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>
8 * entry.S contains the system-call and fault low-level handling routines.
10 * Some of this is documented in Documentation/x86/entry_64.txt
12 * A note on terminology:
13 * - iret frame: Architecture defined interrupt frame from SS to RIP
14 * at the top of the kernel process stack.
17 * - ENTRY/END: Define functions in the symbol table.
18 * - TRACE_IRQ_*: Trace hardirq state for lock debugging.
19 * - idtentry: Define exception entry points.
21 #include <linux/linkage.h>
22 #include <asm/segment.h>
23 #include <asm/cache.h>
24 #include <asm/errno.h>
26 #include <asm/asm-offsets.h>
28 #include <asm/unistd.h>
29 #include <asm/thread_info.h>
30 #include <asm/hw_irq.h>
31 #include <asm/page_types.h>
32 #include <asm/irqflags.h>
33 #include <asm/paravirt.h>
34 #include <asm/percpu.h>
37 #include <asm/pgtable_types.h>
38 #include <linux/err.h>
40 /* Avoid __ASSEMBLER__'ifying <linux/audit.h> just for this. */
41 #include <linux/elf-em.h>
42 #define AUDIT_ARCH_X86_64 (EM_X86_64|__AUDIT_ARCH_64BIT|__AUDIT_ARCH_LE)
43 #define __AUDIT_ARCH_64BIT 0x80000000
44 #define __AUDIT_ARCH_LE 0x40000000
47 .section .entry.text, "ax"
49 #ifdef CONFIG_PARAVIRT
50 ENTRY(native_usergs_sysret64)
53 ENDPROC(native_usergs_sysret64)
54 #endif /* CONFIG_PARAVIRT */
56 .macro TRACE_IRQS_IRETQ
57 #ifdef CONFIG_TRACE_IRQFLAGS
58 bt $9, EFLAGS(%rsp) /* interrupts off? */
66 * When dynamic function tracer is enabled it will add a breakpoint
67 * to all locations that it is about to modify, sync CPUs, update
68 * all the code, sync CPUs, then remove the breakpoints. In this time
69 * if lockdep is enabled, it might jump back into the debug handler
70 * outside the updating of the IST protection. (TRACE_IRQS_ON/OFF).
72 * We need to change the IDT table before calling TRACE_IRQS_ON/OFF to
73 * make sure the stack pointer does not get reset back to the top
74 * of the debug stack, and instead just reuses the current stack.
76 #if defined(CONFIG_DYNAMIC_FTRACE) && defined(CONFIG_TRACE_IRQFLAGS)
78 .macro TRACE_IRQS_OFF_DEBUG
79 call debug_stack_set_zero
81 call debug_stack_reset
84 .macro TRACE_IRQS_ON_DEBUG
85 call debug_stack_set_zero
87 call debug_stack_reset
90 .macro TRACE_IRQS_IRETQ_DEBUG
91 bt $9, EFLAGS(%rsp) /* interrupts off? */
98 # define TRACE_IRQS_OFF_DEBUG TRACE_IRQS_OFF
99 # define TRACE_IRQS_ON_DEBUG TRACE_IRQS_ON
100 # define TRACE_IRQS_IRETQ_DEBUG TRACE_IRQS_IRETQ
104 * 64-bit SYSCALL instruction entry. Up to 6 arguments in registers.
106 * 64-bit SYSCALL saves rip to rcx, clears rflags.RF, then saves rflags to r11,
107 * then loads new ss, cs, and rip from previously programmed MSRs.
108 * rflags gets masked by a value from another MSR (so CLD and CLAC
109 * are not needed). SYSCALL does not save anything on the stack
110 * and does not change rsp.
112 * Registers on entry:
113 * rax system call number
115 * r11 saved rflags (note: r11 is callee-clobbered register in C ABI)
119 * r10 arg3 (needs to be moved to rcx to conform to C ABI)
122 * (note: r12-r15, rbp, rbx are callee-preserved in C ABI)
124 * Only called from user space.
126 * When user can change pt_regs->foo always force IRET. That is because
127 * it deals with uncanonical addresses better. SYSRET has trouble
128 * with them due to bugs in both AMD and Intel CPUs.
131 ENTRY(entry_SYSCALL_64)
133 * Interrupts are off on entry.
134 * We do not frame this tiny irq-off block with TRACE_IRQS_OFF/ON,
135 * it is too small to ever cause noticeable irq latency.
139 * A hypervisor implementation might want to use a label
140 * after the swapgs, so that it can do the swapgs
141 * for the guest and jump here on syscall.
143 GLOBAL(entry_SYSCALL_64_after_swapgs)
145 movq %rsp, PER_CPU_VAR(rsp_scratch)
146 movq PER_CPU_VAR(cpu_current_top_of_stack), %rsp
150 /* Construct struct pt_regs on stack */
151 pushq $__USER_DS /* pt_regs->ss */
152 pushq PER_CPU_VAR(rsp_scratch) /* pt_regs->sp */
153 pushq %r11 /* pt_regs->flags */
154 pushq $__USER_CS /* pt_regs->cs */
155 pushq %rcx /* pt_regs->ip */
156 pushq %rax /* pt_regs->orig_ax */
157 pushq %rdi /* pt_regs->di */
158 pushq %rsi /* pt_regs->si */
159 pushq %rdx /* pt_regs->dx */
160 pushq %rcx /* pt_regs->cx */
161 pushq $-ENOSYS /* pt_regs->ax */
162 pushq %r8 /* pt_regs->r8 */
163 pushq %r9 /* pt_regs->r9 */
164 pushq %r10 /* pt_regs->r10 */
165 pushq %r11 /* pt_regs->r11 */
166 sub $(6*8), %rsp /* pt_regs->bp, bx, r12-15 not saved */
169 * If we need to do entry work or if we guess we'll need to do
170 * exit work, go straight to the slow path.
