sys_call
727 * sigreturn is special because it needs to restore all registers on return.
728 * This cannot be done with SYSRET, so use the IRET return path instead.
730 ENTRY(stub_rt_sigreturn)
736 FIXUP_TOP_OF_STACK %r11
737 call sys_rt_sigreturn
738 movq %rax,RAX(%rsp) # fixme, this could be done at the higher layer
740 jmp int_ret_from_sys_call
742 END(stub_rt_sigreturn)
745 * Build the entry stubs and pointer table with some assembler magic.
746 * We pack 7 stubs into a single 32-byte chunk, which will fit in a
747 * single cache line on all modern x86 implementations.
749 .section .init.rodata,"a"
753 .p2align CONFIG_X86_L1_CACHE_SHIFT
754 ENTRY(irq_entries_start)
756 vector=FIRST_EXTERNAL_VECTOR
757 .rept (NR_VECTORS-FIRST_EXTERNAL_VECTOR+6)/7
760 .if vector < NR_VECTORS
761 .if vector <> FIRST_EXTERNAL_VECTOR
762 CFI_ADJUST_CFA_OFFSET -8
764 1: pushq_cfi $(~vector+0x80) /* Note: always in signed byte range */
765 .if ((vector-FIRST_EXTERNAL_VECTOR)%7) <> 6
774 2: jmp common_interrupt
777 END(irq_entries_start)
784 * Interrupt entry/exit.
786 * Interrupt entry points save only callee clobbered registers in fast path.
788 * Entry runs with interrupts off.
791 /* 0(%rsp): ~(interrupt number) */
792 .macro interrupt func
793 /* reserve pt_regs for scratch regs and rbp */
794 subq $ORIG_RAX-RBP, %rsp
795 CFI_ADJUST_CFA_OFFSET ORIG_RAX-RBP
801 * Interrupt entry/exit should be protected against kprobes
803 .pushsection .kprobes.text, "ax"
805 * The interrupt stubs push (~vector+0x80) onto the stack and
806 * then jump to common_interrupt.
808 .p2align CONFIG_X86_L1_CACHE_SHIFT
811 addq $-0x80,(%rsp) /* Adjust vector to [-256,-1] range */
813 /* 0(%rsp): old_rsp-ARGOFFSET */
815 DISABLE_INTERRUPTS(CLBR_NONE)
817 decl PER_CPU_VAR(irq_count)
819 /* Restore saved previous stack */
821 CFI_DEF_CFA_REGISTER rsi
822 leaq ARGOFFSET-RBP(%rsi), %rsp
823 CFI_DEF_CFA_REGISTER rsp
824 CFI_ADJUST_CFA_OFFSET RBP-ARGOFFSET
827 GET_THREAD_INFO(%rcx)
828 testl $3,CS-ARGOFFSET(%rsp)
831 /* Interrupt came from user space */
833 * Has a correct top of stack, but a partial stack frame
834 * %rcx: thread info. Interrupts off.
836 retint_with_reschedule:
837 movl $_TIF_WORK_MASK,%edi
840 movl TI_flags(%rcx),%edx
845 retint_swapgs: /* return to user-space */
847 * The iretq could re-enable interrupts:
849 DISABLE_INTERRUPTS(CLBR_ANY)
854 retint_restore_args: /* return to kernel space */
855 DISABLE_INTERRUPTS(CLBR_ANY)
857 * The iretq could re-enable interrupts:
868 * Are we returning to a stack segment from the LDT? Note: in
869 * 64-bit mode SS:RSP on the exception stack is always valid.
871 #ifdef CONFIG_X86_ESPFIX64
872 testb $4,(SS-RIP)(%rsp)
873 jnz native_irq_return_ldt
876 .global native_irq_return_iret
877 native_irq_return_iret:
879 * This may fault. Non-paranoid faults on return to userspace are
880 * handled by fixup_bad_iret. These include #SS, #GP, and #NP.
881 * Double-faults due to espfix64 are handled in do_double_fault.
882 * Other faults here are fatal.
