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 * Some of this is documented in Documentation/x86/entry_64.txt
14 * NOTE: This code handles signal-recognition, which happens every time
15 * after an interrupt and after each system call.
17 * Normal syscalls and interrupts don't save a full stack frame, this is
18 * only done for syscall tracing, signals or fork/exec et.al.
20 * A note on terminology:
21 * - top of stack: Architecture defined interrupt frame from SS to RIP
22 * at the top of the kernel process stack.
23 * - partial stack frame: partially saved registers up to R11.
24 * - full stack frame: Like partial stack frame, but all register saved.
27 * - CFI macros are used to generate dwarf2 unwind information for better
28 * backtraces. They don't change any code.
29 * - SAVE_ALL/RESTORE_ALL - Save/restore all registers
30 * - SAVE_ARGS/RESTORE_ARGS - Save/restore registers that C functions modify.
31 * There are unfortunately lots of special cases where some registers
32 * not touched. The macro is a big mess that should be cleaned up.
33 * - SAVE_REST/RESTORE_REST - Handle the registers not saved by SAVE_ARGS.
34 * Gives a full stack frame.
35 * - ENTRY/END Define functions in the symbol table.
36 * - FIXUP_TOP_OF_STACK/RESTORE_TOP_OF_STACK - Fix up the hardware stack
37 * frame that is otherwise undefined after a SYSCALL
38 * - TRACE_IRQ_* - Trace hard interrupt state for lock debugging.
39 * - idtentry - Define exception entry points.
42 #include <linux/linkage.h>
43 #include <asm/segment.h>
44 #include <asm/cache.h>
45 #include <asm/errno.h>
46 #include <asm/dwarf2.h>
47 #include <asm/calling.h>
48 #include <asm/asm-offsets.h>
50 #include <asm/unistd.h>
51 #include <asm/thread_info.h>
52 #include <asm/hw_irq.h>
53 #include <asm/page_types.h>
54 #include <asm/irqflags.h>
55 #include <asm/paravirt.h>
56 #include <asm/percpu.h>
58 #include <asm/context_tracking.h>
60 #include <asm/pgtable_types.h>
61 #include <linux/err.h>
63 /* Avoid __ASSEMBLER__'ifying <linux/audit.h> just for this. */
64 #include <linux/elf-em.h>
65 #define AUDIT_ARCH_X86_64 (EM_X86_64|__AUDIT_ARCH_64BIT|__AUDIT_ARCH_LE)
66 #define __AUDIT_ARCH_64BIT 0x80000000
67 #define __AUDIT_ARCH_LE 0x40000000
70 .section .entry.text, "ax"
73 #ifndef CONFIG_PREEMPT
74 #define retint_kernel retint_restore_args
77 #ifdef CONFIG_PARAVIRT
78 ENTRY(native_usergs_sysret64)
81 ENDPROC(native_usergs_sysret64)
82 #endif /* CONFIG_PARAVIRT */
85 .macro TRACE_IRQS_IRETQ offset=ARGOFFSET
86 #ifdef CONFIG_TRACE_IRQFLAGS
87 bt $9,EFLAGS-\offset(%rsp) /* interrupts off? */
95 * When dynamic function tracer is enabled it will add a breakpoint
96 * to all locations that it is about to modify, sync CPUs, update
97 * all the code, sync CPUs, then remove the breakpoints. In this time
98 * if lockdep is enabled, it might jump back into the debug handler
99 * outside the updating of the IST protection. (TRACE_IRQS_ON/OFF).
101 * We need to change the IDT table before calling TRACE_IRQS_ON/OFF to
102 * make sure the stack pointer does not get reset back to the top
103 * of the debug stack, and instead just reuses the current stack.
105 #if defined(CONFIG_DYNAMIC_FTRACE) && defined(CONFIG_TRACE_IRQFLAGS)
107 .macro TRACE_IRQS_OFF_DEBUG
108 call debug_stack_set_zero
110 call debug_stack_reset
113 .macro TRACE_IRQS_ON_DEBUG
114 call debug_stack_set_zero
116 call debug_stack_reset
119 .macro TRACE_IRQS_IRETQ_DEBUG offset=ARGOFFSET
120 bt $9,EFLAGS-\offset(%rsp) /* interrupts off? */
127 # define TRACE_IRQS_OFF_DEBUG TRACE_IRQS_OFF
128 # define TRACE_IRQS_ON_DEBUG TRACE_IRQS_ON
129 # define TRACE_IRQS_IRETQ_DEBUG TRACE_IRQS_IRETQ
133 * C code is not supposed to know about undefined top of stack. Every time
134 * a C function with an pt_regs argument is called from the SYSCALL based
135 * fast path FIXUP_TOP_OF_STACK is needed.
136 * RESTORE_TOP_OF_STACK syncs the syscall state after any possible ptregs
140 /* %rsp:at FRAMEEND */
141 .macro FIXUP_TOP_OF_STACK tmp offset=0
142 movq PER_CPU_VAR(old_rsp),\tmp
143 movq \tmp,RSP+\offset(%rsp)
144 movq $__USER_DS,SS+\offset(%rsp)
145 movq $__USER_CS,CS+\offset(%rsp)
146 movq $-1,RCX+\offset(%rsp)
147 movq R11+\offset(%rsp),\tmp /* get eflags */
148 movq \tmp,EFLAGS+\offset(%rsp)
151 .macro RESTORE_TOP_OF_STACK tmp offset=0
152 movq RSP+\offset(%rsp),\tmp
153 movq \tmp,PER_CPU_VAR(old_rsp)
154 movq EFLAGS+\offset(%rsp),\tmp
155 movq \tmp,R11+\offset(%rsp)
158 .macro FAKE_STACK_FRAME child_rip
159 /* push in order ss, rsp, eflags, cs, rip */
161 pushq_cfi $__KERNEL_DS /* ss */
162 /*CFI_REL_OFFSET ss,0*/
163 pushq_cfi %rax /* rsp */
165 pushq_cfi $(X86_EFLAGS_IF|X86_EFLAGS_FIXED) /* eflags - interrupts on */
166 /*CFI_REL_OFFSET rflags,0*/
167 pushq_cfi $__KERNEL_CS /* cs */
168 /*CFI_REL_OFFSET cs,0*/
169 pushq_cfi \child_rip /* rip */
171 pushq_cfi %rax /* orig rax */
174 .macro UNFAKE_STACK_FRAME
176 CFI_ADJUST_CFA_OFFSET -(6*8)
180 * initial frame state for interrupts (and exceptions without error code)
182 .macro EMPTY_FRAME start=1 offset=0
186 CFI_DEF_CFA rsp,8+\offset
188 CFI_DEF_CFA_OFFSET 8+\offset
193 * initial frame state for interrupts (and exceptions without error code)
195 .macro INTR_FRAME start=1 offset=0
196 EMPTY_FRAME \start, SS+8+\offset-RIP
197 /*CFI_REL_OFFSET ss, SS+\offset-RIP*/
198 CFI_REL_OFFSET rsp, RSP+\offset-RIP
199 /*CFI_REL_OFFSET rflags, EFLAGS+\offset-RIP*/
200 /*CFI_REL_OFFSET cs, CS+\offset-RIP*/
201 CFI_REL_OFFSET rip, RIP+\offset-RIP
205 * initial frame state for exceptions with error code (and interrupts
206 * with vector already pushed)
208 .macro XCPT_FRAME start=1 offset=0
209 INTR_FRAME \start, RIP+\offset-ORIG_RAX
213 * frame that enables calling into C.
