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 RIP+\offset(%rsp),\tmp /* get rip */
147 movq \tmp,RCX+\offset(%rsp) /* copy it to rcx as sysret would do */
148 movq R11+\offset(%rsp),\tmp /* get eflags */
149 movq \tmp,EFLAGS+\offset(%rsp)
152 .macro RESTORE_TOP_OF_STACK tmp offset=0
153 movq RSP+\offset(%rsp),\tmp
154 movq \tmp,PER_CPU_VAR(old_rsp)
155 movq EFLAGS+\offset(%rsp),\tmp
156 movq \tmp,R11+\offset(%rsp)
160 * initial frame state for interrupts (and exceptions without error code)
162 .macro EMPTY_FRAME start=1 offset=0
166 CFI_DEF_CFA rsp,8+\offset
168 CFI_DEF_CFA_OFFSET 8+\offset
173 * initial frame state for interrupts (and exceptions without error code)
175 .macro INTR_FRAME start=1 offset=0
176 EMPTY_FRAME \start, SS+8+\offset-RIP
177 /*CFI_REL_OFFSET ss, SS+\offset-RIP*/
178 CFI_REL_OFFSET rsp, RSP+\offset-RIP
179 /*CFI_REL_OFFSET rflags, EFLAGS+\offset-RIP*/
180 /*CFI_REL_OFFSET cs, CS+\offset-RIP*/
181 CFI_REL_OFFSET rip, RIP+\offset-RIP
185 * initial frame state for exceptions with error code (and interrupts
186 * with vector already pushed)
188 .macro XCPT_FRAME start=1 offset=0
189 INTR_FRAME \start, RIP+\offset-ORIG_RAX
193 * frame that enables calling into C.
195 .macro PARTIAL_FRAME start=1 offset=0
196 XCPT_FRAME \start, ORIG_RAX+\offset-ARGOFFSET
197 CFI_REL_OFFSET rdi, RDI+\offset-ARGOFFSET
198 CFI_REL_OFFSET rsi, RSI+\offset-ARGOFFSET
199 CFI_REL_OFFSET rdx, RDX+\offset-ARGOFFSET
200 CFI_REL_OFFSET rcx, RCX+\offset-ARGOFFSET
201 CFI_REL_OFFSET rax, RAX+\offset-ARGOFFSET
202 CFI_REL_OFFSET r8, R8+\offset-ARGOFFSET
203 CFI_REL_OFFSET r9, R9+\offset-ARGOFFSET
204 CFI_REL_OFFSET r10, R10+\offset-ARGOFFSET
205 CFI_REL_OFFSET r11, R11+\offset-ARGOFFSET
209 * frame that enables passing a complete pt_regs to a C function.
211 .macro DEFAULT_FRAME start=1 offset=0
212 PARTIAL_FRAME \start, R11+\offset-R15
213 CFI_REL_OFFSET rbx, RBX+\offset
214 CFI_REL_OFFSET rbp, RBP+\offset
215 CFI_REL_OFFSET r12, R12+\offset
216 CFI_REL_OFFSET r13, R13+\offset
217 CFI_REL_OFFSET r14, R14+\offset
218 CFI_REL_OFFSET r15, R15+\offset
224 movq %rdi, RDI+8(%rsp)
225 movq %rsi, RSI+8(%rsp)
231 movq %r10, R10+8(%rsp)
232 movq %r11, R11+8(%rsp)
234 movq %rbp, RBP+8(%rsp)
235 movq %r12, R12+8(%rsp)
236 movq %r13, R13+8(%rsp)
237 movq %r14, R14+8(%rsp)
238 movq %r15, R15+8(%rsp)
240 movl $MSR_GS_BASE,%ecx
243 js 1f /* negative -> in kernel */
251 * A newly forked process directly context switches into this address.
253 * rdi: prev task we switched from
258 LOCK ; btr $TIF_FORK,TI_flags(%r8)
261 popfq_cfi # reset kernel eflags
263 call schedule_tail # rdi: 'prev' task parameter
265 GET_THREAD_INFO(%rcx)
269 testl $3, CS-ARGOFFSET(%rsp) # from kernel_thread?
272 testl $_TIF_IA32, TI_flags(%rcx) # 32-bit compat task needs IRET
273 jnz int_ret_from_sys_call
275 RESTORE_TOP_OF_STACK %rdi, -ARGOFFSET
276 jmp ret_from_sys_call # go to the SYSRET fastpath
279 subq $REST_SKIP, %rsp # leave space for volatiles
280 CFI_ADJUST_CFA_OFFSET REST_SKIP
285 jmp int_ret_from_sys_call
290 * System call entry. Up to 6 arguments in registers are supported.
292 * SYSCALL does not save anything on the stack and does not change the
293 * stack pointer. However, it does mask the flags register for us, so
294 * CLD and CLAC are not needed.
299 * rax system call number
301 * rcx return address for syscall/sysret, C arg3
304 * r10 arg3 (--> moved to rcx for C)
307 * r11 eflags for syscall/sysret, temporary for C
308 * r12-r15,rbp,rbx saved by C code, not touched.
310 * Interrupts are off on entry.
311 * Only called from user space.
313 * XXX if we had a free scratch register we could save the RSP into the stack frame
314 * and report it properly in ps. Unfortunately we haven't.
316 * When user can change the frames always force IRET. That is because
317 * it deals with uncanonical addresses better. SYSRET has trouble
318 * with them due to bugs in both AMD and Intel CPUs.