886 #ifdef CONFIG_X86_ESPFIX64
887 native_irq_return_ldt:
891 movq PER_CPU_VAR(espfix_waddr),%rdi
892 movq %rax,(0*8)(%rdi) /* RAX */
893 movq (2*8)(%rsp),%rax /* RIP */
894 movq %rax,(1*8)(%rdi)
895 movq (3*8)(%rsp),%rax /* CS */
896 movq %rax,(2*8)(%rdi)
897 movq (4*8)(%rsp),%rax /* RFLAGS */
898 movq %rax,(3*8)(%rdi)
899 movq (6*8)(%rsp),%rax /* SS */
900 movq %rax,(5*8)(%rdi)
901 movq (5*8)(%rsp),%rax /* RSP */
902 movq %rax,(4*8)(%rdi)
903 andl $0xffff0000,%eax
905 orq PER_CPU_VAR(espfix_stack),%rax
909 jmp native_irq_return_iret
912 /* edi: workmask, edx: work */
915 bt $TIF_NEED_RESCHED,%edx
918 ENABLE_INTERRUPTS(CLBR_NONE)
922 GET_THREAD_INFO(%rcx)
923 DISABLE_INTERRUPTS(CLBR_NONE)
928 testl $_TIF_DO_NOTIFY_MASK,%edx
931 ENABLE_INTERRUPTS(CLBR_NONE)
933 movq $-1,ORIG_RAX(%rsp)
934 xorl %esi,%esi # oldset
935 movq %rsp,%rdi # &pt_regs
936 call do_notify_resume
938 DISABLE_INTERRUPTS(CLBR_NONE)
940 GET_THREAD_INFO(%rcx)
941 jmp retint_with_reschedule
943 #ifdef CONFIG_PREEMPT
944 /* Returning to kernel space. Check if we need preemption */
945 /* rcx: threadinfo. interrupts off. */
947 cmpl $0,TI_preempt_count(%rcx)
948 jnz retint_restore_args
949 bt $TIF_NEED_RESCHED,TI_flags(%rcx)
950 jnc retint_restore_args
951 bt $9,EFLAGS-ARGOFFSET(%rsp) /* interrupts off? */
952 jnc retint_restore_args
953 call preempt_schedule_irq
957 END(common_interrupt)
960 * End of kprobes section
967 .macro apicinterrupt num sym do_sym
978 apicinterrupt IRQ_MOVE_CLEANUP_VECTOR \
979 irq_move_cleanup_interrupt smp_irq_move_cleanup_interrupt
980 apicinterrupt REBOOT_VECTOR \
981 reboot_interrupt smp_reboot_interrupt
985 apicinterrupt UV_BAU_MESSAGE \
986 uv_bau_message_intr1 uv_bau_message_interrupt
988 apicinterrupt LOCAL_TIMER_VECTOR \
989 apic_timer_interrupt smp_apic_timer_interrupt
990 apicinterrupt X86_PLATFORM_IPI_VECTOR \
991 x86_platform_ipi smp_x86_platform_ipi
994 .irp idx,0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15, \
995 16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31
996 .if NUM_INVALIDATE_TLB_VECTORS > \idx
997 apicinterrupt (INVALIDATE_TLB_VECTOR_START)+\idx \
998 invalidate_interrupt\idx smp_invalidate_interrupt
1003 apicinterrupt THRESHOLD_APIC_VECTOR \
1004 threshold_interrupt smp_threshold_interrupt
1005 apicinterrupt THERMAL_APIC_VECTOR \
1006 thermal_interrupt smp_thermal_interrupt
1009 apicinterrupt CALL_FUNCTION_SINGLE_VECTOR \
1010 call_function_single_interrupt smp_call_function_single_interrupt
1011 apicinterrupt CALL_FUNCTION_VECTOR \
1012 call_function_interrupt smp_call_function_interrupt
1013 apicinterrupt RESCHEDULE_VECTOR \
1014 reschedule_interrupt smp_reschedule_interrupt
1017 apicinterrupt ERROR_APIC_VECTOR \
1018 error_interrupt smp_error_interrupt
1019 apicinterrupt SPURIOUS_APIC_VECTOR \
1020 spurious_interrupt smp_spurious_interrupt
1022 #ifdef CONFIG_IRQ_WORK
1023 apicinterrupt IRQ_WORK_VECTOR \
1024 irq_work_interrupt smp_irq_work_interrupt
1028 * Exception entry points.