172 testl $_TIF_WORK_SYSCALL_ENTRY|_TIF_ALLWORK_MASK, ASM_THREAD_INFO(TI_flags, %rsp, SIZEOF_PTREGS)
173 jnz entry_SYSCALL64_slow_path
175 entry_SYSCALL_64_fastpath:
177 * Easy case: enable interrupts and issue the syscall. If the syscall
178 * needs pt_regs, we'll call a stub that disables interrupts again
179 * and jumps to the slow path.
182 ENABLE_INTERRUPTS(CLBR_NONE)
183 #if __SYSCALL_MASK == ~0
184 cmpq $__NR_syscall_max, %rax
186 andl $__SYSCALL_MASK, %eax
187 cmpl $__NR_syscall_max, %eax
189 ja 1f /* return -ENOSYS (already in pt_regs->ax) */
193 * This call instruction is handled specially in stub_ptregs_64.
194 * It might end up jumping to the slow path. If it jumps, RAX
195 * and all argument registers are clobbered.
197 call *sys_call_table(, %rax, 8)
198 .Lentry_SYSCALL_64_after_fastpath_call:
204 * If we get here, then we know that pt_regs is clean for SYSRET64.
205 * If we see that no exit work is required (which we are required
206 * to check with IRQs off), then we can go straight to SYSRET64.
208 DISABLE_INTERRUPTS(CLBR_NONE)
210 testl $_TIF_ALLWORK_MASK, ASM_THREAD_INFO(TI_flags, %rsp, SIZEOF_PTREGS)
214 TRACE_IRQS_ON /* user mode is traced as IRQs on */
216 movq EFLAGS(%rsp), %r11
217 RESTORE_C_REGS_EXCEPT_RCX_R11
223 * The fast path looked good when we started, but something changed
224 * along the way and we need to switch to the slow path. Calling
225 * raise(3) will trigger this, for example. IRQs are off.
228 ENABLE_INTERRUPTS(CLBR_NONE)
231 call syscall_return_slowpath /* returns with IRQs disabled */
232 jmp return_from_SYSCALL_64
234 entry_SYSCALL64_slow_path:
238 call do_syscall_64 /* returns with IRQs disabled */
240 return_from_SYSCALL_64:
242 TRACE_IRQS_IRETQ /* we're about to change IF */
245 * Try to use SYSRET instead of IRET if we're returning to
246 * a completely clean 64-bit userspace context.
250 cmpq %rcx, %r11 /* RCX == RIP */
251 jne opportunistic_sysret_failed
254 * On Intel CPUs, SYSRET with non-canonical RCX/RIP will #GP
255 * in kernel space. This essentially lets the user take over
256 * the kernel, since userspace controls RSP.
258 * If width of "canonical tail" ever becomes variable, this will need
259 * to be updated to remain correct on both old and new CPUs.
261 .ifne __VIRTUAL_MASK_SHIFT - 47
262 .error "virtual address width changed -- SYSRET checks need update"
265 /* Change top 16 bits to be the sign-extension of 47th bit */
266 shl $(64 - (__VIRTUAL_MASK_SHIFT+1)), %rcx
267 sar $(64 - (__VIRTUAL_MASK_SHIFT+1)), %rcx
269 /* If this changed %rcx, it was not canonical */
271 jne opportunistic_sysret_failed
273 cmpq $__USER_CS, CS(%rsp) /* CS must match SYSRET */
274 jne opportunistic_sysret_failed
277 cmpq %r11, EFLAGS(%rsp) /* R11 == RFLAGS */
278 jne opportunistic_sysret_failed
281 * SYSRET can't restore RF. SYSRET can restore TF, but unlike IRET,
282 * restoring TF results in a trap from userspace immediately after
283 * SYSRET. This would cause an infinite loop whenever #DB happens
284 * with register state that satisfies the opportunistic SYSRET
285 * conditions. For example, single-stepping this user code:
287 * movq $stuck_here, %rcx
292 * would never get past 'stuck_here'.
294 testq $(X86_EFLAGS_RF|X86_EFLAGS_TF), %r11
295 jnz opportunistic_sysret_failed
297 /* nothing to check for RSP */
299 cmpq $__USER_DS, SS(%rsp) /* SS must match SYSRET */
300 jne opportunistic_sysret_failed
303 * We win! This label is here just for ease of understanding
304 * perf profiles. Nothing jumps here.
306 syscall_return_via_sysret:
307 /* rcx and r11 are already restored (see code above) */
308 RESTORE_C_REGS_EXCEPT_RCX_R11
312 opportunistic_sysret_failed:
314 jmp restore_c_regs_and_iret
315 END(entry_SYSCALL_64)
317 ENTRY(stub_ptregs_64)
319 * Syscalls marked as needing ptregs land here.