324 CFI_DEF_CFA rsp,KERNEL_STACK_OFFSET
326 /*CFI_REGISTER rflags,r11*/
329 * A hypervisor implementation might want to use a label
330 * after the swapgs, so that it can do the swapgs
331 * for the guest and jump here on syscall.
333 GLOBAL(system_call_after_swapgs)
335 movq %rsp,PER_CPU_VAR(old_rsp)
336 movq PER_CPU_VAR(kernel_stack),%rsp
338 * No need to follow this irqs off/on section - it's straight
341 ENABLE_INTERRUPTS(CLBR_NONE)
342 SAVE_ARGS 8, 0, rax_enosys=1
343 movq_cfi rax,(ORIG_RAX-ARGOFFSET)
344 movq %rcx,RIP-ARGOFFSET(%rsp)
345 CFI_REL_OFFSET rip,RIP-ARGOFFSET
346 testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET)
348 system_call_fastpath:
349 #if __SYSCALL_MASK == ~0
350 cmpq $__NR_syscall_max,%rax
352 andl $__SYSCALL_MASK,%eax
353 cmpl $__NR_syscall_max,%eax
355 ja ret_from_sys_call /* and return regs->ax */
357 call *sys_call_table(,%rax,8) # XXX: rip relative
358 movq %rax,RAX-ARGOFFSET(%rsp)
360 * Syscall return path ending with SYSRET (fast path)
361 * Has incomplete stack frame and undefined top of stack.
364 testl $_TIF_ALLWORK_MASK,TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET)
365 jnz int_ret_from_sys_call_fixup /* Go the the slow path */
368 DISABLE_INTERRUPTS(CLBR_NONE)
372 * sysretq will re-enable interrupts:
375 movq RIP-ARGOFFSET(%rsp),%rcx
377 RESTORE_ARGS 1,-ARG_SKIP,0
378 /*CFI_REGISTER rflags,r11*/
379 movq PER_CPU_VAR(old_rsp), %rsp
384 int_ret_from_sys_call_fixup:
385 FIXUP_TOP_OF_STACK %r11, -ARGOFFSET
386 jmp int_ret_from_sys_call
388 /* Do syscall tracing */
390 leaq -REST_SKIP(%rsp), %rdi
391 movq $AUDIT_ARCH_X86_64, %rsi
392 call syscall_trace_enter_phase1
394 jnz tracesys_phase2 /* if needed, run the slow path */
395 LOAD_ARGS 0 /* else restore clobbered regs */
396 jmp system_call_fastpath /* and return to the fast path */
400 FIXUP_TOP_OF_STACK %rdi
402 movq $AUDIT_ARCH_X86_64, %rsi
404 call syscall_trace_enter_phase2
407 * Reload arg registers from stack in case ptrace changed them.
408 * We don't reload %rax because syscall_trace_entry_phase2() returned
409 * the value it wants us to use in the table lookup.
411 LOAD_ARGS ARGOFFSET, 1
413 #if __SYSCALL_MASK == ~0
414 cmpq $__NR_syscall_max,%rax
416 andl $__SYSCALL_MASK,%eax
417 cmpl $__NR_syscall_max,%eax
419 ja int_ret_from_sys_call /* RAX(%rsp) is already set */
420 movq %r10,%rcx /* fixup for C */
421 call *sys_call_table(,%rax,8)
422 movq %rax,RAX-ARGOFFSET(%rsp)
423 /* Use IRET because user could have changed frame */
426 * Syscall return path ending with IRET.
427 * Has correct top of stack, but partial stack frame.
429 GLOBAL(int_ret_from_sys_call)
430 DISABLE_INTERRUPTS(CLBR_NONE)
432 movl $_TIF_ALLWORK_MASK,%edi
433 /* edi: mask to check */
434 GLOBAL(int_with_check)
436 GET_THREAD_INFO(%rcx)
437 movl TI_flags(%rcx),%edx
440 andl $~TS_COMPAT,TI_status(%rcx)
443 /* Either reschedule or signal or syscall exit tracking needed. */
444 /* First do a reschedule test. */
445 /* edx: work, edi: workmask */
447 bt $TIF_NEED_RESCHED,%edx
450 ENABLE_INTERRUPTS(CLBR_NONE)
454 DISABLE_INTERRUPTS(CLBR_NONE)
458 /* handle signals and tracing -- both require a full stack frame */
461 ENABLE_INTERRUPTS(CLBR_NONE)
462 int_check_syscall_exit_work:
464 /* Check for syscall exit trace */
465 testl $_TIF_WORK_SYSCALL_EXIT,%edx
468 leaq 8(%rsp),%rdi # &ptregs -> arg1
469 call syscall_trace_leave
471 andl $~(_TIF_WORK_SYSCALL_EXIT|_TIF_SYSCALL_EMU),%edi
475 testl $_TIF_DO_NOTIFY_MASK,%edx
477 movq %rsp,%rdi # &ptregs -> arg1
478 xorl %esi,%esi # oldset -> arg2
479 call do_notify_resume
480 1: movl $_TIF_WORK_MASK,%edi
483 DISABLE_INTERRUPTS(CLBR_NONE)
489 .macro FORK_LIKE func
492 popq %r11 /* save return address */
495 pushq %r11 /* put it back on stack */
496 FIXUP_TOP_OF_STACK %r11, 8
497 DEFAULT_FRAME 0 8 /* offset 8: return address */
499 RESTORE_TOP_OF_STACK %r11, 8
500 ret $REST_SKIP /* pop extended registers */
505 .macro FIXED_FRAME label,func
508 PARTIAL_FRAME 0 8 /* offset 8: return address */
509 FIXUP_TOP_OF_STACK %r11, 8-ARGOFFSET
511 RESTORE_TOP_OF_STACK %r11, 8-ARGOFFSET
520 FIXED_FRAME stub_iopl, sys_iopl
527 FIXUP_TOP_OF_STACK %r11
531 jmp int_ret_from_sys_call
540 FIXUP_TOP_OF_STACK %r11
542 RESTORE_TOP_OF_STACK %r11
545 jmp int_ret_from_sys_call
550 * sigreturn is special because it needs to restore all registers on return.