1030 .macro zeroentry sym do_sym
1033 PARAVIRT_ADJUST_EXCEPTION_FRAME
1034 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1035 subq $ORIG_RAX-R15, %rsp
1036 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1039 movq %rsp,%rdi /* pt_regs pointer */
1040 xorl %esi,%esi /* no error code */
1042 jmp error_exit /* %ebx: no swapgs flag */
1047 .macro paranoidzeroentry sym do_sym
1050 PARAVIRT_ADJUST_EXCEPTION_FRAME
1051 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1052 subq $ORIG_RAX-R15, %rsp
1053 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1056 movq %rsp,%rdi /* pt_regs pointer */
1057 xorl %esi,%esi /* no error code */
1059 jmp paranoid_exit /* %ebx: no swapgs flag */
1064 #define INIT_TSS_IST(x) PER_CPU_VAR(init_tss) + (TSS_ist + ((x) - 1) * 8)
1065 .macro paranoidzeroentry_ist sym do_sym ist
1068 PARAVIRT_ADJUST_EXCEPTION_FRAME
1069 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1070 subq $ORIG_RAX-R15, %rsp
1071 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1074 movq %rsp,%rdi /* pt_regs pointer */
1075 xorl %esi,%esi /* no error code */
1076 subq $EXCEPTION_STKSZ, INIT_TSS_IST(\ist)
1078 addq $EXCEPTION_STKSZ, INIT_TSS_IST(\ist)
1079 jmp paranoid_exit /* %ebx: no swapgs flag */
1084 .macro errorentry sym do_sym
1087 PARAVIRT_ADJUST_EXCEPTION_FRAME
1088 subq $ORIG_RAX-R15, %rsp
1089 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1092 movq %rsp,%rdi /* pt_regs pointer */
1093 movq ORIG_RAX(%rsp),%rsi /* get error code */
1094 movq $-1,ORIG_RAX(%rsp) /* no syscall to restart */
1096 jmp error_exit /* %ebx: no swapgs flag */
1101 /* error code is on the stack already */
1102 .macro paranoiderrorentry sym do_sym
1105 PARAVIRT_ADJUST_EXCEPTION_FRAME
1106 subq $ORIG_RAX-R15, %rsp
1107 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1111 movq %rsp,%rdi /* pt_regs pointer */
1112 movq ORIG_RAX(%rsp),%rsi /* get error code */
1113 movq $-1,ORIG_RAX(%rsp) /* no syscall to restart */
1115 jmp paranoid_exit /* %ebx: no swapgs flag */
1120 zeroentry divide_error do_divide_error
1121 zeroentry overflow do_overflow
1122 zeroentry bounds do_bounds
1123 zeroentry invalid_op do_invalid_op
1124 zeroentry device_not_available do_device_not_available
1125 paranoiderrorentry double_fault do_double_fault
1126 zeroentry coprocessor_segment_overrun do_coprocessor_segment_overrun
1127 errorentry invalid_TSS do_invalid_TSS
1128 errorentry segment_not_present do_segment_not_present
1129 zeroentry spurious_interrupt_bug do_spurious_interrupt_bug
1130 zeroentry coprocessor_error do_coprocessor_error
1131 errorentry alignment_check do_alignment_check
1132 zeroentry simd_coprocessor_error do_simd_coprocessor_error
1135 /* Reload gs selector with exception handling */
1136 /* edi: new selector */
1137 ENTRY(native_load_gs_index)
1140 DISABLE_INTERRUPTS(CLBR_ANY & ~CLBR_RDI)
1144 2: mfence /* workaround */
1149 END(native_load_gs_index)
1151 .section __ex_table,"a"
1153 .quad gs_change,bad_gs
1155 .section .fixup,"ax"
1156 /* running with kernelgs */
1158 SWAPGS /* switch back to user gs */
1164 ENTRY(kernel_thread_helper)
1165 pushq $0 # fake return address
1168 * Here we are in the child and the registers are set as they were
1169 * at kernel_thread() invocation in the parent.
1175 ud2 # padding for call trace
1177 END(kernel_thread_helper)
1180 * execve(). This function needs to use IRET, not SYSRET, to set up all state properly.
1182 * C extern interface:
1183 * extern long execve(const char *name, char **argv, char **envp)
1185 * asm input arguments:
1186 * rdi: name, rsi: argv, rdx: envp
1188 * We want to fallback into:
1189 * extern long sys_execve(const char *name, char **argv,char **envp, struct pt_regs *regs)
1191 * do_sys_execve asm fallback arguments:
1192 * rdi: name, rsi: argv, rdx: envp, rcx: fake frame on the stack
1194 ENTRY(kernel_execve)
1200 movq %rax, RAX(%rsp)
1203 je int_ret_from_sys_call
1210 /* Call softirq on interrupt stack. Interrupts are off. */
1214 CFI_REL_OFFSET rbp,0
1216 CFI_DEF_CFA_REGISTER rbp
1217 incl PER_CPU_VAR(irq_count)
1218 cmove PER_CPU_VAR(irq_stack_ptr),%rsp
1219 push %rbp # backlink for old unwinder
1223 CFI_DEF_CFA_REGISTER rsp
1224 CFI_ADJUST_CFA_OFFSET -8
1225 decl PER_CPU_VAR(irq_count)
1231 zeroentry xen_hypervisor_callback xen_do_hypervisor_callback
1234 * A note on the "critical region" in our callback handler.
