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>
52 #include <asm/irqflags.h>
53 #include <asm/paravirt.h>
54 #include <asm/ftrace.h>
56 /* Avoid __ASSEMBLER__'ifying <linux/audit.h> just for this. */
57 #include <linux/elf-em.h>
58 #define AUDIT_ARCH_X86_64 (EM_X86_64|__AUDIT_ARCH_64BIT|__AUDIT_ARCH_LE)
59 #define __AUDIT_ARCH_64BIT 0x80000000
60 #define __AUDIT_ARCH_LE 0x40000000
63 #ifdef CONFIG_FUNCTION_TRACER
64 #ifdef CONFIG_DYNAMIC_FTRACE
70 cmpl $0, function_trace_stop
77 subq $MCOUNT_INSN_SIZE, %rdi
85 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
86 .globl ftrace_graph_call
96 #else /* ! CONFIG_DYNAMIC_FTRACE */
98 cmpl $0, function_trace_stop
101 cmpq $ftrace_stub, ftrace_trace_function
104 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
105 cmpq $ftrace_stub, ftrace_graph_return
106 jnz ftrace_graph_caller
108 cmpq $ftrace_graph_entry_stub, ftrace_graph_entry
109 jnz ftrace_graph_caller
119 movq 0x38(%rsp), %rdi
121 subq $MCOUNT_INSN_SIZE, %rdi
123 call *ftrace_trace_function
129 #endif /* CONFIG_DYNAMIC_FTRACE */
130 #endif /* CONFIG_FUNCTION_TRACER */
132 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
133 ENTRY(ftrace_graph_caller)
134 cmpl $0, function_trace_stop
140 movq 0x38(%rsp), %rsi
141 subq $MCOUNT_INSN_SIZE, %rsi
143 call prepare_ftrace_return
148 END(ftrace_graph_caller)
151 .globl return_to_handler
165 call ftrace_return_to_handler
182 #ifndef CONFIG_PREEMPT
183 #define retint_kernel retint_restore_args
186 #ifdef CONFIG_PARAVIRT
187 ENTRY(native_usergs_sysret64)
190 #endif /* CONFIG_PARAVIRT */
193 .macro TRACE_IRQS_IRETQ offset=ARGOFFSET
194 #ifdef CONFIG_TRACE_IRQFLAGS
195 bt $9,EFLAGS-\offset(%rsp) /* interrupts off? */
203 * C code is not supposed to know about undefined top of stack. Every time
204 * a C function with an pt_regs argument is called from the SYSCALL based
205 * fast path FIXUP_TOP_OF_STACK is needed.
206 * RESTORE_TOP_OF_STACK syncs the syscall state after any possible ptregs
210 /* %rsp:at FRAMEEND */
211 .macro FIXUP_TOP_OF_STACK tmp offset=0
212 movq %gs:pda_oldrsp,\tmp
213 movq \tmp,RSP+\offset(%rsp)
214 movq $__USER_DS,SS+\offset(%rsp)
215 movq $__USER_CS,CS+\offset(%rsp)
216 movq $-1,RCX+\offset(%rsp)
217 movq R11+\offset(%rsp),\tmp /* get eflags */
218 movq \tmp,EFLAGS+\offset(%rsp)
221 .macro RESTORE_TOP_OF_STACK tmp offset=0
222 movq RSP+\offset(%rsp),\tmp
223 movq \tmp,%gs:pda_oldrsp
224 movq EFLAGS+\offset(%rsp),\tmp
225 movq \tmp,R11+\offset(%rsp)
228 .macro FAKE_STACK_FRAME child_rip
229 /* push in order ss, rsp, eflags, cs, rip */
231 pushq $__KERNEL_DS /* ss */
232 CFI_ADJUST_CFA_OFFSET 8
233 /*CFI_REL_OFFSET ss,0*/
235 CFI_ADJUST_CFA_OFFSET 8
237 pushq $X86_EFLAGS_IF /* eflags - interrupts on */
238 CFI_ADJUST_CFA_OFFSET 8
239 /*CFI_REL_OFFSET rflags,0*/
240 pushq $__KERNEL_CS /* cs */
241 CFI_ADJUST_CFA_OFFSET 8
242 /*CFI_REL_OFFSET cs,0*/
243 pushq \child_rip /* rip */
244 CFI_ADJUST_CFA_OFFSET 8
246 pushq %rax /* orig rax */
247 CFI_ADJUST_CFA_OFFSET 8
250 .macro UNFAKE_STACK_FRAME
252 CFI_ADJUST_CFA_OFFSET -(6*8)
256 * initial frame state for interrupts (and exceptions without error code)
258 .macro EMPTY_FRAME start=1 offset=0
262 CFI_DEF_CFA rsp,8+\offset
264 CFI_DEF_CFA_OFFSET 8+\offset
269 * initial frame state for interrupts (and exceptions without error code)
271 .macro INTR_FRAME start=1 offset=0
272 EMPTY_FRAME \start, SS+8+\offset-RIP
273 /*CFI_REL_OFFSET ss, SS+\offset-RIP*/
274 CFI_REL_OFFSET rsp, RSP+\offset-RIP
275 /*CFI_REL_OFFSET rflags, EFLAGS+\offset-RIP*/
276 /*CFI_REL_OFFSET cs, CS+\offset-RIP*/
277 CFI_REL_OFFSET rip, RIP+\offset-RIP
281 * initial frame state for exceptions with error code (and interrupts
282 * with vector already pushed)
284 .macro XCPT_FRAME start=1 offset=0
285 INTR_FRAME \start, RIP+\offset-ORIG_RAX
286 /*CFI_REL_OFFSET orig_rax, ORIG_RAX-ORIG_RAX*/
290 * frame that enables calling into C.
