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>
57 #ifndef CONFIG_PREEMPT
58 #define retint_kernel retint_restore_args
61 #ifdef CONFIG_PARAVIRT
62 ENTRY(native_irq_enable_syscall_ret)
63 movq %gs:pda_oldrsp,%rsp
66 #endif /* CONFIG_PARAVIRT */
69 .macro TRACE_IRQS_IRETQ offset=ARGOFFSET
70 #ifdef CONFIG_TRACE_IRQFLAGS
71 bt $9,EFLAGS-\offset(%rsp) /* interrupts off? */
79 * C code is not supposed to know about undefined top of stack. Every time
80 * a C function with an pt_regs argument is called from the SYSCALL based
81 * fast path FIXUP_TOP_OF_STACK is needed.
82 * RESTORE_TOP_OF_STACK syncs the syscall state after any possible ptregs
86 /* %rsp:at FRAMEEND */
87 .macro FIXUP_TOP_OF_STACK tmp
88 movq %gs:pda_oldrsp,\tmp
90 movq $__USER_DS,SS(%rsp)
91 movq $__USER_CS,CS(%rsp)
93 movq R11(%rsp),\tmp /* get eflags */
94 movq \tmp,EFLAGS(%rsp)
97 .macro RESTORE_TOP_OF_STACK tmp,offset=0
98 movq RSP-\offset(%rsp),\tmp
99 movq \tmp,%gs:pda_oldrsp
100 movq EFLAGS-\offset(%rsp),\tmp
101 movq \tmp,R11-\offset(%rsp)
104 .macro FAKE_STACK_FRAME child_rip
105 /* push in order ss, rsp, eflags, cs, rip */
108 CFI_ADJUST_CFA_OFFSET 8
109 /*CFI_REL_OFFSET ss,0*/
111 CFI_ADJUST_CFA_OFFSET 8
113 pushq $(1<<9) /* eflags - interrupts on */
114 CFI_ADJUST_CFA_OFFSET 8
115 /*CFI_REL_OFFSET rflags,0*/
116 pushq $__KERNEL_CS /* cs */
117 CFI_ADJUST_CFA_OFFSET 8
118 /*CFI_REL_OFFSET cs,0*/
119 pushq \child_rip /* rip */
120 CFI_ADJUST_CFA_OFFSET 8
122 pushq %rax /* orig rax */
123 CFI_ADJUST_CFA_OFFSET 8
126 .macro UNFAKE_STACK_FRAME
128 CFI_ADJUST_CFA_OFFSET -(6*8)
131 .macro CFI_DEFAULT_STACK start=1
137 CFI_DEF_CFA_OFFSET SS+8
139 CFI_REL_OFFSET r15,R15
140 CFI_REL_OFFSET r14,R14
141 CFI_REL_OFFSET r13,R13
142 CFI_REL_OFFSET r12,R12
143 CFI_REL_OFFSET rbp,RBP
144 CFI_REL_OFFSET rbx,RBX
145 CFI_REL_OFFSET r11,R11
146 CFI_REL_OFFSET r10,R10
149 CFI_REL_OFFSET rax,RAX
150 CFI_REL_OFFSET rcx,RCX
151 CFI_REL_OFFSET rdx,RDX
152 CFI_REL_OFFSET rsi,RSI
153 CFI_REL_OFFSET rdi,RDI
154 CFI_REL_OFFSET rip,RIP
155 /*CFI_REL_OFFSET cs,CS*/
156 /*CFI_REL_OFFSET rflags,EFLAGS*/
157 CFI_REL_OFFSET rsp,RSP
158 /*CFI_REL_OFFSET ss,SS*/
161 * A newly forked process directly context switches into this.
166 push kernel_eflags(%rip)
167 CFI_ADJUST_CFA_OFFSET 4
168 popf # reset kernel eflags
169 CFI_ADJUST_CFA_OFFSET -4
171 GET_THREAD_INFO(%rcx)
172 testl $(_TIF_SYSCALL_TRACE|_TIF_SYSCALL_AUDIT),threadinfo_flags(%rcx)
176 testl $3,CS-ARGOFFSET(%rsp) # from kernel_thread?