320 * If we are on the fast path, we need to save the extra regs,
321 * which we achieve by trying again on the slow path. If we are on
322 * the slow path, the extra regs are already saved.
324 * RAX stores a pointer to the C function implementing the syscall.
327 cmpq $.Lentry_SYSCALL_64_after_fastpath_call, (%rsp)
331 * Called from fast path -- disable IRQs again, pop return address
332 * and jump to slow path
334 DISABLE_INTERRUPTS(CLBR_NONE)
337 jmp entry_SYSCALL64_slow_path
341 jmp *%rax /* called from C */
344 .macro ptregs_stub func
346 leaq \func(%rip), %rax
351 /* Instantiate ptregs_stub for each ptregs-using syscall */
352 #define __SYSCALL_64_QUAL_(sym)
353 #define __SYSCALL_64_QUAL_ptregs(sym) ptregs_stub sym
354 #define __SYSCALL_64(nr, sym, qual) __SYSCALL_64_QUAL_##qual(sym)
355 #include <asm/syscalls_64.h>
358 * A newly forked process directly context switches into this address.
360 * rdi: prev task we switched from
363 LOCK ; btr $TIF_FORK, TI_flags(%r8)
366 popfq /* reset kernel eflags */
368 call schedule_tail /* rdi: 'prev' task parameter */
370 testb $3, CS(%rsp) /* from kernel_thread? */
374 * We came from kernel_thread. This code path is quite twisted, and
375 * someone should clean it up.
377 * copy_thread_tls stashes the function pointer in RBX and the
378 * parameter to be passed in RBP. The called function is permitted
379 * to call do_execve and thereby jump to user mode.
386 * Fall through as though we're exiting a syscall. This makes a
387 * twisted sort of sense if we just called do_execve.
392 call syscall_return_slowpath /* returns with IRQs disabled */
393 TRACE_IRQS_ON /* user mode is traced as IRQS on */
395 jmp restore_regs_and_iret
399 * Build the entry stubs with some assembler magic.
400 * We pack 1 stub into every 8-byte block.
403 ENTRY(irq_entries_start)
404 vector=FIRST_EXTERNAL_VECTOR
405 .rept (FIRST_SYSTEM_VECTOR - FIRST_EXTERNAL_VECTOR)
406 pushq $(~vector+0x80) /* Note: always in signed byte range */
411 END(irq_entries_start)
414 * Interrupt entry/exit.
416 * Interrupt entry points save only callee clobbered registers in fast path.
418 * Entry runs with interrupts off.
421 /* 0(%rsp): ~(interrupt number) */
422 .macro interrupt func
424 ALLOC_PT_GPREGS_ON_STACK
432 * IRQ from user mode. Switch to kernel gsbase and inform context
433 * tracking that we're in kernel mode.
438 * We need to tell lockdep that IRQs are off. We can't do this until
439 * we fix gsbase, and we should do it before enter_from_user_mode
440 * (which can take locks). Since TRACE_IRQS_OFF idempotent,
441 * the simplest way to handle it is to just call it twice if
442 * we enter from user mode. There's no reason to optimize this since
443 * TRACE_IRQS_OFF is a no-op if lockdep is off.
447 CALL_enter_from_user_mode
451 * Save previous stack pointer, optionally switch to interrupt stack.
452 * irq_count is used to check if a CPU is already on an interrupt stack
453 * or not. While this is essentially redundant with preempt_count it is
454 * a little cheaper to use a separate counter in the PDA (short of
455 * moving irq_enter into assembly, which would be too much work)
458 incl PER_CPU_VAR(irq_count)
459 cmovzq PER_CPU_VAR(irq_stack_ptr), %rsp
461 /* We entered an interrupt context - irqs are off: */
464 call \func /* rdi points to pt_regs */
468 * The interrupt stubs push (~vector+0x80) onto the stack and
469 * then jump to common_interrupt.
471 .p2align CONFIG_X86_L1_CACHE_SHIFT
474 addq $-0x80, (%rsp) /* Adjust vector to [-256, -1] range */
476 /* 0(%rsp): old RSP */
478 DISABLE_INTERRUPTS(CLBR_NONE)
480 decl PER_CPU_VAR(irq_count)
482 /* Restore saved previous stack */
488 /* Interrupt came from user space */
491 call prepare_exit_to_usermode
494 jmp restore_regs_and_iret
496 /* Returning to kernel space */
498 #ifdef CONFIG_PREEMPT
499 /* Interrupts are off */
500 /* Check if we need preemption */
501 bt $9, EFLAGS(%rsp) /* were interrupts off? */
503 0: cmpl $0, PER_CPU_VAR(__preempt_count)
505 call preempt_schedule_irq
510 * The iretq could re-enable interrupts:
515 * At this label, code paths which return to kernel and to user,
516 * which come from interrupts/exception and from syscalls, merge.
518 GLOBAL(restore_regs_and_iret)
520 restore_c_regs_and_iret:
522 REMOVE_PT_GPREGS_FROM_STACK 8
527 * Are we returning to a stack segment from the LDT? Note: in
528 * 64-bit mode SS:RSP on the exception stack is always valid.
