1 /* SPDX-License-Identifier: GPL-2.0 */
3 * linux/arch/x86_64/entry.S
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 * Copyright (C) 2000, 2001, 2002 Andi Kleen SuSE Labs
7 * Copyright (C) 2000 Pavel Machek <pavel@suse.cz>
9 * entry.S contains the system-call and fault low-level handling routines.
11 * Some of this is documented in Documentation/x86/entry_64.txt
13 * A note on terminology:
14 * - iret frame: Architecture defined interrupt frame from SS to RIP
15 * at the top of the kernel process stack.
18 * - ENTRY/END: Define functions in the symbol table.
19 * - TRACE_IRQ_*: Trace hardirq state for lock debugging.
20 * - idtentry: Define exception entry points.
22 #include <linux/linkage.h>
23 #include <asm/segment.h>
24 #include <asm/cache.h>
25 #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 <asm/export.h>
39 #include <asm/frame.h>
40 #include <asm/nospec-branch.h>
41 #include <linux/err.h>
46 .section .entry.text, "ax"
48 #ifdef CONFIG_PARAVIRT
49 ENTRY(native_usergs_sysret64)
53 END(native_usergs_sysret64)
54 #endif /* CONFIG_PARAVIRT */
56 .macro TRACE_IRQS_FLAGS flags:req
57 #ifdef CONFIG_TRACE_IRQFLAGS
58 bt $9, \flags /* interrupts off? */
65 .macro TRACE_IRQS_IRETQ
66 TRACE_IRQS_FLAGS EFLAGS(%rsp)
70 * When dynamic function tracer is enabled it will add a breakpoint
71 * to all locations that it is about to modify, sync CPUs, update
72 * all the code, sync CPUs, then remove the breakpoints. In this time
73 * if lockdep is enabled, it might jump back into the debug handler
74 * outside the updating of the IST protection. (TRACE_IRQS_ON/OFF).
76 * We need to change the IDT table before calling TRACE_IRQS_ON/OFF to
77 * make sure the stack pointer does not get reset back to the top
78 * of the debug stack, and instead just reuses the current stack.
80 #if defined(CONFIG_DYNAMIC_FTRACE) && defined(CONFIG_TRACE_IRQFLAGS)
82 .macro TRACE_IRQS_OFF_DEBUG
83 call debug_stack_set_zero
85 call debug_stack_reset
88 .macro TRACE_IRQS_ON_DEBUG
89 call debug_stack_set_zero
91 call debug_stack_reset
94 .macro TRACE_IRQS_IRETQ_DEBUG
95 bt $9, EFLAGS(%rsp) /* interrupts off? */
102 # define TRACE_IRQS_OFF_DEBUG TRACE_IRQS_OFF
103 # define TRACE_IRQS_ON_DEBUG TRACE_IRQS_ON
104 # define TRACE_IRQS_IRETQ_DEBUG TRACE_IRQS_IRETQ
108 * 64-bit SYSCALL instruction entry. Up to 6 arguments in registers.
110 * This is the only entry point used for 64-bit system calls. The
111 * hardware interface is reasonably well designed and the register to
112 * argument mapping Linux uses fits well with the registers that are
113 * available when SYSCALL is used.
115 * SYSCALL instructions can be found inlined in libc implementations as
116 * well as some other programs and libraries. There are also a handful
117 * of SYSCALL instructions in the vDSO used, for example, as a
118 * clock_gettimeofday fallback.
120 * 64-bit SYSCALL saves rip to rcx, clears rflags.RF, then saves rflags to r11,
121 * then loads new ss, cs, and rip from previously programmed MSRs.
122 * rflags gets masked by a value from another MSR (so CLD and CLAC
123 * are not needed). SYSCALL does not save anything on the stack
124 * and does not change rsp.
126 * Registers on entry:
127 * rax system call number
129 * r11 saved rflags (note: r11 is callee-clobbered register in C ABI)
133 * r10 arg3 (needs to be moved to rcx to conform to C ABI)
136 * (note: r12-r15, rbp, rbx are callee-preserved in C ABI)
138 * Only called from user space.
140 * When user can change pt_regs->foo always force IRET. That is because
141 * it deals with uncanonical addresses better. SYSRET has trouble
142 * with them due to bugs in both AMD and Intel CPUs.
145 .pushsection .entry_trampoline, "ax"
148 * The code in here gets remapped into cpu_entry_area's trampoline. This means
149 * that the assembler and linker have the wrong idea as to where this code
150 * lives (and, in fact, it's mapped more than once, so it's not even at a
151 * fixed address). So we can't reference any symbols outside the entry
152 * trampoline and expect it to work.
