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
71 /* taken from glibc */
83 subq $MCOUNT_INSN_SIZE, %rdi
103 #else /* ! CONFIG_DYNAMIC_FTRACE */
105 cmpq $ftrace_stub, ftrace_trace_function
112 /* taken from glibc */
122 movq 0x38(%rsp), %rdi
124 subq $MCOUNT_INSN_SIZE, %rdi
126 call *ftrace_trace_function
139 #endif /* CONFIG_DYNAMIC_FTRACE */
140 #endif /* CONFIG_FUNCTION_TRACER */
142 #ifndef CONFIG_PREEMPT
143 #define retint_kernel retint_restore_args
146 #ifdef CONFIG_PARAVIRT
147 ENTRY(native_usergs_sysret64)
150 #endif /* CONFIG_PARAVIRT */
153 .macro TRACE_IRQS_IRETQ offset=ARGOFFSET
154 #ifdef CONFIG_TRACE_IRQFLAGS
155 bt $9,EFLAGS-\offset(%rsp) /* interrupts off? */
163 * C code is not supposed to know about undefined top of stack. Every time
164 * a C function with an pt_regs argument is called from the SYSCALL based
165 * fast path FIXUP_TOP_OF_STACK is needed.
166 * RESTORE_TOP_OF_STACK syncs the syscall state after any possible ptregs
170 /* %rsp:at FRAMEEND */
171 .macro FIXUP_TOP_OF_STACK tmp offset=0
172 movq %gs:pda_oldrsp,\tmp
173 movq \tmp,RSP+\offset(%rsp)
174 movq $__USER_DS,SS+\offset(%rsp)
175 movq $__USER_CS,CS+\offset(%rsp)
176 movq $-1,RCX+\offset(%rsp)
177 movq R11+\offset(%rsp),\tmp /* get eflags */
178 movq \tmp,EFLAGS+\offset(%rsp)
181 .macro RESTORE_TOP_OF_STACK tmp offset=0
182 movq RSP+\offset(%rsp),\tmp
183 movq \tmp,%gs:pda_oldrsp
184 movq EFLAGS+\offset(%rsp),\tmp
185 movq \tmp,R11+\offset(%rsp)
188 .macro FAKE_STACK_FRAME child_rip
189 /* push in order ss, rsp, eflags, cs, rip */
191 pushq $__KERNEL_DS /* ss */
192 CFI_ADJUST_CFA_OFFSET 8
193 /*CFI_REL_OFFSET ss,0*/
195 CFI_ADJUST_CFA_OFFSET 8
197 pushq $X86_EFLAGS_IF /* eflags - interrupts on */
198 CFI_ADJUST_CFA_OFFSET 8
199 /*CFI_REL_OFFSET rflags,0*/
200 pushq $__KERNEL_CS /* cs */
201 CFI_ADJUST_CFA_OFFSET 8
202 /*CFI_REL_OFFSET cs,0*/
203 pushq \child_rip /* rip */
204 CFI_ADJUST_CFA_OFFSET 8
206 pushq %rax /* orig rax */
207 CFI_ADJUST_CFA_OFFSET 8
210 .macro UNFAKE_STACK_FRAME
212 CFI_ADJUST_CFA_OFFSET -(6*8)
216 * initial frame state for interrupts (and exceptions without error code)
218 .macro EMPTY_FRAME start=1 offset=0
222 CFI_DEF_CFA rsp,8+\offset
224 CFI_DEF_CFA_OFFSET 8+\offset
229 * initial frame state for interrupts (and exceptions without error code)
231 .macro INTR_FRAME start=1 offset=0
232 EMPTY_FRAME \start, SS+8+\offset-RIP
233 /*CFI_REL_OFFSET ss, SS+\offset-RIP*/
234 CFI_REL_OFFSET rsp, RSP+\offset-RIP
235 /*CFI_REL_OFFSET rflags, EFLAGS+\offset-RIP*/
236 /*CFI_REL_OFFSET cs, CS+\offset-RIP*/
237 CFI_REL_OFFSET rip, RIP+\offset-RIP
241 * initial frame state for exceptions with error code (and interrupts
242 * with vector already pushed)
244 .macro XCPT_FRAME start=1 offset=0
245 INTR_FRAME \start, RIP+\offset-ORIG_RAX
246 /*CFI_REL_OFFSET orig_rax, ORIG_RAX-ORIG_RAX*/
250 * frame that enables calling into C.
