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>
51 #include <asm/page_types.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
85 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
86 GLOBAL(ftrace_graph_call)
94 #else /* ! CONFIG_DYNAMIC_FTRACE */
96 cmpl $0, function_trace_stop
99 cmpq $ftrace_stub, ftrace_trace_function
102 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
103 cmpq $ftrace_stub, ftrace_graph_return
104 jnz ftrace_graph_caller
106 cmpq $ftrace_graph_entry_stub, ftrace_graph_entry
107 jnz ftrace_graph_caller
116 movq 0x38(%rsp), %rdi
118 subq $MCOUNT_INSN_SIZE, %rdi
120 call *ftrace_trace_function
126 #endif /* CONFIG_DYNAMIC_FTRACE */
127 #endif /* CONFIG_FUNCTION_TRACER */
129 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
130 ENTRY(ftrace_graph_caller)
131 cmpl $0, function_trace_stop
137 movq 0x38(%rsp), %rsi
138 subq $MCOUNT_INSN_SIZE, %rsi
140 call prepare_ftrace_return
145 END(ftrace_graph_caller)
147 GLOBAL(return_to_handler)
150 /* Save the return values */
154 call ftrace_return_to_handler
164 #ifndef CONFIG_PREEMPT
165 #define retint_kernel retint_restore_args
168 #ifdef CONFIG_PARAVIRT
169 ENTRY(native_usergs_sysret64)
172 ENDPROC(native_usergs_sysret64)
173 #endif /* CONFIG_PARAVIRT */
176 .macro TRACE_IRQS_IRETQ offset=ARGOFFSET
177 #ifdef CONFIG_TRACE_IRQFLAGS
178 bt $9,EFLAGS-\offset(%rsp) /* interrupts off? */
186 * C code is not supposed to know about undefined top of stack. Every time
187 * a C function with an pt_regs argument is called from the SYSCALL based
188 * fast path FIXUP_TOP_OF_STACK is needed.
189 * RESTORE_TOP_OF_STACK syncs the syscall state after any possible ptregs
193 /* %rsp:at FRAMEEND */
194 .macro FIXUP_TOP_OF_STACK tmp offset=0
195 movq PER_CPU_VAR(old_rsp),\tmp
196 movq \tmp,RSP+\offset(%rsp)
197 movq $__USER_DS,SS+\offset(%rsp)
198 movq $__USER_CS,CS+\offset(%rsp)
199 movq $-1,RCX+\offset(%rsp)
200 movq R11+\offset(%rsp),\tmp /* get eflags */
201 movq \tmp,EFLAGS+\offset(%rsp)
204 .macro RESTORE_TOP_OF_STACK tmp offset=0
205 movq RSP+\offset(%rsp),\tmp
206 movq \tmp,PER_CPU_VAR(old_rsp)
207 movq EFLAGS+\offset(%rsp),\tmp
208 movq \tmp,R11+\offset(%rsp)
211 .macro FAKE_STACK_FRAME child_rip
212 /* push in order ss, rsp, eflags, cs, rip */
214 pushq $__KERNEL_DS /* ss */
215 CFI_ADJUST_CFA_OFFSET 8
216 /*CFI_REL_OFFSET ss,0*/
218 CFI_ADJUST_CFA_OFFSET 8
220 pushq $X86_EFLAGS_IF /* eflags - interrupts on */
221 CFI_ADJUST_CFA_OFFSET 8
222 /*CFI_REL_OFFSET rflags,0*/
223 pushq $__KERNEL_CS /* cs */
224 CFI_ADJUST_CFA_OFFSET 8
225 /*CFI_REL_OFFSET cs,0*/
226 pushq \child_rip /* rip */
227 CFI_ADJUST_CFA_OFFSET 8
229 pushq %rax /* orig rax */
230 CFI_ADJUST_CFA_OFFSET 8
233 .macro UNFAKE_STACK_FRAME
235 CFI_ADJUST_CFA_OFFSET -(6*8)
239 * initial frame state for interrupts (and exceptions without error code)
241 .macro EMPTY_FRAME start=1 offset=0
245 CFI_DEF_CFA rsp,8+\offset
247 CFI_DEF_CFA_OFFSET 8+\offset
252 * initial frame state for interrupts (and exceptions without error code)
254 .macro INTR_FRAME start=1 offset=0
255 EMPTY_FRAME \start, SS+8+\offset-RIP
256 /*CFI_REL_OFFSET ss, SS+\offset-RIP*/
257 CFI_REL_OFFSET rsp, RSP+\offset-RIP
258 /*CFI_REL_OFFSET rflags, EFLAGS+\offset-RIP*/
259 /*CFI_REL_OFFSET cs, CS+\offset-RIP*/
260 CFI_REL_OFFSET rip, RIP+\offset-RIP
264 * initial frame state for exceptions with error code (and interrupts
265 * with vector already pushed)
267 .macro XCPT_FRAME start=1 offset=0
268 INTR_FRAME \start, RIP+\offset-ORIG_RAX
269 /*CFI_REL_OFFSET orig_rax, ORIG_RAX-ORIG_RAX*/
273 * frame that enables calling into C.