530 #ifdef CONFIG_X86_ESPFIX64
531 testb $4, (SS-RIP)(%rsp)
532 jnz native_irq_return_ldt
535 .global native_irq_return_iret
536 native_irq_return_iret:
538 * This may fault. Non-paranoid faults on return to userspace are
539 * handled by fixup_bad_iret. These include #SS, #GP, and #NP.
540 * Double-faults due to espfix64 are handled in do_double_fault.
541 * Other faults here are fatal.
545 #ifdef CONFIG_X86_ESPFIX64
546 native_irq_return_ldt:
550 movq PER_CPU_VAR(espfix_waddr), %rdi
551 movq %rax, (0*8)(%rdi) /* RAX */
552 movq (2*8)(%rsp), %rax /* RIP */
553 movq %rax, (1*8)(%rdi)
554 movq (3*8)(%rsp), %rax /* CS */
555 movq %rax, (2*8)(%rdi)
556 movq (4*8)(%rsp), %rax /* RFLAGS */
557 movq %rax, (3*8)(%rdi)
558 movq (6*8)(%rsp), %rax /* SS */
559 movq %rax, (5*8)(%rdi)
560 movq (5*8)(%rsp), %rax /* RSP */
561 movq %rax, (4*8)(%rdi)
562 andl $0xffff0000, %eax
564 orq PER_CPU_VAR(espfix_stack), %rax
568 jmp native_irq_return_iret
570 END(common_interrupt)
575 .macro apicinterrupt3 num sym do_sym
585 #ifdef CONFIG_TRACING
586 #define trace(sym) trace_##sym
587 #define smp_trace(sym) smp_trace_##sym
589 .macro trace_apicinterrupt num sym
590 apicinterrupt3 \num trace(\sym) smp_trace(\sym)
593 .macro trace_apicinterrupt num sym do_sym
597 .macro apicinterrupt num sym do_sym
598 apicinterrupt3 \num \sym \do_sym
599 trace_apicinterrupt \num \sym
603 apicinterrupt3 IRQ_MOVE_CLEANUP_VECTOR irq_move_cleanup_interrupt smp_irq_move_cleanup_interrupt
604 apicinterrupt3 REBOOT_VECTOR reboot_interrupt smp_reboot_interrupt
608 apicinterrupt3 UV_BAU_MESSAGE uv_bau_message_intr1 uv_bau_message_interrupt
611 apicinterrupt LOCAL_TIMER_VECTOR apic_timer_interrupt smp_apic_timer_interrupt
612 apicinterrupt X86_PLATFORM_IPI_VECTOR x86_platform_ipi smp_x86_platform_ipi
614 #ifdef CONFIG_HAVE_KVM
615 apicinterrupt3 POSTED_INTR_VECTOR kvm_posted_intr_ipi smp_kvm_posted_intr_ipi
616 apicinterrupt3 POSTED_INTR_WAKEUP_VECTOR kvm_posted_intr_wakeup_ipi smp_kvm_posted_intr_wakeup_ipi
619 #ifdef CONFIG_X86_MCE_THRESHOLD
620 apicinterrupt THRESHOLD_APIC_VECTOR threshold_interrupt smp_threshold_interrupt
623 #ifdef CONFIG_X86_MCE_AMD
624 apicinterrupt DEFERRED_ERROR_VECTOR deferred_error_interrupt smp_deferred_error_interrupt
627 #ifdef CONFIG_X86_THERMAL_VECTOR
628 apicinterrupt THERMAL_APIC_VECTOR thermal_interrupt smp_thermal_interrupt
632 apicinterrupt CALL_FUNCTION_SINGLE_VECTOR call_function_single_interrupt smp_call_function_single_interrupt
633 apicinterrupt CALL_FUNCTION_VECTOR call_function_interrupt smp_call_function_interrupt
634 apicinterrupt RESCHEDULE_VECTOR reschedule_interrupt smp_reschedule_interrupt
637 apicinterrupt ERROR_APIC_VECTOR error_interrupt smp_error_interrupt
638 apicinterrupt SPURIOUS_APIC_VECTOR spurious_interrupt smp_spurious_interrupt
640 #ifdef CONFIG_IRQ_WORK
641 apicinterrupt IRQ_WORK_VECTOR irq_work_interrupt smp_irq_work_interrupt
645 * Exception entry points.
647 #define CPU_TSS_IST(x) PER_CPU_VAR(cpu_tss) + (TSS_ist + ((x) - 1) * 8)
649 .macro idtentry sym do_sym has_error_code:req paranoid=0 shift_ist=-1
652 .if \shift_ist != -1 && \paranoid == 0
653 .error "using shift_ist requires paranoid=1"
657 PARAVIRT_ADJUST_EXCEPTION_FRAME
659 .ifeq \has_error_code
660 pushq $-1 /* ORIG_RAX: no syscall to restart */
663 ALLOC_PT_GPREGS_ON_STACK
667 testb $3, CS(%rsp) /* If coming from userspace, switch stacks */
674 /* returned flag: ebx=0: need swapgs on exit, ebx=1: don't need it */
678 TRACE_IRQS_OFF_DEBUG /* reload IDT in case of recursion */
684 movq %rsp, %rdi /* pt_regs pointer */
687 movq ORIG_RAX(%rsp), %rsi /* get error code */
688 movq $-1, ORIG_RAX(%rsp) /* no syscall to restart */
690 xorl %esi, %esi /* no error code */
694 subq $EXCEPTION_STKSZ, CPU_TSS_IST(\shift_ist)
700 addq $EXCEPTION_STKSZ, CPU_TSS_IST(\shift_ist)
703 /* these procedures expect "no swapgs" flag in ebx */
712 * Paranoid entry from userspace. Switch stacks and treat it
713 * as a normal entry. This means that paranoid handlers
714 * run in real process context if user_mode(regs).