252 .macro PARTIAL_FRAME start=1 offset=0
253 XCPT_FRAME \start, ORIG_RAX+\offset-ARGOFFSET
254 CFI_REL_OFFSET rdi, RDI+\offset-ARGOFFSET
255 CFI_REL_OFFSET rsi, RSI+\offset-ARGOFFSET
256 CFI_REL_OFFSET rdx, RDX+\offset-ARGOFFSET
257 CFI_REL_OFFSET rcx, RCX+\offset-ARGOFFSET
258 CFI_REL_OFFSET rax, RAX+\offset-ARGOFFSET
259 CFI_REL_OFFSET r8, R8+\offset-ARGOFFSET
260 CFI_REL_OFFSET r9, R9+\offset-ARGOFFSET
261 CFI_REL_OFFSET r10, R10+\offset-ARGOFFSET
262 CFI_REL_OFFSET r11, R11+\offset-ARGOFFSET
266 * frame that enables passing a complete pt_regs to a C function.
268 .macro DEFAULT_FRAME start=1 offset=0
269 PARTIAL_FRAME \start, R11+\offset-R15
270 CFI_REL_OFFSET rbx, RBX+\offset
271 CFI_REL_OFFSET rbp, RBP+\offset
272 CFI_REL_OFFSET r12, R12+\offset
273 CFI_REL_OFFSET r13, R13+\offset
274 CFI_REL_OFFSET r14, R14+\offset
275 CFI_REL_OFFSET r15, R15+\offset
278 /* save partial stack frame */
282 movq_cfi rdi, RDI+16-ARGOFFSET
283 movq_cfi rsi, RSI+16-ARGOFFSET
284 movq_cfi rdx, RDX+16-ARGOFFSET
285 movq_cfi rcx, RCX+16-ARGOFFSET
286 movq_cfi rax, RAX+16-ARGOFFSET
287 movq_cfi r8, R8+16-ARGOFFSET
288 movq_cfi r9, R9+16-ARGOFFSET
289 movq_cfi r10, R10+16-ARGOFFSET
290 movq_cfi r11, R11+16-ARGOFFSET
292 leaq -ARGOFFSET+16(%rsp),%rdi /* arg1 for handler */
293 movq_cfi rbp, 8 /* push %rbp */
294 leaq 8(%rsp), %rbp /* mov %rsp, %ebp */
299 * irqcount is used to check if a CPU is already on an interrupt stack
300 * or not. While this is essentially redundant with preempt_count it is
301 * a little cheaper to use a separate counter in the PDA (short of
302 * moving irq_enter into assembly, which would be too much work)
304 1: incl %gs:pda_irqcount
306 popq_cfi %rax /* move return address... */
307 mov %gs:pda_irqstackptr,%rsp
309 pushq_cfi %rax /* ... to the new stack */
311 * We entered an interrupt context - irqs are off:
319 PARTIAL_FRAME 1 REST_SKIP+8
320 movq 5*8+16(%rsp), %r11 /* save return address */
327 movq %r11, 8(%rsp) /* return address */
328 FIXUP_TOP_OF_STACK %r11, 16
333 /* save complete stack frame */
353 movl $MSR_GS_BASE,%ecx
356 js 1f /* negative -> in kernel */
364 * A newly forked process directly context switches into this.
369 push kernel_eflags(%rip)
370 CFI_ADJUST_CFA_OFFSET 8
371 popf # reset kernel eflags
372 CFI_ADJUST_CFA_OFFSET -8
374 GET_THREAD_INFO(%rcx)
375 testl $(_TIF_SYSCALL_TRACE|_TIF_SYSCALL_AUDIT),TI_flags(%rcx)
379 testl $3,CS-ARGOFFSET(%rsp) # from kernel_thread?
380 je int_ret_from_sys_call
381 testl $_TIF_IA32,TI_flags(%rcx)
382 jnz int_ret_from_sys_call
383 RESTORE_TOP_OF_STACK %rdi, -ARGOFFSET
384 jmp ret_from_sys_call
387 call syscall_trace_leave
388 GET_THREAD_INFO(%rcx)
394 * System call entry. Upto 6 arguments in registers are supported.
396 * SYSCALL does not save anything on the stack and does not change the
402 * rax system call number
404 * rcx return address for syscall/sysret, C arg3
407 * r10 arg3 (--> moved to rcx for C)
410 * r11 eflags for syscall/sysret, temporary for C
411 * r12-r15,rbp,rbx saved by C code, not touched.
413 * Interrupts are off on entry.
414 * Only called from user space.
416 * XXX if we had a free scratch register we could save the RSP into the stack frame
417 * and report it properly in ps. Unfortunately we haven't.
419 * When user can change the frames always force IRET. That is because
420 * it deals with uncanonical addresses better. SYSRET has trouble
421 * with them due to bugs in both AMD and Intel CPUs.