292 .macro PARTIAL_FRAME start=1 offset=0
293 XCPT_FRAME \start, ORIG_RAX+\offset-ARGOFFSET
294 CFI_REL_OFFSET rdi, RDI+\offset-ARGOFFSET
295 CFI_REL_OFFSET rsi, RSI+\offset-ARGOFFSET
296 CFI_REL_OFFSET rdx, RDX+\offset-ARGOFFSET
297 CFI_REL_OFFSET rcx, RCX+\offset-ARGOFFSET
298 CFI_REL_OFFSET rax, RAX+\offset-ARGOFFSET
299 CFI_REL_OFFSET r8, R8+\offset-ARGOFFSET
300 CFI_REL_OFFSET r9, R9+\offset-ARGOFFSET
301 CFI_REL_OFFSET r10, R10+\offset-ARGOFFSET
302 CFI_REL_OFFSET r11, R11+\offset-ARGOFFSET
306 * frame that enables passing a complete pt_regs to a C function.
308 .macro DEFAULT_FRAME start=1 offset=0
309 PARTIAL_FRAME \start, R11+\offset-R15
310 CFI_REL_OFFSET rbx, RBX+\offset
311 CFI_REL_OFFSET rbp, RBP+\offset
312 CFI_REL_OFFSET r12, R12+\offset
313 CFI_REL_OFFSET r13, R13+\offset
314 CFI_REL_OFFSET r14, R14+\offset
315 CFI_REL_OFFSET r15, R15+\offset
318 /* save partial stack frame */
322 movq_cfi rdi, RDI+16-ARGOFFSET
323 movq_cfi rsi, RSI+16-ARGOFFSET
324 movq_cfi rdx, RDX+16-ARGOFFSET
325 movq_cfi rcx, RCX+16-ARGOFFSET
326 movq_cfi rax, RAX+16-ARGOFFSET
327 movq_cfi r8, R8+16-ARGOFFSET
328 movq_cfi r9, R9+16-ARGOFFSET
329 movq_cfi r10, R10+16-ARGOFFSET
330 movq_cfi r11, R11+16-ARGOFFSET
332 leaq -ARGOFFSET+16(%rsp),%rdi /* arg1 for handler */
333 movq_cfi rbp, 8 /* push %rbp */
334 leaq 8(%rsp), %rbp /* mov %rsp, %ebp */
339 * irqcount is used to check if a CPU is already on an interrupt stack
340 * or not. While this is essentially redundant with preempt_count it is
341 * a little cheaper to use a separate counter in the PDA (short of
342 * moving irq_enter into assembly, which would be too much work)
344 1: incl %gs:pda_irqcount
346 popq_cfi %rax /* move return address... */
347 mov %gs:pda_irqstackptr,%rsp
349 pushq_cfi %rbp /* backlink for unwinder */
350 pushq_cfi %rax /* ... to the new stack */
352 * We entered an interrupt context - irqs are off:
360 PARTIAL_FRAME 1 REST_SKIP+8
361 movq 5*8+16(%rsp), %r11 /* save return address */
368 movq %r11, 8(%rsp) /* return address */
369 FIXUP_TOP_OF_STACK %r11, 16
374 /* save complete stack frame */
394 movl $MSR_GS_BASE,%ecx
397 js 1f /* negative -> in kernel */
405 * A newly forked process directly context switches into this address.
407 * rdi: prev task we switched from
412 push kernel_eflags(%rip)
413 CFI_ADJUST_CFA_OFFSET 8
414 popf # reset kernel eflags
415 CFI_ADJUST_CFA_OFFSET -8
417 call schedule_tail # rdi: 'prev' task parameter
419 GET_THREAD_INFO(%rcx)
424 testl $3, CS-ARGOFFSET(%rsp) # from kernel_thread?
425 je int_ret_from_sys_call
427 testl $_TIF_IA32, TI_flags(%rcx) # 32-bit compat task needs IRET
428 jnz int_ret_from_sys_call
430 RESTORE_TOP_OF_STACK %rdi, -ARGOFFSET
431 jmp ret_from_sys_call # go to the SYSRET fastpath
438 * System call entry. Upto 6 arguments in registers are supported.
440 * SYSCALL does not save anything on the stack and does not change the
446 * rax system call number
448 * rcx return address for syscall/sysret, C arg3
451 * r10 arg3 (--> moved to rcx for C)
454 * r11 eflags for syscall/sysret, temporary for C
455 * r12-r15,rbp,rbx saved by C code, not touched.
457 * Interrupts are off on entry.
458 * Only called from user space.