551 * This cannot be done with SYSRET, so use the IRET return path instead.
553 ENTRY(stub_rt_sigreturn)
558 FIXUP_TOP_OF_STACK %r11
559 call sys_rt_sigreturn
560 movq %rax,RAX(%rsp) # fixme, this could be done at the higher layer
562 jmp int_ret_from_sys_call
564 END(stub_rt_sigreturn)
566 #ifdef CONFIG_X86_X32_ABI
567 ENTRY(stub_x32_rt_sigreturn)
572 FIXUP_TOP_OF_STACK %r11
573 call sys32_x32_rt_sigreturn
574 movq %rax,RAX(%rsp) # fixme, this could be done at the higher layer
576 jmp int_ret_from_sys_call
578 END(stub_x32_rt_sigreturn)
580 ENTRY(stub_x32_execve)
585 FIXUP_TOP_OF_STACK %r11
586 call compat_sys_execve
587 RESTORE_TOP_OF_STACK %r11
590 jmp int_ret_from_sys_call
594 ENTRY(stub_x32_execveat)
599 FIXUP_TOP_OF_STACK %r11
600 call compat_sys_execveat
601 RESTORE_TOP_OF_STACK %r11
604 jmp int_ret_from_sys_call
606 END(stub_x32_execveat)
611 * Build the entry stubs and pointer table with some assembler magic.
612 * We pack 7 stubs into a single 32-byte chunk, which will fit in a
613 * single cache line on all modern x86 implementations.
615 .section .init.rodata,"a"
619 .p2align CONFIG_X86_L1_CACHE_SHIFT
620 ENTRY(irq_entries_start)
622 vector=FIRST_EXTERNAL_VECTOR
623 .rept (FIRST_SYSTEM_VECTOR-FIRST_EXTERNAL_VECTOR+6)/7
626 .if vector < FIRST_SYSTEM_VECTOR
627 .if vector <> FIRST_EXTERNAL_VECTOR
628 CFI_ADJUST_CFA_OFFSET -8
630 1: pushq_cfi $(~vector+0x80) /* Note: always in signed byte range */
631 .if ((vector-FIRST_EXTERNAL_VECTOR)%7) <> 6
640 2: jmp common_interrupt
643 END(irq_entries_start)
650 * Interrupt entry/exit.
652 * Interrupt entry points save only callee clobbered registers in fast path.
654 * Entry runs with interrupts off.
657 /* 0(%rsp): ~(interrupt number) */
658 .macro interrupt func
659 /* reserve pt_regs for scratch regs and rbp */
660 subq $ORIG_RAX-RBP, %rsp
661 CFI_ADJUST_CFA_OFFSET ORIG_RAX-RBP
663 /* start from rbp in pt_regs and jump over */
664 movq_cfi rdi, (RDI-RBP)
665 movq_cfi rsi, (RSI-RBP)
666 movq_cfi rdx, (RDX-RBP)
667 movq_cfi rcx, (RCX-RBP)
668 movq_cfi rax, (RAX-RBP)
669 movq_cfi r8, (R8-RBP)
670 movq_cfi r9, (R9-RBP)
671 movq_cfi r10, (R10-RBP)
672 movq_cfi r11, (R11-RBP)
674 /* Save rbp so that we can unwind from get_irq_regs() */
677 /* Save previous stack value */
680 leaq -RBP(%rsp),%rdi /* arg1 for handler */
681 testl $3, CS-RBP(%rsi)
685 * irq_count is used to check if a CPU is already on an interrupt stack
686 * or not. While this is essentially redundant with preempt_count it is
687 * a little cheaper to use a separate counter in the PDA (short of
688 * moving irq_enter into assembly, which would be too much work)
690 1: incl PER_CPU_VAR(irq_count)
691 cmovzq PER_CPU_VAR(irq_stack_ptr),%rsp
692 CFI_DEF_CFA_REGISTER rsi
694 /* Store previous stack value */
696 CFI_ESCAPE 0x0f /* DW_CFA_def_cfa_expression */, 6, \
697 0x77 /* DW_OP_breg7 */, 0, \
698 0x06 /* DW_OP_deref */, \
699 0x08 /* DW_OP_const1u */, SS+8-RBP, \
700 0x22 /* DW_OP_plus */
701 /* We entered an interrupt context - irqs are off: */
708 * The interrupt stubs push (~vector+0x80) onto the stack and
709 * then jump to common_interrupt.