427 CFI_DEF_CFA rsp,PDA_STACKOFFSET
429 /*CFI_REGISTER rflags,r11*/
432 * A hypervisor implementation might want to use a label
433 * after the swapgs, so that it can do the swapgs
434 * for the guest and jump here on syscall.
436 ENTRY(system_call_after_swapgs)
438 movq %rsp,%gs:pda_oldrsp
439 movq %gs:pda_kernelstack,%rsp
441 * No need to follow this irqs off/on section - it's straight
444 ENABLE_INTERRUPTS(CLBR_NONE)
446 movq %rax,ORIG_RAX-ARGOFFSET(%rsp)
447 movq %rcx,RIP-ARGOFFSET(%rsp)
448 CFI_REL_OFFSET rip,RIP-ARGOFFSET
449 GET_THREAD_INFO(%rcx)
450 testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%rcx)
452 system_call_fastpath:
453 cmpq $__NR_syscall_max,%rax
456 call *sys_call_table(,%rax,8) # XXX: rip relative
457 movq %rax,RAX-ARGOFFSET(%rsp)
459 * Syscall return path ending with SYSRET (fast path)
460 * Has incomplete stack frame and undefined top of stack.
463 movl $_TIF_ALLWORK_MASK,%edi
467 GET_THREAD_INFO(%rcx)
468 DISABLE_INTERRUPTS(CLBR_NONE)
470 movl TI_flags(%rcx),%edx
475 * sysretq will re-enable interrupts:
478 movq RIP-ARGOFFSET(%rsp),%rcx
480 RESTORE_ARGS 0,-ARG_SKIP,1
481 /*CFI_REGISTER rflags,r11*/
482 movq %gs:pda_oldrsp, %rsp
486 /* Handle reschedules */
487 /* edx: work, edi: workmask */
489 bt $TIF_NEED_RESCHED,%edx
492 ENABLE_INTERRUPTS(CLBR_NONE)
494 CFI_ADJUST_CFA_OFFSET 8
497 CFI_ADJUST_CFA_OFFSET -8
500 /* Handle a signal */
503 ENABLE_INTERRUPTS(CLBR_NONE)
504 #ifdef CONFIG_AUDITSYSCALL
505 bt $TIF_SYSCALL_AUDIT,%edx
508 /* edx: work flags (arg3) */
509 leaq -ARGOFFSET(%rsp),%rdi # &pt_regs -> arg1
510 xorl %esi,%esi # oldset -> arg2
512 FIXUP_TOP_OF_STACK %r11
513 call do_notify_resume
514 RESTORE_TOP_OF_STACK %r11
516 movl $_TIF_WORK_MASK,%edi
517 /* Use IRET because user could have changed frame. This
518 works because ptregscall_common has called FIXUP_TOP_OF_STACK. */
519 DISABLE_INTERRUPTS(CLBR_NONE)
524 movq $-ENOSYS,RAX-ARGOFFSET(%rsp)
525 jmp ret_from_sys_call
527 #ifdef CONFIG_AUDITSYSCALL
529 * Fast path for syscall audit without full syscall trace.
530 * We just call audit_syscall_entry() directly, and then
531 * jump back to the normal fast path.
534 movq %r10,%r9 /* 6th arg: 4th syscall arg */
535 movq %rdx,%r8 /* 5th arg: 3rd syscall arg */
536 movq %rsi,%rcx /* 4th arg: 2nd syscall arg */
537 movq %rdi,%rdx /* 3rd arg: 1st syscall arg */
538 movq %rax,%rsi /* 2nd arg: syscall number */
539 movl $AUDIT_ARCH_X86_64,%edi /* 1st arg: audit arch */
540 call audit_syscall_entry
541 LOAD_ARGS 0 /* reload call-clobbered registers */
542 jmp system_call_fastpath
545 * Return fast path for syscall audit. Call audit_syscall_exit()
546 * directly and then jump back to the fast path with TIF_SYSCALL_AUDIT
550 movq %rax,%rsi /* second arg, syscall return value */
551 cmpq $0,%rax /* is it < 0? */
552 setl %al /* 1 if so, 0 if not */
553 movzbl %al,%edi /* zero-extend that into %edi */
554 inc %edi /* first arg, 0->1(AUDITSC_SUCCESS), 1->2(AUDITSC_FAILURE) */
555 call audit_syscall_exit
556 movl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT),%edi
558 #endif /* CONFIG_AUDITSYSCALL */
560 /* Do syscall tracing */
562 #ifdef CONFIG_AUDITSYSCALL
563 testl $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags(%rcx)
567 movq $-ENOSYS,RAX(%rsp) /* ptrace can change this for a bad syscall */
568 FIXUP_TOP_OF_STACK %rdi
570 call syscall_trace_enter
572 * Reload arg registers from stack in case ptrace changed them.