460 * XXX if we had a free scratch register we could save the RSP into the stack frame
461 * and report it properly in ps. Unfortunately we haven't.
463 * When user can change the frames always force IRET. That is because
464 * it deals with uncanonical addresses better. SYSRET has trouble
465 * with them due to bugs in both AMD and Intel CPUs.
471 CFI_DEF_CFA rsp,PDA_STACKOFFSET
473 /*CFI_REGISTER rflags,r11*/
476 * A hypervisor implementation might want to use a label
477 * after the swapgs, so that it can do the swapgs
478 * for the guest and jump here on syscall.
480 ENTRY(system_call_after_swapgs)
482 movq %rsp,%gs:pda_oldrsp
483 movq %gs:pda_kernelstack,%rsp
485 * No need to follow this irqs off/on section - it's straight
488 ENABLE_INTERRUPTS(CLBR_NONE)
490 movq %rax,ORIG_RAX-ARGOFFSET(%rsp)
491 movq %rcx,RIP-ARGOFFSET(%rsp)
492 CFI_REL_OFFSET rip,RIP-ARGOFFSET
493 GET_THREAD_INFO(%rcx)
494 testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%rcx)
496 system_call_fastpath:
497 cmpq $__NR_syscall_max,%rax
500 call *sys_call_table(,%rax,8) # XXX: rip relative
501 movq %rax,RAX-ARGOFFSET(%rsp)
503 * Syscall return path ending with SYSRET (fast path)
504 * Has incomplete stack frame and undefined top of stack.
507 movl $_TIF_ALLWORK_MASK,%edi
511 GET_THREAD_INFO(%rcx)
512 DISABLE_INTERRUPTS(CLBR_NONE)
514 movl TI_flags(%rcx),%edx
519 * sysretq will re-enable interrupts:
522 movq RIP-ARGOFFSET(%rsp),%rcx
524 RESTORE_ARGS 0,-ARG_SKIP,1
525 /*CFI_REGISTER rflags,r11*/
526 movq %gs:pda_oldrsp, %rsp
530 /* Handle reschedules */
531 /* edx: work, edi: workmask */
533 bt $TIF_NEED_RESCHED,%edx
536 ENABLE_INTERRUPTS(CLBR_NONE)
538 CFI_ADJUST_CFA_OFFSET 8
541 CFI_ADJUST_CFA_OFFSET -8
544 /* Handle a signal */
547 ENABLE_INTERRUPTS(CLBR_NONE)
548 #ifdef CONFIG_AUDITSYSCALL
549 bt $TIF_SYSCALL_AUDIT,%edx
552 /* edx: work flags (arg3) */
553 leaq -ARGOFFSET(%rsp),%rdi # &pt_regs -> arg1
554 xorl %esi,%esi # oldset -> arg2
556 FIXUP_TOP_OF_STACK %r11
557 call do_notify_resume
558 RESTORE_TOP_OF_STACK %r11
560 movl $_TIF_WORK_MASK,%edi
561 /* Use IRET because user could have changed frame. This
562 works because ptregscall_common has called FIXUP_TOP_OF_STACK. */
563 DISABLE_INTERRUPTS(CLBR_NONE)
568 movq $-ENOSYS,RAX-ARGOFFSET(%rsp)
569 jmp ret_from_sys_call
571 #ifdef CONFIG_AUDITSYSCALL
573 * Fast path for syscall audit without full syscall trace.
574 * We just call audit_syscall_entry() directly, and then
575 * jump back to the normal fast path.
578 movq %r10,%r9 /* 6th arg: 4th syscall arg */
579 movq %rdx,%r8 /* 5th arg: 3rd syscall arg */
580 movq %rsi,%rcx /* 4th arg: 2nd syscall arg */
581 movq %rdi,%rdx /* 3rd arg: 1st syscall arg */
582 movq %rax,%rsi /* 2nd arg: syscall number */
583 movl $AUDIT_ARCH_X86_64,%edi /* 1st arg: audit arch */
584 call audit_syscall_entry
585 LOAD_ARGS 0 /* reload call-clobbered registers */
586 jmp system_call_fastpath
589 * Return fast path for syscall audit. Call audit_syscall_exit()
590 * directly and then jump back to the fast path with TIF_SYSCALL_AUDIT
594 movq %rax,%rsi /* second arg, syscall return value */
595 cmpq $0,%rax /* is it < 0? */
596 setl %al /* 1 if so, 0 if not */
597 movzbl %al,%edi /* zero-extend that into %edi */
598 inc %edi /* first arg, 0->1(AUDITSC_SUCCESS), 1->2(AUDITSC_FAILURE) */
599 call audit_syscall_exit
600 movl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT),%edi
602 #endif /* CONFIG_AUDITSYSCALL */
604 /* Do syscall tracing */
606 #ifdef CONFIG_AUDITSYSCALL
607 testl $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags(%rcx)
611 movq $-ENOSYS,RAX(%rsp) /* ptrace can change this for a bad syscall */
612 FIXUP_TOP_OF_STACK %rdi
614 call syscall_trace_enter
616 * Reload arg registers from stack in case ptrace changed them.
617 * We don't reload %rax because syscall_trace_enter() returned
618 * the value it wants us to use in the table lookup.