720 movq %rsp, %rdi /* pt_regs pointer */
722 movq %rax, %rsp /* switch stack */
724 movq %rsp, %rdi /* pt_regs pointer */
727 movq ORIG_RAX(%rsp), %rsi /* get error code */
728 movq $-1, ORIG_RAX(%rsp) /* no syscall to restart */
730 xorl %esi, %esi /* no error code */
735 jmp error_exit /* %ebx: no swapgs flag */
740 #ifdef CONFIG_TRACING
741 .macro trace_idtentry sym do_sym has_error_code:req
742 idtentry trace(\sym) trace(\do_sym) has_error_code=\has_error_code
743 idtentry \sym \do_sym has_error_code=\has_error_code
746 .macro trace_idtentry sym do_sym has_error_code:req
747 idtentry \sym \do_sym has_error_code=\has_error_code
751 idtentry divide_error do_divide_error has_error_code=0
752 idtentry overflow do_overflow has_error_code=0
753 idtentry bounds do_bounds has_error_code=0
754 idtentry invalid_op do_invalid_op has_error_code=0
755 idtentry device_not_available do_device_not_available has_error_code=0
756 idtentry double_fault do_double_fault has_error_code=1 paranoid=2
757 idtentry coprocessor_segment_overrun do_coprocessor_segment_overrun has_error_code=0
758 idtentry invalid_TSS do_invalid_TSS has_error_code=1
759 idtentry segment_not_present do_segment_not_present has_error_code=1
760 idtentry spurious_interrupt_bug do_spurious_interrupt_bug has_error_code=0
761 idtentry coprocessor_error do_coprocessor_error has_error_code=0
762 idtentry alignment_check do_alignment_check has_error_code=1
763 idtentry simd_coprocessor_error do_simd_coprocessor_error has_error_code=0
767 * Reload gs selector with exception handling
770 ENTRY(native_load_gs_index)
772 DISABLE_INTERRUPTS(CLBR_ANY & ~CLBR_RDI)
776 2: mfence /* workaround */
780 END(native_load_gs_index)
782 _ASM_EXTABLE(gs_change, bad_gs)
783 .section .fixup, "ax"
784 /* running with kernelgs */
786 SWAPGS /* switch back to user gs */
792 /* Call softirq on interrupt stack. Interrupts are off. */
793 ENTRY(do_softirq_own_stack)
796 incl PER_CPU_VAR(irq_count)
797 cmove PER_CPU_VAR(irq_stack_ptr), %rsp
798 push %rbp /* frame pointer backlink */
801 decl PER_CPU_VAR(irq_count)
803 END(do_softirq_own_stack)
806 idtentry xen_hypervisor_callback xen_do_hypervisor_callback has_error_code=0
809 * A note on the "critical region" in our callback handler.
810 * We want to avoid stacking callback handlers due to events occurring
811 * during handling of the last event. To do this, we keep events disabled
812 * until we've done all processing. HOWEVER, we must enable events before
813 * popping the stack frame (can't be done atomically) and so it would still
814 * be possible to get enough handler activations to overflow the stack.
815 * Although unlikely, bugs of that kind are hard to track down, so we'd
816 * like to avoid the possibility.
817 * So, on entry to the handler we detect whether we interrupted an
818 * existing activation in its critical region -- if so, we pop the current
819 * activation and restart the handler using the previous one.
821 ENTRY(xen_do_hypervisor_callback) /* do_hypervisor_callback(struct *pt_regs) */
824 * Since we don't modify %rdi, evtchn_do_upall(struct *pt_regs) will
825 * see the correct pointer to the pt_regs
827 movq %rdi, %rsp /* we don't return, adjust the stack frame */
828 11: incl PER_CPU_VAR(irq_count)
830 cmovzq PER_CPU_VAR(irq_stack_ptr), %rsp
831 pushq %rbp /* frame pointer backlink */
832 call xen_evtchn_do_upcall
834 decl PER_CPU_VAR(irq_count)
835 #ifndef CONFIG_PREEMPT
836 call xen_maybe_preempt_hcall
839 END(xen_do_hypervisor_callback)
842 * Hypervisor uses this for application faults while it executes.
843 * We get here for two reasons:
844 * 1. Fault while reloading DS, ES, FS or GS
845 * 2. Fault while executing IRET
846 * Category 1 we do not need to fix up as Xen has already reloaded all segment
847 * registers that could be reloaded and zeroed the others.
848 * Category 2 we fix up by killing the current process. We cannot use the
849 * normal Linux return path in this case because if we use the IRET hypercall
850 * to pop the stack frame we end up in an infinite loop of failsafe callbacks.
851 * We distinguish between categories by comparing each saved segment register
852 * with its current contents: any discrepancy means we in category 1.