573 * We don't reload %rax because syscall_trace_enter() returned
574 * the value it wants us to use in the table lookup.
576 LOAD_ARGS ARGOFFSET, 1
578 cmpq $__NR_syscall_max,%rax
579 ja int_ret_from_sys_call /* RAX(%rsp) set to -ENOSYS above */
580 movq %r10,%rcx /* fixup for C */
581 call *sys_call_table(,%rax,8)
582 movq %rax,RAX-ARGOFFSET(%rsp)
583 /* Use IRET because user could have changed frame */
586 * Syscall return path ending with IRET.
587 * Has correct top of stack, but partial stack frame.
589 .globl int_ret_from_sys_call
590 .globl int_with_check
591 int_ret_from_sys_call:
592 DISABLE_INTERRUPTS(CLBR_NONE)
594 testl $3,CS-ARGOFFSET(%rsp)
595 je retint_restore_args
596 movl $_TIF_ALLWORK_MASK,%edi
597 /* edi: mask to check */
600 GET_THREAD_INFO(%rcx)
601 movl TI_flags(%rcx),%edx
604 andl $~TS_COMPAT,TI_status(%rcx)
607 /* Either reschedule or signal or syscall exit tracking needed. */
608 /* First do a reschedule test. */
609 /* edx: work, edi: workmask */
611 bt $TIF_NEED_RESCHED,%edx
614 ENABLE_INTERRUPTS(CLBR_NONE)
616 CFI_ADJUST_CFA_OFFSET 8
619 CFI_ADJUST_CFA_OFFSET -8
620 DISABLE_INTERRUPTS(CLBR_NONE)
624 /* handle signals and tracing -- both require a full stack frame */
627 ENABLE_INTERRUPTS(CLBR_NONE)
629 /* Check for syscall exit trace */
630 testl $_TIF_WORK_SYSCALL_EXIT,%edx
633 CFI_ADJUST_CFA_OFFSET 8
634 leaq 8(%rsp),%rdi # &ptregs -> arg1
635 call syscall_trace_leave
637 CFI_ADJUST_CFA_OFFSET -8
638 andl $~(_TIF_WORK_SYSCALL_EXIT|_TIF_SYSCALL_EMU),%edi
642 testl $_TIF_DO_NOTIFY_MASK,%edx
644 movq %rsp,%rdi # &ptregs -> arg1
645 xorl %esi,%esi # oldset -> arg2
646 call do_notify_resume
647 1: movl $_TIF_WORK_MASK,%edi
650 DISABLE_INTERRUPTS(CLBR_NONE)
657 * Certain special system calls that need to save a complete full stack frame.
659 .macro PTREGSCALL label,func,arg
661 PARTIAL_FRAME 1 8 /* offset 8: return address */
662 subq $REST_SKIP, %rsp
663 CFI_ADJUST_CFA_OFFSET REST_SKIP
665 DEFAULT_FRAME 0 8 /* offset 8: return address */
666 leaq 8(%rsp), \arg /* pt_regs pointer */
668 jmp ptregscall_common
673 PTREGSCALL stub_clone, sys_clone, %r8
674 PTREGSCALL stub_fork, sys_fork, %rdi
675 PTREGSCALL stub_vfork, sys_vfork, %rdi
676 PTREGSCALL stub_sigaltstack, sys_sigaltstack, %rdx
677 PTREGSCALL stub_iopl, sys_iopl, %rsi
679 ENTRY(ptregscall_common)
680 DEFAULT_FRAME 1 8 /* offset 8: return address */
681 RESTORE_TOP_OF_STACK %r11, 8
682 movq_cfi_restore R15+8, r15
683 movq_cfi_restore R14+8, r14
684 movq_cfi_restore R13+8, r13
685 movq_cfi_restore R12+8, r12
686 movq_cfi_restore RBP+8, rbp
687 movq_cfi_restore RBX+8, rbx
688 ret $REST_SKIP /* pop extended registers */
690 END(ptregscall_common)
695 CFI_ADJUST_CFA_OFFSET -8
696 CFI_REGISTER rip, r11
698 FIXUP_TOP_OF_STACK %r11
701 RESTORE_TOP_OF_STACK %r11
704 jmp int_ret_from_sys_call
709 * sigreturn is special because it needs to restore all registers on return.
710 * This cannot be done with SYSRET, so use the IRET return path instead.