711 .p2align CONFIG_X86_L1_CACHE_SHIFT
715 addq $-0x80,(%rsp) /* Adjust vector to [-256,-1] range */
717 /* 0(%rsp): old_rsp-ARGOFFSET */
719 DISABLE_INTERRUPTS(CLBR_NONE)
721 decl PER_CPU_VAR(irq_count)
723 /* Restore saved previous stack */
725 CFI_DEF_CFA rsi,SS+8-RBP /* reg/off reset after def_cfa_expr */
726 leaq ARGOFFSET-RBP(%rsi), %rsp
727 CFI_DEF_CFA_REGISTER rsp
728 CFI_ADJUST_CFA_OFFSET RBP-ARGOFFSET
731 GET_THREAD_INFO(%rcx)
732 testl $3,CS-ARGOFFSET(%rsp)
735 /* Interrupt came from user space */
737 * Has a correct top of stack, but a partial stack frame
738 * %rcx: thread info. Interrupts off.
740 retint_with_reschedule:
741 movl $_TIF_WORK_MASK,%edi
744 movl TI_flags(%rcx),%edx
749 retint_swapgs: /* return to user-space */
751 * The iretq could re-enable interrupts:
753 DISABLE_INTERRUPTS(CLBR_ANY)
757 * Try to use SYSRET instead of IRET if we're returning to
758 * a completely clean 64-bit userspace context.
760 movq (RCX-R11)(%rsp), %rcx
761 cmpq %rcx,(RIP-R11)(%rsp) /* RCX == RIP */
762 jne opportunistic_sysret_failed
765 * On Intel CPUs, sysret with non-canonical RCX/RIP will #GP
766 * in kernel space. This essentially lets the user take over
767 * the kernel, since userspace controls RSP. It's not worth
768 * testing for canonicalness exactly -- this check detects any
769 * of the 17 high bits set, which is true for non-canonical
770 * or kernel addresses. (This will pessimize vsyscall=native.
773 * If virtual addresses ever become wider, this will need
774 * to be updated to remain correct on both old and new CPUs.
776 .ifne __VIRTUAL_MASK_SHIFT - 47
777 .error "virtual address width changed -- sysret checks need update"
779 shr $__VIRTUAL_MASK_SHIFT, %rcx
780 jnz opportunistic_sysret_failed
782 cmpq $__USER_CS,(CS-R11)(%rsp) /* CS must match SYSRET */
783 jne opportunistic_sysret_failed
785 movq (R11-ARGOFFSET)(%rsp), %r11
786 cmpq %r11,(EFLAGS-ARGOFFSET)(%rsp) /* R11 == RFLAGS */
787 jne opportunistic_sysret_failed
789 testq $X86_EFLAGS_RF,%r11 /* sysret can't restore RF */
790 jnz opportunistic_sysret_failed
792 /* nothing to check for RSP */
794 cmpq $__USER_DS,(SS-ARGOFFSET)(%rsp) /* SS must match SYSRET */
795 jne opportunistic_sysret_failed
798 * We win! This label is here just for ease of understanding
799 * perf profiles. Nothing jumps here.
801 irq_return_via_sysret:
804 movq (RSP-RIP)(%rsp),%rsp
808 opportunistic_sysret_failed:
812 retint_restore_args: /* return to kernel space */
813 DISABLE_INTERRUPTS(CLBR_ANY)
815 * The iretq could re-enable interrupts:
826 * Are we returning to a stack segment from the LDT? Note: in
827 * 64-bit mode SS:RSP on the exception stack is always valid.
829 #ifdef CONFIG_X86_ESPFIX64
830 testb $4,(SS-RIP)(%rsp)
831 jnz native_irq_return_ldt
834 .global native_irq_return_iret
835 native_irq_return_iret:
837 * This may fault. Non-paranoid faults on return to userspace are
838 * handled by fixup_bad_iret. These include #SS, #GP, and #NP.
839 * Double-faults due to espfix64 are handled in do_double_fault.
840 * Other faults here are fatal.
844 #ifdef CONFIG_X86_ESPFIX64
845 native_irq_return_ldt:
849 movq PER_CPU_VAR(espfix_waddr),%rdi
850 movq %rax,(0*8)(%rdi) /* RAX */
851 movq (2*8)(%rsp),%rax /* RIP */
852 movq %rax,(1*8)(%rdi)
853 movq (3*8)(%rsp),%rax /* CS */
854 movq %rax,(2*8)(%rdi)
855 movq (4*8)(%rsp),%rax /* RFLAGS */
856 movq %rax,(3*8)(%rdi)
857 movq (6*8)(%rsp),%rax /* SS */
858 movq %rax,(5*8)(%rdi)
859 movq (5*8)(%rsp),%rax /* RSP */
860 movq %rax,(4*8)(%rdi)
861 andl $0xffff0000,%eax
863 orq PER_CPU_VAR(espfix_stack),%rax
867 jmp native_irq_return_iret
870 /* edi: workmask, edx: work */
873 bt $TIF_NEED_RESCHED,%edx
876 ENABLE_INTERRUPTS(CLBR_NONE)
880 GET_THREAD_INFO(%rcx)