215 .macro PARTIAL_FRAME start=1 offset=0
216 XCPT_FRAME \start, ORIG_RAX+\offset-ARGOFFSET
217 CFI_REL_OFFSET rdi, RDI+\offset-ARGOFFSET
218 CFI_REL_OFFSET rsi, RSI+\offset-ARGOFFSET
219 CFI_REL_OFFSET rdx, RDX+\offset-ARGOFFSET
220 CFI_REL_OFFSET rcx, RCX+\offset-ARGOFFSET
221 CFI_REL_OFFSET rax, RAX+\offset-ARGOFFSET
222 CFI_REL_OFFSET r8, R8+\offset-ARGOFFSET
223 CFI_REL_OFFSET r9, R9+\offset-ARGOFFSET
224 CFI_REL_OFFSET r10, R10+\offset-ARGOFFSET
225 CFI_REL_OFFSET r11, R11+\offset-ARGOFFSET
229 * frame that enables passing a complete pt_regs to a C function.
231 .macro DEFAULT_FRAME start=1 offset=0
232 PARTIAL_FRAME \start, R11+\offset-R15
233 CFI_REL_OFFSET rbx, RBX+\offset
234 CFI_REL_OFFSET rbp, RBP+\offset
235 CFI_REL_OFFSET r12, R12+\offset
236 CFI_REL_OFFSET r13, R13+\offset
237 CFI_REL_OFFSET r14, R14+\offset
238 CFI_REL_OFFSET r15, R15+\offset
241 /* save partial stack frame */
244 /* start from rbp in pt_regs and jump over */
245 movq_cfi rdi, (RDI-RBP)
246 movq_cfi rsi, (RSI-RBP)
247 movq_cfi rdx, (RDX-RBP)
248 movq_cfi rcx, (RCX-RBP)
249 movq_cfi rax, (RAX-RBP)
250 movq_cfi r8, (R8-RBP)
251 movq_cfi r9, (R9-RBP)
252 movq_cfi r10, (R10-RBP)
253 movq_cfi r11, (R11-RBP)
255 /* Save rbp so that we can unwind from get_irq_regs() */
258 /* Save previous stack value */
261 leaq -RBP(%rsp),%rdi /* arg1 for handler */
262 testl $3, CS-RBP(%rsi)
266 * irq_count is used to check if a CPU is already on an interrupt stack
267 * or not. While this is essentially redundant with preempt_count it is
268 * a little cheaper to use a separate counter in the PDA (short of
269 * moving irq_enter into assembly, which would be too much work)
271 1: incl PER_CPU_VAR(irq_count)
272 cmovzq PER_CPU_VAR(irq_stack_ptr),%rsp
273 CFI_DEF_CFA_REGISTER rsi
275 /* Store previous stack value */
277 CFI_ESCAPE 0x0f /* DW_CFA_def_cfa_expression */, 6, \
278 0x77 /* DW_OP_breg7 */, 0, \
279 0x06 /* DW_OP_deref */, \
280 0x08 /* DW_OP_const1u */, SS+8-RBP, \
281 0x22 /* DW_OP_plus */
282 /* We entered an interrupt context - irqs are off: */
289 movq %rdi, RDI+8(%rsp)
290 movq %rsi, RSI+8(%rsp)
296 movq %r10, R10+8(%rsp)
297 movq %r11, R11+8(%rsp)
299 movq %rbp, RBP+8(%rsp)
300 movq %r12, R12+8(%rsp)
301 movq %r13, R13+8(%rsp)
302 movq %r14, R14+8(%rsp)
303 movq %r15, R15+8(%rsp)
305 movl $MSR_GS_BASE,%ecx
308 js 1f /* negative -> in kernel */
316 * A newly forked process directly context switches into this address.
318 * rdi: prev task we switched from
323 LOCK ; btr $TIF_FORK,TI_flags(%r8)
326 popfq_cfi # reset kernel eflags
328 call schedule_tail # rdi: 'prev' task parameter
330 GET_THREAD_INFO(%rcx)
334 testl $3, CS-ARGOFFSET(%rsp) # from kernel_thread?
337 testl $_TIF_IA32, TI_flags(%rcx) # 32-bit compat task needs IRET
338 jnz int_ret_from_sys_call
340 RESTORE_TOP_OF_STACK %rdi, -ARGOFFSET
341 jmp ret_from_sys_call # go to the SYSRET fastpath
344 subq $REST_SKIP, %rsp # leave space for volatiles
345 CFI_ADJUST_CFA_OFFSET REST_SKIP
350 jmp int_ret_from_sys_call
355 * System call entry. Up to 6 arguments in registers are supported.
357 * SYSCALL does not save anything on the stack and does not change the
358 * stack pointer. However, it does mask the flags register for us, so
359 * CLD and CLAC are not needed.
364 * rax system call number
366 * rcx return address for syscall/sysret, C arg3
369 * r10 arg3 (--> moved to rcx for C)
372 * r11 eflags for syscall/sysret, temporary for C
373 * r12-r15,rbp,rbx saved by C code, not touched.