712 ENTRY(stub_rt_sigreturn)
715 CFI_ADJUST_CFA_OFFSET -8
718 FIXUP_TOP_OF_STACK %r11
719 call sys_rt_sigreturn
720 movq %rax,RAX(%rsp) # fixme, this could be done at the higher layer
722 jmp int_ret_from_sys_call
724 END(stub_rt_sigreturn)
727 * Build the entry stubs and pointer table with some assembler magic.
728 * We pack 7 stubs into a single 32-byte chunk, which will fit in a
729 * single cache line on all modern x86 implementations.
731 .section .init.rodata,"a"
735 .p2align CONFIG_X86_L1_CACHE_SHIFT
736 ENTRY(irq_entries_start)
738 vector=FIRST_EXTERNAL_VECTOR
739 .rept (NR_VECTORS-FIRST_EXTERNAL_VECTOR+6)/7
742 .if vector < NR_VECTORS
743 .if vector <> FIRST_EXTERNAL_VECTOR
744 CFI_ADJUST_CFA_OFFSET -8
746 1: pushq $(~vector+0x80) /* Note: always in signed byte range */
747 CFI_ADJUST_CFA_OFFSET 8
748 .if ((vector-FIRST_EXTERNAL_VECTOR)%7) <> 6
757 2: jmp common_interrupt
760 END(irq_entries_start)
767 * Interrupt entry/exit.
769 * Interrupt entry points save only callee clobbered registers in fast path.
771 * Entry runs with interrupts off.
774 /* 0(%rsp): ~(interrupt number) */
775 .macro interrupt func
777 CFI_ADJUST_CFA_OFFSET 10*8
784 * The interrupt stubs push (~vector+0x80) onto the stack and
785 * then jump to common_interrupt.
787 .p2align CONFIG_X86_L1_CACHE_SHIFT
790 addq $-0x80,(%rsp) /* Adjust vector to [-256,-1] range */
792 /* 0(%rsp): oldrsp-ARGOFFSET */
794 DISABLE_INTERRUPTS(CLBR_NONE)
796 decl %gs:pda_irqcount
798 CFI_DEF_CFA_REGISTER rsp
799 CFI_ADJUST_CFA_OFFSET -8
801 GET_THREAD_INFO(%rcx)
802 testl $3,CS-ARGOFFSET(%rsp)
805 /* Interrupt came from user space */
807 * Has a correct top of stack, but a partial stack frame
808 * %rcx: thread info. Interrupts off.
810 retint_with_reschedule:
811 movl $_TIF_WORK_MASK,%edi
814 movl TI_flags(%rcx),%edx
819 retint_swapgs: /* return to user-space */
821 * The iretq could re-enable interrupts:
823 DISABLE_INTERRUPTS(CLBR_ANY)
828 retint_restore_args: /* return to kernel space */
829 DISABLE_INTERRUPTS(CLBR_ANY)
831 * The iretq could re-enable interrupts:
840 .section __ex_table, "a"
841 .quad irq_return, bad_iret
844 #ifdef CONFIG_PARAVIRT
848 .section __ex_table,"a"
849 .quad native_iret, bad_iret
856 * The iret traps when the %cs or %ss being restored is bogus.
857 * We've lost the original trap vector and error code.
858 * #GPF is the most likely one to get for an invalid selector.
859 * So pretend we completed the iret and took the #GPF in user mode.
861 * We are now running with the kernel GS after exception recovery.