275 .macro PARTIAL_FRAME start=1 offset=0
276 XCPT_FRAME \start, ORIG_RAX+\offset-ARGOFFSET
277 CFI_REL_OFFSET rdi, RDI+\offset-ARGOFFSET
278 CFI_REL_OFFSET rsi, RSI+\offset-ARGOFFSET
279 CFI_REL_OFFSET rdx, RDX+\offset-ARGOFFSET
280 CFI_REL_OFFSET rcx, RCX+\offset-ARGOFFSET
281 CFI_REL_OFFSET rax, RAX+\offset-ARGOFFSET
282 CFI_REL_OFFSET r8, R8+\offset-ARGOFFSET
283 CFI_REL_OFFSET r9, R9+\offset-ARGOFFSET
284 CFI_REL_OFFSET r10, R10+\offset-ARGOFFSET
285 CFI_REL_OFFSET r11, R11+\offset-ARGOFFSET
289 * frame that enables passing a complete pt_regs to a C function.
291 .macro DEFAULT_FRAME start=1 offset=0
292 PARTIAL_FRAME \start, R11+\offset-R15
293 CFI_REL_OFFSET rbx, RBX+\offset
294 CFI_REL_OFFSET rbp, RBP+\offset
295 CFI_REL_OFFSET r12, R12+\offset
296 CFI_REL_OFFSET r13, R13+\offset
297 CFI_REL_OFFSET r14, R14+\offset
298 CFI_REL_OFFSET r15, R15+\offset
301 /* save partial stack frame */
305 movq_cfi rdi, RDI+16-ARGOFFSET
306 movq_cfi rsi, RSI+16-ARGOFFSET
307 movq_cfi rdx, RDX+16-ARGOFFSET
308 movq_cfi rcx, RCX+16-ARGOFFSET
309 movq_cfi rax, RAX+16-ARGOFFSET
310 movq_cfi r8, R8+16-ARGOFFSET
311 movq_cfi r9, R9+16-ARGOFFSET
312 movq_cfi r10, R10+16-ARGOFFSET
313 movq_cfi r11, R11+16-ARGOFFSET
315 leaq -ARGOFFSET+16(%rsp),%rdi /* arg1 for handler */
316 movq_cfi rbp, 8 /* push %rbp */
317 leaq 8(%rsp), %rbp /* mov %rsp, %ebp */
322 * irq_count is used to check if a CPU is already on an interrupt stack
323 * or not. While this is essentially redundant with preempt_count it is
324 * a little cheaper to use a separate counter in the PDA (short of
325 * moving irq_enter into assembly, which would be too much work)
327 1: incl PER_CPU_VAR(irq_count)
329 popq_cfi %rax /* move return address... */
330 mov PER_CPU_VAR(irq_stack_ptr),%rsp
332 pushq_cfi %rbp /* backlink for unwinder */
333 pushq_cfi %rax /* ... to the new stack */
335 * We entered an interrupt context - irqs are off:
343 PARTIAL_FRAME 1 REST_SKIP+8
344 movq 5*8+16(%rsp), %r11 /* save return address */
351 movq %r11, 8(%rsp) /* return address */
352 FIXUP_TOP_OF_STACK %r11, 16
357 /* save complete stack frame */
358 .pushsection .kprobes.text, "ax"
378 movl $MSR_GS_BASE,%ecx
381 js 1f /* negative -> in kernel */
390 * A newly forked process directly context switches into this address.
392 * rdi: prev task we switched from
397 LOCK ; btr $TIF_FORK,TI_flags(%r8)
399 push kernel_eflags(%rip)
400 CFI_ADJUST_CFA_OFFSET 8
401 popf # reset kernel eflags
402 CFI_ADJUST_CFA_OFFSET -8
404 call schedule_tail # rdi: 'prev' task parameter
406 GET_THREAD_INFO(%rcx)
410 testl $3, CS-ARGOFFSET(%rsp) # from kernel_thread?
411 je int_ret_from_sys_call
413 testl $_TIF_IA32, TI_flags(%rcx) # 32-bit compat task needs IRET
414 jnz int_ret_from_sys_call
416 RESTORE_TOP_OF_STACK %rdi, -ARGOFFSET
417 jmp ret_from_sys_call # go to the SYSRET fastpath
423 * System call entry. Upto 6 arguments in registers are supported.
425 * SYSCALL does not save anything on the stack and does not change the
431 * rax system call number
433 * rcx return address for syscall/sysret, C arg3
436 * r10 arg3 (--> moved to rcx for C)
439 * r11 eflags for syscall/sysret, temporary for C
440 * r12-r15,rbp,rbx saved by C code, not touched.
442 * Interrupts are off on entry.
443 * Only called from user space.
445 * XXX if we had a free scratch register we could save the RSP into the stack frame
446 * and report it properly in ps. Unfortunately we haven't.
448 * When user can change the frames always force IRET. That is because
449 * it deals with uncanonical addresses better. SYSRET has trouble
450 * with them due to bugs in both AMD and Intel CPUs.
456 CFI_DEF_CFA rsp,KERNEL_STACK_OFFSET
458 /*CFI_REGISTER rflags,r11*/
461 * A hypervisor implementation might want to use a label
462 * after the swapgs, so that it can do the swapgs
463 * for the guest and jump here on syscall.