177 je int_ret_from_sys_call
178 testl $_TIF_IA32,threadinfo_flags(%rcx)
179 jnz int_ret_from_sys_call
180 RESTORE_TOP_OF_STACK %rdi,ARGOFFSET
181 jmp ret_from_sys_call
184 call syscall_trace_leave
185 GET_THREAD_INFO(%rcx)
191 * System call entry. Upto 6 arguments in registers are supported.
193 * SYSCALL does not save anything on the stack and does not change the
199 * rax system call number
201 * rcx return address for syscall/sysret, C arg3
204 * r10 arg3 (--> moved to rcx for C)
207 * r11 eflags for syscall/sysret, temporary for C
208 * r12-r15,rbp,rbx saved by C code, not touched.
210 * Interrupts are off on entry.
211 * Only called from user space.
213 * XXX if we had a free scratch register we could save the RSP into the stack frame
214 * and report it properly in ps. Unfortunately we haven't.
216 * When user can change the frames always force IRET. That is because
217 * it deals with uncanonical addresses better. SYSRET has trouble
218 * with them due to bugs in both AMD and Intel CPUs.
224 CFI_DEF_CFA rsp,PDA_STACKOFFSET
226 /*CFI_REGISTER rflags,r11*/
229 * A hypervisor implementation might want to use a label
230 * after the swapgs, so that it can do the swapgs
231 * for the guest and jump here on syscall.
233 ENTRY(system_call_after_swapgs)
235 movq %rsp,%gs:pda_oldrsp
236 movq %gs:pda_kernelstack,%rsp
238 * No need to follow this irqs off/on section - it's straight
241 ENABLE_INTERRUPTS(CLBR_NONE)
243 movq %rax,ORIG_RAX-ARGOFFSET(%rsp)
244 movq %rcx,RIP-ARGOFFSET(%rsp)
245 CFI_REL_OFFSET rip,RIP-ARGOFFSET
246 GET_THREAD_INFO(%rcx)
247 testl $(_TIF_SYSCALL_TRACE|_TIF_SYSCALL_AUDIT|_TIF_SECCOMP),threadinfo_flags(%rcx)
249 cmpq $__NR_syscall_max,%rax
252 call *sys_call_table(,%rax,8) # XXX: rip relative
253 movq %rax,RAX-ARGOFFSET(%rsp)
255 * Syscall return path ending with SYSRET (fast path)
256 * Has incomplete stack frame and undefined top of stack.
259 movl $_TIF_ALLWORK_MASK,%edi
263 GET_THREAD_INFO(%rcx)
264 DISABLE_INTERRUPTS(CLBR_NONE)
266 movl threadinfo_flags(%rcx),%edx
271 * sysretq will re-enable interrupts:
274 movq RIP-ARGOFFSET(%rsp),%rcx
276 RESTORE_ARGS 0,-ARG_SKIP,1
277 /*CFI_REGISTER rflags,r11*/
278 ENABLE_INTERRUPTS_SYSCALL_RET
281 /* Handle reschedules */
282 /* edx: work, edi: workmask */
284 bt $TIF_NEED_RESCHED,%edx
287 ENABLE_INTERRUPTS(CLBR_NONE)
289 CFI_ADJUST_CFA_OFFSET 8
292 CFI_ADJUST_CFA_OFFSET -8
295 /* Handle a signal */
298 ENABLE_INTERRUPTS(CLBR_NONE)
299 testl $_TIF_DO_NOTIFY_MASK,%edx
302 /* Really a signal */
303 /* edx: work flags (arg3) */
304 leaq do_notify_resume(%rip),%rax
305 leaq -ARGOFFSET(%rsp),%rdi # &pt_regs -> arg1
306 xorl %esi,%esi # oldset -> arg2
307 call ptregscall_common
308 1: movl $_TIF_NEED_RESCHED,%edi
309 /* Use IRET because user could have changed frame. This
310 works because ptregscall_common has called FIXUP_TOP_OF_STACK. */
311 DISABLE_INTERRUPTS(CLBR_NONE)
316 movq $-ENOSYS,RAX-ARGOFFSET(%rsp)
317 jmp ret_from_sys_call
319 /* Do syscall tracing */
322 movq $-ENOSYS,RAX(%rsp)
323 FIXUP_TOP_OF_STACK %rdi
325 call syscall_trace_enter
326 LOAD_ARGS ARGOFFSET /* reload args from stack in case ptrace changed it */
328 cmpq $__NR_syscall_max,%rax
332 movq %r10,%rcx /* fixup for C */
333 call *sys_call_table(,%rax,8)
334 1: movq %rax,RAX-ARGOFFSET(%rsp)
335 /* Use IRET because user could have changed frame */
338 * Syscall return path ending with IRET.
