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>
55 #include <asm/percpu.h>
57 /* Avoid __ASSEMBLER__'ifying <linux/audit.h> just for this. */
58 #include <linux/elf-em.h>
59 #define AUDIT_ARCH_X86_64 (EM_X86_64|__AUDIT_ARCH_64BIT|__AUDIT_ARCH_LE)
60 #define __AUDIT_ARCH_64BIT 0x80000000
61 #define __AUDIT_ARCH_LE 0x40000000
64 #ifdef CONFIG_FUNCTION_TRACER
65 #ifdef CONFIG_DYNAMIC_FTRACE
71 cmpl $0, function_trace_stop
78 subq $MCOUNT_INSN_SIZE, %rdi
86 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
87 .globl ftrace_graph_call
97 #else /* ! CONFIG_DYNAMIC_FTRACE */
99 cmpl $0, function_trace_stop
102 cmpq $ftrace_stub, ftrace_trace_function
105 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
106 cmpq $ftrace_stub, ftrace_graph_return
107 jnz ftrace_graph_caller
109 cmpq $ftrace_graph_entry_stub, ftrace_graph_entry
110 jnz ftrace_graph_caller
120 movq 0x38(%rsp), %rdi
122 subq $MCOUNT_INSN_SIZE, %rdi
124 call *ftrace_trace_function
130 #endif /* CONFIG_DYNAMIC_FTRACE */
131 #endif /* CONFIG_FUNCTION_TRACER */
133 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
134 ENTRY(ftrace_graph_caller)
135 cmpl $0, function_trace_stop
141 movq 0x38(%rsp), %rsi
142 subq $MCOUNT_INSN_SIZE, %rsi
144 call prepare_ftrace_return
149 END(ftrace_graph_caller)
152 .globl return_to_handler
166 call ftrace_return_to_handler
183 #ifndef CONFIG_PREEMPT
184 #define retint_kernel retint_restore_args
187 #ifdef CONFIG_PARAVIRT
188 ENTRY(native_usergs_sysret64)
191 #endif /* CONFIG_PARAVIRT */
194 .macro TRACE_IRQS_IRETQ offset=ARGOFFSET
195 #ifdef CONFIG_TRACE_IRQFLAGS
196 bt $9,EFLAGS-\offset(%rsp) /* interrupts off? */
204 * C code is not supposed to know about undefined top of stack. Every time
205 * a C function with an pt_regs argument is called from the SYSCALL based
206 * fast path FIXUP_TOP_OF_STACK is needed.
207 * RESTORE_TOP_OF_STACK syncs the syscall state after any possible ptregs
211 /* %rsp:at FRAMEEND */
212 .macro FIXUP_TOP_OF_STACK tmp offset=0
213 movq PER_CPU_VAR(old_rsp),\tmp
214 movq \tmp,RSP+\offset(%rsp)
215 movq $__USER_DS,SS+\offset(%rsp)
216 movq $__USER_CS,CS+\offset(%rsp)
217 movq $-1,RCX+\offset(%rsp)
218 movq R11+\offset(%rsp),\tmp /* get eflags */
219 movq \tmp,EFLAGS+\offset(%rsp)
222 .macro RESTORE_TOP_OF_STACK tmp offset=0
223 movq RSP+\offset(%rsp),\tmp
224 movq \tmp,PER_CPU_VAR(old_rsp)
225 movq EFLAGS+\offset(%rsp),\tmp
226 movq \tmp,R11+\offset(%rsp)
229 .macro FAKE_STACK_FRAME child_rip
230 /* push in order ss, rsp, eflags, cs, rip */
232 pushq $__KERNEL_DS /* ss */
233 CFI_ADJUST_CFA_OFFSET 8
234 /*CFI_REL_OFFSET ss,0*/
236 CFI_ADJUST_CFA_OFFSET 8
238 pushq $X86_EFLAGS_IF /* eflags - interrupts on */
239 CFI_ADJUST_CFA_OFFSET 8
240 /*CFI_REL_OFFSET rflags,0*/
241 pushq $__KERNEL_CS /* cs */
242 CFI_ADJUST_CFA_OFFSET 8
243 /*CFI_REL_OFFSET cs,0*/
244 pushq \child_rip /* rip */
245 CFI_ADJUST_CFA_OFFSET 8
247 pushq %rax /* orig rax */
248 CFI_ADJUST_CFA_OFFSET 8
251 .macro UNFAKE_STACK_FRAME
253 CFI_ADJUST_CFA_OFFSET -(6*8)
257 * initial frame state for interrupts (and exceptions without error code)
259 .macro EMPTY_FRAME start=1 offset=0
263 CFI_DEF_CFA rsp,8+\offset
265 CFI_DEF_CFA_OFFSET 8+\offset
270 * initial frame state for interrupts (and exceptions without error code)
272 .macro INTR_FRAME start=1 offset=0
273 EMPTY_FRAME \start, SS+8+\offset-RIP
274 /*CFI_REL_OFFSET ss, SS+\offset-RIP*/
275 CFI_REL_OFFSET rsp, RSP+\offset-RIP
276 /*CFI_REL_OFFSET rflags, EFLAGS+\offset-RIP*/
277 /*CFI_REL_OFFSET cs, CS+\offset-RIP*/
278 CFI_REL_OFFSET rip, RIP+\offset-RIP
282 * initial frame state for exceptions with error code (and interrupts
283 * with vector already pushed)
285 .macro XCPT_FRAME start=1 offset=0
286 INTR_FRAME \start, RIP+\offset-ORIG_RAX
287 /*CFI_REL_OFFSET orig_rax, ORIG_RAX-ORIG_RAX*/
291 * frame that enables calling into C.