154 * Instead, we carefully abuse %rip-relative addressing.
155 * _entry_trampoline(%rip) refers to the start of the remapped) entry
156 * trampoline. We can thus find cpu_entry_area with this macro:
159 #define CPU_ENTRY_AREA \
160 _entry_trampoline - CPU_ENTRY_AREA_entry_trampoline(%rip)
162 /* The top word of the SYSENTER stack is hot and is usable as scratch space. */
163 #define RSP_SCRATCH CPU_ENTRY_AREA_entry_stack + \
164 SIZEOF_entry_stack - 8 + CPU_ENTRY_AREA
166 ENTRY(entry_SYSCALL_64_trampoline)
170 /* Stash the user RSP. */
171 movq %rsp, RSP_SCRATCH
173 /* Note: using %rsp as a scratch reg. */
174 SWITCH_TO_KERNEL_CR3 scratch_reg=%rsp
176 /* Load the top of the task stack into RSP */
177 movq CPU_ENTRY_AREA_tss + TSS_sp1 + CPU_ENTRY_AREA, %rsp
179 /* Start building the simulated IRET frame. */
180 pushq $__USER_DS /* pt_regs->ss */
181 pushq RSP_SCRATCH /* pt_regs->sp */
182 pushq %r11 /* pt_regs->flags */
183 pushq $__USER_CS /* pt_regs->cs */
184 pushq %rcx /* pt_regs->ip */
187 * x86 lacks a near absolute jump, and we can't jump to the real
188 * entry text with a relative jump. We could push the target
189 * address and then use retq, but this destroys the pipeline on
190 * many CPUs (wasting over 20 cycles on Sandy Bridge). Instead,
191 * spill RDI and restore it in a second-stage trampoline.
194 movq $entry_SYSCALL_64_stage2, %rdi
196 END(entry_SYSCALL_64_trampoline)
200 ENTRY(entry_SYSCALL_64_stage2)
203 jmp entry_SYSCALL_64_after_hwframe
204 END(entry_SYSCALL_64_stage2)
206 ENTRY(entry_SYSCALL_64)
209 * Interrupts are off on entry.
210 * We do not frame this tiny irq-off block with TRACE_IRQS_OFF/ON,
211 * it is too small to ever cause noticeable irq latency.
216 * This path is not taken when PAGE_TABLE_ISOLATION is disabled so it
217 * is not required to switch CR3.
219 movq %rsp, PER_CPU_VAR(rsp_scratch)
220 movq PER_CPU_VAR(cpu_current_top_of_stack), %rsp
222 /* Construct struct pt_regs on stack */
223 pushq $__USER_DS /* pt_regs->ss */
224 pushq PER_CPU_VAR(rsp_scratch) /* pt_regs->sp */
225 pushq %r11 /* pt_regs->flags */
226 pushq $__USER_CS /* pt_regs->cs */
227 pushq %rcx /* pt_regs->ip */
228 GLOBAL(entry_SYSCALL_64_after_hwframe)
229 pushq %rax /* pt_regs->orig_ax */
230 pushq %rdi /* pt_regs->di */
231 pushq %rsi /* pt_regs->si */
232 pushq %rdx /* pt_regs->dx */
233 pushq %rcx /* pt_regs->cx */
234 pushq $-ENOSYS /* pt_regs->ax */
235 pushq %r8 /* pt_regs->r8 */
236 pushq %r9 /* pt_regs->r9 */
237 pushq %r10 /* pt_regs->r10 */
238 pushq %r11 /* pt_regs->r11 */
239 pushq %rbx /* pt_regs->rbx */
240 pushq %rbp /* pt_regs->rbp */
241 pushq %r12 /* pt_regs->r12 */
242 pushq %r13 /* pt_regs->r13 */
243 pushq %r14 /* pt_regs->r14 */
244 pushq %r15 /* pt_regs->r15 */
251 call do_syscall_64 /* returns with IRQs disabled */
253 TRACE_IRQS_IRETQ /* we're about to change IF */
256 * Try to use SYSRET instead of IRET if we're returning to
257 * a completely clean 64-bit userspace context. If we're not,
258 * go to the slow exit path.
263 cmpq %rcx, %r11 /* SYSRET requires RCX == RIP */
264 jne swapgs_restore_regs_and_return_to_usermode
267 * On Intel CPUs, SYSRET with non-canonical RCX/RIP will #GP
268 * in kernel space. This essentially lets the user take over
269 * the kernel, since userspace controls RSP.