862 * But error_entry expects us to have user GS to match the user %cs,
868 jmp general_protection
872 /* edi: workmask, edx: work */
875 bt $TIF_NEED_RESCHED,%edx
878 ENABLE_INTERRUPTS(CLBR_NONE)
880 CFI_ADJUST_CFA_OFFSET 8
883 CFI_ADJUST_CFA_OFFSET -8
884 GET_THREAD_INFO(%rcx)
885 DISABLE_INTERRUPTS(CLBR_NONE)
890 testl $_TIF_DO_NOTIFY_MASK,%edx
893 ENABLE_INTERRUPTS(CLBR_NONE)
895 movq $-1,ORIG_RAX(%rsp)
896 xorl %esi,%esi # oldset
897 movq %rsp,%rdi # &pt_regs
898 call do_notify_resume
900 DISABLE_INTERRUPTS(CLBR_NONE)
902 GET_THREAD_INFO(%rcx)
903 jmp retint_with_reschedule
905 #ifdef CONFIG_PREEMPT
906 /* Returning to kernel space. Check if we need preemption */
907 /* rcx: threadinfo. interrupts off. */
909 cmpl $0,TI_preempt_count(%rcx)
910 jnz retint_restore_args
911 bt $TIF_NEED_RESCHED,TI_flags(%rcx)
912 jnc retint_restore_args
913 bt $9,EFLAGS-ARGOFFSET(%rsp) /* interrupts off? */
914 jnc retint_restore_args
915 call preempt_schedule_irq
920 END(common_interrupt)
925 .macro apicinterrupt num sym do_sym
929 CFI_ADJUST_CFA_OFFSET 8
937 apicinterrupt IRQ_MOVE_CLEANUP_VECTOR \
938 irq_move_cleanup_interrupt smp_irq_move_cleanup_interrupt
941 apicinterrupt UV_BAU_MESSAGE \
942 uv_bau_message_intr1 uv_bau_message_interrupt
943 apicinterrupt LOCAL_TIMER_VECTOR \
944 apic_timer_interrupt smp_apic_timer_interrupt
947 apicinterrupt INVALIDATE_TLB_VECTOR_START+0 \
948 invalidate_interrupt0 smp_invalidate_interrupt
949 apicinterrupt INVALIDATE_TLB_VECTOR_START+1 \
950 invalidate_interrupt1 smp_invalidate_interrupt
951 apicinterrupt INVALIDATE_TLB_VECTOR_START+2 \
952 invalidate_interrupt2 smp_invalidate_interrupt
953 apicinterrupt INVALIDATE_TLB_VECTOR_START+3 \
954 invalidate_interrupt3 smp_invalidate_interrupt
955 apicinterrupt INVALIDATE_TLB_VECTOR_START+4 \
956 invalidate_interrupt4 smp_invalidate_interrupt
957 apicinterrupt INVALIDATE_TLB_VECTOR_START+5 \
958 invalidate_interrupt5 smp_invalidate_interrupt
959 apicinterrupt INVALIDATE_TLB_VECTOR_START+6 \
960 invalidate_interrupt6 smp_invalidate_interrupt
961 apicinterrupt INVALIDATE_TLB_VECTOR_START+7 \
962 invalidate_interrupt7 smp_invalidate_interrupt
965 apicinterrupt THRESHOLD_APIC_VECTOR \
966 threshold_interrupt mce_threshold_interrupt
967 apicinterrupt THERMAL_APIC_VECTOR \
968 thermal_interrupt smp_thermal_interrupt
971 apicinterrupt CALL_FUNCTION_SINGLE_VECTOR \
972 call_function_single_interrupt smp_call_function_single_interrupt
973 apicinterrupt CALL_FUNCTION_VECTOR \
974 call_function_interrupt smp_call_function_interrupt
975 apicinterrupt RESCHEDULE_VECTOR \
976 reschedule_interrupt smp_reschedule_interrupt
979 apicinterrupt ERROR_APIC_VECTOR \
980 error_interrupt smp_error_interrupt
981 apicinterrupt SPURIOUS_APIC_VECTOR \
982 spurious_interrupt smp_spurious_interrupt
985 * Exception entry points.
987 .macro zeroentry sym do_sym
990 PARAVIRT_ADJUST_EXCEPTION_FRAME
991 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
993 CFI_ADJUST_CFA_OFFSET 15*8
996 movq %rsp,%rdi /* pt_regs pointer */
997 xorl %esi,%esi /* no error code */
999 jmp error_exit /* %ebx: no swapgs flag */
1004 .macro paranoidzeroentry sym do_sym
1007 PARAVIRT_ADJUST_EXCEPTION_FRAME
1008 pushq $-1 /* ORIG_RAX: no syscall to restart */
1009 CFI_ADJUST_CFA_OFFSET 8
1013 movq %rsp,%rdi /* pt_regs pointer */
1014 xorl %esi,%esi /* no error code */
1016 jmp paranoid_exit /* %ebx: no swapgs flag */
1021 .macro paranoidzeroentry_ist sym do_sym ist
1024 PARAVIRT_ADJUST_EXCEPTION_FRAME
1025 pushq $-1 /* ORIG_RAX: no syscall to restart */
1026 CFI_ADJUST_CFA_OFFSET 8
1030 movq %rsp,%rdi /* pt_regs pointer */
1031 xorl %esi,%esi /* no error code */
1032 movq %gs:pda_data_offset, %rbp
1033 subq $EXCEPTION_STKSZ, per_cpu__init_tss + TSS_ist + (\ist - 1) * 8(%rbp)
1035 addq $EXCEPTION_STKSZ, per_cpu__init_tss + TSS_ist + (\ist - 1) * 8(%rbp)
1036 jmp paranoid_exit /* %ebx: no swapgs flag */
1041 .macro errorentry sym do_sym
1044 PARAVIRT_ADJUST_EXCEPTION_FRAME