293 .macro PARTIAL_FRAME start=1 offset=0
294 XCPT_FRAME \start, ORIG_RAX+\offset-ARGOFFSET
295 CFI_REL_OFFSET rdi, RDI+\offset-ARGOFFSET
296 CFI_REL_OFFSET rsi, RSI+\offset-ARGOFFSET
297 CFI_REL_OFFSET rdx, RDX+\offset-ARGOFFSET
298 CFI_REL_OFFSET rcx, RCX+\offset-ARGOFFSET
299 CFI_REL_OFFSET rax, RAX+\offset-ARGOFFSET
300 CFI_REL_OFFSET r8, R8+\offset-ARGOFFSET
301 CFI_REL_OFFSET r9, R9+\offset-ARGOFFSET
302 CFI_REL_OFFSET r10, R10+\offset-ARGOFFSET
303 CFI_REL_OFFSET r11, R11+\offset-ARGOFFSET
307 * frame that enables passing a complete pt_regs to a C function.
309 .macro DEFAULT_FRAME start=1 offset=0
310 PARTIAL_FRAME \start, R11+\offset-R15
311 CFI_REL_OFFSET rbx, RBX+\offset
312 CFI_REL_OFFSET rbp, RBP+\offset
313 CFI_REL_OFFSET r12, R12+\offset
314 CFI_REL_OFFSET r13, R13+\offset
315 CFI_REL_OFFSET r14, R14+\offset
316 CFI_REL_OFFSET r15, R15+\offset
319 /* save partial stack frame */
323 movq_cfi rdi, RDI+16-ARGOFFSET
324 movq_cfi rsi, RSI+16-ARGOFFSET
325 movq_cfi rdx, RDX+16-ARGOFFSET
326 movq_cfi rcx, RCX+16-ARGOFFSET
327 movq_cfi rax, RAX+16-ARGOFFSET
328 movq_cfi r8, R8+16-ARGOFFSET
329 movq_cfi r9, R9+16-ARGOFFSET
330 movq_cfi r10, R10+16-ARGOFFSET
331 movq_cfi r11, R11+16-ARGOFFSET
333 leaq -ARGOFFSET+16(%rsp),%rdi /* arg1 for handler */
334 movq_cfi rbp, 8 /* push %rbp */
335 leaq 8(%rsp), %rbp /* mov %rsp, %ebp */
340 * irq_count is used to check if a CPU is already on an interrupt stack
341 * or not. While this is essentially redundant with preempt_count it is
342 * a little cheaper to use a separate counter in the PDA (short of
343 * moving irq_enter into assembly, which would be too much work)
345 1: incl PER_CPU_VAR(irq_count)
347 popq_cfi %rax /* move return address... */
348 mov PER_CPU_VAR(irq_stack_ptr),%rsp
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 LOCK ; btr $TIF_FORK,TI_flags(%r8)
414 push kernel_eflags(%rip)
415 CFI_ADJUST_CFA_OFFSET 8
416 popf # reset kernel eflags
417 CFI_ADJUST_CFA_OFFSET -8
419 call schedule_tail # rdi: 'prev' task parameter
421 GET_THREAD_INFO(%rcx)
426 testl $3, CS-ARGOFFSET(%rsp) # from kernel_thread?
427 je int_ret_from_sys_call
429 testl $_TIF_IA32, TI_flags(%rcx) # 32-bit compat task needs IRET
430 jnz int_ret_from_sys_call
432 RESTORE_TOP_OF_STACK %rdi, -ARGOFFSET
433 jmp ret_from_sys_call # go to the SYSRET fastpath
440 * System call entry. Upto 6 arguments in registers are supported.
442 * SYSCALL does not save anything on the stack and does not change the
448 * rax system call number
450 * rcx return address for syscall/sysret, C arg3
453 * r10 arg3 (--> moved to rcx for C)
456 * r11 eflags for syscall/sysret, temporary for C
457 * r12-r15,rbp,rbx saved by C code, not touched.
459 * Interrupts are off on entry.
460 * Only called from user space.
462 * XXX if we had a free scratch register we could save the RSP into the stack frame
463 * and report it properly in ps. Unfortunately we haven't.
465 * When user can change the frames always force IRET. That is because
466 * it deals with uncanonical addresses better. SYSRET has trouble
467 * with them due to bugs in both AMD and Intel CPUs.
473 CFI_DEF_CFA rsp,KERNEL_STACK_OFFSET
475 /*CFI_REGISTER rflags,r11*/
478 * A hypervisor implementation might want to use a label
479 * after the swapgs, so that it can do the swapgs
480 * for the guest and jump here on syscall.