375 * Interrupts are off on entry.
376 * Only called from user space.
378 * XXX if we had a free scratch register we could save the RSP into the stack frame
379 * and report it properly in ps. Unfortunately we haven't.
381 * When user can change the frames always force IRET. That is because
382 * it deals with uncanonical addresses better. SYSRET has trouble
383 * with them due to bugs in both AMD and Intel CPUs.
389 CFI_DEF_CFA rsp,KERNEL_STACK_OFFSET
391 /*CFI_REGISTER rflags,r11*/
394 * A hypervisor implementation might want to use a label
395 * after the swapgs, so that it can do the swapgs
396 * for the guest and jump here on syscall.
398 GLOBAL(system_call_after_swapgs)
400 movq %rsp,PER_CPU_VAR(old_rsp)
401 movq PER_CPU_VAR(kernel_stack),%rsp
403 * No need to follow this irqs off/on section - it's straight
406 ENABLE_INTERRUPTS(CLBR_NONE)
407 SAVE_ARGS 8, 0, rax_enosys=1
408 movq_cfi rax,(ORIG_RAX-ARGOFFSET)
409 movq %rcx,RIP-ARGOFFSET(%rsp)
410 CFI_REL_OFFSET rip,RIP-ARGOFFSET
411 testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET)
413 system_call_fastpath:
414 #if __SYSCALL_MASK == ~0
415 cmpq $__NR_syscall_max,%rax
417 andl $__SYSCALL_MASK,%eax
418 cmpl $__NR_syscall_max,%eax
420 ja ret_from_sys_call /* and return regs->ax */
422 call *sys_call_table(,%rax,8) # XXX: rip relative
423 movq %rax,RAX-ARGOFFSET(%rsp)
425 * Syscall return path ending with SYSRET (fast path)
426 * Has incomplete stack frame and undefined top of stack.
429 movl $_TIF_ALLWORK_MASK,%edi
433 DISABLE_INTERRUPTS(CLBR_NONE)
435 movl TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET),%edx
440 * sysretq will re-enable interrupts:
443 movq RIP-ARGOFFSET(%rsp),%rcx
445 RESTORE_ARGS 1,-ARG_SKIP,0
446 /*CFI_REGISTER rflags,r11*/
447 movq PER_CPU_VAR(old_rsp), %rsp
451 /* Handle reschedules */
452 /* edx: work, edi: workmask */
454 bt $TIF_NEED_RESCHED,%edx
457 ENABLE_INTERRUPTS(CLBR_NONE)
463 /* Handle a signal */
466 ENABLE_INTERRUPTS(CLBR_NONE)
467 #ifdef CONFIG_AUDITSYSCALL
468 bt $TIF_SYSCALL_AUDIT,%edx
472 * We have a signal, or exit tracing or single-step.
473 * These all wind up with the iret return path anyway,
474 * so just join that path right now.
476 FIXUP_TOP_OF_STACK %r11, -ARGOFFSET
477 jmp int_check_syscall_exit_work
479 #ifdef CONFIG_AUDITSYSCALL
481 * Return fast path for syscall audit. Call __audit_syscall_exit()
482 * directly and then jump back to the fast path with TIF_SYSCALL_AUDIT
486 movq RAX-ARGOFFSET(%rsp),%rsi /* second arg, syscall return value */
487 cmpq $-MAX_ERRNO,%rsi /* is it < -MAX_ERRNO? */
488 setbe %al /* 1 if so, 0 if not */
489 movzbl %al,%edi /* zero-extend that into %edi */
490 call __audit_syscall_exit
491 movl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT),%edi
493 #endif /* CONFIG_AUDITSYSCALL */
495 /* Do syscall tracing */
497 leaq -REST_SKIP(%rsp), %rdi
498 movq $AUDIT_ARCH_X86_64, %rsi
499 call syscall_trace_enter_phase1
501 jnz tracesys_phase2 /* if needed, run the slow path */
502 LOAD_ARGS 0 /* else restore clobbered regs */
503 jmp system_call_fastpath /* and return to the fast path */
507 FIXUP_TOP_OF_STACK %rdi
509 movq $AUDIT_ARCH_X86_64, %rsi
511 call syscall_trace_enter_phase2
514 * Reload arg registers from stack in case ptrace changed them.
515 * We don't reload %rax because syscall_trace_entry_phase2() returned
516 * the value it wants us to use in the table lookup.
518 LOAD_ARGS ARGOFFSET, 1
520 #if __SYSCALL_MASK == ~0
521 cmpq $__NR_syscall_max,%rax
523 andl $__SYSCALL_MASK,%eax
524 cmpl $__NR_syscall_max,%eax
526 ja int_ret_from_sys_call /* RAX(%rsp) is already set */
527 movq %r10,%rcx /* fixup for C */
528 call *sys_call_table(,%rax,8)
529 movq %rax,RAX-ARGOFFSET(%rsp)
530 /* Use IRET because user could have changed frame */
533 * Syscall return path ending with IRET.
534 * Has correct top of stack, but partial stack frame.