620 LOAD_ARGS ARGOFFSET, 1
622 cmpq $__NR_syscall_max,%rax
623 ja int_ret_from_sys_call /* RAX(%rsp) set to -ENOSYS above */
624 movq %r10,%rcx /* fixup for C */
625 call *sys_call_table(,%rax,8)
626 movq %rax,RAX-ARGOFFSET(%rsp)
627 /* Use IRET because user could have changed frame */
630 * Syscall return path ending with IRET.
631 * Has correct top of stack, but partial stack frame.
633 .globl int_ret_from_sys_call
634 .globl int_with_check
635 int_ret_from_sys_call:
636 DISABLE_INTERRUPTS(CLBR_NONE)
638 testl $3,CS-ARGOFFSET(%rsp)
639 je retint_restore_args
640 movl $_TIF_ALLWORK_MASK,%edi
641 /* edi: mask to check */
644 GET_THREAD_INFO(%rcx)
645 movl TI_flags(%rcx),%edx
648 andl $~TS_COMPAT,TI_status(%rcx)
651 /* Either reschedule or signal or syscall exit tracking needed. */
652 /* First do a reschedule test. */
653 /* edx: work, edi: workmask */
655 bt $TIF_NEED_RESCHED,%edx
658 ENABLE_INTERRUPTS(CLBR_NONE)
660 CFI_ADJUST_CFA_OFFSET 8
663 CFI_ADJUST_CFA_OFFSET -8
664 DISABLE_INTERRUPTS(CLBR_NONE)
668 /* handle signals and tracing -- both require a full stack frame */
671 ENABLE_INTERRUPTS(CLBR_NONE)
673 /* Check for syscall exit trace */
674 testl $_TIF_WORK_SYSCALL_EXIT,%edx
677 CFI_ADJUST_CFA_OFFSET 8
678 leaq 8(%rsp),%rdi # &ptregs -> arg1
679 call syscall_trace_leave
681 CFI_ADJUST_CFA_OFFSET -8
682 andl $~(_TIF_WORK_SYSCALL_EXIT|_TIF_SYSCALL_EMU),%edi
686 testl $_TIF_DO_NOTIFY_MASK,%edx
688 movq %rsp,%rdi # &ptregs -> arg1
689 xorl %esi,%esi # oldset -> arg2
690 call do_notify_resume
691 1: movl $_TIF_WORK_MASK,%edi
694 DISABLE_INTERRUPTS(CLBR_NONE)
701 * Certain special system calls that need to save a complete full stack frame.
703 .macro PTREGSCALL label,func,arg
705 PARTIAL_FRAME 1 8 /* offset 8: return address */
706 subq $REST_SKIP, %rsp
707 CFI_ADJUST_CFA_OFFSET REST_SKIP
709 DEFAULT_FRAME 0 8 /* offset 8: return address */
710 leaq 8(%rsp), \arg /* pt_regs pointer */
712 jmp ptregscall_common
717 PTREGSCALL stub_clone, sys_clone, %r8
718 PTREGSCALL stub_fork, sys_fork, %rdi
719 PTREGSCALL stub_vfork, sys_vfork, %rdi
720 PTREGSCALL stub_sigaltstack, sys_sigaltstack, %rdx
721 PTREGSCALL stub_iopl, sys_iopl, %rsi
723 ENTRY(ptregscall_common)
724 DEFAULT_FRAME 1 8 /* offset 8: return address */
725 RESTORE_TOP_OF_STACK %r11, 8
726 movq_cfi_restore R15+8, r15
727 movq_cfi_restore R14+8, r14
728 movq_cfi_restore R13+8, r13
729 movq_cfi_restore R12+8, r12
730 movq_cfi_restore RBP+8, rbp
731 movq_cfi_restore RBX+8, rbx
732 ret $REST_SKIP /* pop extended registers */
734 END(ptregscall_common)
739 CFI_ADJUST_CFA_OFFSET -8
740 CFI_REGISTER rip, r11
742 FIXUP_TOP_OF_STACK %r11
745 RESTORE_TOP_OF_STACK %r11
748 jmp int_ret_from_sys_call
753 * sigreturn is special because it needs to restore all registers on return.
754 * This cannot be done with SYSRET, so use the IRET return path instead.
756 ENTRY(stub_rt_sigreturn)
759 CFI_ADJUST_CFA_OFFSET -8
762 FIXUP_TOP_OF_STACK %r11
763 call sys_rt_sigreturn
764 movq %rax,RAX(%rsp) # fixme, this could be done at the higher layer
766 jmp int_ret_from_sys_call
768 END(stub_rt_sigreturn)
771 * Build the entry stubs and pointer table with some assembler magic.
772 * We pack 7 stubs into a single 32-byte chunk, which will fit in a
773 * single cache line on all modern x86 implementations.