482 ENTRY(system_call_after_swapgs)
484 movq %rsp,PER_CPU_VAR(old_rsp)
485 movq PER_CPU_VAR(kernel_stack),%rsp
487 * No need to follow this irqs off/on section - it's straight
490 ENABLE_INTERRUPTS(CLBR_NONE)
492 movq %rax,ORIG_RAX-ARGOFFSET(%rsp)
493 movq %rcx,RIP-ARGOFFSET(%rsp)
494 CFI_REL_OFFSET rip,RIP-ARGOFFSET
495 GET_THREAD_INFO(%rcx)
496 testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%rcx)
498 system_call_fastpath:
499 cmpq $__NR_syscall_max,%rax
502 call *sys_call_table(,%rax,8) # XXX: rip relative
503 movq %rax,RAX-ARGOFFSET(%rsp)
505 * Syscall return path ending with SYSRET (fast path)
506 * Has incomplete stack frame and undefined top of stack.
509 movl $_TIF_ALLWORK_MASK,%edi
513 GET_THREAD_INFO(%rcx)
514 DISABLE_INTERRUPTS(CLBR_NONE)
516 movl TI_flags(%rcx),%edx
521 * sysretq will re-enable interrupts:
524 movq RIP-ARGOFFSET(%rsp),%rcx
526 RESTORE_ARGS 0,-ARG_SKIP,1
527 /*CFI_REGISTER rflags,r11*/
528 movq PER_CPU_VAR(old_rsp), %rsp
532 /* Handle reschedules */
533 /* edx: work, edi: workmask */
535 bt $TIF_NEED_RESCHED,%edx
538 ENABLE_INTERRUPTS(CLBR_NONE)
540 CFI_ADJUST_CFA_OFFSET 8
543 CFI_ADJUST_CFA_OFFSET -8
546 /* Handle a signal */
549 ENABLE_INTERRUPTS(CLBR_NONE)
550 #ifdef CONFIG_AUDITSYSCALL
551 bt $TIF_SYSCALL_AUDIT,%edx
554 /* edx: work flags (arg3) */
555 leaq -ARGOFFSET(%rsp),%rdi # &pt_regs -> arg1
556 xorl %esi,%esi # oldset -> arg2
558 FIXUP_TOP_OF_STACK %r11
559 call do_notify_resume
560 RESTORE_TOP_OF_STACK %r11
562 movl $_TIF_WORK_MASK,%edi
563 /* Use IRET because user could have changed frame. This
564 works because ptregscall_common has called FIXUP_TOP_OF_STACK. */
565 DISABLE_INTERRUPTS(CLBR_NONE)
570 movq $-ENOSYS,RAX-ARGOFFSET(%rsp)
571 jmp ret_from_sys_call
573 #ifdef CONFIG_AUDITSYSCALL
575 * Fast path for syscall audit without full syscall trace.
576 * We just call audit_syscall_entry() directly, and then
577 * jump back to the normal fast path.
580 movq %r10,%r9 /* 6th arg: 4th syscall arg */
581 movq %rdx,%r8 /* 5th arg: 3rd syscall arg */
582 movq %rsi,%rcx /* 4th arg: 2nd syscall arg */
583 movq %rdi,%rdx /* 3rd arg: 1st syscall arg */
584 movq %rax,%rsi /* 2nd arg: syscall number */
585 movl $AUDIT_ARCH_X86_64,%edi /* 1st arg: audit arch */
586 call audit_syscall_entry
587 LOAD_ARGS 0 /* reload call-clobbered registers */
588 jmp system_call_fastpath
591 * Return fast path for syscall audit. Call audit_syscall_exit()
592 * directly and then jump back to the fast path with TIF_SYSCALL_AUDIT
596 movq %rax,%rsi /* second arg, syscall return value */
597 cmpq $0,%rax /* is it < 0? */
598 setl %al /* 1 if so, 0 if not */
599 movzbl %al,%edi /* zero-extend that into %edi */
600 inc %edi /* first arg, 0->1(AUDITSC_SUCCESS), 1->2(AUDITSC_FAILURE) */
601 call audit_syscall_exit
602 movl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT),%edi
604 #endif /* CONFIG_AUDITSYSCALL */
606 /* Do syscall tracing */
608 #ifdef CONFIG_AUDITSYSCALL
609 testl $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags(%rcx)
613 movq $-ENOSYS,RAX(%rsp) /* ptrace can change this for a bad syscall */
614 FIXUP_TOP_OF_STACK %rdi
616 call syscall_trace_enter
618 * Reload arg registers from stack in case ptrace changed them.
619 * We don't reload %rax because syscall_trace_enter() returned
620 * the value it wants us to use in the table lookup.
622 LOAD_ARGS ARGOFFSET, 1
624 cmpq $__NR_syscall_max,%rax
625 ja int_ret_from_sys_call /* RAX(%rsp) set to -ENOSYS above */
626 movq %r10,%rcx /* fixup for C */
627 call *sys_call_table(,%rax,8)
628 movq %rax,RAX-ARGOFFSET(%rsp)
629 /* Use IRET because user could have changed frame */
632 * Syscall return path ending with IRET.