536 GLOBAL(int_ret_from_sys_call)
537 DISABLE_INTERRUPTS(CLBR_NONE)
539 movl $_TIF_ALLWORK_MASK,%edi
540 /* edi: mask to check */
541 GLOBAL(int_with_check)
543 GET_THREAD_INFO(%rcx)
544 movl TI_flags(%rcx),%edx
547 andl $~TS_COMPAT,TI_status(%rcx)
550 /* Either reschedule or signal or syscall exit tracking needed. */
551 /* First do a reschedule test. */
552 /* edx: work, edi: workmask */
554 bt $TIF_NEED_RESCHED,%edx
557 ENABLE_INTERRUPTS(CLBR_NONE)
561 DISABLE_INTERRUPTS(CLBR_NONE)
565 /* handle signals and tracing -- both require a full stack frame */
568 ENABLE_INTERRUPTS(CLBR_NONE)
569 int_check_syscall_exit_work:
571 /* Check for syscall exit trace */
572 testl $_TIF_WORK_SYSCALL_EXIT,%edx
575 leaq 8(%rsp),%rdi # &ptregs -> arg1
576 call syscall_trace_leave
578 andl $~(_TIF_WORK_SYSCALL_EXIT|_TIF_SYSCALL_EMU),%edi
582 testl $_TIF_DO_NOTIFY_MASK,%edx
584 movq %rsp,%rdi # &ptregs -> arg1
585 xorl %esi,%esi # oldset -> arg2
586 call do_notify_resume
587 1: movl $_TIF_WORK_MASK,%edi
590 DISABLE_INTERRUPTS(CLBR_NONE)
596 .macro FORK_LIKE func
599 popq %r11 /* save return address */
602 pushq %r11 /* put it back on stack */
603 FIXUP_TOP_OF_STACK %r11, 8
604 DEFAULT_FRAME 0 8 /* offset 8: return address */
606 RESTORE_TOP_OF_STACK %r11, 8
607 ret $REST_SKIP /* pop extended registers */
612 .macro FIXED_FRAME label,func
615 PARTIAL_FRAME 0 8 /* offset 8: return address */
616 FIXUP_TOP_OF_STACK %r11, 8-ARGOFFSET
618 RESTORE_TOP_OF_STACK %r11, 8-ARGOFFSET
627 FIXED_FRAME stub_iopl, sys_iopl
629 ENTRY(ptregscall_common)
630 DEFAULT_FRAME 1 8 /* offset 8: return address */
631 RESTORE_TOP_OF_STACK %r11, 8
632 movq_cfi_restore R15+8, r15
633 movq_cfi_restore R14+8, r14
634 movq_cfi_restore R13+8, r13
635 movq_cfi_restore R12+8, r12
636 movq_cfi_restore RBP+8, rbp
637 movq_cfi_restore RBX+8, rbx
638 ret $REST_SKIP /* pop extended registers */
640 END(ptregscall_common)
647 FIXUP_TOP_OF_STACK %r11
651 jmp int_ret_from_sys_call
656 * sigreturn is special because it needs to restore all registers on return.
657 * This cannot be done with SYSRET, so use the IRET return path instead.
659 ENTRY(stub_rt_sigreturn)
664 FIXUP_TOP_OF_STACK %r11
665 call sys_rt_sigreturn
666 movq %rax,RAX(%rsp) # fixme, this could be done at the higher layer
668 jmp int_ret_from_sys_call
670 END(stub_rt_sigreturn)
672 #ifdef CONFIG_X86_X32_ABI
673 ENTRY(stub_x32_rt_sigreturn)
678 FIXUP_TOP_OF_STACK %r11
679 call sys32_x32_rt_sigreturn
680 movq %rax,RAX(%rsp) # fixme, this could be done at the higher layer
682 jmp int_ret_from_sys_call
684 END(stub_x32_rt_sigreturn)
686 ENTRY(stub_x32_execve)
691 FIXUP_TOP_OF_STACK %r11
692 call compat_sys_execve
693 RESTORE_TOP_OF_STACK %r11
696 jmp int_ret_from_sys_call
703 * Build the entry stubs and pointer table with some assembler magic.
704 * We pack 7 stubs into a single 32-byte chunk, which will fit in a
705 * single cache line on all modern x86 implementations.
707 .section .init.rodata,"a"
711 .p2align CONFIG_X86_L1_CACHE_SHIFT
712 ENTRY(irq_entries_start)
714 vector=FIRST_EXTERNAL_VECTOR
715 .rept (NR_VECTORS-FIRST_EXTERNAL_VECTOR+6)/7
718 .if vector < NR_VECTORS
719 .if vector <> FIRST_EXTERNAL_VECTOR
720 CFI_ADJUST_CFA_OFFSET -8
722 1: pushq_cfi $(~vector+0x80) /* Note: always in signed byte range */
723 .if ((vector-FIRST_EXTERNAL_VECTOR)%7) <> 6
732 2: jmp common_interrupt
735 END(irq_entries_start)
742 * Interrupt entry/exit.
744 * Interrupt entry points save only callee clobbered registers in fast path.
746 * Entry runs with interrupts off.
749 /* 0(%rsp): ~(interrupt number) */
750 .macro interrupt func
751 /* reserve pt_regs for scratch regs and rbp */
752 subq $ORIG_RAX-RBP, %rsp
753 CFI_ADJUST_CFA_OFFSET ORIG_RAX-RBP
759 * The interrupt stubs push (~vector+0x80) onto the stack and
760 * then jump to common_interrupt.
762 .p2align CONFIG_X86_L1_CACHE_SHIFT
766 addq $-0x80,(%rsp) /* Adjust vector to [-256,-1] range */
768 /* 0(%rsp): old_rsp-ARGOFFSET */
770 DISABLE_INTERRUPTS(CLBR_NONE)
772 decl PER_CPU_VAR(irq_count)
774 /* Restore saved previous stack */
776 CFI_DEF_CFA rsi,SS+8-RBP /* reg/off reset after def_cfa_expr */
777 leaq ARGOFFSET-RBP(%rsi), %rsp
778 CFI_DEF_CFA_REGISTER rsp
779 CFI_ADJUST_CFA_OFFSET RBP-ARGOFFSET
782 GET_THREAD_INFO(%rcx)
783 testl $3,CS-ARGOFFSET(%rsp)
786 /* Interrupt came from user space */
788 * Has a correct top of stack, but a partial stack frame
789 * %rcx: thread info. Interrupts off.
791 retint_with_reschedule:
792 movl $_TIF_WORK_MASK,%edi
795 movl TI_flags(%rcx),%edx
800 retint_swapgs: /* return to user-space */
802 * The iretq could re-enable interrupts:
804 DISABLE_INTERRUPTS(CLBR_ANY)
809 retint_restore_args: /* return to kernel space */
810 DISABLE_INTERRUPTS(CLBR_ANY)
812 * The iretq could re-enable interrupts:
823 * Are we returning to a stack segment from the LDT? Note: in
824 * 64-bit mode SS:RSP on the exception stack is always valid.