775 .section .init.rodata,"a"
779 .p2align CONFIG_X86_L1_CACHE_SHIFT
780 ENTRY(irq_entries_start)
782 vector=FIRST_EXTERNAL_VECTOR
783 .rept (NR_VECTORS-FIRST_EXTERNAL_VECTOR+6)/7
786 .if vector < NR_VECTORS
787 .if vector <> FIRST_EXTERNAL_VECTOR
788 CFI_ADJUST_CFA_OFFSET -8
790 1: pushq $(~vector+0x80) /* Note: always in signed byte range */
791 CFI_ADJUST_CFA_OFFSET 8
792 .if ((vector-FIRST_EXTERNAL_VECTOR)%7) <> 6
801 2: jmp common_interrupt
804 END(irq_entries_start)
811 * Interrupt entry/exit.
813 * Interrupt entry points save only callee clobbered registers in fast path.
815 * Entry runs with interrupts off.
818 /* 0(%rsp): ~(interrupt number) */
819 .macro interrupt func
821 CFI_ADJUST_CFA_OFFSET 10*8
828 * The interrupt stubs push (~vector+0x80) onto the stack and
829 * then jump to common_interrupt.
831 .p2align CONFIG_X86_L1_CACHE_SHIFT
834 addq $-0x80,(%rsp) /* Adjust vector to [-256,-1] range */
836 /* 0(%rsp): oldrsp-ARGOFFSET */
838 DISABLE_INTERRUPTS(CLBR_NONE)
840 decl %gs:pda_irqcount
842 CFI_DEF_CFA_REGISTER rsp
843 CFI_ADJUST_CFA_OFFSET -8
845 GET_THREAD_INFO(%rcx)
846 testl $3,CS-ARGOFFSET(%rsp)
849 /* Interrupt came from user space */
851 * Has a correct top of stack, but a partial stack frame
852 * %rcx: thread info. Interrupts off.
854 retint_with_reschedule:
855 movl $_TIF_WORK_MASK,%edi
858 movl TI_flags(%rcx),%edx
863 retint_swapgs: /* return to user-space */
865 * The iretq could re-enable interrupts:
867 DISABLE_INTERRUPTS(CLBR_ANY)
872 retint_restore_args: /* return to kernel space */
873 DISABLE_INTERRUPTS(CLBR_ANY)
875 * The iretq could re-enable interrupts:
884 .section __ex_table, "a"
885 .quad irq_return, bad_iret
888 #ifdef CONFIG_PARAVIRT
892 .section __ex_table,"a"
893 .quad native_iret, bad_iret
900 * The iret traps when the %cs or %ss being restored is bogus.
901 * We've lost the original trap vector and error code.
902 * #GPF is the most likely one to get for an invalid selector.
903 * So pretend we completed the iret and took the #GPF in user mode.
905 * We are now running with the kernel GS after exception recovery.
906 * But error_entry expects us to have user GS to match the user %cs,
912 jmp general_protection
916 /* edi: workmask, edx: work */
919 bt $TIF_NEED_RESCHED,%edx
922 ENABLE_INTERRUPTS(CLBR_NONE)
924 CFI_ADJUST_CFA_OFFSET 8
927 CFI_ADJUST_CFA_OFFSET -8
928 GET_THREAD_INFO(%rcx)
929 DISABLE_INTERRUPTS(CLBR_NONE)
934 testl $_TIF_DO_NOTIFY_MASK,%edx
937 ENABLE_INTERRUPTS(CLBR_NONE)
939 movq $-1,ORIG_RAX(%rsp)
940 xorl %esi,%esi # oldset
941 movq %rsp,%rdi # &pt_regs
942 call do_notify_resume
944 DISABLE_INTERRUPTS(CLBR_NONE)
946 GET_THREAD_INFO(%rcx)
947 jmp retint_with_reschedule
949 #ifdef CONFIG_PREEMPT
950 /* Returning to kernel space. Check if we need preemption */
951 /* rcx: threadinfo. interrupts off. */
953 cmpl $0,TI_preempt_count(%rcx)
954 jnz retint_restore_args
955 bt $TIF_NEED_RESCHED,TI_flags(%rcx)
956 jnc retint_restore_args
957 bt $9,EFLAGS-ARGOFFSET(%rsp) /* interrupts off? */
958 jnc retint_restore_args
959 call preempt_schedule_irq
964 END(common_interrupt)
969 .macro apicinterrupt num sym do_sym
973 CFI_ADJUST_CFA_OFFSET 8
981 apicinterrupt IRQ_MOVE_CLEANUP_VECTOR \
982 irq_move_cleanup_interrupt smp_irq_move_cleanup_interrupt
985 apicinterrupt UV_BAU_MESSAGE \
986 uv_bau_message_intr1 uv_bau_message_interrupt
987 apicinterrupt LOCAL_TIMER_VECTOR \
988 apic_timer_interrupt smp_apic_timer_interrupt
991 apicinterrupt INVALIDATE_TLB_VECTOR_START+0 \
992 invalidate_interrupt0 smp_invalidate_interrupt
993 apicinterrupt INVALIDATE_TLB_VECTOR_START+1 \
994 invalidate_interrupt1 smp_invalidate_interrupt
995 apicinterrupt INVALIDATE_TLB_VECTOR_START+2 \
996 invalidate_interrupt2 smp_invalidate_interrupt
997 apicinterrupt INVALIDATE_TLB_VECTOR_START+3 \
998 invalidate_interrupt3 smp_invalidate_interrupt
999 apicinterrupt INVALIDATE_TLB_VECTOR_START+4 \
1000 invalidate_interrupt4 smp_invalidate_interrupt
1001 apicinterrupt INVALIDATE_TLB_VECTOR_START+5 \
1002 invalidate_interrupt5 smp_invalidate_interrupt
1003 apicinterrupt INVALIDATE_TLB_VECTOR_START+6 \
1004 invalidate_interrupt6 smp_invalidate_interrupt
1005 apicinterrupt INVALIDATE_TLB_VECTOR_START+7 \
1006 invalidate_interrupt7 smp_invalidate_interrupt
1009 apicinterrupt THRESHOLD_APIC_VECTOR \
1010 threshold_interrupt mce_threshold_interrupt
1011 apicinterrupt THERMAL_APIC_VECTOR \
1012 thermal_interrupt smp_thermal_interrupt
1015 apicinterrupt CALL_FUNCTION_SINGLE_VECTOR \
1016 call_function_single_interrupt smp_call_function_single_interrupt
1017 apicinterrupt CALL_FUNCTION_VECTOR \
1018 call_function_interrupt smp_call_function_interrupt
1019 apicinterrupt RESCHEDULE_VECTOR \
1020 reschedule_interrupt smp_reschedule_interrupt
1023 apicinterrupt ERROR_APIC_VECTOR \
1024 error_interrupt smp_error_interrupt
1025 apicinterrupt SPURIOUS_APIC_VECTOR \
1026 spurious_interrupt smp_spurious_interrupt
1029 * Exception entry points.