339 * Has correct top of stack, but partial stack frame.
341 .globl int_ret_from_sys_call
342 int_ret_from_sys_call:
343 DISABLE_INTERRUPTS(CLBR_NONE)
345 testl $3,CS-ARGOFFSET(%rsp)
346 je retint_restore_args
347 movl $_TIF_ALLWORK_MASK,%edi
348 /* edi: mask to check */
351 GET_THREAD_INFO(%rcx)
352 movl threadinfo_flags(%rcx),%edx
355 andl $~TS_COMPAT,threadinfo_status(%rcx)
358 /* Either reschedule or signal or syscall exit tracking needed. */
359 /* First do a reschedule test. */
360 /* edx: work, edi: workmask */
362 bt $TIF_NEED_RESCHED,%edx
365 ENABLE_INTERRUPTS(CLBR_NONE)
367 CFI_ADJUST_CFA_OFFSET 8
370 CFI_ADJUST_CFA_OFFSET -8
371 DISABLE_INTERRUPTS(CLBR_NONE)
375 /* handle signals and tracing -- both require a full stack frame */
378 ENABLE_INTERRUPTS(CLBR_NONE)
380 /* Check for syscall exit trace */
381 testl $(_TIF_SYSCALL_TRACE|_TIF_SYSCALL_AUDIT|_TIF_SINGLESTEP),%edx
384 CFI_ADJUST_CFA_OFFSET 8
385 leaq 8(%rsp),%rdi # &ptregs -> arg1
386 call syscall_trace_leave
388 CFI_ADJUST_CFA_OFFSET -8
389 andl $~(_TIF_SYSCALL_TRACE|_TIF_SYSCALL_AUDIT|_TIF_SINGLESTEP),%edi
393 testl $_TIF_DO_NOTIFY_MASK,%edx
395 movq %rsp,%rdi # &ptregs -> arg1
396 xorl %esi,%esi # oldset -> arg2
397 call do_notify_resume
398 1: movl $_TIF_NEED_RESCHED,%edi
401 DISABLE_INTERRUPTS(CLBR_NONE)
408 * Certain special system calls that need to save a complete full stack frame.
411 .macro PTREGSCALL label,func,arg
414 leaq \func(%rip),%rax
415 leaq -ARGOFFSET+8(%rsp),\arg /* 8 for return address */
416 jmp ptregscall_common
422 PTREGSCALL stub_clone, sys_clone, %r8
423 PTREGSCALL stub_fork, sys_fork, %rdi
424 PTREGSCALL stub_vfork, sys_vfork, %rdi
425 PTREGSCALL stub_rt_sigsuspend, sys_rt_sigsuspend, %rdx
426 PTREGSCALL stub_sigaltstack, sys_sigaltstack, %rdx
427 PTREGSCALL stub_iopl, sys_iopl, %rsi
429 ENTRY(ptregscall_common)
431 CFI_ADJUST_CFA_OFFSET -8
432 CFI_REGISTER rip, r11
435 CFI_REGISTER rip, r15
436 FIXUP_TOP_OF_STACK %r11
438 RESTORE_TOP_OF_STACK %r11
440 CFI_REGISTER rip, r11
443 CFI_ADJUST_CFA_OFFSET 8
444 CFI_REL_OFFSET rip, 0
447 END(ptregscall_common)
452 CFI_ADJUST_CFA_OFFSET -8
453 CFI_REGISTER rip, r11
455 FIXUP_TOP_OF_STACK %r11
457 RESTORE_TOP_OF_STACK %r11
460 jmp int_ret_from_sys_call
465 * sigreturn is special because it needs to restore all registers on return.
466 * This cannot be done with SYSRET, so use the IRET return path instead.