633 * Has correct top of stack, but partial stack frame.
635 .globl int_ret_from_sys_call
636 .globl int_with_check
637 int_ret_from_sys_call:
638 DISABLE_INTERRUPTS(CLBR_NONE)
640 testl $3,CS-ARGOFFSET(%rsp)
641 je retint_restore_args
642 movl $_TIF_ALLWORK_MASK,%edi
643 /* edi: mask to check */
646 GET_THREAD_INFO(%rcx)
647 movl TI_flags(%rcx),%edx
650 andl $~TS_COMPAT,TI_status(%rcx)
653 /* Either reschedule or signal or syscall exit tracking needed. */
654 /* First do a reschedule test. */
655 /* edx: work, edi: workmask */
657 bt $TIF_NEED_RESCHED,%edx
660 ENABLE_INTERRUPTS(CLBR_NONE)
662 CFI_ADJUST_CFA_OFFSET 8
665 CFI_ADJUST_CFA_OFFSET -8
666 DISABLE_INTERRUPTS(CLBR_NONE)
670 /* handle signals and tracing -- both require a full stack frame */
673 ENABLE_INTERRUPTS(CLBR_NONE)
675 /* Check for syscall exit trace */
676 testl $_TIF_WORK_SYSCALL_EXIT,%edx
679 CFI_ADJUST_CFA_OFFSET 8
680 leaq 8(%rsp),%rdi # &ptregs -> arg1
681 call syscall_trace_leave
683 CFI_ADJUST_CFA_OFFSET -8
684 andl $~(_TIF_WORK_SYSCALL_EXIT|_TIF_SYSCALL_EMU),%edi
688 testl $_TIF_DO_NOTIFY_MASK,%edx
690 movq %rsp,%rdi # &ptregs -> arg1
691 xorl %esi,%esi # oldset -> arg2
692 call do_notify_resume
693 1: movl $_TIF_WORK_MASK,%edi
696 DISABLE_INTERRUPTS(CLBR_NONE)
703 * Certain special system calls that need to save a complete full stack frame.
705 .macro PTREGSCALL label,func,arg
707 PARTIAL_FRAME 1 8 /* offset 8: return address */
708 subq $REST_SKIP, %rsp
709 CFI_ADJUST_CFA_OFFSET REST_SKIP
711 DEFAULT_FRAME 0 8 /* offset 8: return address */
712 leaq 8(%rsp), \arg /* pt_regs pointer */
714 jmp ptregscall_common
719 PTREGSCALL stub_clone, sys_clone, %r8
720 PTREGSCALL stub_fork, sys_fork, %rdi
721 PTREGSCALL stub_vfork, sys_vfork, %rdi
722 PTREGSCALL stub_sigaltstack, sys_sigaltstack, %rdx
723 PTREGSCALL stub_iopl, sys_iopl, %rsi
725 ENTRY(ptregscall_common)
726 DEFAULT_FRAME 1 8 /* offset 8: return address */
727 RESTORE_TOP_OF_STACK %r11, 8
728 movq_cfi_restore R15+8, r15
729 movq_cfi_restore R14+8, r14
730 movq_cfi_restore R13+8, r13
731 movq_cfi_restore R12+8, r12
732 movq_cfi_restore RBP+8, rbp
733 movq_cfi_restore RBX+8, rbx
734 ret $REST_SKIP /* pop extended registers */
736 END(ptregscall_common)
741 CFI_ADJUST_CFA_OFFSET -8
742 CFI_REGISTER rip, r11
744 FIXUP_TOP_OF_STACK %r11
747 RESTORE_TOP_OF_STACK %r11
750 jmp int_ret_from_sys_call
755 * sigreturn is special because it needs to restore all registers on return.
756 * This cannot be done with SYSRET, so use the IRET return path instead.
758 ENTRY(stub_rt_sigreturn)
761 CFI_ADJUST_CFA_OFFSET -8
764 FIXUP_TOP_OF_STACK %r11
765 call sys_rt_sigreturn
766 movq %rax,RAX(%rsp) # fixme, this could be done at the higher layer
768 jmp int_ret_from_sys_call
770 END(stub_rt_sigreturn)
773 * Build the entry stubs and pointer table with some assembler magic.
774 * We pack 7 stubs into a single 32-byte chunk, which will fit in a
775 * single cache line on all modern x86 implementations.
777 .section .init.rodata,"a"
781 .p2align CONFIG_X86_L1_CACHE_SHIFT
782 ENTRY(irq_entries_start)
784 vector=FIRST_EXTERNAL_VECTOR
785 .rept (NR_VECTORS-FIRST_EXTERNAL_VECTOR+6)/7
788 .if vector < NR_VECTORS
789 .if vector <> FIRST_EXTERNAL_VECTOR
790 CFI_ADJUST_CFA_OFFSET -8
792 1: pushq $(~vector+0x80) /* Note: always in signed byte range */
793 CFI_ADJUST_CFA_OFFSET 8
794 .if ((vector-FIRST_EXTERNAL_VECTOR)%7) <> 6
803 2: jmp common_interrupt
806 END(irq_entries_start)
813 * Interrupt entry/exit.