881 DISABLE_INTERRUPTS(CLBR_NONE)
886 testl $_TIF_DO_NOTIFY_MASK,%edx
889 ENABLE_INTERRUPTS(CLBR_NONE)
891 movq $-1,ORIG_RAX(%rsp)
892 xorl %esi,%esi # oldset
893 movq %rsp,%rdi # &pt_regs
894 call do_notify_resume
896 DISABLE_INTERRUPTS(CLBR_NONE)
898 GET_THREAD_INFO(%rcx)
899 jmp retint_with_reschedule
901 #ifdef CONFIG_PREEMPT
902 /* Returning to kernel space. Check if we need preemption */
903 /* rcx: threadinfo. interrupts off. */
905 cmpl $0,PER_CPU_VAR(__preempt_count)
906 jnz retint_restore_args
907 bt $9,EFLAGS-ARGOFFSET(%rsp) /* interrupts off? */
908 jnc retint_restore_args
909 call preempt_schedule_irq
913 END(common_interrupt)
918 .macro apicinterrupt3 num sym do_sym
930 #ifdef CONFIG_TRACING
931 #define trace(sym) trace_##sym
932 #define smp_trace(sym) smp_trace_##sym
934 .macro trace_apicinterrupt num sym
935 apicinterrupt3 \num trace(\sym) smp_trace(\sym)
938 .macro trace_apicinterrupt num sym do_sym
942 .macro apicinterrupt num sym do_sym
943 apicinterrupt3 \num \sym \do_sym
944 trace_apicinterrupt \num \sym
948 apicinterrupt3 IRQ_MOVE_CLEANUP_VECTOR \
949 irq_move_cleanup_interrupt smp_irq_move_cleanup_interrupt
950 apicinterrupt3 REBOOT_VECTOR \
951 reboot_interrupt smp_reboot_interrupt
955 apicinterrupt3 UV_BAU_MESSAGE \
956 uv_bau_message_intr1 uv_bau_message_interrupt
958 apicinterrupt LOCAL_TIMER_VECTOR \
959 apic_timer_interrupt smp_apic_timer_interrupt
960 apicinterrupt X86_PLATFORM_IPI_VECTOR \
961 x86_platform_ipi smp_x86_platform_ipi
963 #ifdef CONFIG_HAVE_KVM
964 apicinterrupt3 POSTED_INTR_VECTOR \
965 kvm_posted_intr_ipi smp_kvm_posted_intr_ipi
968 #ifdef CONFIG_X86_MCE_THRESHOLD
969 apicinterrupt THRESHOLD_APIC_VECTOR \
970 threshold_interrupt smp_threshold_interrupt
973 #ifdef CONFIG_X86_THERMAL_VECTOR
974 apicinterrupt THERMAL_APIC_VECTOR \
975 thermal_interrupt smp_thermal_interrupt
979 apicinterrupt CALL_FUNCTION_SINGLE_VECTOR \
980 call_function_single_interrupt smp_call_function_single_interrupt
981 apicinterrupt CALL_FUNCTION_VECTOR \
982 call_function_interrupt smp_call_function_interrupt
983 apicinterrupt RESCHEDULE_VECTOR \
984 reschedule_interrupt smp_reschedule_interrupt
987 apicinterrupt ERROR_APIC_VECTOR \
988 error_interrupt smp_error_interrupt
989 apicinterrupt SPURIOUS_APIC_VECTOR \
990 spurious_interrupt smp_spurious_interrupt
992 #ifdef CONFIG_IRQ_WORK
993 apicinterrupt IRQ_WORK_VECTOR \
994 irq_work_interrupt smp_irq_work_interrupt
998 * Exception entry points.
1000 #define INIT_TSS_IST(x) PER_CPU_VAR(init_tss) + (TSS_ist + ((x) - 1) * 8)
1002 .macro idtentry sym do_sym has_error_code:req paranoid=0 shift_ist=-1
1005 .if \shift_ist != -1 && \paranoid == 0
1006 .error "using shift_ist requires paranoid=1"
1016 PARAVIRT_ADJUST_EXCEPTION_FRAME
1018 .ifeq \has_error_code
1019 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1022 subq $ORIG_RAX-R15, %rsp
1023 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1028 testl $3, CS(%rsp) /* If coming from userspace, switch */
1029 jnz 1f /* stacks. */
1039 .if \shift_ist != -1
1040 TRACE_IRQS_OFF_DEBUG /* reload IDT in case of recursion */
1046 movq %rsp,%rdi /* pt_regs pointer */
1049 movq ORIG_RAX(%rsp),%rsi /* get error code */
1050 movq $-1,ORIG_RAX(%rsp) /* no syscall to restart */
1052 xorl %esi,%esi /* no error code */
1055 .if \shift_ist != -1
1056 subq $EXCEPTION_STKSZ, INIT_TSS_IST(\shift_ist)
1061 .if \shift_ist != -1
1062 addq $EXCEPTION_STKSZ, INIT_TSS_IST(\shift_ist)
1066 jmp paranoid_exit /* %ebx: no swapgs flag */
1068 jmp error_exit /* %ebx: no swapgs flag */
1074 * Paranoid entry from userspace. Switch stacks and treat it
1075 * as a normal entry. This means that paranoid handlers
1076 * run in real process context if user_mode(regs).