854 ENTRY(xen_failsafe_callback)
867 /* All segments match their saved values => Category 2 (Bad IRET). */
874 jmp general_protection
875 1: /* Segment mismatch => Category 1 (Bad segment). Retry the IRET. */
879 pushq $-1 /* orig_ax = -1 => not a system call */
880 ALLOC_PT_GPREGS_ON_STACK
884 END(xen_failsafe_callback)
886 apicinterrupt3 HYPERVISOR_CALLBACK_VECTOR \
887 xen_hvm_callback_vector xen_evtchn_do_upcall
889 #endif /* CONFIG_XEN */
891 #if IS_ENABLED(CONFIG_HYPERV)
892 apicinterrupt3 HYPERVISOR_CALLBACK_VECTOR \
893 hyperv_callback_vector hyperv_vector_handler
894 #endif /* CONFIG_HYPERV */
896 idtentry debug do_debug has_error_code=0 paranoid=1 shift_ist=DEBUG_STACK
897 idtentry int3 do_int3 has_error_code=0 paranoid=1 shift_ist=DEBUG_STACK
898 idtentry stack_segment do_stack_segment has_error_code=1
901 idtentry xen_debug do_debug has_error_code=0
902 idtentry xen_int3 do_int3 has_error_code=0
903 idtentry xen_stack_segment do_stack_segment has_error_code=1
906 idtentry general_protection do_general_protection has_error_code=1
907 trace_idtentry page_fault do_page_fault has_error_code=1
909 #ifdef CONFIG_KVM_GUEST
910 idtentry async_page_fault do_async_page_fault has_error_code=1
913 #ifdef CONFIG_X86_MCE
914 idtentry machine_check has_error_code=0 paranoid=1 do_sym=*machine_check_vector(%rip)
918 * Save all registers in pt_regs, and switch gs if needed.
919 * Use slow, but surefire "are we in kernel?" check.
920 * Return: ebx=0: need swapgs on exit, ebx=1: otherwise
922 ENTRY(paranoid_entry)
927 movl $MSR_GS_BASE, %ecx
930 js 1f /* negative -> in kernel */
937 * "Paranoid" exit path from exception stack. This is invoked
938 * only on return from non-NMI IST interrupts that came
941 * We may be returning to very strange contexts (e.g. very early
942 * in syscall entry), so checking for preemption here would
943 * be complicated. Fortunately, we there's no good reason
944 * to try to handle preemption here.
946 * On entry, ebx is "no swapgs" flag (1: don't need swapgs, 0: need it)
949 DISABLE_INTERRUPTS(CLBR_NONE)
951 testl %ebx, %ebx /* swapgs needed? */
952 jnz paranoid_exit_no_swapgs
955 jmp paranoid_exit_restore
956 paranoid_exit_no_swapgs:
957 TRACE_IRQS_IRETQ_DEBUG
958 paranoid_exit_restore:
961 REMOVE_PT_GPREGS_FROM_STACK 8
966 * Save all registers in pt_regs, and switch gs if needed.
967 * Return: EBX=0: came from user mode; EBX=1: otherwise
975 jz .Lerror_kernelspace
977 .Lerror_entry_from_usermode_swapgs:
979 * We entered from user mode or we're pretending to have entered
980 * from user mode due to an IRET fault.
984 .Lerror_entry_from_usermode_after_swapgs:
986 * We need to tell lockdep that IRQs are off. We can't do this until
987 * we fix gsbase, and we should do it before enter_from_user_mode
988 * (which can take locks).
991 CALL_enter_from_user_mode
999 * There are two places in the kernel that can potentially fault with
1000 * usergs. Handle them here. B stepping K8s sometimes report a
1001 * truncated RIP for IRET exceptions returning to compat mode. Check
1002 * for these here too.
1004 .Lerror_kernelspace:
1006 leaq native_irq_return_iret(%rip), %rcx
1007 cmpq %rcx, RIP+8(%rsp)
1009 movl %ecx, %eax /* zero extend */
1010 cmpq %rax, RIP+8(%rsp)
1012 cmpq $gs_change, RIP+8(%rsp)
1013 jne .Lerror_entry_done
1016 * hack: gs_change can fail with user gsbase. If this happens, fix up
1017 * gsbase and proceed. We'll fix up the exception and land in
1018 * gs_change's error handler with kernel gsbase.
1020 jmp .Lerror_entry_from_usermode_swapgs
1023 /* Fix truncated RIP */
1024 movq %rcx, RIP+8(%rsp)
1029 * We came from an IRET to user mode, so we have user gsbase.
1030 * Switch to kernel gsbase:
1035 * Pretend that the exception came from user mode: set up pt_regs
1036 * as if we faulted immediately after IRET and clear EBX so that
1037 * error_exit knows that we will be returning to user mode.
1043 jmp .Lerror_entry_from_usermode_after_swapgs
1048 * On entry, EBS is a "return to kernel mode" flag:
1049 * 1: already in kernel mode, don't need SWAPGS
1050 * 0: user gsbase is loaded, we need SWAPGS and standard preparation for return to usermode
1054 DISABLE_INTERRUPTS(CLBR_NONE)
1061 /* Runs on exception stack */
1064 * Fix up the exception frame if we're on Xen.
1065 * PARAVIRT_ADJUST_EXCEPTION_FRAME is guaranteed to push at most
1066 * one value to the stack on native, so it may clobber the rdx
1067 * scratch slot, but it won't clobber any of the important
1070 * Xen is a different story, because the Xen frame itself overlaps
1071 * the "NMI executing" variable.