815 * Interrupt entry points save only callee clobbered registers in fast path.
817 * Entry runs with interrupts off.
820 /* 0(%rsp): ~(interrupt number) */
821 .macro interrupt func
823 CFI_ADJUST_CFA_OFFSET 10*8
830 * The interrupt stubs push (~vector+0x80) onto the stack and
831 * then jump to common_interrupt.
833 .p2align CONFIG_X86_L1_CACHE_SHIFT
836 addq $-0x80,(%rsp) /* Adjust vector to [-256,-1] range */
838 /* 0(%rsp): old_rsp-ARGOFFSET */
840 DISABLE_INTERRUPTS(CLBR_NONE)
842 decl PER_CPU_VAR(irq_count)
844 CFI_DEF_CFA_REGISTER rsp
845 CFI_ADJUST_CFA_OFFSET -8
847 GET_THREAD_INFO(%rcx)
848 testl $3,CS-ARGOFFSET(%rsp)
851 /* Interrupt came from user space */
853 * Has a correct top of stack, but a partial stack frame
854 * %rcx: thread info. Interrupts off.
856 retint_with_reschedule:
857 movl $_TIF_WORK_MASK,%edi
860 movl TI_flags(%rcx),%edx
865 retint_swapgs: /* return to user-space */
867 * The iretq could re-enable interrupts:
869 DISABLE_INTERRUPTS(CLBR_ANY)
874 retint_restore_args: /* return to kernel space */
875 DISABLE_INTERRUPTS(CLBR_ANY)
877 * The iretq could re-enable interrupts:
886 .section __ex_table, "a"
887 .quad irq_return, bad_iret
890 #ifdef CONFIG_PARAVIRT
894 .section __ex_table,"a"
895 .quad native_iret, bad_iret
902 * The iret traps when the %cs or %ss being restored is bogus.
903 * We've lost the original trap vector and error code.
904 * #GPF is the most likely one to get for an invalid selector.
905 * So pretend we completed the iret and took the #GPF in user mode.
907 * We are now running with the kernel GS after exception recovery.
908 * But error_entry expects us to have user GS to match the user %cs,
914 jmp general_protection
918 /* edi: workmask, edx: work */
921 bt $TIF_NEED_RESCHED,%edx
924 ENABLE_INTERRUPTS(CLBR_NONE)
926 CFI_ADJUST_CFA_OFFSET 8
929 CFI_ADJUST_CFA_OFFSET -8
930 GET_THREAD_INFO(%rcx)
931 DISABLE_INTERRUPTS(CLBR_NONE)
936 testl $_TIF_DO_NOTIFY_MASK,%edx
939 ENABLE_INTERRUPTS(CLBR_NONE)
941 movq $-1,ORIG_RAX(%rsp)
942 xorl %esi,%esi # oldset
943 movq %rsp,%rdi # &pt_regs
944 call do_notify_resume
946 DISABLE_INTERRUPTS(CLBR_NONE)
948 GET_THREAD_INFO(%rcx)
949 jmp retint_with_reschedule
951 #ifdef CONFIG_PREEMPT
952 /* Returning to kernel space. Check if we need preemption */
953 /* rcx: threadinfo. interrupts off. */
955 cmpl $0,TI_preempt_count(%rcx)
956 jnz retint_restore_args
957 bt $TIF_NEED_RESCHED,TI_flags(%rcx)
958 jnc retint_restore_args
959 bt $9,EFLAGS-ARGOFFSET(%rsp) /* interrupts off? */
960 jnc retint_restore_args
961 call preempt_schedule_irq
966 END(common_interrupt)
971 .macro apicinterrupt num sym do_sym
975 CFI_ADJUST_CFA_OFFSET 8
983 apicinterrupt IRQ_MOVE_CLEANUP_VECTOR \
984 irq_move_cleanup_interrupt smp_irq_move_cleanup_interrupt
988 apicinterrupt UV_BAU_MESSAGE \
989 uv_bau_message_intr1 uv_bau_message_interrupt
991 apicinterrupt LOCAL_TIMER_VECTOR \
992 apic_timer_interrupt smp_apic_timer_interrupt
995 apicinterrupt INVALIDATE_TLB_VECTOR_START+0 \
996 invalidate_interrupt0 smp_invalidate_interrupt
997 apicinterrupt INVALIDATE_TLB_VECTOR_START+1 \
998 invalidate_interrupt1 smp_invalidate_interrupt
999 apicinterrupt INVALIDATE_TLB_VECTOR_START+2 \
1000 invalidate_interrupt2 smp_invalidate_interrupt
1001 apicinterrupt INVALIDATE_TLB_VECTOR_START+3 \
1002 invalidate_interrupt3 smp_invalidate_interrupt
1003 apicinterrupt INVALIDATE_TLB_VECTOR_START+4 \
1004 invalidate_interrupt4 smp_invalidate_interrupt
1005 apicinterrupt INVALIDATE_TLB_VECTOR_START+5 \
1006 invalidate_interrupt5 smp_invalidate_interrupt
1007 apicinterrupt INVALIDATE_TLB_VECTOR_START+6 \
1008 invalidate_interrupt6 smp_invalidate_interrupt
1009 apicinterrupt INVALIDATE_TLB_VECTOR_START+7 \
1010 invalidate_interrupt7 smp_invalidate_interrupt
1013 apicinterrupt THRESHOLD_APIC_VECTOR \
1014 threshold_interrupt mce_threshold_interrupt
1015 apicinterrupt THERMAL_APIC_VECTOR \
1016 thermal_interrupt smp_thermal_interrupt
1019 apicinterrupt CALL_FUNCTION_SINGLE_VECTOR \
1020 call_function_single_interrupt smp_call_function_single_interrupt
1021 apicinterrupt CALL_FUNCTION_VECTOR \
1022 call_function_interrupt smp_call_function_interrupt
1023 apicinterrupt RESCHEDULE_VECTOR \
1024 reschedule_interrupt smp_reschedule_interrupt
1027 apicinterrupt ERROR_APIC_VECTOR \
1028 error_interrupt smp_error_interrupt
1029 apicinterrupt SPURIOUS_APIC_VECTOR \
1030 spurious_interrupt smp_spurious_interrupt
1033 * Exception entry points.