465 ENTRY(system_call_after_swapgs)
467 movq %rsp,PER_CPU_VAR(old_rsp)
468 movq PER_CPU_VAR(kernel_stack),%rsp
470 * No need to follow this irqs off/on section - it's straight
473 ENABLE_INTERRUPTS(CLBR_NONE)
475 movq %rax,ORIG_RAX-ARGOFFSET(%rsp)
476 movq %rcx,RIP-ARGOFFSET(%rsp)
477 CFI_REL_OFFSET rip,RIP-ARGOFFSET
478 GET_THREAD_INFO(%rcx)
479 testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%rcx)
481 system_call_fastpath:
482 cmpq $__NR_syscall_max,%rax
485 call *sys_call_table(,%rax,8) # XXX: rip relative
486 movq %rax,RAX-ARGOFFSET(%rsp)
488 * Syscall return path ending with SYSRET (fast path)
489 * Has incomplete stack frame and undefined top of stack.
492 movl $_TIF_ALLWORK_MASK,%edi
496 GET_THREAD_INFO(%rcx)
497 DISABLE_INTERRUPTS(CLBR_NONE)
499 movl TI_flags(%rcx),%edx
504 * sysretq will re-enable interrupts:
507 movq RIP-ARGOFFSET(%rsp),%rcx
509 RESTORE_ARGS 0,-ARG_SKIP,1
510 /*CFI_REGISTER rflags,r11*/
511 movq PER_CPU_VAR(old_rsp), %rsp
515 /* Handle reschedules */
516 /* edx: work, edi: workmask */
518 bt $TIF_NEED_RESCHED,%edx
521 ENABLE_INTERRUPTS(CLBR_NONE)
523 CFI_ADJUST_CFA_OFFSET 8
526 CFI_ADJUST_CFA_OFFSET -8
529 /* Handle a signal */
532 ENABLE_INTERRUPTS(CLBR_NONE)
533 #ifdef CONFIG_AUDITSYSCALL
534 bt $TIF_SYSCALL_AUDIT,%edx
537 /* edx: work flags (arg3) */
538 leaq -ARGOFFSET(%rsp),%rdi # &pt_regs -> arg1
539 xorl %esi,%esi # oldset -> arg2
541 FIXUP_TOP_OF_STACK %r11
542 call do_notify_resume
543 RESTORE_TOP_OF_STACK %r11
545 movl $_TIF_WORK_MASK,%edi
546 /* Use IRET because user could have changed frame. This
547 works because ptregscall_common has called FIXUP_TOP_OF_STACK. */
548 DISABLE_INTERRUPTS(CLBR_NONE)
553 movq $-ENOSYS,RAX-ARGOFFSET(%rsp)
554 jmp ret_from_sys_call
556 #ifdef CONFIG_AUDITSYSCALL
558 * Fast path for syscall audit without full syscall trace.
559 * We just call audit_syscall_entry() directly, and then
560 * jump back to the normal fast path.
563 movq %r10,%r9 /* 6th arg: 4th syscall arg */
564 movq %rdx,%r8 /* 5th arg: 3rd syscall arg */
565 movq %rsi,%rcx /* 4th arg: 2nd syscall arg */
566 movq %rdi,%rdx /* 3rd arg: 1st syscall arg */
567 movq %rax,%rsi /* 2nd arg: syscall number */
568 movl $AUDIT_ARCH_X86_64,%edi /* 1st arg: audit arch */
569 call audit_syscall_entry
570 LOAD_ARGS 0 /* reload call-clobbered registers */
571 jmp system_call_fastpath
574 * Return fast path for syscall audit. Call audit_syscall_exit()
575 * directly and then jump back to the fast path with TIF_SYSCALL_AUDIT
579 movq %rax,%rsi /* second arg, syscall return value */
580 cmpq $0,%rax /* is it < 0? */
581 setl %al /* 1 if so, 0 if not */
582 movzbl %al,%edi /* zero-extend that into %edi */
583 inc %edi /* first arg, 0->1(AUDITSC_SUCCESS), 1->2(AUDITSC_FAILURE) */
584 call audit_syscall_exit
585 movl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT),%edi
587 #endif /* CONFIG_AUDITSYSCALL */
589 /* Do syscall tracing */
591 #ifdef CONFIG_AUDITSYSCALL
592 testl $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags(%rcx)
596 movq $-ENOSYS,RAX(%rsp) /* ptrace can change this for a bad syscall */
597 FIXUP_TOP_OF_STACK %rdi
599 call syscall_trace_enter
601 * Reload arg registers from stack in case ptrace changed them.
602 * We don't reload %rax because syscall_trace_enter() returned
603 * the value it wants us to use in the table lookup.
605 LOAD_ARGS ARGOFFSET, 1
607 cmpq $__NR_syscall_max,%rax
608 ja int_ret_from_sys_call /* RAX(%rsp) set to -ENOSYS above */
609 movq %r10,%rcx /* fixup for C */
610 call *sys_call_table(,%rax,8)
611 movq %rax,RAX-ARGOFFSET(%rsp)
612 /* Use IRET because user could have changed frame */
615 * Syscall return path ending with IRET.