826 #ifdef CONFIG_X86_ESPFIX64
827 testb $4,(SS-RIP)(%rsp)
828 jnz native_irq_return_ldt
831 native_irq_return_iret:
833 _ASM_EXTABLE(native_irq_return_iret, bad_iret)
835 #ifdef CONFIG_X86_ESPFIX64
836 native_irq_return_ldt:
840 movq PER_CPU_VAR(espfix_waddr),%rdi
841 movq %rax,(0*8)(%rdi) /* RAX */
842 movq (2*8)(%rsp),%rax /* RIP */
843 movq %rax,(1*8)(%rdi)
844 movq (3*8)(%rsp),%rax /* CS */
845 movq %rax,(2*8)(%rdi)
846 movq (4*8)(%rsp),%rax /* RFLAGS */
847 movq %rax,(3*8)(%rdi)
848 movq (6*8)(%rsp),%rax /* SS */
849 movq %rax,(5*8)(%rdi)
850 movq (5*8)(%rsp),%rax /* RSP */
851 movq %rax,(4*8)(%rdi)
852 andl $0xffff0000,%eax
854 orq PER_CPU_VAR(espfix_stack),%rax
858 jmp native_irq_return_iret
864 * The iret traps when the %cs or %ss being restored is bogus.
865 * We've lost the original trap vector and error code.
866 * #GPF is the most likely one to get for an invalid selector.
867 * So pretend we completed the iret and took the #GPF in user mode.
869 * We are now running with the kernel GS after exception recovery.
870 * But error_entry expects us to have user GS to match the user %cs,
876 jmp general_protection
880 /* edi: workmask, edx: work */
883 bt $TIF_NEED_RESCHED,%edx
886 ENABLE_INTERRUPTS(CLBR_NONE)
890 GET_THREAD_INFO(%rcx)
891 DISABLE_INTERRUPTS(CLBR_NONE)
896 testl $_TIF_DO_NOTIFY_MASK,%edx
899 ENABLE_INTERRUPTS(CLBR_NONE)
901 movq $-1,ORIG_RAX(%rsp)
902 xorl %esi,%esi # oldset
903 movq %rsp,%rdi # &pt_regs
904 call do_notify_resume
906 DISABLE_INTERRUPTS(CLBR_NONE)
908 GET_THREAD_INFO(%rcx)
909 jmp retint_with_reschedule
911 #ifdef CONFIG_PREEMPT
912 /* Returning to kernel space. Check if we need preemption */
913 /* rcx: threadinfo. interrupts off. */
915 cmpl $0,PER_CPU_VAR(__preempt_count)
916 jnz retint_restore_args
917 bt $9,EFLAGS-ARGOFFSET(%rsp) /* interrupts off? */
918 jnc retint_restore_args
919 call preempt_schedule_irq
923 END(common_interrupt)
926 * If IRET takes a fault on the espfix stack, then we
927 * end up promoting it to a doublefault. In that case,
928 * modify the stack to make it look like we just entered
929 * the #GP handler from user space, similar to bad_iret.
931 #ifdef CONFIG_X86_ESPFIX64
935 movq RSP(%rdi),%rax /* Trap on the espfix stack? */
936 sarq $PGDIR_SHIFT,%rax
937 cmpl $ESPFIX_PGD_ENTRY,%eax
938 jne do_double_fault /* No, just deliver the fault */
939 cmpl $__KERNEL_CS,CS(%rdi)
942 cmpq $native_irq_return_iret,%rax
943 jne do_double_fault /* This shouldn't happen... */
944 movq PER_CPU_VAR(kernel_stack),%rax
945 subq $(6*8-KERNEL_STACK_OFFSET),%rax /* Reset to original stack */
947 movq $0,(%rax) /* Missing (lost) #GP error code */
948 movq $general_protection,RIP(%rdi)
951 END(__do_double_fault)
953 # define __do_double_fault do_double_fault
959 .macro apicinterrupt3 num sym do_sym
971 #ifdef CONFIG_TRACING
972 #define trace(sym) trace_##sym
973 #define smp_trace(sym) smp_trace_##sym
975 .macro trace_apicinterrupt num sym
976 apicinterrupt3 \num trace(\sym) smp_trace(\sym)
979 .macro trace_apicinterrupt num sym do_sym
983 .macro apicinterrupt num sym do_sym
984 apicinterrupt3 \num \sym \do_sym
985 trace_apicinterrupt \num \sym
989 apicinterrupt3 IRQ_MOVE_CLEANUP_VECTOR \
990 irq_move_cleanup_interrupt smp_irq_move_cleanup_interrupt
991 apicinterrupt3 REBOOT_VECTOR \
992 reboot_interrupt smp_reboot_interrupt
996 apicinterrupt3 UV_BAU_MESSAGE \
997 uv_bau_message_intr1 uv_bau_message_interrupt
999 apicinterrupt LOCAL_TIMER_VECTOR \
1000 apic_timer_interrupt smp_apic_timer_interrupt
1001 apicinterrupt X86_PLATFORM_IPI_VECTOR \
1002 x86_platform_ipi smp_x86_platform_ipi
1004 #ifdef CONFIG_HAVE_KVM
1005 apicinterrupt3 POSTED_INTR_VECTOR \
1006 kvm_posted_intr_ipi smp_kvm_posted_intr_ipi
1009 #ifdef CONFIG_X86_MCE_THRESHOLD
1010 apicinterrupt THRESHOLD_APIC_VECTOR \
1011 threshold_interrupt smp_threshold_interrupt
1014 #ifdef CONFIG_X86_THERMAL_VECTOR
1015 apicinterrupt THERMAL_APIC_VECTOR \
1016 thermal_interrupt smp_thermal_interrupt
1020 apicinterrupt CALL_FUNCTION_SINGLE_VECTOR \
1021 call_function_single_interrupt smp_call_function_single_interrupt
1022 apicinterrupt CALL_FUNCTION_VECTOR \
1023 call_function_interrupt smp_call_function_interrupt
1024 apicinterrupt RESCHEDULE_VECTOR \
1025 reschedule_interrupt smp_reschedule_interrupt
1028 apicinterrupt ERROR_APIC_VECTOR \
1029 error_interrupt smp_error_interrupt
1030 apicinterrupt SPURIOUS_APIC_VECTOR \
1031 spurious_interrupt smp_spurious_interrupt
1033 #ifdef CONFIG_IRQ_WORK
1034 apicinterrupt IRQ_WORK_VECTOR \
1035 irq_work_interrupt smp_irq_work_interrupt
1039 * Exception entry points.