1031 .macro zeroentry sym do_sym
1034 PARAVIRT_ADJUST_EXCEPTION_FRAME
1035 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1037 CFI_ADJUST_CFA_OFFSET 15*8
1040 movq %rsp,%rdi /* pt_regs pointer */
1041 xorl %esi,%esi /* no error code */
1043 jmp error_exit /* %ebx: no swapgs flag */
1048 .macro paranoidzeroentry sym do_sym
1051 PARAVIRT_ADJUST_EXCEPTION_FRAME
1052 pushq $-1 /* ORIG_RAX: no syscall to restart */
1053 CFI_ADJUST_CFA_OFFSET 8
1057 movq %rsp,%rdi /* pt_regs pointer */
1058 xorl %esi,%esi /* no error code */
1060 jmp paranoid_exit /* %ebx: no swapgs flag */
1065 .macro paranoidzeroentry_ist sym do_sym ist
1068 PARAVIRT_ADJUST_EXCEPTION_FRAME
1069 pushq $-1 /* ORIG_RAX: no syscall to restart */
1070 CFI_ADJUST_CFA_OFFSET 8
1074 movq %rsp,%rdi /* pt_regs pointer */
1075 xorl %esi,%esi /* no error code */
1076 movq %gs:pda_data_offset, %rbp
1077 subq $EXCEPTION_STKSZ, per_cpu__init_tss + TSS_ist + (\ist - 1) * 8(%rbp)
1079 addq $EXCEPTION_STKSZ, per_cpu__init_tss + TSS_ist + (\ist - 1) * 8(%rbp)
1080 jmp paranoid_exit /* %ebx: no swapgs flag */
1085 .macro errorentry sym do_sym
1088 PARAVIRT_ADJUST_EXCEPTION_FRAME
1090 CFI_ADJUST_CFA_OFFSET 15*8
1093 movq %rsp,%rdi /* pt_regs pointer */
1094 movq ORIG_RAX(%rsp),%rsi /* get error code */
1095 movq $-1,ORIG_RAX(%rsp) /* no syscall to restart */
1097 jmp error_exit /* %ebx: no swapgs flag */
1102 /* error code is on the stack already */
1103 .macro paranoiderrorentry sym do_sym
1106 PARAVIRT_ADJUST_EXCEPTION_FRAME
1108 CFI_ADJUST_CFA_OFFSET 15*8
1112 movq %rsp,%rdi /* pt_regs pointer */
1113 movq ORIG_RAX(%rsp),%rsi /* get error code */
1114 movq $-1,ORIG_RAX(%rsp) /* no syscall to restart */
1116 jmp paranoid_exit /* %ebx: no swapgs flag */
1121 zeroentry divide_error do_divide_error
1122 zeroentry overflow do_overflow
1123 zeroentry bounds do_bounds
1124 zeroentry invalid_op do_invalid_op
1125 zeroentry device_not_available do_device_not_available
1126 paranoiderrorentry double_fault do_double_fault
1127 zeroentry coprocessor_segment_overrun do_coprocessor_segment_overrun
1128 errorentry invalid_TSS do_invalid_TSS
1129 errorentry segment_not_present do_segment_not_present
1130 zeroentry spurious_interrupt_bug do_spurious_interrupt_bug
1131 zeroentry coprocessor_error do_coprocessor_error
1132 errorentry alignment_check do_alignment_check
1133 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 CFI_ADJUST_CFA_OFFSET 8
1141 DISABLE_INTERRUPTS(CLBR_ANY | ~(CLBR_RDI))
1145 2: mfence /* workaround */
1148 CFI_ADJUST_CFA_OFFSET -8
1151 END(native_load_gs_index)
1153 .section __ex_table,"a"
1155 .quad gs_change,bad_gs
1157 .section .fixup,"ax"
1158 /* running with kernelgs */
1160 SWAPGS /* switch back to user gs */
1167 * Create a kernel thread.