1035 .macro zeroentry sym do_sym
1038 PARAVIRT_ADJUST_EXCEPTION_FRAME
1039 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1041 CFI_ADJUST_CFA_OFFSET 15*8
1044 movq %rsp,%rdi /* pt_regs pointer */
1045 xorl %esi,%esi /* no error code */
1047 jmp error_exit /* %ebx: no swapgs flag */
1052 .macro paranoidzeroentry sym do_sym
1055 PARAVIRT_ADJUST_EXCEPTION_FRAME
1056 pushq $-1 /* ORIG_RAX: no syscall to restart */
1057 CFI_ADJUST_CFA_OFFSET 8
1061 movq %rsp,%rdi /* pt_regs pointer */
1062 xorl %esi,%esi /* no error code */
1064 jmp paranoid_exit /* %ebx: no swapgs flag */
1069 .macro paranoidzeroentry_ist sym do_sym ist
1072 PARAVIRT_ADJUST_EXCEPTION_FRAME
1073 pushq $-1 /* ORIG_RAX: no syscall to restart */
1074 CFI_ADJUST_CFA_OFFSET 8
1078 movq %rsp,%rdi /* pt_regs pointer */
1079 xorl %esi,%esi /* no error code */
1080 PER_CPU(init_tss, %rbp)
1081 subq $EXCEPTION_STKSZ, TSS_ist + (\ist - 1) * 8(%rbp)
1083 addq $EXCEPTION_STKSZ, TSS_ist + (\ist - 1) * 8(%rbp)
1084 jmp paranoid_exit /* %ebx: no swapgs flag */
1089 .macro errorentry sym do_sym
1092 PARAVIRT_ADJUST_EXCEPTION_FRAME
1094 CFI_ADJUST_CFA_OFFSET 15*8
1097 movq %rsp,%rdi /* pt_regs pointer */
1098 movq ORIG_RAX(%rsp),%rsi /* get error code */
1099 movq $-1,ORIG_RAX(%rsp) /* no syscall to restart */
1101 jmp error_exit /* %ebx: no swapgs flag */
1106 /* error code is on the stack already */
1107 .macro paranoiderrorentry sym do_sym
1110 PARAVIRT_ADJUST_EXCEPTION_FRAME
1112 CFI_ADJUST_CFA_OFFSET 15*8
1116 movq %rsp,%rdi /* pt_regs pointer */
1117 movq ORIG_RAX(%rsp),%rsi /* get error code */
1118 movq $-1,ORIG_RAX(%rsp) /* no syscall to restart */
1120 jmp paranoid_exit /* %ebx: no swapgs flag */
1125 zeroentry divide_error do_divide_error
1126 zeroentry overflow do_overflow
1127 zeroentry bounds do_bounds
1128 zeroentry invalid_op do_invalid_op
1129 zeroentry device_not_available do_device_not_available
1130 paranoiderrorentry double_fault do_double_fault
1131 zeroentry coprocessor_segment_overrun do_coprocessor_segment_overrun
1132 errorentry invalid_TSS do_invalid_TSS
1133 errorentry segment_not_present do_segment_not_present
1134 zeroentry spurious_interrupt_bug do_spurious_interrupt_bug
1135 zeroentry coprocessor_error do_coprocessor_error
1136 errorentry alignment_check do_alignment_check
1137 zeroentry simd_coprocessor_error do_simd_coprocessor_error
1139 /* Reload gs selector with exception handling */
1140 /* edi: new selector */
1141 ENTRY(native_load_gs_index)
1144 CFI_ADJUST_CFA_OFFSET 8
1145 DISABLE_INTERRUPTS(CLBR_ANY | ~(CLBR_RDI))
1149 2: mfence /* workaround */
1152 CFI_ADJUST_CFA_OFFSET -8
1155 END(native_load_gs_index)
1157 .section __ex_table,"a"
1159 .quad gs_change,bad_gs
1161 .section .fixup,"ax"
1162 /* running with kernelgs */
1164 SWAPGS /* switch back to user gs */
1171 * Create a kernel thread.