1041 #define INIT_TSS_IST(x) PER_CPU_VAR(init_tss) + (TSS_ist + ((x) - 1) * 8)
1043 .macro idtentry sym do_sym has_error_code:req paranoid=0 shift_ist=-1
1046 .if \shift_ist != -1 && \paranoid == 0
1047 .error "using shift_ist requires paranoid=1"
1057 PARAVIRT_ADJUST_EXCEPTION_FRAME
1059 .ifeq \has_error_code
1060 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1063 subq $ORIG_RAX-R15, %rsp
1064 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1075 .if \shift_ist != -1
1076 TRACE_IRQS_OFF_DEBUG /* reload IDT in case of recursion */
1082 movq %rsp,%rdi /* pt_regs pointer */
1085 movq ORIG_RAX(%rsp),%rsi /* get error code */
1086 movq $-1,ORIG_RAX(%rsp) /* no syscall to restart */
1088 xorl %esi,%esi /* no error code */
1091 .if \shift_ist != -1
1092 subq $EXCEPTION_STKSZ, INIT_TSS_IST(\shift_ist)
1097 .if \shift_ist != -1
1098 addq $EXCEPTION_STKSZ, INIT_TSS_IST(\shift_ist)
1102 jmp paranoid_exit /* %ebx: no swapgs flag */
1104 jmp error_exit /* %ebx: no swapgs flag */
1111 #ifdef CONFIG_TRACING
1112 .macro trace_idtentry sym do_sym has_error_code:req
1113 idtentry trace(\sym) trace(\do_sym) has_error_code=\has_error_code
1114 idtentry \sym \do_sym has_error_code=\has_error_code
1117 .macro trace_idtentry sym do_sym has_error_code:req
1118 idtentry \sym \do_sym has_error_code=\has_error_code
1122 idtentry divide_error do_divide_error has_error_code=0
1123 idtentry overflow do_overflow has_error_code=0
1124 idtentry bounds do_bounds has_error_code=0
1125 idtentry invalid_op do_invalid_op has_error_code=0
1126 idtentry device_not_available do_device_not_available has_error_code=0
1127 idtentry double_fault __do_double_fault has_error_code=1 paranoid=1
1128 idtentry coprocessor_segment_overrun do_coprocessor_segment_overrun has_error_code=0
1129 idtentry invalid_TSS do_invalid_TSS has_error_code=1
1130 idtentry segment_not_present do_segment_not_present has_error_code=1
1131 idtentry spurious_interrupt_bug do_spurious_interrupt_bug has_error_code=0
1132 idtentry coprocessor_error do_coprocessor_error has_error_code=0
1133 idtentry alignment_check do_alignment_check has_error_code=1
1134 idtentry simd_coprocessor_error do_simd_coprocessor_error has_error_code=0
1137 /* Reload gs selector with exception handling */
1138 /* edi: new selector */
1139 ENTRY(native_load_gs_index)
1142 DISABLE_INTERRUPTS(CLBR_ANY & ~CLBR_RDI)
1146 2: mfence /* workaround */
1151 END(native_load_gs_index)
1153 _ASM_EXTABLE(gs_change,bad_gs)
1154 .section .fixup,"ax"
1155 /* running with kernelgs */
1157 SWAPGS /* switch back to user gs */
1163 /* Call softirq on interrupt stack. Interrupts are off. */
1164 ENTRY(do_softirq_own_stack)
1167 CFI_REL_OFFSET rbp,0
1169 CFI_DEF_CFA_REGISTER rbp
1170 incl PER_CPU_VAR(irq_count)
1171 cmove PER_CPU_VAR(irq_stack_ptr),%rsp
1172 push %rbp # backlink for old unwinder
1176 CFI_DEF_CFA_REGISTER rsp
1177 CFI_ADJUST_CFA_OFFSET -8
1178 decl PER_CPU_VAR(irq_count)
1181 END(do_softirq_own_stack)
1184 idtentry xen_hypervisor_callback xen_do_hypervisor_callback has_error_code=0
1187 * A note on the "critical region" in our callback handler.
1188 * We want to avoid stacking callback handlers due to events occurring
1189 * during handling of the last event. To do this, we keep events disabled
1190 * until we've done all processing. HOWEVER, we must enable events before
1191 * popping the stack frame (can't be done atomically) and so it would still
1192 * be possible to get enough handler activations to overflow the stack.
1193 * Although unlikely, bugs of that kind are hard to track down, so we'd
1194 * like to avoid the possibility.
1195 * So, on entry to the handler we detect whether we interrupted an
1196 * existing activation in its critical region -- if so, we pop the current
1197 * activation and restart the handler using the previous one.
1199 ENTRY(xen_do_hypervisor_callback) # do_hypervisor_callback(struct *pt_regs)
1202 * Since we don't modify %rdi, evtchn_do_upall(struct *pt_regs) will
1203 * see the correct pointer to the pt_regs
1205 movq %rdi, %rsp # we don't return, adjust the stack frame
1208 11: incl PER_CPU_VAR(irq_count)
1210 CFI_DEF_CFA_REGISTER rbp
1211 cmovzq PER_CPU_VAR(irq_stack_ptr),%rsp
1212 pushq %rbp # backlink for old unwinder
1213 call xen_evtchn_do_upcall
1215 CFI_DEF_CFA_REGISTER rsp
1216 decl PER_CPU_VAR(irq_count)
1219 END(xen_do_hypervisor_callback)
1222 * Hypervisor uses this for application faults while it executes.
1223 * We get here for two reasons:
1224 * 1. Fault while reloading DS, ES, FS or GS
1225 * 2. Fault while executing IRET
1226 * Category 1 we do not need to fix up as Xen has already reloaded all segment
1227 * registers that could be reloaded and zeroed the others.
1228 * Category 2 we fix up by killing the current process. We cannot use the
1229 * normal Linux return path in this case because if we use the IRET hypercall
1230 * to pop the stack frame we end up in an infinite loop of failsafe callbacks.
1231 * We distinguish between categories by comparing each saved segment register
1232 * with its current contents: any discrepancy means we in category 1.