616 * Has correct top of stack, but partial stack frame.
618 GLOBAL(int_ret_from_sys_call)
619 DISABLE_INTERRUPTS(CLBR_NONE)
621 testl $3,CS-ARGOFFSET(%rsp)
622 je retint_restore_args
623 movl $_TIF_ALLWORK_MASK,%edi
624 /* edi: mask to check */
625 GLOBAL(int_with_check)
627 GET_THREAD_INFO(%rcx)
628 movl TI_flags(%rcx),%edx
631 andl $~TS_COMPAT,TI_status(%rcx)
634 /* Either reschedule or signal or syscall exit tracking needed. */
635 /* First do a reschedule test. */
636 /* edx: work, edi: workmask */
638 bt $TIF_NEED_RESCHED,%edx
641 ENABLE_INTERRUPTS(CLBR_NONE)
643 CFI_ADJUST_CFA_OFFSET 8
646 CFI_ADJUST_CFA_OFFSET -8
647 DISABLE_INTERRUPTS(CLBR_NONE)
651 /* handle signals and tracing -- both require a full stack frame */
654 ENABLE_INTERRUPTS(CLBR_NONE)
656 /* Check for syscall exit trace */
657 testl $_TIF_WORK_SYSCALL_EXIT,%edx
660 CFI_ADJUST_CFA_OFFSET 8
661 leaq 8(%rsp),%rdi # &ptregs -> arg1
662 call syscall_trace_leave
664 CFI_ADJUST_CFA_OFFSET -8
665 andl $~(_TIF_WORK_SYSCALL_EXIT|_TIF_SYSCALL_EMU),%edi
669 testl $_TIF_DO_NOTIFY_MASK,%edx
671 movq %rsp,%rdi # &ptregs -> arg1
672 xorl %esi,%esi # oldset -> arg2
673 call do_notify_resume
674 1: movl $_TIF_WORK_MASK,%edi
677 DISABLE_INTERRUPTS(CLBR_NONE)
684 * Certain special system calls that need to save a complete full stack frame.
686 .macro PTREGSCALL label,func,arg
688 PARTIAL_FRAME 1 8 /* offset 8: return address */
689 subq $REST_SKIP, %rsp
690 CFI_ADJUST_CFA_OFFSET REST_SKIP
692 DEFAULT_FRAME 0 8 /* offset 8: return address */
693 leaq 8(%rsp), \arg /* pt_regs pointer */
695 jmp ptregscall_common
700 PTREGSCALL stub_clone, sys_clone, %r8
701 PTREGSCALL stub_fork, sys_fork, %rdi
702 PTREGSCALL stub_vfork, sys_vfork, %rdi
703 PTREGSCALL stub_sigaltstack, sys_sigaltstack, %rdx
704 PTREGSCALL stub_iopl, sys_iopl, %rsi
706 ENTRY(ptregscall_common)
707 DEFAULT_FRAME 1 8 /* offset 8: return address */
708 RESTORE_TOP_OF_STACK %r11, 8
709 movq_cfi_restore R15+8, r15
710 movq_cfi_restore R14+8, r14
711 movq_cfi_restore R13+8, r13
712 movq_cfi_restore R12+8, r12
713 movq_cfi_restore RBP+8, rbp
714 movq_cfi_restore RBX+8, rbx
715 ret $REST_SKIP /* pop extended registers */
717 END(ptregscall_common)
722 CFI_ADJUST_CFA_OFFSET -8
723 CFI_REGISTER rip, r11
725 FIXUP_TOP_OF_STACK %r11
728 RESTORE_TOP_OF_STACK %r11
731 jmp int_ret_from_sys_call
736 * sigreturn is special because it needs to restore all registers on return.
737 * This cannot be done with SYSRET, so use the IRET return path instead.
739 ENTRY(stub_rt_sigreturn)
742 CFI_ADJUST_CFA_OFFSET -8
745 FIXUP_TOP_OF_STACK %r11
746 call sys_rt_sigreturn
747 movq %rax,RAX(%rsp) # fixme, this could be done at the higher layer
749 jmp int_ret_from_sys_call
751 END(stub_rt_sigreturn)
754 * Build the entry stubs and pointer table with some assembler magic.
755 * We pack 7 stubs into a single 32-byte chunk, which will fit in a
756 * single cache line on all modern x86 implementations.
758 .section .init.rodata,"a"
762 .p2align CONFIG_X86_L1_CACHE_SHIFT
763 ENTRY(irq_entries_start)
765 vector=FIRST_EXTERNAL_VECTOR
766 .rept (NR_VECTORS-FIRST_EXTERNAL_VECTOR+6)/7
769 .if vector < NR_VECTORS
770 .if vector <> FIRST_EXTERNAL_VECTOR
771 CFI_ADJUST_CFA_OFFSET -8
773 1: pushq $(~vector+0x80) /* Note: always in signed byte range */
774 CFI_ADJUST_CFA_OFFSET 8
775 .if ((vector-FIRST_EXTERNAL_VECTOR)%7) <> 6
784 2: jmp common_interrupt
787 END(irq_entries_start)
794 * Interrupt entry/exit.