1073 PARAVIRT_ADJUST_EXCEPTION_FRAME
1076 * We allow breakpoints in NMIs. If a breakpoint occurs, then
1077 * the iretq it performs will take us out of NMI context.
1078 * This means that we can have nested NMIs where the next
1079 * NMI is using the top of the stack of the previous NMI. We
1080 * can't let it execute because the nested NMI will corrupt the
1081 * stack of the previous NMI. NMI handlers are not re-entrant
1084 * To handle this case we do the following:
1085 * Check the a special location on the stack that contains
1086 * a variable that is set when NMIs are executing.
1087 * The interrupted task's stack is also checked to see if it
1089 * If the variable is not set and the stack is not the NMI
1091 * o Set the special variable on the stack
1092 * o Copy the interrupt frame into an "outermost" location on the
1094 * o Copy the interrupt frame into an "iret" location on the stack
1095 * o Continue processing the NMI
1096 * If the variable is set or the previous stack is the NMI stack:
1097 * o Modify the "iret" location to jump to the repeat_nmi
1098 * o return back to the first NMI
1100 * Now on exit of the first NMI, we first clear the stack variable
1101 * The NMI stack will tell any nested NMIs at that point that it is
1102 * nested. Then we pop the stack normally with iret, and if there was
1103 * a nested NMI that updated the copy interrupt stack frame, a
1104 * jump will be made to the repeat_nmi code that will handle the second
1107 * However, espfix prevents us from directly returning to userspace
1108 * with a single IRET instruction. Similarly, IRET to user mode
1109 * can fault. We therefore handle NMIs from user space like
1110 * other IST entries.
1113 /* Use %rdx as our temp variable throughout */
1116 testb $3, CS-RIP+8(%rsp)
1117 jz .Lnmi_from_kernel
1120 * NMI from user mode. We need to run on the thread stack, but we
1121 * can't go through the normal entry paths: NMIs are masked, and
1122 * we don't want to enable interrupts, because then we'll end
1123 * up in an awkward situation in which IRQs are on but NMIs
1126 * We also must not push anything to the stack before switching
1127 * stacks lest we corrupt the "NMI executing" variable.
1133 movq PER_CPU_VAR(cpu_current_top_of_stack), %rsp
1134 pushq 5*8(%rdx) /* pt_regs->ss */
1135 pushq 4*8(%rdx) /* pt_regs->rsp */
1136 pushq 3*8(%rdx) /* pt_regs->flags */
1137 pushq 2*8(%rdx) /* pt_regs->cs */
1138 pushq 1*8(%rdx) /* pt_regs->rip */
1139 pushq $-1 /* pt_regs->orig_ax */
1140 pushq %rdi /* pt_regs->di */
1141 pushq %rsi /* pt_regs->si */
1142 pushq (%rdx) /* pt_regs->dx */
1143 pushq %rcx /* pt_regs->cx */
1144 pushq %rax /* pt_regs->ax */
1145 pushq %r8 /* pt_regs->r8 */
1146 pushq %r9 /* pt_regs->r9 */
1147 pushq %r10 /* pt_regs->r10 */
1148 pushq %r11 /* pt_regs->r11 */
1149 pushq %rbx /* pt_regs->rbx */
1150 pushq %rbp /* pt_regs->rbp */
1151 pushq %r12 /* pt_regs->r12 */
1152 pushq %r13 /* pt_regs->r13 */
1153 pushq %r14 /* pt_regs->r14 */
1154 pushq %r15 /* pt_regs->r15 */
1157 * At this point we no longer need to worry about stack damage
1158 * due to nesting -- we're on the normal thread stack and we're
1159 * done with the NMI stack.
1167 * Return back to user mode. We must *not* do the normal exit
1168 * work, because we don't want to enable interrupts. Fortunately,
1169 * do_nmi doesn't modify pt_regs.
1172 jmp restore_c_regs_and_iret
1176 * Here's what our stack frame will look like:
1177 * +---------------------------------------------------------+
1179 * | original Return RSP |
1180 * | original RFLAGS |
1183 * +---------------------------------------------------------+
1184 * | temp storage for rdx |
1185 * +---------------------------------------------------------+
1186 * | "NMI executing" variable |
1187 * +---------------------------------------------------------+
1188 * | iret SS } Copied from "outermost" frame |
1189 * | iret Return RSP } on each loop iteration; overwritten |
1190 * | iret RFLAGS } by a nested NMI to force another |
1191 * | iret CS } iteration if needed. |
1193 * +---------------------------------------------------------+
1194 * | outermost SS } initialized in first_nmi; |
1195 * | outermost Return RSP } will not be changed before |
1196 * | outermost RFLAGS } NMI processing is done. |
1197 * | outermost CS } Copied to "iret" frame on each |
1198 * | outermost RIP } iteration. |
1199 * +---------------------------------------------------------+
1201 * +---------------------------------------------------------+
1203 * The "original" frame is used by hardware. Before re-enabling
1204 * NMIs, we need to be done with it, and we need to leave enough
1205 * space for the asm code here.
1207 * We return by executing IRET while RSP points to the "iret" frame.
1208 * That will either return for real or it will loop back into NMI
1211 * The "outermost" frame is copied to the "iret" frame on each
1212 * iteration of the loop, so each iteration starts with the "iret"
1213 * frame pointing to the final return target.
