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:
589 .section __ex_table, "a"
590 .quad irq_return, bad_iret
593 #ifdef CONFIG_PARAVIRT
597 .section __ex_table,"a"
598 .quad native_iret, bad_iret
605 * The iret traps when the %cs or %ss being restored is bogus.
606 * We've lost the original trap vector and error code.
607 * #GPF is the most likely one to get for an invalid selector.
608 * So pretend we completed the iret and took the #GPF in user mode.
610 * We are now running with the kernel GS after exception recovery.
611 * But error_entry expects us to have user GS to match the user %cs,
617 jmp general_protection
621 /* edi: workmask, edx: work */
624 bt $TIF_NEED_RESCHED,%edx
627 ENABLE_INTERRUPTS(CLBR_NONE)
629 CFI_ADJUST_CFA_OFFSET 8
632 CFI_ADJUST_CFA_OFFSET -8
633 GET_THREAD_INFO(%rcx)
634 DISABLE_INTERRUPTS(CLBR_NONE)
639 testl $_TIF_DO_NOTIFY_MASK,%edx
642 ENABLE_INTERRUPTS(CLBR_NONE)
644 movq $-1,ORIG_RAX(%rsp)
645 xorl %esi,%esi # oldset
646 movq %rsp,%rdi # &pt_regs
647 call do_notify_resume
649 DISABLE_INTERRUPTS(CLBR_NONE)
651 movl $_TIF_NEED_RESCHED,%edi
652 GET_THREAD_INFO(%rcx)
655 #ifdef CONFIG_PREEMPT
656 /* Returning to kernel space. Check if we need preemption */
657 /* rcx: threadinfo. interrupts off. */
659 cmpl $0,threadinfo_preempt_count(%rcx)
660 jnz retint_restore_args
661 bt $TIF_NEED_RESCHED,threadinfo_flags(%rcx)
662 jnc retint_restore_args
663 bt $9,EFLAGS-ARGOFFSET(%rsp) /* interrupts off? */
664 jnc retint_restore_args
665 call preempt_schedule_irq
670 END(common_interrupt)
675 .macro apicinterrupt num,func
678 CFI_ADJUST_CFA_OFFSET 8
684 ENTRY(thermal_interrupt)
685 apicinterrupt THERMAL_APIC_VECTOR,smp_thermal_interrupt
686 END(thermal_interrupt)
688 ENTRY(threshold_interrupt)
689 apicinterrupt THRESHOLD_APIC_VECTOR,mce_threshold_interrupt
690 END(threshold_interrupt)
693 ENTRY(reschedule_interrupt)
694 apicinterrupt RESCHEDULE_VECTOR,smp_reschedule_interrupt
695 END(reschedule_interrupt)
697 .macro INVALIDATE_ENTRY num
698 ENTRY(invalidate_interrupt\num)
699 apicinterrupt INVALIDATE_TLB_VECTOR_START+\num,smp_invalidate_interrupt
700 END(invalidate_interrupt\num)
712 ENTRY(call_function_interrupt)
713 apicinterrupt CALL_FUNCTION_VECTOR,smp_call_function_interrupt
714 END(call_function_interrupt)
715 ENTRY(irq_move_cleanup_interrupt)
716 apicinterrupt IRQ_MOVE_CLEANUP_VECTOR,smp_irq_move_cleanup_interrupt
717 END(irq_move_cleanup_interrupt)
720 ENTRY(apic_timer_interrupt)
721 apicinterrupt LOCAL_TIMER_VECTOR,smp_apic_timer_interrupt
722 END(apic_timer_interrupt)
724 ENTRY(error_interrupt)
725 apicinterrupt ERROR_APIC_VECTOR,smp_error_interrupt
728 ENTRY(spurious_interrupt)
729 apicinterrupt SPURIOUS_APIC_VECTOR,smp_spurious_interrupt
730 END(spurious_interrupt)
733 * Exception entry points.
737 pushq $0 /* push error code/oldrax */
738 CFI_ADJUST_CFA_OFFSET 8
739 pushq %rax /* push real oldrax to the rdi slot */
740 CFI_ADJUST_CFA_OFFSET 8
747 .macro errorentry sym
750 CFI_ADJUST_CFA_OFFSET 8
757 /* error code is on the stack already */
758 /* handle NMI like exceptions that can happen everywhere */
759 .macro paranoidentry sym, ist=0, irqtrace=1
763 movl $MSR_GS_BASE,%ecx
771 movq %gs:pda_data_offset, %rbp
774 movq ORIG_RAX(%rsp),%rsi
775 movq $-1,ORIG_RAX(%rsp)
777 subq $EXCEPTION_STKSZ, per_cpu__init_tss + TSS_ist + (\ist - 1) * 8(%rbp)
781 addq $EXCEPTION_STKSZ, per_cpu__init_tss + TSS_ist + (\ist - 1) * 8(%rbp)
783 DISABLE_INTERRUPTS(CLBR_NONE)
790 * "Paranoid" exit path from exception stack.