796 * Interrupt entry points save only callee clobbered registers in fast path.
798 * Entry runs with interrupts off.
801 /* 0(%rsp): ~(interrupt number) */
802 .macro interrupt func
804 CFI_ADJUST_CFA_OFFSET 10*8
811 * The interrupt stubs push (~vector+0x80) onto the stack and
812 * then jump to common_interrupt.
814 .p2align CONFIG_X86_L1_CACHE_SHIFT
817 addq $-0x80,(%rsp) /* Adjust vector to [-256,-1] range */
819 /* 0(%rsp): old_rsp-ARGOFFSET */
821 DISABLE_INTERRUPTS(CLBR_NONE)
823 decl PER_CPU_VAR(irq_count)
825 CFI_DEF_CFA_REGISTER rsp
826 CFI_ADJUST_CFA_OFFSET -8
828 GET_THREAD_INFO(%rcx)
829 testl $3,CS-ARGOFFSET(%rsp)
832 /* Interrupt came from user space */
834 * Has a correct top of stack, but a partial stack frame
835 * %rcx: thread info. Interrupts off.
837 retint_with_reschedule:
838 movl $_TIF_WORK_MASK,%edi
841 movl TI_flags(%rcx),%edx
846 retint_swapgs: /* return to user-space */
848 * The iretq could re-enable interrupts:
850 DISABLE_INTERRUPTS(CLBR_ANY)
855 retint_restore_args: /* return to kernel space */
856 DISABLE_INTERRUPTS(CLBR_ANY)
858 * The iretq could re-enable interrupts:
867 .section __ex_table, "a"
868 .quad irq_return, bad_iret
871 #ifdef CONFIG_PARAVIRT
875 .section __ex_table,"a"
876 .quad native_iret, bad_iret
883 * The iret traps when the %cs or %ss being restored is bogus.
884 * We've lost the original trap vector and error code.
885 * #GPF is the most likely one to get for an invalid selector.
886 * So pretend we completed the iret and took the #GPF in user mode.
888 * We are now running with the kernel GS after exception recovery.
889 * But error_entry expects us to have user GS to match the user %cs,
895 jmp general_protection
899 /* edi: workmask, edx: work */
902 bt $TIF_NEED_RESCHED,%edx
905 ENABLE_INTERRUPTS(CLBR_NONE)
907 CFI_ADJUST_CFA_OFFSET 8
910 CFI_ADJUST_CFA_OFFSET -8
911 GET_THREAD_INFO(%rcx)
912 DISABLE_INTERRUPTS(CLBR_NONE)
917 testl $_TIF_DO_NOTIFY_MASK,%edx
920 ENABLE_INTERRUPTS(CLBR_NONE)
922 movq $-1,ORIG_RAX(%rsp)
923 xorl %esi,%esi # oldset
924 movq %rsp,%rdi # &pt_regs
925 call do_notify_resume
927 DISABLE_INTERRUPTS(CLBR_NONE)
929 GET_THREAD_INFO(%rcx)
930 jmp retint_with_reschedule
932 #ifdef CONFIG_PREEMPT
933 /* Returning to kernel space. Check if we need preemption */
934 /* rcx: threadinfo. interrupts off. */
936 cmpl $0,TI_preempt_count(%rcx)
937 jnz retint_restore_args
938 bt $TIF_NEED_RESCHED,TI_flags(%rcx)
939 jnc retint_restore_args
940 bt $9,EFLAGS-ARGOFFSET(%rsp) /* interrupts off? */
941 jnc retint_restore_args
942 call preempt_schedule_irq
947 END(common_interrupt)
952 .macro apicinterrupt num sym do_sym
956 CFI_ADJUST_CFA_OFFSET 8
964 apicinterrupt IRQ_MOVE_CLEANUP_VECTOR \
965 irq_move_cleanup_interrupt smp_irq_move_cleanup_interrupt
969 apicinterrupt UV_BAU_MESSAGE \
970 uv_bau_message_intr1 uv_bau_message_interrupt
972 apicinterrupt LOCAL_TIMER_VECTOR \
973 apic_timer_interrupt smp_apic_timer_interrupt
974 apicinterrupt GENERIC_INTERRUPT_VECTOR \
975 generic_interrupt smp_generic_interrupt
978 apicinterrupt INVALIDATE_TLB_VECTOR_START+0 \
979 invalidate_interrupt0 smp_invalidate_interrupt
980 apicinterrupt INVALIDATE_TLB_VECTOR_START+1 \
981 invalidate_interrupt1 smp_invalidate_interrupt
982 apicinterrupt INVALIDATE_TLB_VECTOR_START+2 \
983 invalidate_interrupt2 smp_invalidate_interrupt
984 apicinterrupt INVALIDATE_TLB_VECTOR_START+3 \
985 invalidate_interrupt3 smp_invalidate_interrupt
986 apicinterrupt INVALIDATE_TLB_VECTOR_START+4 \
987 invalidate_interrupt4 smp_invalidate_interrupt
988 apicinterrupt INVALIDATE_TLB_VECTOR_START+5 \
989 invalidate_interrupt5 smp_invalidate_interrupt
990 apicinterrupt INVALIDATE_TLB_VECTOR_START+6 \
991 invalidate_interrupt6 smp_invalidate_interrupt
992 apicinterrupt INVALIDATE_TLB_VECTOR_START+7 \
993 invalidate_interrupt7 smp_invalidate_interrupt
996 apicinterrupt THRESHOLD_APIC_VECTOR \
997 threshold_interrupt mce_threshold_interrupt
998 apicinterrupt THERMAL_APIC_VECTOR \
999 thermal_interrupt smp_thermal_interrupt
1002 apicinterrupt CALL_FUNCTION_SINGLE_VECTOR \
1003 call_function_single_interrupt smp_call_function_single_interrupt
1004 apicinterrupt CALL_FUNCTION_VECTOR \
1005 call_function_interrupt smp_call_function_interrupt
1006 apicinterrupt RESCHEDULE_VECTOR \
1007 reschedule_interrupt smp_reschedule_interrupt
1010 apicinterrupt ERROR_APIC_VECTOR \
1011 error_interrupt smp_error_interrupt
1012 apicinterrupt SPURIOUS_APIC_VECTOR \
1013 spurious_interrupt smp_spurious_interrupt
1016 * Exception entry points.