1083 movq %rsp,%rdi /* pt_regs pointer */
1085 movq %rax,%rsp /* switch stack */
1087 movq %rsp,%rdi /* pt_regs pointer */
1090 movq ORIG_RAX(%rsp),%rsi /* get error code */
1091 movq $-1,ORIG_RAX(%rsp) /* no syscall to restart */
1093 xorl %esi,%esi /* no error code */
1098 jmp error_exit /* %ebx: no swapgs flag */
1105 #ifdef CONFIG_TRACING
1106 .macro trace_idtentry sym do_sym has_error_code:req
1107 idtentry trace(\sym) trace(\do_sym) has_error_code=\has_error_code
1108 idtentry \sym \do_sym has_error_code=\has_error_code
1111 .macro trace_idtentry sym do_sym has_error_code:req
1112 idtentry \sym \do_sym has_error_code=\has_error_code
1116 idtentry divide_error do_divide_error has_error_code=0
1117 idtentry overflow do_overflow has_error_code=0
1118 idtentry bounds do_bounds has_error_code=0
1119 idtentry invalid_op do_invalid_op has_error_code=0
1120 idtentry device_not_available do_device_not_available has_error_code=0
1121 idtentry double_fault do_double_fault has_error_code=1 paranoid=2
1122 idtentry coprocessor_segment_overrun do_coprocessor_segment_overrun has_error_code=0
1123 idtentry invalid_TSS do_invalid_TSS has_error_code=1
1124 idtentry segment_not_present do_segment_not_present has_error_code=1
1125 idtentry spurious_interrupt_bug do_spurious_interrupt_bug has_error_code=0
1126 idtentry coprocessor_error do_coprocessor_error has_error_code=0
1127 idtentry alignment_check do_alignment_check has_error_code=1
1128 idtentry simd_coprocessor_error do_simd_coprocessor_error has_error_code=0
1131 /* Reload gs selector with exception handling */
1132 /* edi: new selector */
1133 ENTRY(native_load_gs_index)
1136 DISABLE_INTERRUPTS(CLBR_ANY & ~CLBR_RDI)
1140 2: mfence /* workaround */
1145 END(native_load_gs_index)
1147 _ASM_EXTABLE(gs_change,bad_gs)
1148 .section .fixup,"ax"
1149 /* running with kernelgs */
1151 SWAPGS /* switch back to user gs */
1157 /* Call softirq on interrupt stack. Interrupts are off. */
1158 ENTRY(do_softirq_own_stack)
1161 CFI_REL_OFFSET rbp,0
1163 CFI_DEF_CFA_REGISTER rbp
1164 incl PER_CPU_VAR(irq_count)
1165 cmove PER_CPU_VAR(irq_stack_ptr),%rsp
1166 push %rbp # backlink for old unwinder
1170 CFI_DEF_CFA_REGISTER rsp
1171 CFI_ADJUST_CFA_OFFSET -8
1172 decl PER_CPU_VAR(irq_count)
1175 END(do_softirq_own_stack)
1178 idtentry xen_hypervisor_callback xen_do_hypervisor_callback has_error_code=0
1181 * A note on the "critical region" in our callback handler.
1182 * We want to avoid stacking callback handlers due to events occurring
1183 * during handling of the last event. To do this, we keep events disabled
1184 * until we've done all processing. HOWEVER, we must enable events before
1185 * popping the stack frame (can't be done atomically) and so it would still
1186 * be possible to get enough handler activations to overflow the stack.
1187 * Although unlikely, bugs of that kind are hard to track down, so we'd
1188 * like to avoid the possibility.
1189 * So, on entry to the handler we detect whether we interrupted an
1190 * existing activation in its critical region -- if so, we pop the current
1191 * activation and restart the handler using the previous one.
1193 ENTRY(xen_do_hypervisor_callback) # do_hypervisor_callback(struct *pt_regs)
1196 * Since we don't modify %rdi, evtchn_do_upall(struct *pt_regs) will
1197 * see the correct pointer to the pt_regs
1199 movq %rdi, %rsp # we don't return, adjust the stack frame
1202 11: incl PER_CPU_VAR(irq_count)
1204 CFI_DEF_CFA_REGISTER rbp
1205 cmovzq PER_CPU_VAR(irq_stack_ptr),%rsp
1206 pushq %rbp # backlink for old unwinder
1207 call xen_evtchn_do_upcall
1209 CFI_DEF_CFA_REGISTER rsp
1210 decl PER_CPU_VAR(irq_count)
1213 END(xen_do_hypervisor_callback)
1216 * Hypervisor uses this for application faults while it executes.
1217 * We get here for two reasons:
1218 * 1. Fault while reloading DS, ES, FS or GS
1219 * 2. Fault while executing IRET
1220 * Category 1 we do not need to fix up as Xen has already reloaded all segment
1221 * registers that could be reloaded and zeroed the others.
1222 * Category 2 we fix up by killing the current process. We cannot use the
1223 * normal Linux return path in this case because if we use the IRET hypercall
1224 * to pop the stack frame we end up in an infinite loop of failsafe callbacks.
1225 * We distinguish between categories by comparing each saved segment register
1226 * with its current contents: any discrepancy means we in category 1.