791 * Paranoid because this is used by NMIs and cannot take
792 * any kernel state for granted.
793 * We don't do kernel preemption checks here, because only
794 * NMI should be common and it does not enable IRQs and
795 * cannot get reschedule ticks.
797 * "trace" is 0 for the NMI handler only, because irq-tracing
798 * is fundamentally NMI-unsafe. (we cannot change the soft and
799 * hard flags at once, atomically)
801 .macro paranoidexit trace=1
802 /* ebx: no swapgs flag */
804 testl %ebx,%ebx /* swapgs needed? */
805 jnz paranoid_restore\trace
807 jnz paranoid_userspace\trace
808 paranoid_swapgs\trace:
813 paranoid_restore\trace:
816 paranoid_userspace\trace:
817 GET_THREAD_INFO(%rcx)
818 movl threadinfo_flags(%rcx),%ebx
819 andl $_TIF_WORK_MASK,%ebx
820 jz paranoid_swapgs\trace
821 movq %rsp,%rdi /* &pt_regs */
823 movq %rax,%rsp /* switch stack for scheduling */
824 testl $_TIF_NEED_RESCHED,%ebx
825 jnz paranoid_schedule\trace
826 movl %ebx,%edx /* arg3: thread flags */
830 ENABLE_INTERRUPTS(CLBR_NONE)
831 xorl %esi,%esi /* arg2: oldset */
832 movq %rsp,%rdi /* arg1: &pt_regs */
833 call do_notify_resume
834 DISABLE_INTERRUPTS(CLBR_NONE)
838 jmp paranoid_userspace\trace
839 paranoid_schedule\trace:
843 ENABLE_INTERRUPTS(CLBR_ANY)
845 DISABLE_INTERRUPTS(CLBR_ANY)
849 jmp paranoid_userspace\trace
854 * Exception entry point. This expects an error code/orig_rax on the stack
855 * and the exception handler in %rax.
857 KPROBE_ENTRY(error_entry)
860 /* rdi slot contains rax, oldrax contains error code */
863 CFI_ADJUST_CFA_OFFSET (14*8)
865 CFI_REL_OFFSET rsi,RSI
866 movq 14*8(%rsp),%rsi /* load rax from rdi slot */
869 CFI_REL_OFFSET rdx,RDX
871 CFI_REL_OFFSET rcx,RCX
872 movq %rsi,10*8(%rsp) /* store rax */
873 CFI_REL_OFFSET rax,RAX
879 CFI_REL_OFFSET r10,R10
881 CFI_REL_OFFSET r11,R11
883 CFI_REL_OFFSET rbx,RBX
885 CFI_REL_OFFSET rbp,RBP
887 CFI_REL_OFFSET r12,R12
889 CFI_REL_OFFSET r13,R13
891 CFI_REL_OFFSET r14,R14
893 CFI_REL_OFFSET r15,R15
901 CFI_REL_OFFSET rdi,RDI
903 movq ORIG_RAX(%rsp),%rsi /* get error code */
904 movq $-1,ORIG_RAX(%rsp)
906 /* ebx: no swapgs flag (1: don't need swapgs, 0: need it) */
910 DISABLE_INTERRUPTS(CLBR_NONE)
912 GET_THREAD_INFO(%rcx)
916 movl threadinfo_flags(%rcx),%edx
917 movl $_TIF_WORK_MASK,%edi
925 /* There are two places in the kernel that can potentially fault with
926 usergs. Handle them here. The exception handlers after
927 iret run with kernel gs again, so don't set the user space flag.
928 B stepping K8s sometimes report an truncated RIP for IRET
929 exceptions returning to compat mode. Check for these here too. */
930 leaq irq_return(%rip),%rbp
933 movl %ebp,%ebp /* zero extend */
936 cmpq $gs_change,RIP(%rsp)
939 KPROBE_END(error_entry)
941 /* Reload gs selector with exception handling */
942 /* edi: new selector */
946 CFI_ADJUST_CFA_OFFSET 8
947 DISABLE_INTERRUPTS(CLBR_ANY | ~(CLBR_RDI))
951 2: mfence /* workaround */
954 CFI_ADJUST_CFA_OFFSET -8
957 ENDPROC(load_gs_index)
959 .section __ex_table,"a"
961 .quad gs_change,bad_gs
964 /* running with kernelgs */
966 SWAPGS /* switch back to user gs */
973 * Create a kernel thread.