1234 ENTRY(xen_failsafe_callback)
1236 /*CFI_REL_OFFSET gs,GS*/
1237 /*CFI_REL_OFFSET fs,FS*/
1238 /*CFI_REL_OFFSET es,ES*/
1239 /*CFI_REL_OFFSET ds,DS*/
1240 CFI_REL_OFFSET r11,8
1241 CFI_REL_OFFSET rcx,0
1255 /* All segments match their saved values => Category 2 (Bad IRET). */
1261 CFI_ADJUST_CFA_OFFSET -0x30
1262 pushq_cfi $0 /* RIP */
1265 jmp general_protection
1267 1: /* Segment mismatch => Category 1 (Bad segment). Retry the IRET. */
1273 CFI_ADJUST_CFA_OFFSET -0x30
1274 pushq_cfi $-1 /* orig_ax = -1 => not a system call */
1278 END(xen_failsafe_callback)
1280 apicinterrupt3 HYPERVISOR_CALLBACK_VECTOR \
1281 xen_hvm_callback_vector xen_evtchn_do_upcall
1283 #endif /* CONFIG_XEN */
1285 #if IS_ENABLED(CONFIG_HYPERV)
1286 apicinterrupt3 HYPERVISOR_CALLBACK_VECTOR \
1287 hyperv_callback_vector hyperv_vector_handler
1288 #endif /* CONFIG_HYPERV */
1290 idtentry debug do_debug has_error_code=0 paranoid=1 shift_ist=DEBUG_STACK
1291 idtentry int3 do_int3 has_error_code=0 paranoid=1 shift_ist=DEBUG_STACK
1292 idtentry stack_segment do_stack_segment has_error_code=1 paranoid=1
1294 idtentry xen_debug do_debug has_error_code=0
1295 idtentry xen_int3 do_int3 has_error_code=0
1296 idtentry xen_stack_segment do_stack_segment has_error_code=1
1298 idtentry general_protection do_general_protection has_error_code=1
1299 trace_idtentry page_fault do_page_fault has_error_code=1
1300 #ifdef CONFIG_KVM_GUEST
1301 idtentry async_page_fault do_async_page_fault has_error_code=1
1303 #ifdef CONFIG_X86_MCE
1304 idtentry machine_check has_error_code=0 paranoid=1 do_sym=*machine_check_vector(%rip)
1308 * "Paranoid" exit path from exception stack.
1309 * Paranoid because this is used by NMIs and cannot take
1310 * any kernel state for granted.
1311 * We don't do kernel preemption checks here, because only
1312 * NMI should be common and it does not enable IRQs and
1313 * cannot get reschedule ticks.
1315 * "trace" is 0 for the NMI handler only, because irq-tracing
1316 * is fundamentally NMI-unsafe. (we cannot change the soft and
1317 * hard flags at once, atomically)
1320 /* ebx: no swapgs flag */
1321 ENTRY(paranoid_exit)
1323 DISABLE_INTERRUPTS(CLBR_NONE)
1324 TRACE_IRQS_OFF_DEBUG
1325 testl %ebx,%ebx /* swapgs needed? */
1326 jnz paranoid_restore
1328 jnz paranoid_userspace
1335 TRACE_IRQS_IRETQ_DEBUG 0
1339 GET_THREAD_INFO(%rcx)
1340 movl TI_flags(%rcx),%ebx
1341 andl $_TIF_WORK_MASK,%ebx
1343 movq %rsp,%rdi /* &pt_regs */
1345 movq %rax,%rsp /* switch stack for scheduling */
1346 testl $_TIF_NEED_RESCHED,%ebx
1347 jnz paranoid_schedule
1348 movl %ebx,%edx /* arg3: thread flags */
1350 ENABLE_INTERRUPTS(CLBR_NONE)
1351 xorl %esi,%esi /* arg2: oldset */
1352 movq %rsp,%rdi /* arg1: &pt_regs */
1353 call do_notify_resume
1354 DISABLE_INTERRUPTS(CLBR_NONE)
1356 jmp paranoid_userspace
1359 ENABLE_INTERRUPTS(CLBR_ANY)
1361 DISABLE_INTERRUPTS(CLBR_ANY)
1363 jmp paranoid_userspace
1368 * Exception entry point. This expects an error code/orig_rax on the stack.
1369 * returns in "no swapgs flag" in %ebx.
1373 CFI_ADJUST_CFA_OFFSET 15*8
1374 /* oldrax contains error code */
1376 movq %rdi, RDI+8(%rsp)
1377 movq %rsi, RSI+8(%rsp)
1378 movq %rdx, RDX+8(%rsp)
1379 movq %rcx, RCX+8(%rsp)
1380 movq %rax, RAX+8(%rsp)
1381 movq %r8, R8+8(%rsp)
1382 movq %r9, R9+8(%rsp)
1383 movq %r10, R10+8(%rsp)
1384 movq %r11, R11+8(%rsp)
1386 movq %rbp, RBP+8(%rsp)
1387 movq %r12, R12+8(%rsp)
1388 movq %r13, R13+8(%rsp)
1389 movq %r14, R14+8(%rsp)
1390 movq %r15, R15+8(%rsp)
1393 je error_kernelspace
1401 * There are two places in the kernel that can potentially fault with
1402 * usergs. Handle them here. The exception handlers after iret run with
1403 * kernel gs again, so don't set the user space flag. B stepping K8s
1404 * sometimes report an truncated RIP for IRET exceptions returning to
1405 * compat mode. Check for these here too.
1408 CFI_REL_OFFSET rcx, RCX+8
1410 leaq native_irq_return_iret(%rip),%rcx
1411 cmpq %rcx,RIP+8(%rsp)
1413 movl %ecx,%eax /* zero extend */
1414 cmpq %rax,RIP+8(%rsp)
1416 cmpq $gs_change,RIP+8(%rsp)
1421 /* Fix truncated RIP */
1422 movq %rcx,RIP+8(%rsp)
1428 /* ebx: no swapgs flag (1: don't need swapgs, 0: need it) */
1433 DISABLE_INTERRUPTS(CLBR_NONE)
1435 GET_THREAD_INFO(%rcx)
1438 LOCKDEP_SYS_EXIT_IRQ
1439 movl TI_flags(%rcx),%edx
1440 movl $_TIF_WORK_MASK,%edi
1448 * Test if a given stack is an NMI stack or not.
1450 .macro test_in_nmi reg stack nmi_ret normal_ret
1453 subq $EXCEPTION_STKSZ, %\reg
1459 /* runs on exception stack */
1462 PARAVIRT_ADJUST_EXCEPTION_FRAME
1464 * We allow breakpoints in NMIs. If a breakpoint occurs, then
1465 * the iretq it performs will take us out of NMI context.