1169 * C extern interface:
1170 * extern long kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
1172 * asm input arguments:
1173 * rdi: fn, rsi: arg, rdx: flags
1175 ENTRY(kernel_thread)
1177 FAKE_STACK_FRAME $child_rip
1180 # rdi: flags, rsi: usp, rdx: will be &pt_regs
1182 orq kernel_thread_flags(%rip),%rdi
1195 * It isn't worth to check for reschedule here,
1196 * so internally to the x86_64 port you can rely on kernel_thread()
1197 * not to reschedule the child before returning, this avoids the need
1198 * of hacks for example to fork off the per-CPU idle tasks.
1199 * [Hopefully no generic code relies on the reschedule -AK]
1208 pushq $0 # fake return address
1211 * Here we are in the child and the registers are set as they were
1212 * at kernel_thread() invocation in the parent.
1220 ud2 # padding for call trace
1225 * execve(). This function needs to use IRET, not SYSRET, to set up all state properly.
1227 * C extern interface:
1228 * extern long execve(char *name, char **argv, char **envp)
1230 * asm input arguments:
1231 * rdi: name, rsi: argv, rdx: envp
1233 * We want to fallback into:
1234 * extern long sys_execve(char *name, char **argv,char **envp, struct pt_regs *regs)
1236 * do_sys_execve asm fallback arguments:
1237 * rdi: name, rsi: argv, rdx: envp, rcx: fake frame on the stack
1239 ENTRY(kernel_execve)
1245 movq %rax, RAX(%rsp)
1248 je int_ret_from_sys_call
1255 /* Call softirq on interrupt stack. Interrupts are off. */
1259 CFI_ADJUST_CFA_OFFSET 8
1260 CFI_REL_OFFSET rbp,0
1262 CFI_DEF_CFA_REGISTER rbp
1263 incl %gs:pda_irqcount
1264 cmove %gs:pda_irqstackptr,%rsp
1265 push %rbp # backlink for old unwinder
1268 CFI_DEF_CFA_REGISTER rsp
1269 CFI_ADJUST_CFA_OFFSET -8
1270 decl %gs:pda_irqcount
1276 zeroentry xen_hypervisor_callback xen_do_hypervisor_callback
1279 * A note on the "critical region" in our callback handler.
1280 * We want to avoid stacking callback handlers due to events occurring
1281 * during handling of the last event. To do this, we keep events disabled
1282 * until we've done all processing. HOWEVER, we must enable events before
1283 * popping the stack frame (can't be done atomically) and so it would still
1284 * be possible to get enough handler activations to overflow the stack.
1285 * Although unlikely, bugs of that kind are hard to track down, so we'd
1286 * like to avoid the possibility.
1287 * So, on entry to the handler we detect whether we interrupted an
1288 * existing activation in its critical region -- if so, we pop the current
1289 * activation and restart the handler using the previous one.
1291 ENTRY(xen_do_hypervisor_callback) # do_hypervisor_callback(struct *pt_regs)
1294 * Since we don't modify %rdi, evtchn_do_upall(struct *pt_regs) will
1295 * see the correct pointer to the pt_regs
1297 movq %rdi, %rsp # we don't return, adjust the stack frame
1300 11: incl %gs:pda_irqcount
1302 CFI_DEF_CFA_REGISTER rbp
1303 cmovzq %gs:pda_irqstackptr,%rsp
1304 pushq %rbp # backlink for old unwinder
1305 call xen_evtchn_do_upcall
1307 CFI_DEF_CFA_REGISTER rsp
1308 decl %gs:pda_irqcount
1311 END(do_hypervisor_callback)
1314 * Hypervisor uses this for application faults while it executes.
1315 * We get here for two reasons:
1316 * 1. Fault while reloading DS, ES, FS or GS
1317 * 2. Fault while executing IRET
1318 * Category 1 we do not need to fix up as Xen has already reloaded all segment
1319 * registers that could be reloaded and zeroed the others.
1320 * Category 2 we fix up by killing the current process. We cannot use the
1321 * normal Linux return path in this case because if we use the IRET hypercall
1322 * to pop the stack frame we end up in an infinite loop of failsafe callbacks.
1323 * We distinguish between categories by comparing each saved segment register
1324 * with its current contents: any discrepancy means we in category 1.
1326 ENTRY(xen_failsafe_callback)
1328 /*CFI_REL_OFFSET gs,GS*/
1329 /*CFI_REL_OFFSET fs,FS*/
1330 /*CFI_REL_OFFSET es,ES*/
1331 /*CFI_REL_OFFSET ds,DS*/
1332 CFI_REL_OFFSET r11,8
1333 CFI_REL_OFFSET rcx,0
1347 /* All segments match their saved values => Category 2 (Bad IRET). */
1353 CFI_ADJUST_CFA_OFFSET -0x30
1354 pushq_cfi $0 /* RIP */
1357 jmp general_protection
1359 1: /* Segment mismatch => Category 1 (Bad segment). Retry the IRET. */
1365 CFI_ADJUST_CFA_OFFSET -0x30
1370 END(xen_failsafe_callback)
1372 #endif /* CONFIG_XEN */
1375 * Some functions should be protected against kprobes
1377 .pushsection .kprobes.text, "ax"
1379 paranoidzeroentry_ist debug do_debug DEBUG_STACK
1380 paranoidzeroentry_ist int3 do_int3 DEBUG_STACK
1381 paranoiderrorentry stack_segment do_stack_segment
1382 errorentry general_protection do_general_protection
1383 errorentry page_fault do_page_fault
1384 #ifdef CONFIG_X86_MCE
1385 paranoidzeroentry machine_check do_machine_check
1389 * "Paranoid" exit path from exception stack.