271 * If width of "canonical tail" ever becomes variable, this will need
272 * to be updated to remain correct on both old and new CPUs.
274 * Change top bits to match most significant bit (47th or 56th bit
275 * depending on paging mode) in the address.
277 shl $(64 - (__VIRTUAL_MASK_SHIFT+1)), %rcx
278 sar $(64 - (__VIRTUAL_MASK_SHIFT+1)), %rcx
280 /* If this changed %rcx, it was not canonical */
282 jne swapgs_restore_regs_and_return_to_usermode
284 cmpq $__USER_CS, CS(%rsp) /* CS must match SYSRET */
285 jne swapgs_restore_regs_and_return_to_usermode
288 cmpq %r11, EFLAGS(%rsp) /* R11 == RFLAGS */
289 jne swapgs_restore_regs_and_return_to_usermode
292 * SYSCALL clears RF when it saves RFLAGS in R11 and SYSRET cannot
293 * restore RF properly. If the slowpath sets it for whatever reason, we
294 * need to restore it correctly.
296 * SYSRET can restore TF, but unlike IRET, restoring TF results in a
297 * trap from userspace immediately after SYSRET. This would cause an
298 * infinite loop whenever #DB happens with register state that satisfies
299 * the opportunistic SYSRET conditions. For example, single-stepping
302 * movq $stuck_here, %rcx
307 * would never get past 'stuck_here'.
309 testq $(X86_EFLAGS_RF|X86_EFLAGS_TF), %r11
310 jnz swapgs_restore_regs_and_return_to_usermode
312 /* nothing to check for RSP */
314 cmpq $__USER_DS, SS(%rsp) /* SS must match SYSRET */
315 jne swapgs_restore_regs_and_return_to_usermode
318 * We win! This label is here just for ease of understanding
319 * perf profiles. Nothing jumps here.
321 syscall_return_via_sysret:
322 /* rcx and r11 are already restored (see code above) */
325 popq %rsi /* skip r11 */
330 popq %rsi /* skip rcx */
335 * Now all regs are restored except RSP and RDI.
336 * Save old stack pointer and switch to trampoline stack.
339 movq PER_CPU_VAR(cpu_tss_rw + TSS_sp0), %rsp
341 pushq RSP-RDI(%rdi) /* RSP */
342 pushq (%rdi) /* RDI */
345 * We are on the trampoline stack. All regs except RDI are live.
346 * We can do future final exit work right here.
348 SWITCH_TO_USER_CR3_STACK scratch_reg=%rdi
353 END(entry_SYSCALL_64)
359 ENTRY(__switch_to_asm)
362 * Save callee-saved registers
363 * This must match the order in inactive_task_frame
373 movq %rsp, TASK_threadsp(%rdi)
374 movq TASK_threadsp(%rsi), %rsp
376 #ifdef CONFIG_CC_STACKPROTECTOR
377 movq TASK_stack_canary(%rsi), %rbx
378 movq %rbx, PER_CPU_VAR(irq_stack_union)+stack_canary_offset
381 #ifdef CONFIG_RETPOLINE
383 * When switching from a shallower to a deeper call stack
384 * the RSB may either underflow or use entries populated
385 * with userspace addresses. On CPUs where those concerns
386 * exist, overwrite the RSB with entries which capture
387 * speculative execution to prevent attack.
390 FILL_RETURN_BUFFER RSB_CLEAR_LOOPS, X86_FEATURE_RSB_CTXSW
393 /* restore callee-saved registers */
405 * A newly forked process directly context switches into this address.
407 * rax: prev task we switched from
408 * rbx: kernel thread func (NULL for user thread)
409 * r12: kernel thread arg
414 call schedule_tail /* rdi: 'prev' task parameter */
416 testq %rbx, %rbx /* from kernel_thread? */
417 jnz 1f /* kernel threads are uncommon */
422 call syscall_return_slowpath /* returns with IRQs disabled */
423 TRACE_IRQS_ON /* user mode is traced as IRQS on */
424 jmp swapgs_restore_regs_and_return_to_usermode
431 * A kernel thread is allowed to return here after successfully
432 * calling do_execve(). Exit to userspace to complete the execve()
440 * Build the entry stubs with some assembler magic.
441 * We pack 1 stub into every 8-byte block.