1018 .macro zeroentry sym do_sym
1021 PARAVIRT_ADJUST_EXCEPTION_FRAME
1022 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1024 CFI_ADJUST_CFA_OFFSET 15*8
1027 movq %rsp,%rdi /* pt_regs pointer */
1028 xorl %esi,%esi /* no error code */
1030 jmp error_exit /* %ebx: no swapgs flag */
1035 .macro paranoidzeroentry sym do_sym
1038 PARAVIRT_ADJUST_EXCEPTION_FRAME
1039 pushq $-1 /* ORIG_RAX: no syscall to restart */
1040 CFI_ADJUST_CFA_OFFSET 8
1044 movq %rsp,%rdi /* pt_regs pointer */
1045 xorl %esi,%esi /* no error code */
1047 jmp paranoid_exit /* %ebx: no swapgs flag */
1052 .macro paranoidzeroentry_ist sym do_sym ist
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 */
1063 PER_CPU(init_tss, %rbp)
1064 subq $EXCEPTION_STKSZ, TSS_ist + (\ist - 1) * 8(%rbp)
1066 addq $EXCEPTION_STKSZ, TSS_ist + (\ist - 1) * 8(%rbp)
1067 jmp paranoid_exit /* %ebx: no swapgs flag */
1072 .macro errorentry sym do_sym
1075 PARAVIRT_ADJUST_EXCEPTION_FRAME
1077 CFI_ADJUST_CFA_OFFSET 15*8
1080 movq %rsp,%rdi /* pt_regs pointer */
1081 movq ORIG_RAX(%rsp),%rsi /* get error code */
1082 movq $-1,ORIG_RAX(%rsp) /* no syscall to restart */
1084 jmp error_exit /* %ebx: no swapgs flag */
1089 /* error code is on the stack already */
1090 .macro paranoiderrorentry sym do_sym
1093 PARAVIRT_ADJUST_EXCEPTION_FRAME
1095 CFI_ADJUST_CFA_OFFSET 15*8
1099 movq %rsp,%rdi /* pt_regs pointer */
1100 movq ORIG_RAX(%rsp),%rsi /* get error code */
1101 movq $-1,ORIG_RAX(%rsp) /* no syscall to restart */
1103 jmp paranoid_exit /* %ebx: no swapgs flag */
1108 zeroentry divide_error do_divide_error
1109 zeroentry overflow do_overflow
1110 zeroentry bounds do_bounds
1111 zeroentry invalid_op do_invalid_op
1112 zeroentry device_not_available do_device_not_available
1113 paranoiderrorentry double_fault do_double_fault
1114 zeroentry coprocessor_segment_overrun do_coprocessor_segment_overrun
1115 errorentry invalid_TSS do_invalid_TSS
1116 errorentry segment_not_present do_segment_not_present
1117 zeroentry spurious_interrupt_bug do_spurious_interrupt_bug
1118 zeroentry coprocessor_error do_coprocessor_error
1119 errorentry alignment_check do_alignment_check
1120 zeroentry simd_coprocessor_error do_simd_coprocessor_error
1122 /* Reload gs selector with exception handling */
1123 /* edi: new selector */
1124 ENTRY(native_load_gs_index)
1127 CFI_ADJUST_CFA_OFFSET 8
1128 DISABLE_INTERRUPTS(CLBR_ANY & ~CLBR_RDI)
1132 2: mfence /* workaround */
1135 CFI_ADJUST_CFA_OFFSET -8
1138 END(native_load_gs_index)
1140 .section __ex_table,"a"
1142 .quad gs_change,bad_gs
1144 .section .fixup,"ax"
1145 /* running with kernelgs */
1147 SWAPGS /* switch back to user gs */
1154 * Create a kernel thread.