1390 * Paranoid because this is used by NMIs and cannot take
1391 * any kernel state for granted.
1392 * We don't do kernel preemption checks here, because only
1393 * NMI should be common and it does not enable IRQs and
1394 * cannot get reschedule ticks.
1396 * "trace" is 0 for the NMI handler only, because irq-tracing
1397 * is fundamentally NMI-unsafe. (we cannot change the soft and
1398 * hard flags at once, atomically)
1401 /* ebx: no swapgs flag */
1402 ENTRY(paranoid_exit)
1404 DISABLE_INTERRUPTS(CLBR_NONE)
1406 testl %ebx,%ebx /* swapgs needed? */
1407 jnz paranoid_restore
1409 jnz paranoid_userspace
1417 GET_THREAD_INFO(%rcx)
1418 movl TI_flags(%rcx),%ebx
1419 andl $_TIF_WORK_MASK,%ebx
1421 movq %rsp,%rdi /* &pt_regs */
1423 movq %rax,%rsp /* switch stack for scheduling */
1424 testl $_TIF_NEED_RESCHED,%ebx
1425 jnz paranoid_schedule
1426 movl %ebx,%edx /* arg3: thread flags */
1428 ENABLE_INTERRUPTS(CLBR_NONE)
1429 xorl %esi,%esi /* arg2: oldset */
1430 movq %rsp,%rdi /* arg1: &pt_regs */
1431 call do_notify_resume
1432 DISABLE_INTERRUPTS(CLBR_NONE)
1434 jmp paranoid_userspace
1437 ENABLE_INTERRUPTS(CLBR_ANY)
1439 DISABLE_INTERRUPTS(CLBR_ANY)
1441 jmp paranoid_userspace
1446 * Exception entry point. This expects an error code/orig_rax on the stack.
1447 * returns in "no swapgs flag" in %ebx.
1451 CFI_ADJUST_CFA_OFFSET 15*8
1452 /* oldrax contains error code */
1471 je error_kernelspace
1480 * There are two places in the kernel that can potentially fault with
1481 * usergs. Handle them here. The exception handlers after iret run with
1482 * kernel gs again, so don't set the user space flag. B stepping K8s
1483 * sometimes report an truncated RIP for IRET exceptions returning to
1484 * compat mode. Check for these here too.
1488 leaq irq_return(%rip),%rcx
1489 cmpq %rcx,RIP+8(%rsp)
1491 movl %ecx,%ecx /* zero extend */
1492 cmpq %rcx,RIP+8(%rsp)
1494 cmpq $gs_change,RIP+8(%rsp)
1500 /* ebx: no swapgs flag (1: don't need swapgs, 0: need it) */
1505 DISABLE_INTERRUPTS(CLBR_NONE)
1507 GET_THREAD_INFO(%rcx)
1510 LOCKDEP_SYS_EXIT_IRQ
1511 movl TI_flags(%rcx),%edx
1512 movl $_TIF_WORK_MASK,%edi
1520 /* runs on exception stack */
1523 PARAVIRT_ADJUST_EXCEPTION_FRAME
1526 CFI_ADJUST_CFA_OFFSET 15*8
1529 /* paranoidentry do_nmi, 0; without TRACE_IRQS_OFF */
1533 #ifdef CONFIG_TRACE_IRQFLAGS
1534 /* paranoidexit; without TRACE_IRQS_OFF */
1535 /* ebx: no swapgs flag */
1536 DISABLE_INTERRUPTS(CLBR_NONE)
1537 testl %ebx,%ebx /* swapgs needed? */
1547 GET_THREAD_INFO(%rcx)
1548 movl TI_flags(%rcx),%ebx
1549 andl $_TIF_WORK_MASK,%ebx
1551 movq %rsp,%rdi /* &pt_regs */
1553 movq %rax,%rsp /* switch stack for scheduling */
1554 testl $_TIF_NEED_RESCHED,%ebx
1556 movl %ebx,%edx /* arg3: thread flags */
1557 ENABLE_INTERRUPTS(CLBR_NONE)
1558 xorl %esi,%esi /* arg2: oldset */
1559 movq %rsp,%rdi /* arg1: &pt_regs */
1560 call do_notify_resume
1561 DISABLE_INTERRUPTS(CLBR_NONE)
1564 ENABLE_INTERRUPTS(CLBR_ANY)
1566 DISABLE_INTERRUPTS(CLBR_ANY)
1575 ENTRY(ignore_sysret)
1583 * End of kprobes section