444 ENTRY(irq_entries_start)
445 vector=FIRST_EXTERNAL_VECTOR
446 .rept (FIRST_SYSTEM_VECTOR - FIRST_EXTERNAL_VECTOR)
447 UNWIND_HINT_IRET_REGS
448 pushq $(~vector+0x80) /* Note: always in signed byte range */
453 END(irq_entries_start)
455 .macro DEBUG_ENTRY_ASSERT_IRQS_OFF
456 #ifdef CONFIG_DEBUG_ENTRY
459 testl $X86_EFLAGS_IF, %eax
468 * Enters the IRQ stack if we're not already using it. NMI-safe. Clobbers
469 * flags and puts old RSP into old_rsp, and leaves all other GPRs alone.
470 * Requires kernel GSBASE.
472 * The invariant is that, if irq_count != -1, then the IRQ stack is in use.
474 .macro ENTER_IRQ_STACK regs=1 old_rsp
475 DEBUG_ENTRY_ASSERT_IRQS_OFF
479 UNWIND_HINT_REGS base=\old_rsp
482 incl PER_CPU_VAR(irq_count)
483 jnz .Lirq_stack_push_old_rsp_\@
486 * Right now, if we just incremented irq_count to zero, we've
487 * claimed the IRQ stack but we haven't switched to it yet.
489 * If anything is added that can interrupt us here without using IST,
490 * it must be *extremely* careful to limit its stack usage. This
491 * could include kprobes and a hypothetical future IST-less #DB
494 * The OOPS unwinder relies on the word at the top of the IRQ
495 * stack linking back to the previous RSP for the entire time we're
496 * on the IRQ stack. For this to work reliably, we need to write
497 * it before we actually move ourselves to the IRQ stack.
500 movq \old_rsp, PER_CPU_VAR(irq_stack_union + IRQ_STACK_SIZE - 8)
501 movq PER_CPU_VAR(irq_stack_ptr), %rsp
503 #ifdef CONFIG_DEBUG_ENTRY
505 * If the first movq above becomes wrong due to IRQ stack layout
506 * changes, the only way we'll notice is if we try to unwind right
507 * here. Assert that we set up the stack right to catch this type
510 cmpq -8(%rsp), \old_rsp
511 je .Lirq_stack_okay\@
516 .Lirq_stack_push_old_rsp_\@:
520 UNWIND_HINT_REGS indirect=1
525 * Undoes ENTER_IRQ_STACK.
527 .macro LEAVE_IRQ_STACK regs=1
528 DEBUG_ENTRY_ASSERT_IRQS_OFF
529 /* We need to be off the IRQ stack before decrementing irq_count. */
537 * As in ENTER_IRQ_STACK, irq_count == 0, we are still claiming
538 * the irq stack but we're not on it.
541 decl PER_CPU_VAR(irq_count)
545 * Interrupt entry/exit.
547 * Interrupt entry points save only callee clobbered registers in fast path.
549 * Entry runs with interrupts off.
552 /* 0(%rsp): ~(interrupt number) */
553 .macro interrupt func
556 testb $3, CS-ORIG_RAX(%rsp)
559 call switch_to_thread_stack
562 ALLOC_PT_GPREGS_ON_STACK
571 * IRQ from user mode.
573 * We need to tell lockdep that IRQs are off. We can't do this until
574 * we fix gsbase, and we should do it before enter_from_user_mode
575 * (which can take locks). Since TRACE_IRQS_OFF idempotent,
576 * the simplest way to handle it is to just call it twice if
577 * we enter from user mode. There's no reason to optimize this since
578 * TRACE_IRQS_OFF is a no-op if lockdep is off.
582 CALL_enter_from_user_mode
585 ENTER_IRQ_STACK old_rsp=%rdi
586 /* We entered an interrupt context - irqs are off: */
589 call \func /* rdi points to pt_regs */
593 * The interrupt stubs push (~vector+0x80) onto the stack and
594 * then jump to common_interrupt.
596 .p2align CONFIG_X86_L1_CACHE_SHIFT
599 addq $-0x80, (%rsp) /* Adjust vector to [-256, -1] range */
601 /* 0(%rsp): old RSP */
603 DISABLE_INTERRUPTS(CLBR_ANY)
611 /* Interrupt came from user space */
614 call prepare_exit_to_usermode
617 GLOBAL(swapgs_restore_regs_and_return_to_usermode)
618 #ifdef CONFIG_DEBUG_ENTRY
619 /* Assert that pt_regs indicates user mode. */
636 * The stack is now user RDI, orig_ax, RIP, CS, EFLAGS, RSP, SS.
637 * Save old stack pointer and switch to trampoline stack.