1235 * We want to avoid stacking callback handlers due to events occurring
1236 * during handling of the last event. To do this, we keep events disabled
1237 * until we've done all processing. HOWEVER, we must enable events before
1238 * popping the stack frame (can't be done atomically) and so it would still
1239 * be possible to get enough handler activations to overflow the stack.
1240 * Although unlikely, bugs of that kind are hard to track down, so we'd
1241 * like to avoid the possibility.
1242 * So, on entry to the handler we detect whether we interrupted an
1243 * existing activation in its critical region -- if so, we pop the current
1244 * activation and restart the handler using the previous one.
1246 ENTRY(xen_do_hypervisor_callback) # do_hypervisor_callback(struct *pt_regs)
1249 * Since we don't modify %rdi, evtchn_do_upall(struct *pt_regs) will
1250 * see the correct pointer to the pt_regs
1252 movq %rdi, %rsp # we don't return, adjust the stack frame
1255 11: incl PER_CPU_VAR(irq_count)
1257 CFI_DEF_CFA_REGISTER rbp
1258 cmovzq PER_CPU_VAR(irq_stack_ptr),%rsp
1259 pushq %rbp # backlink for old unwinder
1260 call xen_evtchn_do_upcall
1262 CFI_DEF_CFA_REGISTER rsp
1263 decl PER_CPU_VAR(irq_count)
1266 END(xen_do_hypervisor_callback)
1269 * Hypervisor uses this for application faults while it executes.
1270 * We get here for two reasons:
1271 * 1. Fault while reloading DS, ES, FS or GS
1272 * 2. Fault while executing IRET
1273 * Category 1 we do not need to fix up as Xen has already reloaded all segment
1274 * registers that could be reloaded and zeroed the others.
1275 * Category 2 we fix up by killing the current process. We cannot use the
1276 * normal Linux return path in this case because if we use the IRET hypercall
1277 * to pop the stack frame we end up in an infinite loop of failsafe callbacks.
1278 * We distinguish between categories by comparing each saved segment register
1279 * with its current contents: any discrepancy means we in category 1.
1281 ENTRY(xen_failsafe_callback)
1283 /*CFI_REL_OFFSET gs,GS*/
1284 /*CFI_REL_OFFSET fs,FS*/
1285 /*CFI_REL_OFFSET es,ES*/
1286 /*CFI_REL_OFFSET ds,DS*/
1287 CFI_REL_OFFSET r11,8
1288 CFI_REL_OFFSET rcx,0
1302 /* All segments match their saved values => Category 2 (Bad IRET). */
1308 CFI_ADJUST_CFA_OFFSET -0x30
1309 pushq_cfi $0 /* RIP */
1312 jmp general_protection
1314 1: /* Segment mismatch => Category 1 (Bad segment). Retry the IRET. */
1320 CFI_ADJUST_CFA_OFFSET -0x30
1321 pushq_cfi $-1 /* orig_ax = -1 => not a system call */
1325 END(xen_failsafe_callback)
1327 apicinterrupt XEN_HVM_EVTCHN_CALLBACK \
1328 xen_hvm_callback_vector xen_evtchn_do_upcall
1330 #endif /* CONFIG_XEN */
1333 * Some functions should be protected against kprobes
1335 .pushsection .kprobes.text, "ax"
1337 paranoidzeroentry_ist debug do_debug DEBUG_STACK
1338 paranoidzeroentry_ist int3 do_int3 DEBUG_STACK
1339 errorentry stack_segment do_stack_segment
1341 zeroentry xen_debug do_debug
1342 zeroentry xen_int3 do_int3
1343 errorentry xen_stack_segment do_stack_segment
1345 errorentry general_protection do_general_protection
1346 errorentry page_fault do_page_fault
1347 #ifdef CONFIG_KVM_GUEST
1348 errorentry async_page_fault do_async_page_fault
1350 #ifdef CONFIG_X86_MCE
1351 paranoidzeroentry machine_check *machine_check_vector(%rip)
1355 * "Paranoid" exit path from exception stack.
1356 * Paranoid because this is used by NMIs and cannot take
1357 * any kernel state for granted.