1173 * C extern interface:
1174 * extern long kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
1176 * asm input arguments:
1177 * rdi: fn, rsi: arg, rdx: flags
1179 ENTRY(kernel_thread)
1181 FAKE_STACK_FRAME $child_rip
1184 # rdi: flags, rsi: usp, rdx: will be &pt_regs
1186 orq kernel_thread_flags(%rip),%rdi
1199 * It isn't worth to check for reschedule here,
1200 * so internally to the x86_64 port you can rely on kernel_thread()
1201 * not to reschedule the child before returning, this avoids the need
1202 * of hacks for example to fork off the per-CPU idle tasks.
1203 * [Hopefully no generic code relies on the reschedule -AK]
1212 pushq $0 # fake return address
1215 * Here we are in the child and the registers are set as they were
1216 * at kernel_thread() invocation in the parent.
1224 ud2 # padding for call trace
1229 * execve(). This function needs to use IRET, not SYSRET, to set up all state properly.
1231 * C extern interface:
1232 * extern long execve(char *name, char **argv, char **envp)
1234 * asm input arguments:
1235 * rdi: name, rsi: argv, rdx: envp
1237 * We want to fallback into:
1238 * extern long sys_execve(char *name, char **argv,char **envp, struct pt_regs *regs)
1240 * do_sys_execve asm fallback arguments:
1241 * rdi: name, rsi: argv, rdx: envp, rcx: fake frame on the stack
1243 ENTRY(kernel_execve)
1249 movq %rax, RAX(%rsp)
1252 je int_ret_from_sys_call
1259 /* Call softirq on interrupt stack. Interrupts are off. */
1263 CFI_ADJUST_CFA_OFFSET 8
1264 CFI_REL_OFFSET rbp,0
1266 CFI_DEF_CFA_REGISTER rbp
1267 incl PER_CPU_VAR(irq_count)
1268 cmove PER_CPU_VAR(irq_stack_ptr),%rsp
1269 push %rbp # backlink for old unwinder
1272 CFI_DEF_CFA_REGISTER rsp
1273 CFI_ADJUST_CFA_OFFSET -8
1274 decl PER_CPU_VAR(irq_count)
1280 zeroentry xen_hypervisor_callback xen_do_hypervisor_callback
1283 * A note on the "critical region" in our callback handler.
1284 * We want to avoid stacking callback handlers due to events occurring
1285 * during handling of the last event. To do this, we keep events disabled
1286 * until we've done all processing. HOWEVER, we must enable events before
1287 * popping the stack frame (can't be done atomically) and so it would still
1288 * be possible to get enough handler activations to overflow the stack.
1289 * Although unlikely, bugs of that kind are hard to track down, so we'd
1290 * like to avoid the possibility.
1291 * So, on entry to the handler we detect whether we interrupted an
1292 * existing activation in its critical region -- if so, we pop the current
1293 * activation and restart the handler using the previous one.
1295 ENTRY(xen_do_hypervisor_callback) # do_hypervisor_callback(struct *pt_regs)
1298 * Since we don't modify %rdi, evtchn_do_upall(struct *pt_regs) will
1299 * see the correct pointer to the pt_regs
1301 movq %rdi, %rsp # we don't return, adjust the stack frame
1304 11: incl PER_CPU_VAR(irq_count)
1306 CFI_DEF_CFA_REGISTER rbp
1307 cmovzq PER_CPU_VAR(irq_stack_ptr),%rsp
1308 pushq %rbp # backlink for old unwinder
1309 call xen_evtchn_do_upcall
1311 CFI_DEF_CFA_REGISTER rsp
1312 decl PER_CPU_VAR(irq_count)
1315 END(do_hypervisor_callback)
1318 * Hypervisor uses this for application faults while it executes.
1319 * We get here for two reasons:
1320 * 1. Fault while reloading DS, ES, FS or GS
1321 * 2. Fault while executing IRET
1322 * Category 1 we do not need to fix up as Xen has already reloaded all segment
1323 * registers that could be reloaded and zeroed the others.
1324 * Category 2 we fix up by killing the current process. We cannot use the
1325 * normal Linux return path in this case because if we use the IRET hypercall
1326 * to pop the stack frame we end up in an infinite loop of failsafe callbacks.
1327 * We distinguish between categories by comparing each saved segment register
1328 * with its current contents: any discrepancy means we in category 1.
1330 ENTRY(xen_failsafe_callback)
1332 /*CFI_REL_OFFSET gs,GS*/
1333 /*CFI_REL_OFFSET fs,FS*/
1334 /*CFI_REL_OFFSET es,ES*/
1335 /*CFI_REL_OFFSET ds,DS*/
1336 CFI_REL_OFFSET r11,8
1337 CFI_REL_OFFSET rcx,0
1351 /* All segments match their saved values => Category 2 (Bad IRET). */
1357 CFI_ADJUST_CFA_OFFSET -0x30
1358 pushq_cfi $0 /* RIP */
1361 jmp general_protection
1363 1: /* Segment mismatch => Category 1 (Bad segment). Retry the IRET. */
1369 CFI_ADJUST_CFA_OFFSET -0x30
1374 END(xen_failsafe_callback)
1376 #endif /* CONFIG_XEN */
1379 * Some functions should be protected against kprobes
1381 .pushsection .kprobes.text, "ax"
1383 paranoidzeroentry_ist debug do_debug DEBUG_STACK
1384 paranoidzeroentry_ist int3 do_int3 DEBUG_STACK
1385 paranoiderrorentry stack_segment do_stack_segment
1386 errorentry general_protection do_general_protection
1387 errorentry page_fault do_page_fault
1388 #ifdef CONFIG_X86_MCE
1389 paranoidzeroentry machine_check do_machine_check
1393 * "Paranoid" exit path from exception stack.