640 movq PER_CPU_VAR(cpu_tss_rw + TSS_sp0), %rsp
642 /* Copy the IRET frame to the trampoline stack. */
643 pushq 6*8(%rdi) /* SS */
644 pushq 5*8(%rdi) /* RSP */
645 pushq 4*8(%rdi) /* EFLAGS */
646 pushq 3*8(%rdi) /* CS */
647 pushq 2*8(%rdi) /* RIP */
649 /* Push user RDI on the trampoline stack. */
653 * We are on the trampoline stack. All regs except RDI are live.
654 * We can do future final exit work right here.
657 SWITCH_TO_USER_CR3_STACK scratch_reg=%rdi
665 /* Returning to kernel space */
667 #ifdef CONFIG_PREEMPT
668 /* Interrupts are off */
669 /* Check if we need preemption */
670 bt $9, EFLAGS(%rsp) /* were interrupts off? */
672 0: cmpl $0, PER_CPU_VAR(__preempt_count)
674 call preempt_schedule_irq
679 * The iretq could re-enable interrupts:
683 GLOBAL(restore_regs_and_return_to_kernel)
684 #ifdef CONFIG_DEBUG_ENTRY
685 /* Assert that pt_regs indicates kernel mode. */
693 addq $8, %rsp /* skip regs->orig_ax */
695 * ARCH_HAS_MEMBARRIER_SYNC_CORE rely on IRET core serialization
696 * when returning from IPI handler.
701 UNWIND_HINT_IRET_REGS
703 * Are we returning to a stack segment from the LDT? Note: in
704 * 64-bit mode SS:RSP on the exception stack is always valid.
706 #ifdef CONFIG_X86_ESPFIX64
707 testb $4, (SS-RIP)(%rsp)
708 jnz native_irq_return_ldt
711 .global native_irq_return_iret
712 native_irq_return_iret:
714 * This may fault. Non-paranoid faults on return to userspace are
715 * handled by fixup_bad_iret. These include #SS, #GP, and #NP.
716 * Double-faults due to espfix64 are handled in do_double_fault.
717 * Other faults here are fatal.
721 #ifdef CONFIG_X86_ESPFIX64
722 native_irq_return_ldt:
724 * We are running with user GSBASE. All GPRs contain their user
725 * values. We have a percpu ESPFIX stack that is eight slots
726 * long (see ESPFIX_STACK_SIZE). espfix_waddr points to the bottom
727 * of the ESPFIX stack.
729 * We clobber RAX and RDI in this code. We stash RDI on the
730 * normal stack and RAX on the ESPFIX stack.
732 * The ESPFIX stack layout we set up looks like this:
734 * --- top of ESPFIX stack ---
739 * RIP <-- RSP points here when we're done
740 * RAX <-- espfix_waddr points here
741 * --- bottom of ESPFIX stack ---
744 pushq %rdi /* Stash user RDI */
745 SWAPGS /* to kernel GS */
746 SWITCH_TO_KERNEL_CR3 scratch_reg=%rdi /* to kernel CR3 */
748 movq PER_CPU_VAR(espfix_waddr), %rdi
749 movq %rax, (0*8)(%rdi) /* user RAX */
750 movq (1*8)(%rsp), %rax /* user RIP */
751 movq %rax, (1*8)(%rdi)
752 movq (2*8)(%rsp), %rax /* user CS */
753 movq %rax, (2*8)(%rdi)
754 movq (3*8)(%rsp), %rax /* user RFLAGS */
755 movq %rax, (3*8)(%rdi)
756 movq (5*8)(%rsp), %rax /* user SS */
757 movq %rax, (5*8)(%rdi)
758 movq (4*8)(%rsp), %rax /* user RSP */
759 movq %rax, (4*8)(%rdi)
760 /* Now RAX == RSP. */
762 andl $0xffff0000, %eax /* RAX = (RSP & 0xffff0000) */
765 * espfix_stack[31:16] == 0. The page tables are set up such that
766 * (espfix_stack | (X & 0xffff0000)) points to a read-only alias of
767 * espfix_waddr for any X. That is, there are 65536 RO aliases of
768 * the same page. Set up RSP so that RSP[31:16] contains the
769 * respective 16 bits of the /userspace/ RSP and RSP nonetheless
770 * still points to an RO alias of the ESPFIX stack.
772 orq PER_CPU_VAR(espfix_stack), %rax
774 SWITCH_TO_USER_CR3_STACK scratch_reg=%rdi
775 SWAPGS /* to user GS */
776 popq %rdi /* Restore user RDI */
779 UNWIND_HINT_IRET_REGS offset=8
782 * At this point, we cannot write to the stack any more, but we can
785 popq %rax /* Restore user RAX */
788 * RSP now points to an ordinary IRET frame, except that the page
789 * is read-only and RSP[31:16] are preloaded with the userspace
790 * values. We can now IRET back to userspace.