1046 CFI_ADJUST_CFA_OFFSET 15*8
1049 movq %rsp,%rdi /* pt_regs pointer */
1050 movq ORIG_RAX(%rsp),%rsi /* get error code */
1051 movq $-1,ORIG_RAX(%rsp) /* no syscall to restart */
1053 jmp error_exit /* %ebx: no swapgs flag */
1058 /* error code is on the stack already */
1059 .macro paranoiderrorentry sym do_sym
1062 PARAVIRT_ADJUST_EXCEPTION_FRAME
1064 CFI_ADJUST_CFA_OFFSET 15*8
1068 movq %rsp,%rdi /* pt_regs pointer */
1069 movq ORIG_RAX(%rsp),%rsi /* get error code */
1070 movq $-1,ORIG_RAX(%rsp) /* no syscall to restart */
1072 jmp paranoid_exit /* %ebx: no swapgs flag */
1077 zeroentry divide_error do_divide_error
1078 zeroentry overflow do_overflow
1079 zeroentry bounds do_bounds
1080 zeroentry invalid_op do_invalid_op
1081 zeroentry device_not_available do_device_not_available
1082 paranoiderrorentry double_fault do_double_fault
1083 zeroentry coprocessor_segment_overrun do_coprocessor_segment_overrun
1084 errorentry invalid_TSS do_invalid_TSS
1085 errorentry segment_not_present do_segment_not_present
1086 zeroentry spurious_interrupt_bug do_spurious_interrupt_bug
1087 zeroentry coprocessor_error do_coprocessor_error
1088 errorentry alignment_check do_alignment_check
1089 zeroentry simd_coprocessor_error do_simd_coprocessor_error
1091 /* Reload gs selector with exception handling */
1092 /* edi: new selector */
1093 ENTRY(native_load_gs_index)
1096 CFI_ADJUST_CFA_OFFSET 8
1097 DISABLE_INTERRUPTS(CLBR_ANY | ~(CLBR_RDI))
1101 2: mfence /* workaround */
1104 CFI_ADJUST_CFA_OFFSET -8
1107 END(native_load_gs_index)
1109 .section __ex_table,"a"
1111 .quad gs_change,bad_gs
1113 .section .fixup,"ax"
1114 /* running with kernelgs */
1116 SWAPGS /* switch back to user gs */
1123 * Create a kernel thread.
1125 * C extern interface:
1126 * extern long kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
1128 * asm input arguments:
1129 * rdi: fn, rsi: arg, rdx: flags
1131 ENTRY(kernel_thread)
1133 FAKE_STACK_FRAME $child_rip
1136 # rdi: flags, rsi: usp, rdx: will be &pt_regs
1138 orq kernel_thread_flags(%rip),%rdi
1151 * It isn't worth to check for reschedule here,
1152 * so internally to the x86_64 port you can rely on kernel_thread()
1153 * not to reschedule the child before returning, this avoids the need
1154 * of hacks for example to fork off the per-CPU idle tasks.
1155 * [Hopefully no generic code relies on the reschedule -AK]
1164 pushq $0 # fake return address
1167 * Here we are in the child and the registers are set as they were
1168 * at kernel_thread() invocation in the parent.
1180 * execve(). This function needs to use IRET, not SYSRET, to set up all state properly.
1182 * C extern interface:
1183 * extern long execve(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(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_ADJUST_CFA_OFFSET 8
1215 CFI_REL_OFFSET rbp,0
1217 CFI_DEF_CFA_REGISTER rbp
1218 incl %gs:pda_irqcount
1219 cmove %gs:pda_irqstackptr,%rsp
1220 push %rbp # backlink for old unwinder
1223 CFI_DEF_CFA_REGISTER rsp
1224 CFI_ADJUST_CFA_OFFSET -8
1225 decl %gs:pda_irqcount
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 %gs:pda_irqcount
1257 CFI_DEF_CFA_REGISTER rbp
1258 cmovzq %gs:pda_irqstackptr,%rsp
1259 pushq %rbp # backlink for old unwinder
1260 call xen_evtchn_do_upcall
1262 CFI_DEF_CFA_REGISTER rsp
1263 decl %gs:pda_irqcount
1266 END(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
1325 END(xen_failsafe_callback)
1327 #endif /* CONFIG_XEN */
1330 * Some functions should be protected against kprobes
1332 .pushsection .kprobes.text, "ax"
1334 paranoidzeroentry_ist debug do_debug DEBUG_STACK
1335 paranoidzeroentry_ist int3 do_int3 DEBUG_STACK
1336 paranoiderrorentry stack_segment do_stack_segment
1337 errorentry general_protection do_general_protection
1338 errorentry page_fault do_page_fault
1339 #ifdef CONFIG_X86_MCE
1340 paranoidzeroentry machine_check do_machine_check
1344 * "Paranoid" exit path from exception stack.