975 * C extern interface:
976 * extern long kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
978 * asm input arguments:
979 * rdi: fn, rsi: arg, rdx: flags
983 FAKE_STACK_FRAME $child_rip
986 # rdi: flags, rsi: usp, rdx: will be &pt_regs
988 orq kernel_thread_flags(%rip),%rdi
1001 * It isn't worth to check for reschedule here,
1002 * so internally to the x86_64 port you can rely on kernel_thread()
1003 * not to reschedule the child before returning, this avoids the need
1004 * of hacks for example to fork off the per-CPU idle tasks.
1005 * [Hopefully no generic code relies on the reschedule -AK]
1011 ENDPROC(kernel_thread)
1014 pushq $0 # fake return address
1017 * Here we are in the child and the registers are set as they were
1018 * at kernel_thread() invocation in the parent.
1030 * execve(). This function needs to use IRET, not SYSRET, to set up all state properly.
1032 * C extern interface:
1033 * extern long execve(char *name, char **argv, char **envp)
1035 * asm input arguments:
1036 * rdi: name, rsi: argv, rdx: envp
1038 * We want to fallback into:
1039 * extern long sys_execve(char *name, char **argv,char **envp, struct pt_regs regs)
1041 * do_sys_execve asm fallback arguments:
1042 * rdi: name, rsi: argv, rdx: envp, fake frame on the stack
1044 ENTRY(kernel_execve)
1049 movq %rax, RAX(%rsp)
1052 je int_ret_from_sys_call
1057 ENDPROC(kernel_execve)
1059 KPROBE_ENTRY(page_fault)
1060 errorentry do_page_fault
1061 KPROBE_END(page_fault)
1063 ENTRY(coprocessor_error)
1064 zeroentry do_coprocessor_error
1065 END(coprocessor_error)
1067 ENTRY(simd_coprocessor_error)
1068 zeroentry do_simd_coprocessor_error
1069 END(simd_coprocessor_error)
1071 ENTRY(device_not_available)
1072 zeroentry math_state_restore
1073 END(device_not_available)
1075 /* runs on exception stack */
1079 CFI_ADJUST_CFA_OFFSET 8
1080 paranoidentry do_debug, DEBUG_STACK
1084 /* runs on exception stack */
1088 CFI_ADJUST_CFA_OFFSET 8
1089 paranoidentry do_nmi, 0, 0
1090 #ifdef CONFIG_TRACE_IRQFLAGS
1101 CFI_ADJUST_CFA_OFFSET 8
1102 paranoidentry do_int3, DEBUG_STACK
1108 zeroentry do_overflow
1116 zeroentry do_invalid_op
1119 ENTRY(coprocessor_segment_overrun)
1120 zeroentry do_coprocessor_segment_overrun
1121 END(coprocessor_segment_overrun)
1124 zeroentry do_reserved
1127 /* runs on exception stack */
1130 paranoidentry do_double_fault
1136 errorentry do_invalid_TSS
1139 ENTRY(segment_not_present)
1140 errorentry do_segment_not_present
1141 END(segment_not_present)
1143 /* runs on exception stack */
1144 ENTRY(stack_segment)
1146 paranoidentry do_stack_segment
1151 KPROBE_ENTRY(general_protection)
1152 errorentry do_general_protection
1153 KPROBE_END(general_protection)
1155 ENTRY(alignment_check)
1156 errorentry do_alignment_check
1157 END(alignment_check)
1160 zeroentry do_divide_error
1163 ENTRY(spurious_interrupt_bug)
1164 zeroentry do_spurious_interrupt_bug
1165 END(spurious_interrupt_bug)
1167 #ifdef CONFIG_X86_MCE
1168 /* runs on exception stack */
1169 ENTRY(machine_check)
1172 CFI_ADJUST_CFA_OFFSET 8
1173 paranoidentry do_machine_check
1179 /* Call softirq on interrupt stack. Interrupts are off. */
1183 CFI_ADJUST_CFA_OFFSET 8
1184 CFI_REL_OFFSET rbp,0
1186 CFI_DEF_CFA_REGISTER rbp
1187 incl %gs:pda_irqcount
1188 cmove %gs:pda_irqstackptr,%rsp
1189 push %rbp # backlink for old unwinder
1192 CFI_DEF_CFA_REGISTER rsp
1193 CFI_ADJUST_CFA_OFFSET -8
1194 decl %gs:pda_irqcount
1197 ENDPROC(call_softirq)
1199 KPROBE_ENTRY(ignore_sysret)
1204 ENDPROC(ignore_sysret)