792 jmp native_irq_return_iret
794 END(common_interrupt)
799 .macro apicinterrupt3 num sym do_sym
801 UNWIND_HINT_IRET_REGS
810 /* Make sure APIC interrupt handlers end up in the irqentry section: */
811 #define PUSH_SECTION_IRQENTRY .pushsection .irqentry.text, "ax"
812 #define POP_SECTION_IRQENTRY .popsection
814 .macro apicinterrupt num sym do_sym
815 PUSH_SECTION_IRQENTRY
816 apicinterrupt3 \num \sym \do_sym
821 apicinterrupt3 IRQ_MOVE_CLEANUP_VECTOR irq_move_cleanup_interrupt smp_irq_move_cleanup_interrupt
822 apicinterrupt3 REBOOT_VECTOR reboot_interrupt smp_reboot_interrupt
826 apicinterrupt3 UV_BAU_MESSAGE uv_bau_message_intr1 uv_bau_message_interrupt
829 apicinterrupt LOCAL_TIMER_VECTOR apic_timer_interrupt smp_apic_timer_interrupt
830 apicinterrupt X86_PLATFORM_IPI_VECTOR x86_platform_ipi smp_x86_platform_ipi
832 #ifdef CONFIG_HAVE_KVM
833 apicinterrupt3 POSTED_INTR_VECTOR kvm_posted_intr_ipi smp_kvm_posted_intr_ipi
834 apicinterrupt3 POSTED_INTR_WAKEUP_VECTOR kvm_posted_intr_wakeup_ipi smp_kvm_posted_intr_wakeup_ipi
835 apicinterrupt3 POSTED_INTR_NESTED_VECTOR kvm_posted_intr_nested_ipi smp_kvm_posted_intr_nested_ipi
838 #ifdef CONFIG_X86_MCE_THRESHOLD
839 apicinterrupt THRESHOLD_APIC_VECTOR threshold_interrupt smp_threshold_interrupt
842 #ifdef CONFIG_X86_MCE_AMD
843 apicinterrupt DEFERRED_ERROR_VECTOR deferred_error_interrupt smp_deferred_error_interrupt
846 #ifdef CONFIG_X86_THERMAL_VECTOR
847 apicinterrupt THERMAL_APIC_VECTOR thermal_interrupt smp_thermal_interrupt
851 apicinterrupt CALL_FUNCTION_SINGLE_VECTOR call_function_single_interrupt smp_call_function_single_interrupt
852 apicinterrupt CALL_FUNCTION_VECTOR call_function_interrupt smp_call_function_interrupt
853 apicinterrupt RESCHEDULE_VECTOR reschedule_interrupt smp_reschedule_interrupt
856 apicinterrupt ERROR_APIC_VECTOR error_interrupt smp_error_interrupt
857 apicinterrupt SPURIOUS_APIC_VECTOR spurious_interrupt smp_spurious_interrupt
859 #ifdef CONFIG_IRQ_WORK
860 apicinterrupt IRQ_WORK_VECTOR irq_work_interrupt smp_irq_work_interrupt
864 * Exception entry points.
866 #define CPU_TSS_IST(x) PER_CPU_VAR(cpu_tss_rw) + (TSS_ist + ((x) - 1) * 8)
869 * Switch to the thread stack. This is called with the IRET frame and
870 * orig_ax on the stack. (That is, RDI..R12 are not on the stack and
871 * space has not been allocated for them.)