1345 * Paranoid because this is used by NMIs and cannot take
1346 * any kernel state for granted.
1347 * We don't do kernel preemption checks here, because only
1348 * NMI should be common and it does not enable IRQs and
1349 * cannot get reschedule ticks.
1351 * "trace" is 0 for the NMI handler only, because irq-tracing
1352 * is fundamentally NMI-unsafe. (we cannot change the soft and
1353 * hard flags at once, atomically)
1356 /* ebx: no swapgs flag */
1357 ENTRY(paranoid_exit)
1359 DISABLE_INTERRUPTS(CLBR_NONE)
1361 testl %ebx,%ebx /* swapgs needed? */
1362 jnz paranoid_restore
1364 jnz paranoid_userspace
1372 GET_THREAD_INFO(%rcx)
1373 movl TI_flags(%rcx),%ebx
1374 andl $_TIF_WORK_MASK,%ebx
1376 movq %rsp,%rdi /* &pt_regs */
1378 movq %rax,%rsp /* switch stack for scheduling */
1379 testl $_TIF_NEED_RESCHED,%ebx
1380 jnz paranoid_schedule
1381 movl %ebx,%edx /* arg3: thread flags */
1383 ENABLE_INTERRUPTS(CLBR_NONE)
1384 xorl %esi,%esi /* arg2: oldset */
1385 movq %rsp,%rdi /* arg1: &pt_regs */
1386 call do_notify_resume
1387 DISABLE_INTERRUPTS(CLBR_NONE)
1389 jmp paranoid_userspace
1392 ENABLE_INTERRUPTS(CLBR_ANY)
1394 DISABLE_INTERRUPTS(CLBR_ANY)
1396 jmp paranoid_userspace
1401 * Exception entry point. This expects an error code/orig_rax on the stack.
1402 * returns in "no swapgs flag" in %ebx.
1406 CFI_ADJUST_CFA_OFFSET 15*8
1407 /* oldrax contains error code */
1426 je error_kernelspace
1435 * There are two places in the kernel that can potentially fault with
1436 * usergs. Handle them here. The exception handlers after iret run with
1437 * kernel gs again, so don't set the user space flag. B stepping K8s
1438 * sometimes report an truncated RIP for IRET exceptions returning to
1439 * compat mode. Check for these here too.
1443 leaq irq_return(%rip),%rcx
1444 cmpq %rcx,RIP+8(%rsp)
1446 movl %ecx,%ecx /* zero extend */
1447 cmpq %rcx,RIP+8(%rsp)
1449 cmpq $gs_change,RIP+8(%rsp)
1455 /* ebx: no swapgs flag (1: don't need swapgs, 0: need it) */
1460 DISABLE_INTERRUPTS(CLBR_NONE)
1462 GET_THREAD_INFO(%rcx)
1465 LOCKDEP_SYS_EXIT_IRQ
1466 movl TI_flags(%rcx),%edx
1467 movl $_TIF_WORK_MASK,%edi
1475 /* runs on exception stack */
1478 PARAVIRT_ADJUST_EXCEPTION_FRAME
1481 CFI_ADJUST_CFA_OFFSET 15*8
1484 /* paranoidentry do_nmi, 0; without TRACE_IRQS_OFF */
1488 #ifdef CONFIG_TRACE_IRQFLAGS
1489 /* paranoidexit; without TRACE_IRQS_OFF */
1490 /* ebx: no swapgs flag */
1491 DISABLE_INTERRUPTS(CLBR_NONE)
1492 testl %ebx,%ebx /* swapgs needed? */
1502 GET_THREAD_INFO(%rcx)
1503 movl TI_flags(%rcx),%ebx
1504 andl $_TIF_WORK_MASK,%ebx
1506 movq %rsp,%rdi /* &pt_regs */
1508 movq %rax,%rsp /* switch stack for scheduling */
1509 testl $_TIF_NEED_RESCHED,%ebx
1511 movl %ebx,%edx /* arg3: thread flags */
1512 ENABLE_INTERRUPTS(CLBR_NONE)
1513 xorl %esi,%esi /* arg2: oldset */
1514 movq %rsp,%rdi /* arg1: &pt_regs */
1515 call do_notify_resume
1516 DISABLE_INTERRUPTS(CLBR_NONE)
1519 ENABLE_INTERRUPTS(CLBR_ANY)
1521 DISABLE_INTERRUPTS(CLBR_ANY)
1530 ENTRY(ignore_sysret)
1538 * End of kprobes section