468 ENTRY(stub_rt_sigreturn)
471 CFI_ADJUST_CFA_OFFSET -8
474 FIXUP_TOP_OF_STACK %r11
475 call sys_rt_sigreturn
476 movq %rax,RAX(%rsp) # fixme, this could be done at the higher layer
478 jmp int_ret_from_sys_call
480 END(stub_rt_sigreturn)
483 * initial frame state for interrupts and exceptions
488 CFI_DEF_CFA rsp,SS+8-\ref
489 /*CFI_REL_OFFSET ss,SS-\ref*/
490 CFI_REL_OFFSET rsp,RSP-\ref
491 /*CFI_REL_OFFSET rflags,EFLAGS-\ref*/
492 /*CFI_REL_OFFSET cs,CS-\ref*/
493 CFI_REL_OFFSET rip,RIP-\ref
496 /* initial frame state for interrupts (and exceptions without error code) */
497 #define INTR_FRAME _frame RIP
498 /* initial frame state for exceptions with error code (and interrupts with
499 vector already pushed) */
500 #define XCPT_FRAME _frame ORIG_RAX
503 * Interrupt entry/exit.
505 * Interrupt entry points save only callee clobbered registers in fast path.
507 * Entry runs with interrupts off.
510 /* 0(%rsp): interrupt number */
511 .macro interrupt func
514 leaq -ARGOFFSET(%rsp),%rdi # arg1 for handler
516 CFI_ADJUST_CFA_OFFSET 8
517 CFI_REL_OFFSET rbp, 0
519 CFI_DEF_CFA_REGISTER rbp
523 /* irqcount is used to check if a CPU is already on an interrupt
524 stack or not. While this is essentially redundant with preempt_count
525 it is a little cheaper to use a separate counter in the PDA
526 (short of moving irq_enter into assembly, which would be too
528 1: incl %gs:pda_irqcount
529 cmoveq %gs:pda_irqstackptr,%rsp
530 push %rbp # backlink for old unwinder
532 * We entered an interrupt context - irqs are off:
538 ENTRY(common_interrupt)
541 /* 0(%rsp): oldrsp-ARGOFFSET */
543 DISABLE_INTERRUPTS(CLBR_NONE)
545 decl %gs:pda_irqcount
547 CFI_DEF_CFA_REGISTER rsp
548 CFI_ADJUST_CFA_OFFSET -8
550 GET_THREAD_INFO(%rcx)
551 testl $3,CS-ARGOFFSET(%rsp)
554 /* Interrupt came from user space */
556 * Has a correct top of stack, but a partial stack frame
557 * %rcx: thread info. Interrupts off.
559 retint_with_reschedule:
560 movl $_TIF_WORK_MASK,%edi
563 movl threadinfo_flags(%rcx),%edx
568 retint_swapgs: /* return to user-space */
570 * The iretq could re-enable interrupts:
572 DISABLE_INTERRUPTS(CLBR_ANY)
577 retint_restore_args: /* return to kernel space */
578 DISABLE_INTERRUPTS(CLBR_ANY)
580 * The iretq could re-enable interrupts:
586 #ifdef CONFIG_PARAVIRT
592 .section __ex_table,"a"
593 .quad native_iret, bad_iret
596 /* force a signal here? this matches i386 behaviour */
597 /* running with kernel gs */
599 movq $11,%rdi /* SIGSEGV */
601 ENABLE_INTERRUPTS(CLBR_ANY | ~(CLBR_RDI))
605 /* edi: workmask, edx: work */
608 bt $TIF_NEED_RESCHED,%edx
611 ENABLE_INTERRUPTS(CLBR_NONE)
613 CFI_ADJUST_CFA_OFFSET 8
616 CFI_ADJUST_CFA_OFFSET -8
617 GET_THREAD_INFO(%rcx)
618 DISABLE_INTERRUPTS(CLBR_NONE)
623 testl $_TIF_DO_NOTIFY_MASK,%edx
626 ENABLE_INTERRUPTS(CLBR_NONE)
628 movq $-1,ORIG_RAX(%rsp)