873 ENTRY(switch_to_thread_stack)
877 /* Need to switch before accessing the thread stack. */
878 SWITCH_TO_KERNEL_CR3 scratch_reg=%rdi
880 movq PER_CPU_VAR(cpu_current_top_of_stack), %rsp
881 UNWIND_HINT sp_offset=16 sp_reg=ORC_REG_DI
883 pushq 7*8(%rdi) /* regs->ss */
884 pushq 6*8(%rdi) /* regs->rsp */
885 pushq 5*8(%rdi) /* regs->eflags */
886 pushq 4*8(%rdi) /* regs->cs */
887 pushq 3*8(%rdi) /* regs->ip */
888 pushq 2*8(%rdi) /* regs->orig_ax */
889 pushq 8(%rdi) /* return address */
894 END(switch_to_thread_stack)
896 .macro idtentry sym do_sym has_error_code:req paranoid=0 shift_ist=-1
898 UNWIND_HINT_IRET_REGS offset=\has_error_code*8
901 .if \shift_ist != -1 && \paranoid == 0
902 .error "using shift_ist requires paranoid=1"
907 .if \has_error_code == 0
908 pushq $-1 /* ORIG_RAX: no syscall to restart */
911 ALLOC_PT_GPREGS_ON_STACK
914 testb $3, CS(%rsp) /* If coming from userspace, switch stacks */
915 jnz .Lfrom_usermode_switch_stack_\@
924 /* returned flag: ebx=0: need swapgs on exit, ebx=1: don't need it */
928 TRACE_IRQS_OFF_DEBUG /* reload IDT in case of recursion */
934 movq %rsp, %rdi /* pt_regs pointer */
937 movq ORIG_RAX(%rsp), %rsi /* get error code */
938 movq $-1, ORIG_RAX(%rsp) /* no syscall to restart */
940 xorl %esi, %esi /* no error code */
944 subq $EXCEPTION_STKSZ, CPU_TSS_IST(\shift_ist)
950 addq $EXCEPTION_STKSZ, CPU_TSS_IST(\shift_ist)
953 /* these procedures expect "no swapgs" flag in ebx */
962 * Entry from userspace. Switch stacks and treat it
963 * as a normal entry. This means that paranoid handlers
964 * run in real process context if user_mode(regs).
966 .Lfrom_usermode_switch_stack_\@:
969 movq %rsp, %rdi /* pt_regs pointer */
972 movq ORIG_RAX(%rsp), %rsi /* get error code */
973 movq $-1, ORIG_RAX(%rsp) /* no syscall to restart */
975 xorl %esi, %esi /* no error code */
980 jmp error_exit /* %ebx: no swapgs flag */
985 idtentry divide_error do_divide_error has_error_code=0
986 idtentry overflow do_overflow has_error_code=0
987 idtentry bounds do_bounds has_error_code=0
988 idtentry invalid_op do_invalid_op has_error_code=0
989 idtentry device_not_available do_device_not_available has_error_code=0
990 idtentry double_fault do_double_fault has_error_code=1 paranoid=2
991 idtentry coprocessor_segment_overrun do_coprocessor_segment_overrun has_error_code=0
992 idtentry invalid_TSS do_invalid_TSS has_error_code=1
993 idtentry segment_not_present do_segment_not_present has_error_code=1
994 idtentry spurious_interrupt_bug do_spurious_interrupt_bug has_error_code=0
995 idtentry coprocessor_error do_coprocessor_error has_error_code=0
996 idtentry alignment_check do_alignment_check has_error_code=1
997 idtentry simd_coprocessor_error do_simd_coprocessor_error has_error_code=0
1001 * Reload gs selector with exception handling
1004 ENTRY(native_load_gs_index)
1007 DISABLE_INTERRUPTS(CLBR_ANY & ~CLBR_RDI)
1012 2: ALTERNATIVE "", "mfence", X86_BUG_SWAPGS_FENCE
1014 TRACE_IRQS_FLAGS (%rsp)
1018 ENDPROC(native_load_gs_index)
1019 EXPORT_SYMBOL(native_load_gs_index)
1021 _ASM_EXTABLE(.Lgs_change, bad_gs)
1022 .section .fixup, "ax"
1023 /* running with kernelgs */
1025 SWAPGS /* switch back to user gs */
1027 /* This can't be a string because the preprocessor needs to see it. */
1028 movl $__USER_DS, %eax
1031 ALTERNATIVE "", "ZAP_GS", X86_BUG_NULL_SEG
1037 /* Call softirq on interrupt stack. Interrupts are off. */
1038 ENTRY(do_softirq_own_stack)
1041 ENTER_IRQ_STACK regs=0 old_rsp=%r11
1043 LEAVE_IRQ_STACK regs=0
1046 ENDPROC(do_softirq_own_stack)
1049 idtentry hypervisor_callback xen_do_hypervisor_callback has_error_code=0
1052 * A note on the "critical region" in our callback handler.
1053 * We want to avoid stacking callback handlers due to events occurring
1054 * during handling of the last event. To do this, we keep events disabled
1055 * until we've done all processing. HOWEVER, we must enable events before
1056 * popping the stack frame (can't be done atomically) and so it would still
1057 * be possible to get enough handler activations to overflow the stack.
1058 * Although unlikely, bugs of that kind are hard to track down, so we'd
1059 * like to avoid the possibility.
1060 * So, on entry to the handler we detect whether we interrupted an
1061 * existing activation in its critical region -- if so, we pop the current
1062 * activation and restart the handler using the previous one.