1358 * We don't do kernel preemption checks here, because only
1359 * NMI should be common and it does not enable IRQs and
1360 * cannot get reschedule ticks.
1362 * "trace" is 0 for the NMI handler only, because irq-tracing
1363 * is fundamentally NMI-unsafe. (we cannot change the soft and
1364 * hard flags at once, atomically)
1367 /* ebx: no swapgs flag */
1368 ENTRY(paranoid_exit)
1370 DISABLE_INTERRUPTS(CLBR_NONE)
1372 testl %ebx,%ebx /* swapgs needed? */
1373 jnz paranoid_restore
1375 jnz paranoid_userspace
1386 GET_THREAD_INFO(%rcx)
1387 movl TI_flags(%rcx),%ebx
1388 andl $_TIF_WORK_MASK,%ebx
1390 movq %rsp,%rdi /* &pt_regs */
1392 movq %rax,%rsp /* switch stack for scheduling */
1393 testl $_TIF_NEED_RESCHED,%ebx
1394 jnz paranoid_schedule
1395 movl %ebx,%edx /* arg3: thread flags */
1397 ENABLE_INTERRUPTS(CLBR_NONE)
1398 xorl %esi,%esi /* arg2: oldset */
1399 movq %rsp,%rdi /* arg1: &pt_regs */
1400 call do_notify_resume
1401 DISABLE_INTERRUPTS(CLBR_NONE)
1403 jmp paranoid_userspace
1406 ENABLE_INTERRUPTS(CLBR_ANY)
1408 DISABLE_INTERRUPTS(CLBR_ANY)
1410 jmp paranoid_userspace
1415 * Exception entry point. This expects an error code/orig_rax on the stack.
1416 * returns in "no swapgs flag" in %ebx.
1420 CFI_ADJUST_CFA_OFFSET 15*8
1421 /* oldrax contains error code */
1440 je error_kernelspace
1448 * There are two places in the kernel that can potentially fault with
1449 * usergs. Handle them here. B stepping K8s sometimes report a
1450 * truncated RIP for IRET exceptions returning to compat mode. Check
1451 * for these here too.
1455 leaq native_irq_return_iret(%rip),%rcx
1456 cmpq %rcx,RIP+8(%rsp)
1458 movl %ecx,%eax /* zero extend */
1459 cmpq %rax,RIP+8(%rsp)
1461 cmpq $gs_change,RIP+8(%rsp)
1466 /* Fix truncated RIP */
1467 movq %rcx,RIP+8(%rsp)
1475 decl %ebx /* Return to usergs */
1481 /* ebx: no swapgs flag (1: don't need swapgs, 0: need it) */
1486 DISABLE_INTERRUPTS(CLBR_NONE)
1488 GET_THREAD_INFO(%rcx)
1491 LOCKDEP_SYS_EXIT_IRQ
1492 movl TI_flags(%rcx),%edx
1493 movl $_TIF_WORK_MASK,%edi
1501 /* runs on exception stack */
1504 PARAVIRT_ADJUST_EXCEPTION_FRAME
1506 subq $ORIG_RAX-R15, %rsp
1507 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1510 /* paranoidentry do_nmi, 0; without TRACE_IRQS_OFF */
1514 #ifdef CONFIG_TRACE_IRQFLAGS
1515 /* paranoidexit; without TRACE_IRQS_OFF */
1516 /* ebx: no swapgs flag */
1517 DISABLE_INTERRUPTS(CLBR_NONE)
1518 testl %ebx,%ebx /* swapgs needed? */
1528 GET_THREAD_INFO(%rcx)
1529 movl TI_flags(%rcx),%ebx
1530 andl $_TIF_WORK_MASK,%ebx
1532 movq %rsp,%rdi /* &pt_regs */
1534 movq %rax,%rsp /* switch stack for scheduling */
1535 testl $_TIF_NEED_RESCHED,%ebx
1537 movl %ebx,%edx /* arg3: thread flags */
1538 ENABLE_INTERRUPTS(CLBR_NONE)
1539 xorl %esi,%esi /* arg2: oldset */
1540 movq %rsp,%rdi /* arg1: &pt_regs */
1541 call do_notify_resume
1542 DISABLE_INTERRUPTS(CLBR_NONE)
1545 ENABLE_INTERRUPTS(CLBR_ANY)
1547 DISABLE_INTERRUPTS(CLBR_ANY)
1556 ENTRY(ignore_sysret)
1564 * End of kprobes section