629 xorl %esi,%esi # oldset
630 movq %rsp,%rdi # &pt_regs
631 call do_notify_resume
633 DISABLE_INTERRUPTS(CLBR_NONE)
635 movl $_TIF_NEED_RESCHED,%edi
636 GET_THREAD_INFO(%rcx)
639 #ifdef CONFIG_PREEMPT
640 /* Returning to kernel space. Check if we need preemption */
641 /* rcx: threadinfo. interrupts off. */
643 cmpl $0,threadinfo_preempt_count(%rcx)
644 jnz retint_restore_args
645 bt $TIF_NEED_RESCHED,threadinfo_flags(%rcx)
646 jnc retint_restore_args
647 bt $9,EFLAGS-ARGOFFSET(%rsp) /* interrupts off? */
648 jnc retint_restore_args
649 call preempt_schedule_irq
654 END(common_interrupt)
659 .macro apicinterrupt num,func
662 CFI_ADJUST_CFA_OFFSET 8
668 ENTRY(thermal_interrupt)
669 apicinterrupt THERMAL_APIC_VECTOR,smp_thermal_interrupt
670 END(thermal_interrupt)
672 ENTRY(threshold_interrupt)
673 apicinterrupt THRESHOLD_APIC_VECTOR,mce_threshold_interrupt
674 END(threshold_interrupt)
677 ENTRY(reschedule_interrupt)
678 apicinterrupt RESCHEDULE_VECTOR,smp_reschedule_interrupt
679 END(reschedule_interrupt)
681 .macro INVALIDATE_ENTRY num
682 ENTRY(invalidate_interrupt\num)
683 apicinterrupt INVALIDATE_TLB_VECTOR_START+\num,smp_invalidate_interrupt
684 END(invalidate_interrupt\num)
696 ENTRY(call_function_interrupt)
697 apicinterrupt CALL_FUNCTION_VECTOR,smp_call_function_interrupt
698 END(call_function_interrupt)
699 ENTRY(irq_move_cleanup_interrupt)
700 apicinterrupt IRQ_MOVE_CLEANUP_VECTOR,smp_irq_move_cleanup_interrupt
701 END(irq_move_cleanup_interrupt)
704 ENTRY(apic_timer_interrupt)
705 apicinterrupt LOCAL_TIMER_VECTOR,smp_apic_timer_interrupt
706 END(apic_timer_interrupt)
708 ENTRY(error_interrupt)
709 apicinterrupt ERROR_APIC_VECTOR,smp_error_interrupt
712 ENTRY(spurious_interrupt)
713 apicinterrupt SPURIOUS_APIC_VECTOR,smp_spurious_interrupt
714 END(spurious_interrupt)
717 * Exception entry points.
721 pushq $0 /* push error code/oldrax */
722 CFI_ADJUST_CFA_OFFSET 8
723 pushq %rax /* push real oldrax to the rdi slot */
724 CFI_ADJUST_CFA_OFFSET 8
731 .macro errorentry sym
734 CFI_ADJUST_CFA_OFFSET 8
741 /* error code is on the stack already */
742 /* handle NMI like exceptions that can happen everywhere */
743 .macro paranoidentry sym, ist=0, irqtrace=1
747 movl $MSR_GS_BASE,%ecx
755 movq %gs:pda_data_offset, %rbp
758 movq ORIG_RAX(%rsp),%rsi
759 movq $-1,ORIG_RAX(%rsp)
761 subq $EXCEPTION_STKSZ, per_cpu__init_tss + TSS_ist + (\ist - 1) * 8(%rbp)
765 addq $EXCEPTION_STKSZ, per_cpu__init_tss + TSS_ist + (\ist - 1) * 8(%rbp)
767 DISABLE_INTERRUPTS(CLBR_NONE)
774 * "Paranoid" exit path from exception stack.
775 * Paranoid because this is used by NMIs and cannot take
776 * any kernel state for granted.
777 * We don't do kernel preemption checks here, because only
778 * NMI should be common and it does not enable IRQs and
779 * cannot get reschedule ticks.