1394 * Paranoid because this is used by NMIs and cannot take
1395 * any kernel state for granted.
1396 * We don't do kernel preemption checks here, because only
1397 * NMI should be common and it does not enable IRQs and
1398 * cannot get reschedule ticks.
1400 * "trace" is 0 for the NMI handler only, because irq-tracing
1401 * is fundamentally NMI-unsafe. (we cannot change the soft and
1402 * hard flags at once, atomically)
1405 /* ebx: no swapgs flag */
1406 ENTRY(paranoid_exit)
1408 DISABLE_INTERRUPTS(CLBR_NONE)
1410 testl %ebx,%ebx /* swapgs needed? */
1411 jnz paranoid_restore
1413 jnz paranoid_userspace
1421 GET_THREAD_INFO(%rcx)
1422 movl TI_flags(%rcx),%ebx
1423 andl $_TIF_WORK_MASK,%ebx
1425 movq %rsp,%rdi /* &pt_regs */
1427 movq %rax,%rsp /* switch stack for scheduling */
1428 testl $_TIF_NEED_RESCHED,%ebx
1429 jnz paranoid_schedule
1430 movl %ebx,%edx /* arg3: thread flags */
1432 ENABLE_INTERRUPTS(CLBR_NONE)
1433 xorl %esi,%esi /* arg2: oldset */
1434 movq %rsp,%rdi /* arg1: &pt_regs */
1435 call do_notify_resume
1436 DISABLE_INTERRUPTS(CLBR_NONE)
1438 jmp paranoid_userspace
1441 ENABLE_INTERRUPTS(CLBR_ANY)
1443 DISABLE_INTERRUPTS(CLBR_ANY)
1445 jmp paranoid_userspace
1450 * Exception entry point. This expects an error code/orig_rax on the stack.
1451 * returns in "no swapgs flag" in %ebx.
1455 CFI_ADJUST_CFA_OFFSET 15*8
1456 /* oldrax contains error code */
1475 je error_kernelspace
1484 * There are two places in the kernel that can potentially fault with
1485 * usergs. Handle them here. The exception handlers after iret run with
1486 * kernel gs again, so don't set the user space flag. B stepping K8s
1487 * sometimes report an truncated RIP for IRET exceptions returning to
1488 * compat mode. Check for these here too.
1492 leaq irq_return(%rip),%rcx
1493 cmpq %rcx,RIP+8(%rsp)
1495 movl %ecx,%ecx /* zero extend */
1496 cmpq %rcx,RIP+8(%rsp)
1498 cmpq $gs_change,RIP+8(%rsp)
1504 /* ebx: no swapgs flag (1: don't need swapgs, 0: need it) */
1509 DISABLE_INTERRUPTS(CLBR_NONE)
1511 GET_THREAD_INFO(%rcx)
1514 LOCKDEP_SYS_EXIT_IRQ
1515 movl TI_flags(%rcx),%edx
1516 movl $_TIF_WORK_MASK,%edi
1524 /* runs on exception stack */
1527 PARAVIRT_ADJUST_EXCEPTION_FRAME
1530 CFI_ADJUST_CFA_OFFSET 15*8
1533 /* paranoidentry do_nmi, 0; without TRACE_IRQS_OFF */
1537 #ifdef CONFIG_TRACE_IRQFLAGS
1538 /* paranoidexit; without TRACE_IRQS_OFF */
1539 /* ebx: no swapgs flag */
1540 DISABLE_INTERRUPTS(CLBR_NONE)
1541 testl %ebx,%ebx /* swapgs needed? */
1551 GET_THREAD_INFO(%rcx)
1552 movl TI_flags(%rcx),%ebx
1553 andl $_TIF_WORK_MASK,%ebx
1555 movq %rsp,%rdi /* &pt_regs */
1557 movq %rax,%rsp /* switch stack for scheduling */
1558 testl $_TIF_NEED_RESCHED,%ebx
1560 movl %ebx,%edx /* arg3: thread flags */
1561 ENABLE_INTERRUPTS(CLBR_NONE)
1562 xorl %esi,%esi /* arg2: oldset */
1563 movq %rsp,%rdi /* arg1: &pt_regs */
1564 call do_notify_resume
1565 DISABLE_INTERRUPTS(CLBR_NONE)
1568 ENABLE_INTERRUPTS(CLBR_ANY)
1570 DISABLE_INTERRUPTS(CLBR_ANY)
1579 ENTRY(ignore_sysret)
1587 * End of kprobes section