1156 * C extern interface:
1157 * extern long kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
1159 * asm input arguments:
1160 * rdi: fn, rsi: arg, rdx: flags
1162 ENTRY(kernel_thread)
1164 FAKE_STACK_FRAME $child_rip
1167 # rdi: flags, rsi: usp, rdx: will be &pt_regs
1169 orq kernel_thread_flags(%rip),%rdi
1182 * It isn't worth to check for reschedule here,
1183 * so internally to the x86_64 port you can rely on kernel_thread()
1184 * not to reschedule the child before returning, this avoids the need
1185 * of hacks for example to fork off the per-CPU idle tasks.
1186 * [Hopefully no generic code relies on the reschedule -AK]
1195 pushq $0 # fake return address
1198 * Here we are in the child and the registers are set as they were
1199 * at kernel_thread() invocation in the parent.
1207 ud2 # padding for call trace
1212 * execve(). This function needs to use IRET, not SYSRET, to set up all state properly.
1214 * C extern interface:
1215 * extern long execve(char *name, char **argv, char **envp)
1217 * asm input arguments:
1218 * rdi: name, rsi: argv, rdx: envp
1220 * We want to fallback into:
1221 * extern long sys_execve(char *name, char **argv,char **envp, struct pt_regs *regs)
1223 * do_sys_execve asm fallback arguments:
1224 * rdi: name, rsi: argv, rdx: envp, rcx: fake frame on the stack
1226 ENTRY(kernel_execve)
1232 movq %rax, RAX(%rsp)
1235 je int_ret_from_sys_call
1242 /* Call softirq on interrupt stack. Interrupts are off. */
1246 CFI_ADJUST_CFA_OFFSET 8
1247 CFI_REL_OFFSET rbp,0
1249 CFI_DEF_CFA_REGISTER rbp
1250 incl PER_CPU_VAR(irq_count)
1251 cmove PER_CPU_VAR(irq_stack_ptr),%rsp
1252 push %rbp # backlink for old unwinder
1255 CFI_DEF_CFA_REGISTER rsp
1256 CFI_ADJUST_CFA_OFFSET -8
1257 decl PER_CPU_VAR(irq_count)
1263 zeroentry xen_hypervisor_callback xen_do_hypervisor_callback
1266 * A note on the "critical region" in our callback handler.
1267 * We want to avoid stacking callback handlers due to events occurring
1268 * during handling of the last event. To do this, we keep events disabled
1269 * until we've done all processing. HOWEVER, we must enable events before
1270 * popping the stack frame (can't be done atomically) and so it would still
1271 * be possible to get enough handler activations to overflow the stack.
1272 * Although unlikely, bugs of that kind are hard to track down, so we'd
1273 * like to avoid the possibility.
1274 * So, on entry to the handler we detect whether we interrupted an
1275 * existing activation in its critical region -- if so, we pop the current
1276 * activation and restart the handler using the previous one.
1278 ENTRY(xen_do_hypervisor_callback) # do_hypervisor_callback(struct *pt_regs)
1281 * Since we don't modify %rdi, evtchn_do_upall(struct *pt_regs) will
1282 * see the correct pointer to the pt_regs
1284 movq %rdi, %rsp # we don't return, adjust the stack frame
1287 11: incl PER_CPU_VAR(irq_count)
1289 CFI_DEF_CFA_REGISTER rbp
1290 cmovzq PER_CPU_VAR(irq_stack_ptr),%rsp
1291 pushq %rbp # backlink for old unwinder
1292 call xen_evtchn_do_upcall
1294 CFI_DEF_CFA_REGISTER rsp
1295 decl PER_CPU_VAR(irq_count)
1298 END(do_hypervisor_callback)
1301 * Hypervisor uses this for application faults while it executes.
1302 * We get here for two reasons:
1303 * 1. Fault while reloading DS, ES, FS or GS
1304 * 2. Fault while executing IRET
1305 * Category 1 we do not need to fix up as Xen has already reloaded all segment
1306 * registers that could be reloaded and zeroed the others.
1307 * Category 2 we fix up by killing the current process. We cannot use the
1308 * normal Linux return path in this case because if we use the IRET hypercall
1309 * to pop the stack frame we end up in an infinite loop of failsafe callbacks.
1310 * We distinguish between categories by comparing each saved segment register
1311 * with its current contents: any discrepancy means we in category 1.
1313 ENTRY(xen_failsafe_callback)
1315 /*CFI_REL_OFFSET gs,GS*/
1316 /*CFI_REL_OFFSET fs,FS*/
1317 /*CFI_REL_OFFSET es,ES*/
1318 /*CFI_REL_OFFSET ds,DS*/
1319 CFI_REL_OFFSET r11,8
1320 CFI_REL_OFFSET rcx,0
1334 /* All segments match their saved values => Category 2 (Bad IRET). */
1340 CFI_ADJUST_CFA_OFFSET -0x30
1341 pushq_cfi $0 /* RIP */
1344 jmp general_protection
1346 1: /* Segment mismatch => Category 1 (Bad segment). Retry the IRET. */
1352 CFI_ADJUST_CFA_OFFSET -0x30
1357 END(xen_failsafe_callback)
1359 #endif /* CONFIG_XEN */
1362 * Some functions should be protected against kprobes
1364 .pushsection .kprobes.text, "ax"
1366 paranoidzeroentry_ist debug do_debug DEBUG_STACK
1367 paranoidzeroentry_ist int3 do_int3 DEBUG_STACK
1368 paranoiderrorentry stack_segment do_stack_segment
1370 zeroentry xen_debug do_debug
1371 zeroentry xen_int3 do_int3
1372 errorentry xen_stack_segment do_stack_segment
1374 errorentry general_protection do_general_protection
1375 errorentry page_fault do_page_fault
1376 #ifdef CONFIG_X86_MCE
1377 paranoidzeroentry machine_check do_machine_check
1381 * "Paranoid" exit path from exception stack.