1217 * Determine whether we're a nested NMI.
1219 * If we interrupted kernel code between repeat_nmi and
1220 * end_repeat_nmi, then we are a nested NMI. We must not
1221 * modify the "iret" frame because it's being written by
1222 * the outer NMI. That's okay; the outer NMI handler is
1223 * about to about to call do_nmi anyway, so we can just
1224 * resume the outer NMI.
1227 movq $repeat_nmi, %rdx
1230 movq $end_repeat_nmi, %rdx
1236 * Now check "NMI executing". If it's set, then we're nested.
1237 * This will not detect if we interrupted an outer NMI just
1244 * Now test if the previous stack was an NMI stack. This covers
1245 * the case where we interrupt an outer NMI after it clears
1246 * "NMI executing" but before IRET. We need to be careful, though:
1247 * there is one case in which RSP could point to the NMI stack
1248 * despite there being no NMI active: naughty userspace controls
1249 * RSP at the very beginning of the SYSCALL targets. We can
1250 * pull a fast one on naughty userspace, though: we program
1251 * SYSCALL to mask DF, so userspace cannot cause DF to be set
1252 * if it controls the kernel's RSP. We set DF before we clear
1256 /* Compare the NMI stack (rdx) with the stack we came from (4*8(%rsp)) */
1257 cmpq %rdx, 4*8(%rsp)
1258 /* If the stack pointer is above the NMI stack, this is a normal NMI */
1261 subq $EXCEPTION_STKSZ, %rdx
1262 cmpq %rdx, 4*8(%rsp)
1263 /* If it is below the NMI stack, it is a normal NMI */
1266 /* Ah, it is within the NMI stack. */
1268 testb $(X86_EFLAGS_DF >> 8), (3*8 + 1)(%rsp)
1269 jz first_nmi /* RSP was user controlled. */
1271 /* This is a nested NMI. */
1275 * Modify the "iret" frame to point to repeat_nmi, forcing another
1276 * iteration of NMI handling.
1279 leaq -10*8(%rsp), %rdx
1286 /* Put stack back */
1292 /* We are returning to kernel mode, so this cannot result in a fault. */
1299 /* Make room for "NMI executing". */
1302 /* Leave room for the "iret" frame */
1305 /* Copy the "original" frame to the "outermost" frame */
1310 /* Everything up to here is safe from nested NMIs */
1312 #ifdef CONFIG_DEBUG_ENTRY
1314 * For ease of testing, unmask NMIs right away. Disabled by
1315 * default because IRET is very expensive.
1318 pushq %rsp /* RSP (minus 8 because of the previous push) */
1319 addq $8, (%rsp) /* Fix up RSP */
1321 pushq $__KERNEL_CS /* CS */
1323 INTERRUPT_RETURN /* continues at repeat_nmi below */
1329 * If there was a nested NMI, the first NMI's iret will return
1330 * here. But NMIs are still enabled and we can take another
1331 * nested NMI. The nested NMI checks the interrupted RIP to see
1332 * if it is between repeat_nmi and end_repeat_nmi, and if so
1333 * it will just return, as we are about to repeat an NMI anyway.
1334 * This makes it safe to copy to the stack frame that a nested
1337 * RSP is pointing to "outermost RIP". gsbase is unknown, but, if
1338 * we're repeating an NMI, gsbase has the same value that it had on
1339 * the first iteration. paranoid_entry will load the kernel
1340 * gsbase if needed before we call do_nmi. "NMI executing"
1343 movq $1, 10*8(%rsp) /* Set "NMI executing". */
1346 * Copy the "outermost" frame to the "iret" frame. NMIs that nest
1347 * here must not modify the "iret" frame while we're writing to
1348 * it or it will end up containing garbage.
1358 * Everything below this point can be preempted by a nested NMI.
1359 * If this happens, then the inner NMI will change the "iret"
1360 * frame to point back to repeat_nmi.
1362 pushq $-1 /* ORIG_RAX: no syscall to restart */
1363 ALLOC_PT_GPREGS_ON_STACK
1366 * Use paranoid_entry to handle SWAPGS, but no need to use paranoid_exit
1367 * as we should not be calling schedule in NMI context.
1368 * Even with normal interrupts enabled. An NMI should not be
1369 * setting NEED_RESCHED or anything that normal interrupts and
1370 * exceptions might do.
1374 /* paranoidentry do_nmi, 0; without TRACE_IRQS_OFF */
1379 testl %ebx, %ebx /* swapgs needed? */
1387 /* Point RSP at the "iret" frame. */
1388 REMOVE_PT_GPREGS_FROM_STACK 6*8
1391 * Clear "NMI executing". Set DF first so that we can easily
1392 * distinguish the remaining code between here and IRET from
1393 * the SYSCALL entry and exit paths. On a native kernel, we
1394 * could just inspect RIP, but, on paravirt kernels,
1395 * INTERRUPT_RETURN can translate into a jump into a
1399 movq $0, 5*8(%rsp) /* clear "NMI executing" */
1402 * INTERRUPT_RETURN reads the "iret" frame and exits the NMI
1403 * stack in a single instruction. We are returning to kernel
1404 * mode, so this cannot result in a fault.
1409 ENTRY(ignore_sysret)