781 * "trace" is 0 for the NMI handler only, because irq-tracing
782 * is fundamentally NMI-unsafe. (we cannot change the soft and
783 * hard flags at once, atomically)
785 .macro paranoidexit trace=1
786 /* ebx: no swapgs flag */
788 testl %ebx,%ebx /* swapgs needed? */
789 jnz paranoid_restore\trace
791 jnz paranoid_userspace\trace
792 paranoid_swapgs\trace:
797 paranoid_restore\trace:
800 paranoid_userspace\trace:
801 GET_THREAD_INFO(%rcx)
802 movl threadinfo_flags(%rcx),%ebx
803 andl $_TIF_WORK_MASK,%ebx
804 jz paranoid_swapgs\trace
805 movq %rsp,%rdi /* &pt_regs */
807 movq %rax,%rsp /* switch stack for scheduling */
808 testl $_TIF_NEED_RESCHED,%ebx
809 jnz paranoid_schedule\trace
810 movl %ebx,%edx /* arg3: thread flags */
814 ENABLE_INTERRUPTS(CLBR_NONE)
815 xorl %esi,%esi /* arg2: oldset */
816 movq %rsp,%rdi /* arg1: &pt_regs */
817 call do_notify_resume
818 DISABLE_INTERRUPTS(CLBR_NONE)
822 jmp paranoid_userspace\trace
823 paranoid_schedule\trace:
827 ENABLE_INTERRUPTS(CLBR_ANY)
829 DISABLE_INTERRUPTS(CLBR_ANY)
833 jmp paranoid_userspace\trace
838 * Exception entry point. This expects an error code/orig_rax on the stack
839 * and the exception handler in %rax.
841 KPROBE_ENTRY(error_entry)
844 /* rdi slot contains rax, oldrax contains error code */
847 CFI_ADJUST_CFA_OFFSET (14*8)
849 CFI_REL_OFFSET rsi,RSI
850 movq 14*8(%rsp),%rsi /* load rax from rdi slot */
853 CFI_REL_OFFSET rdx,RDX
855 CFI_REL_OFFSET rcx,RCX
856 movq %rsi,10*8(%rsp) /* store rax */
857 CFI_REL_OFFSET rax,RAX
863 CFI_REL_OFFSET r10,R10
865 CFI_REL_OFFSET r11,R11
867 CFI_REL_OFFSET rbx,RBX
869 CFI_REL_OFFSET rbp,RBP
871 CFI_REL_OFFSET r12,R12
873 CFI_REL_OFFSET r13,R13
875 CFI_REL_OFFSET r14,R14
877 CFI_REL_OFFSET r15,R15
885 CFI_REL_OFFSET rdi,RDI
887 movq ORIG_RAX(%rsp),%rsi /* get error code */
888 movq $-1,ORIG_RAX(%rsp)
890 /* ebx: no swapgs flag (1: don't need swapgs, 0: need it) */
894 DISABLE_INTERRUPTS(CLBR_NONE)
896 GET_THREAD_INFO(%rcx)
900 movl threadinfo_flags(%rcx),%edx
901 movl $_TIF_WORK_MASK,%edi
909 /* There are two places in the kernel that can potentially fault with
910 usergs. Handle them here. The exception handlers after
911 iret run with kernel gs again, so don't set the user space flag.
912 B stepping K8s sometimes report an truncated RIP for IRET
913 exceptions returning to compat mode. Check for these here too. */
914 leaq iret_label(%rip),%rbp
917 movl %ebp,%ebp /* zero extend */
920 cmpq $gs_change,RIP(%rsp)
923 KPROBE_END(error_entry)
925 /* Reload gs selector with exception handling */
926 /* edi: new selector */
930 CFI_ADJUST_CFA_OFFSET 8
931 DISABLE_INTERRUPTS(CLBR_ANY | ~(CLBR_RDI))
935 2: mfence /* workaround */
938 CFI_ADJUST_CFA_OFFSET -8
941 ENDPROC(load_gs_index)
943 .section __ex_table,"a"
945 .quad gs_change,bad_gs
948 /* running with kernelgs */
950 SWAPGS /* switch back to user gs */
957 * Create a kernel thread.
959 * C extern interface:
960 * extern long kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
962 * asm input arguments:
963 * rdi: fn, rsi: arg, rdx: flags
967 FAKE_STACK_FRAME $child_rip
970 # rdi: flags, rsi: usp, rdx: will be &pt_regs
972 orq kernel_thread_flags(%rip),%rdi
985 * It isn't worth to check for reschedule here,
986 * so internally to the x86_64 port you can rely on kernel_thread()
987 * not to reschedule the child before returning, this avoids the need
988 * of hacks for example to fork off the per-CPU idle tasks.