1382 * Paranoid because this is used by NMIs and cannot take
1383 * any kernel state for granted.
1384 * We don't do kernel preemption checks here, because only
1385 * NMI should be common and it does not enable IRQs and
1386 * cannot get reschedule ticks.
1388 * "trace" is 0 for the NMI handler only, because irq-tracing
1389 * is fundamentally NMI-unsafe. (we cannot change the soft and
1390 * hard flags at once, atomically)
1393 /* ebx: no swapgs flag */
1394 ENTRY(paranoid_exit)
1396 DISABLE_INTERRUPTS(CLBR_NONE)
1398 testl %ebx,%ebx /* swapgs needed? */
1399 jnz paranoid_restore
1401 jnz paranoid_userspace
1412 GET_THREAD_INFO(%rcx)
1413 movl TI_flags(%rcx),%ebx
1414 andl $_TIF_WORK_MASK,%ebx
1416 movq %rsp,%rdi /* &pt_regs */
1418 movq %rax,%rsp /* switch stack for scheduling */
1419 testl $_TIF_NEED_RESCHED,%ebx
1420 jnz paranoid_schedule
1421 movl %ebx,%edx /* arg3: thread flags */
1423 ENABLE_INTERRUPTS(CLBR_NONE)
1424 xorl %esi,%esi /* arg2: oldset */
1425 movq %rsp,%rdi /* arg1: &pt_regs */
1426 call do_notify_resume
1427 DISABLE_INTERRUPTS(CLBR_NONE)
1429 jmp paranoid_userspace
1432 ENABLE_INTERRUPTS(CLBR_ANY)
1434 DISABLE_INTERRUPTS(CLBR_ANY)
1436 jmp paranoid_userspace
1441 * Exception entry point. This expects an error code/orig_rax on the stack.
1442 * returns in "no swapgs flag" in %ebx.
1446 CFI_ADJUST_CFA_OFFSET 15*8
1447 /* oldrax contains error code */
1466 je error_kernelspace
1475 * There are two places in the kernel that can potentially fault with
1476 * usergs. Handle them here. The exception handlers after iret run with
1477 * kernel gs again, so don't set the user space flag. B stepping K8s
1478 * sometimes report an truncated RIP for IRET exceptions returning to
1479 * compat mode. Check for these here too.
1483 leaq irq_return(%rip),%rcx
1484 cmpq %rcx,RIP+8(%rsp)
1486 movl %ecx,%ecx /* zero extend */
1487 cmpq %rcx,RIP+8(%rsp)
1489 cmpq $gs_change,RIP+8(%rsp)
1495 /* ebx: no swapgs flag (1: don't need swapgs, 0: need it) */
1500 DISABLE_INTERRUPTS(CLBR_NONE)
1502 GET_THREAD_INFO(%rcx)
1505 LOCKDEP_SYS_EXIT_IRQ
1506 movl TI_flags(%rcx),%edx
1507 movl $_TIF_WORK_MASK,%edi
1515 /* runs on exception stack */
1518 PARAVIRT_ADJUST_EXCEPTION_FRAME
1521 CFI_ADJUST_CFA_OFFSET 15*8
1524 /* paranoidentry do_nmi, 0; without TRACE_IRQS_OFF */
1528 #ifdef CONFIG_TRACE_IRQFLAGS
1529 /* paranoidexit; without TRACE_IRQS_OFF */
1530 /* ebx: no swapgs flag */
1531 DISABLE_INTERRUPTS(CLBR_NONE)
1532 testl %ebx,%ebx /* swapgs needed? */
1542 GET_THREAD_INFO(%rcx)
1543 movl TI_flags(%rcx),%ebx
1544 andl $_TIF_WORK_MASK,%ebx
1546 movq %rsp,%rdi /* &pt_regs */
1548 movq %rax,%rsp /* switch stack for scheduling */
1549 testl $_TIF_NEED_RESCHED,%ebx
1551 movl %ebx,%edx /* arg3: thread flags */
1552 ENABLE_INTERRUPTS(CLBR_NONE)
1553 xorl %esi,%esi /* arg2: oldset */
1554 movq %rsp,%rdi /* arg1: &pt_regs */
1555 call do_notify_resume
1556 DISABLE_INTERRUPTS(CLBR_NONE)
1559 ENABLE_INTERRUPTS(CLBR_ANY)
1561 DISABLE_INTERRUPTS(CLBR_ANY)
1570 ENTRY(ignore_sysret)
1578 * End of kprobes section