1228 ENTRY(xen_failsafe_callback)
1230 /*CFI_REL_OFFSET gs,GS*/
1231 /*CFI_REL_OFFSET fs,FS*/
1232 /*CFI_REL_OFFSET es,ES*/
1233 /*CFI_REL_OFFSET ds,DS*/
1234 CFI_REL_OFFSET r11,8
1235 CFI_REL_OFFSET rcx,0
1249 /* All segments match their saved values => Category 2 (Bad IRET). */
1255 CFI_ADJUST_CFA_OFFSET -0x30
1256 pushq_cfi $0 /* RIP */
1259 jmp general_protection
1261 1: /* Segment mismatch => Category 1 (Bad segment). Retry the IRET. */
1267 CFI_ADJUST_CFA_OFFSET -0x30
1268 pushq_cfi $-1 /* orig_ax = -1 => not a system call */
1272 END(xen_failsafe_callback)
1274 apicinterrupt3 HYPERVISOR_CALLBACK_VECTOR \
1275 xen_hvm_callback_vector xen_evtchn_do_upcall
1277 #endif /* CONFIG_XEN */
1279 #if IS_ENABLED(CONFIG_HYPERV)
1280 apicinterrupt3 HYPERVISOR_CALLBACK_VECTOR \
1281 hyperv_callback_vector hyperv_vector_handler
1282 #endif /* CONFIG_HYPERV */
1284 idtentry debug do_debug has_error_code=0 paranoid=1 shift_ist=DEBUG_STACK
1285 idtentry int3 do_int3 has_error_code=0 paranoid=1 shift_ist=DEBUG_STACK
1286 idtentry stack_segment do_stack_segment has_error_code=1
1288 idtentry xen_debug do_debug has_error_code=0
1289 idtentry xen_int3 do_int3 has_error_code=0
1290 idtentry xen_stack_segment do_stack_segment has_error_code=1
1292 idtentry general_protection do_general_protection has_error_code=1
1293 trace_idtentry page_fault do_page_fault has_error_code=1
1294 #ifdef CONFIG_KVM_GUEST
1295 idtentry async_page_fault do_async_page_fault has_error_code=1
1297 #ifdef CONFIG_X86_MCE
1298 idtentry machine_check has_error_code=0 paranoid=1 do_sym=*machine_check_vector(%rip)
1302 * "Paranoid" exit path from exception stack. This is invoked
1303 * only on return from non-NMI IST interrupts that came
1304 * from kernel space.
1306 * We may be returning to very strange contexts (e.g. very early
1307 * in syscall entry), so checking for preemption here would
1308 * be complicated. Fortunately, we there's no good reason
1309 * to try to handle preemption here.
1312 /* ebx: no swapgs flag */
1313 ENTRY(paranoid_exit)
1315 DISABLE_INTERRUPTS(CLBR_NONE)
1316 TRACE_IRQS_OFF_DEBUG
1317 testl %ebx,%ebx /* swapgs needed? */
1318 jnz paranoid_restore
1324 TRACE_IRQS_IRETQ_DEBUG 0
1331 * Exception entry point. This expects an error code/orig_rax on the stack.
1332 * returns in "no swapgs flag" in %ebx.
1336 CFI_ADJUST_CFA_OFFSET 15*8
1337 /* oldrax contains error code */
1339 movq %rdi, RDI+8(%rsp)
1340 movq %rsi, RSI+8(%rsp)
1341 movq %rdx, RDX+8(%rsp)
1342 movq %rcx, RCX+8(%rsp)
1343 movq %rax, RAX+8(%rsp)
1344 movq %r8, R8+8(%rsp)
1345 movq %r9, R9+8(%rsp)
1346 movq %r10, R10+8(%rsp)
1347 movq %r11, R11+8(%rsp)
1349 movq %rbp, RBP+8(%rsp)
1350 movq %r12, R12+8(%rsp)
1351 movq %r13, R13+8(%rsp)
1352 movq %r14, R14+8(%rsp)
1353 movq %r15, R15+8(%rsp)
1356 je error_kernelspace
1364 * There are two places in the kernel that can potentially fault with
1365 * usergs. Handle them here. B stepping K8s sometimes report a
1366 * truncated RIP for IRET exceptions returning to compat mode. Check
1367 * for these here too.
1370 CFI_REL_OFFSET rcx, RCX+8
1372 leaq native_irq_return_iret(%rip),%rcx
1373 cmpq %rcx,RIP+8(%rsp)
1375 movl %ecx,%eax /* zero extend */
1376 cmpq %rax,RIP+8(%rsp)
1378 cmpq $gs_change,RIP+8(%rsp)
1383 /* Fix truncated RIP */
1384 movq %rcx,RIP+8(%rsp)
1392 decl %ebx /* Return to usergs */
1398 /* ebx: no swapgs flag (1: don't need swapgs, 0: need it) */
1403 DISABLE_INTERRUPTS(CLBR_NONE)
1405 GET_THREAD_INFO(%rcx)
1408 LOCKDEP_SYS_EXIT_IRQ
1409 movl TI_flags(%rcx),%edx
1410 movl $_TIF_WORK_MASK,%edi
1418 * Test if a given stack is an NMI stack or not.
1420 .macro test_in_nmi reg stack nmi_ret normal_ret
1423 subq $EXCEPTION_STKSZ, %\reg
1429 /* runs on exception stack */
1432 PARAVIRT_ADJUST_EXCEPTION_FRAME
1434 * We allow breakpoints in NMIs. If a breakpoint occurs, then
1435 * the iretq it performs will take us out of NMI context.
1436 * This means that we can have nested NMIs where the next
1437 * NMI is using the top of the stack of the previous NMI. We
1438 * can't let it execute because the nested NMI will corrupt the
1439 * stack of the previous NMI. NMI handlers are not re-entrant
1442 * To handle this case we do the following:
1443 * Check the a special location on the stack that contains
1444 * a variable that is set when NMIs are executing.