989 * [Hopefully no generic code relies on the reschedule -AK]
995 ENDPROC(kernel_thread)
998 pushq $0 # fake return address
1001 * Here we are in the child and the registers are set as they were
1002 * at kernel_thread() invocation in the parent.
1014 * execve(). This function needs to use IRET, not SYSRET, to set up all state properly.
1016 * C extern interface:
1017 * extern long execve(char *name, char **argv, char **envp)
1019 * asm input arguments:
1020 * rdi: name, rsi: argv, rdx: envp
1022 * We want to fallback into:
1023 * extern long sys_execve(char *name, char **argv,char **envp, struct pt_regs regs)
1025 * do_sys_execve asm fallback arguments:
1026 * rdi: name, rsi: argv, rdx: envp, fake frame on the stack
1028 ENTRY(kernel_execve)
1033 movq %rax, RAX(%rsp)
1036 je int_ret_from_sys_call
1041 ENDPROC(kernel_execve)
1043 KPROBE_ENTRY(page_fault)
1044 errorentry do_page_fault
1045 KPROBE_END(page_fault)
1047 ENTRY(coprocessor_error)
1048 zeroentry do_coprocessor_error
1049 END(coprocessor_error)
1051 ENTRY(simd_coprocessor_error)
1052 zeroentry do_simd_coprocessor_error
1053 END(simd_coprocessor_error)
1055 ENTRY(device_not_available)
1056 zeroentry math_state_restore
1057 END(device_not_available)
1059 /* runs on exception stack */
1063 CFI_ADJUST_CFA_OFFSET 8
1064 paranoidentry do_debug, DEBUG_STACK
1068 /* runs on exception stack */
1072 CFI_ADJUST_CFA_OFFSET 8
1073 paranoidentry do_nmi, 0, 0
1074 #ifdef CONFIG_TRACE_IRQFLAGS
1085 CFI_ADJUST_CFA_OFFSET 8
1086 paranoidentry do_int3, DEBUG_STACK
1092 zeroentry do_overflow
1100 zeroentry do_invalid_op
1103 ENTRY(coprocessor_segment_overrun)
1104 zeroentry do_coprocessor_segment_overrun
1105 END(coprocessor_segment_overrun)
1108 zeroentry do_reserved
1111 /* runs on exception stack */
1114 paranoidentry do_double_fault
1120 errorentry do_invalid_TSS
1123 ENTRY(segment_not_present)
1124 errorentry do_segment_not_present
1125 END(segment_not_present)
1127 /* runs on exception stack */
1128 ENTRY(stack_segment)
1130 paranoidentry do_stack_segment
1135 KPROBE_ENTRY(general_protection)
1136 errorentry do_general_protection
1137 KPROBE_END(general_protection)
1139 ENTRY(alignment_check)
1140 errorentry do_alignment_check
1141 END(alignment_check)
1144 zeroentry do_divide_error
1147 ENTRY(spurious_interrupt_bug)
1148 zeroentry do_spurious_interrupt_bug
1149 END(spurious_interrupt_bug)
1151 #ifdef CONFIG_X86_MCE
1152 /* runs on exception stack */
1153 ENTRY(machine_check)
1156 CFI_ADJUST_CFA_OFFSET 8
1157 paranoidentry do_machine_check
1163 /* Call softirq on interrupt stack. Interrupts are off. */
1167 CFI_ADJUST_CFA_OFFSET 8
1168 CFI_REL_OFFSET rbp,0
1170 CFI_DEF_CFA_REGISTER rbp
1171 incl %gs:pda_irqcount
1172 cmove %gs:pda_irqstackptr,%rsp
1173 push %rbp # backlink for old unwinder
1176 CFI_DEF_CFA_REGISTER rsp
1177 CFI_ADJUST_CFA_OFFSET -8
1178 decl %gs:pda_irqcount
1181 ENDPROC(call_softirq)
1183 KPROBE_ENTRY(ignore_sysret)
1188 ENDPROC(ignore_sysret)