1466 * This means that we can have nested NMIs where the next
1467 * NMI is using the top of the stack of the previous NMI. We
1468 * can't let it execute because the nested NMI will corrupt the
1469 * stack of the previous NMI. NMI handlers are not re-entrant
1472 * To handle this case we do the following:
1473 * Check the a special location on the stack that contains
1474 * a variable that is set when NMIs are executing.
1475 * The interrupted task's stack is also checked to see if it
1477 * If the variable is not set and the stack is not the NMI
1479 * o Set the special variable on the stack
1480 * o Copy the interrupt frame into a "saved" location on the stack
1481 * o Copy the interrupt frame into a "copy" location on the stack
1482 * o Continue processing the NMI
1483 * If the variable is set or the previous stack is the NMI stack:
1484 * o Modify the "copy" location to jump to the repeate_nmi
1485 * o return back to the first NMI
1487 * Now on exit of the first NMI, we first clear the stack variable
1488 * The NMI stack will tell any nested NMIs at that point that it is
1489 * nested. Then we pop the stack normally with iret, and if there was
1490 * a nested NMI that updated the copy interrupt stack frame, a
1491 * jump will be made to the repeat_nmi code that will handle the second
1495 /* Use %rdx as out temp variable throughout */
1497 CFI_REL_OFFSET rdx, 0
1500 * If %cs was not the kernel segment, then the NMI triggered in user
1501 * space, which means it is definitely not nested.
1503 cmpl $__KERNEL_CS, 16(%rsp)
1507 * Check the special variable on the stack to see if NMIs are
1514 * Now test if the previous stack was an NMI stack.
1515 * We need the double check. We check the NMI stack to satisfy the
1516 * race when the first NMI clears the variable before returning.
1517 * We check the variable because the first NMI could be in a
1518 * breakpoint routine using a breakpoint stack.
1521 test_in_nmi rdx, 4*8(%rsp), nested_nmi, first_nmi
1526 * Do nothing if we interrupted the fixup in repeat_nmi.
1527 * It's about to repeat the NMI handler, so we are fine
1528 * with ignoring this one.
1530 movq $repeat_nmi, %rdx
1533 movq $end_repeat_nmi, %rdx
1538 /* Set up the interrupted NMIs stack to jump to repeat_nmi */
1539 leaq -1*8(%rsp), %rdx
1541 CFI_ADJUST_CFA_OFFSET 1*8
1542 leaq -10*8(%rsp), %rdx
1543 pushq_cfi $__KERNEL_DS
1546 pushq_cfi $__KERNEL_CS
1547 pushq_cfi $repeat_nmi
1549 /* Put stack back */
1551 CFI_ADJUST_CFA_OFFSET -6*8
1557 /* No need to check faults here */
1563 * Because nested NMIs will use the pushed location that we
1564 * stored in rdx, we must keep that space available.
1565 * Here's what our stack frame will look like:
1566 * +-------------------------+
1568 * | original Return RSP |
1569 * | original RFLAGS |
1572 * +-------------------------+
1573 * | temp storage for rdx |
1574 * +-------------------------+
1575 * | NMI executing variable |
1576 * +-------------------------+
1578 * | copied Return RSP |
1582 * +-------------------------+
1584 * | Saved Return RSP |
1588 * +-------------------------+
1590 * +-------------------------+
1592 * The saved stack frame is used to fix up the copied stack frame
1593 * that a nested NMI may change to make the interrupted NMI iret jump
1594 * to the repeat_nmi. The original stack frame and the temp storage
1595 * is also used by nested NMIs and can not be trusted on exit.
1597 /* Do not pop rdx, nested NMIs will corrupt that part of the stack */
1601 /* Set the NMI executing variable on the stack. */
1605 * Leave room for the "copied" frame
1608 CFI_ADJUST_CFA_OFFSET 5*8
1610 /* Copy the stack frame to the Saved frame */
1612 pushq_cfi 11*8(%rsp)
1614 CFI_DEF_CFA_OFFSET SS+8-RIP
1616 /* Everything up to here is safe from nested NMIs */
1619 * If there was a nested NMI, the first NMI's iret will return
1620 * here. But NMIs are still enabled and we can take another
1621 * nested NMI. The nested NMI checks the interrupted RIP to see
1622 * if it is between repeat_nmi and end_repeat_nmi, and if so
1623 * it will just return, as we are about to repeat an NMI anyway.
1624 * This makes it safe to copy to the stack frame that a nested
1629 * Update the stack variable to say we are still in NMI (the update
1630 * is benign for the non-repeat case, where 1 was pushed just above
1631 * to this very stack slot).
1635 /* Make another copy, this one may be modified by nested NMIs */
1637 CFI_ADJUST_CFA_OFFSET -10*8
1639 pushq_cfi -6*8(%rsp)
1642 CFI_DEF_CFA_OFFSET SS+8-RIP
1646 * Everything below this point can be preempted by a nested
1647 * NMI if the first NMI took an exception and reset our iret stack
1648 * so that we repeat another NMI.
1650 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1651 subq $ORIG_RAX-R15, %rsp
1652 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1654 * Use save_paranoid to handle SWAPGS, but no need to use paranoid_exit
1655 * as we should not be calling schedule in NMI context.
1656 * Even with normal interrupts enabled. An NMI should not be
1657 * setting NEED_RESCHED or anything that normal interrupts and
1658 * exceptions might do.
1664 * Save off the CR2 register. If we take a page fault in the NMI then
1665 * it could corrupt the CR2 value. If the NMI preempts a page fault
1666 * handler before it was able to read the CR2 register, and then the
1667 * NMI itself takes a page fault, the page fault that was preempted
1668 * will read the information from the NMI page fault and not the
1669 * origin fault. Save it off and restore it if it changes.
1670 * Use the r12 callee-saved register.
1674 /* paranoidentry do_nmi, 0; without TRACE_IRQS_OFF */
1679 /* Did the NMI take a page fault? Restore cr2 if it did */
1686 testl %ebx,%ebx /* swapgs needed? */
1691 /* Pop the extra iret frame at once */
1694 /* Clear the NMI executing stack variable */
1700 ENTRY(ignore_sysret)