1445 * The interrupted task's stack is also checked to see if it
1447 * If the variable is not set and the stack is not the NMI
1449 * o Set the special variable on the stack
1450 * o Copy the interrupt frame into a "saved" location on the stack
1451 * o Copy the interrupt frame into a "copy" location on the stack
1452 * o Continue processing the NMI
1453 * If the variable is set or the previous stack is the NMI stack:
1454 * o Modify the "copy" location to jump to the repeate_nmi
1455 * o return back to the first NMI
1457 * Now on exit of the first NMI, we first clear the stack variable
1458 * The NMI stack will tell any nested NMIs at that point that it is
1459 * nested. Then we pop the stack normally with iret, and if there was
1460 * a nested NMI that updated the copy interrupt stack frame, a
1461 * jump will be made to the repeat_nmi code that will handle the second
1465 /* Use %rdx as out temp variable throughout */
1467 CFI_REL_OFFSET rdx, 0
1470 * If %cs was not the kernel segment, then the NMI triggered in user
1471 * space, which means it is definitely not nested.
1473 cmpl $__KERNEL_CS, 16(%rsp)
1477 * Check the special variable on the stack to see if NMIs are
1484 * Now test if the previous stack was an NMI stack.
1485 * We need the double check. We check the NMI stack to satisfy the
1486 * race when the first NMI clears the variable before returning.
1487 * We check the variable because the first NMI could be in a
1488 * breakpoint routine using a breakpoint stack.
1491 test_in_nmi rdx, 4*8(%rsp), nested_nmi, first_nmi
1496 * Do nothing if we interrupted the fixup in repeat_nmi.
1497 * It's about to repeat the NMI handler, so we are fine
1498 * with ignoring this one.
1500 movq $repeat_nmi, %rdx
1503 movq $end_repeat_nmi, %rdx
1508 /* Set up the interrupted NMIs stack to jump to repeat_nmi */
1509 leaq -1*8(%rsp), %rdx
1511 CFI_ADJUST_CFA_OFFSET 1*8
1512 leaq -10*8(%rsp), %rdx
1513 pushq_cfi $__KERNEL_DS
1516 pushq_cfi $__KERNEL_CS
1517 pushq_cfi $repeat_nmi
1519 /* Put stack back */
1521 CFI_ADJUST_CFA_OFFSET -6*8
1527 /* No need to check faults here */
1533 * Because nested NMIs will use the pushed location that we
1534 * stored in rdx, we must keep that space available.
1535 * Here's what our stack frame will look like:
1536 * +-------------------------+
1538 * | original Return RSP |
1539 * | original RFLAGS |
1542 * +-------------------------+
1543 * | temp storage for rdx |
1544 * +-------------------------+
1545 * | NMI executing variable |
1546 * +-------------------------+
1548 * | copied Return RSP |
1552 * +-------------------------+
1554 * | Saved Return RSP |
1558 * +-------------------------+
1560 * +-------------------------+
1562 * The saved stack frame is used to fix up the copied stack frame
1563 * that a nested NMI may change to make the interrupted NMI iret jump
1564 * to the repeat_nmi. The original stack frame and the temp storage
1565 * is also used by nested NMIs and can not be trusted on exit.
1567 /* Do not pop rdx, nested NMIs will corrupt that part of the stack */
1571 /* Set the NMI executing variable on the stack. */
1575 * Leave room for the "copied" frame
1578 CFI_ADJUST_CFA_OFFSET 5*8
1580 /* Copy the stack frame to the Saved frame */
1582 pushq_cfi 11*8(%rsp)
1584 CFI_DEF_CFA_OFFSET SS+8-RIP
1586 /* Everything up to here is safe from nested NMIs */
1589 * If there was a nested NMI, the first NMI's iret will return
1590 * here. But NMIs are still enabled and we can take another
1591 * nested NMI. The nested NMI checks the interrupted RIP to see
1592 * if it is between repeat_nmi and end_repeat_nmi, and if so
1593 * it will just return, as we are about to repeat an NMI anyway.
1594 * This makes it safe to copy to the stack frame that a nested
1599 * Update the stack variable to say we are still in NMI (the update
1600 * is benign for the non-repeat case, where 1 was pushed just above
1601 * to this very stack slot).
1605 /* Make another copy, this one may be modified by nested NMIs */
1607 CFI_ADJUST_CFA_OFFSET -10*8
1609 pushq_cfi -6*8(%rsp)
1612 CFI_DEF_CFA_OFFSET SS+8-RIP
1616 * Everything below this point can be preempted by a nested
1617 * NMI if the first NMI took an exception and reset our iret stack
1618 * so that we repeat another NMI.
1620 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1621 subq $ORIG_RAX-R15, %rsp
1622 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1624 * Use save_paranoid to handle SWAPGS, but no need to use paranoid_exit
1625 * as we should not be calling schedule in NMI context.
1626 * Even with normal interrupts enabled. An NMI should not be
1627 * setting NEED_RESCHED or anything that normal interrupts and
1628 * exceptions might do.
1634 * Save off the CR2 register. If we take a page fault in the NMI then
1635 * it could corrupt the CR2 value. If the NMI preempts a page fault
1636 * handler before it was able to read the CR2 register, and then the
1637 * NMI itself takes a page fault, the page fault that was preempted
1638 * will read the information from the NMI page fault and not the
1639 * origin fault. Save it off and restore it if it changes.
1640 * Use the r12 callee-saved register.
1644 /* paranoidentry do_nmi, 0; without TRACE_IRQS_OFF */
1649 /* Did the NMI take a page fault? Restore cr2 if it did */
1656 testl %ebx,%ebx /* swapgs needed? */
1661 /* Pop the extra iret frame at once */
1664 /* Clear the NMI executing stack variable */
1670 ENTRY(ignore_sysret)