1064 ENTRY(xen_do_hypervisor_callback) /* do_hypervisor_callback(struct *pt_regs) */
1067 * Since we don't modify %rdi, evtchn_do_upall(struct *pt_regs) will
1068 * see the correct pointer to the pt_regs
1071 movq %rdi, %rsp /* we don't return, adjust the stack frame */
1074 ENTER_IRQ_STACK old_rsp=%r10
1075 call xen_evtchn_do_upcall
1078 #ifndef CONFIG_PREEMPT
1079 call xen_maybe_preempt_hcall
1082 END(xen_do_hypervisor_callback)
1085 * Hypervisor uses this for application faults while it executes.
1086 * We get here for two reasons:
1087 * 1. Fault while reloading DS, ES, FS or GS
1088 * 2. Fault while executing IRET
1089 * Category 1 we do not need to fix up as Xen has already reloaded all segment
1090 * registers that could be reloaded and zeroed the others.
1091 * Category 2 we fix up by killing the current process. We cannot use the
1092 * normal Linux return path in this case because if we use the IRET hypercall
1093 * to pop the stack frame we end up in an infinite loop of failsafe callbacks.
1094 * We distinguish between categories by comparing each saved segment register
1095 * with its current contents: any discrepancy means we in category 1.
1097 ENTRY(xen_failsafe_callback)
1100 cmpw %cx, 0x10(%rsp)
1103 cmpw %cx, 0x18(%rsp)
1106 cmpw %cx, 0x20(%rsp)
1109 cmpw %cx, 0x28(%rsp)
1111 /* All segments match their saved values => Category 2 (Bad IRET). */
1116 UNWIND_HINT_IRET_REGS offset=8
1117 jmp general_protection
1118 1: /* Segment mismatch => Category 1 (Bad segment). Retry the IRET. */
1122 UNWIND_HINT_IRET_REGS
1123 pushq $-1 /* orig_ax = -1 => not a system call */
1124 ALLOC_PT_GPREGS_ON_STACK
1127 ENCODE_FRAME_POINTER
1129 END(xen_failsafe_callback)
1131 apicinterrupt3 HYPERVISOR_CALLBACK_VECTOR \
1132 xen_hvm_callback_vector xen_evtchn_do_upcall
1134 #endif /* CONFIG_XEN */
1136 #if IS_ENABLED(CONFIG_HYPERV)
1137 apicinterrupt3 HYPERVISOR_CALLBACK_VECTOR \
1138 hyperv_callback_vector hyperv_vector_handler
1140 apicinterrupt3 HYPERV_REENLIGHTENMENT_VECTOR \
1141 hyperv_reenlightenment_vector hyperv_reenlightenment_intr
1142 #endif /* CONFIG_HYPERV */
1144 idtentry debug do_debug has_error_code=0 paranoid=1 shift_ist=DEBUG_STACK
1145 idtentry int3 do_int3 has_error_code=0 paranoid=1 shift_ist=DEBUG_STACK
1146 idtentry stack_segment do_stack_segment has_error_code=1
1149 idtentry xennmi do_nmi has_error_code=0
1150 idtentry xendebug do_debug has_error_code=0
1151 idtentry xenint3 do_int3 has_error_code=0
1154 idtentry general_protection do_general_protection has_error_code=1
1155 idtentry page_fault do_page_fault has_error_code=1
1157 #ifdef CONFIG_KVM_GUEST
1158 idtentry async_page_fault do_async_page_fault has_error_code=1
1161 #ifdef CONFIG_X86_MCE
1162 idtentry machine_check do_mce has_error_code=0 paranoid=1
1166 * Save all registers in pt_regs, and switch gs if needed.
1167 * Use slow, but surefire "are we in kernel?" check.
1168 * Return: ebx=0: need swapgs on exit, ebx=1: otherwise
1170 ENTRY(paranoid_entry)
1175 ENCODE_FRAME_POINTER 8
1177 movl $MSR_GS_BASE, %ecx
1180 js 1f /* negative -> in kernel */
1185 SAVE_AND_SWITCH_TO_KERNEL_CR3 scratch_reg=%rax save_reg=%r14
1191 * "Paranoid" exit path from exception stack. This is invoked
1192 * only on return from non-NMI IST interrupts that came
1193 * from kernel space.
1195 * We may be returning to very strange contexts (e.g. very early
1196 * in syscall entry), so checking for preemption here would
1197 * be complicated. Fortunately, we there's no good reason
1198 * to try to handle preemption here.
1200 * On entry, ebx is "no swapgs" flag (1: don't need swapgs, 0: need it)
1202 ENTRY(paranoid_exit)
1204 DISABLE_INTERRUPTS(CLBR_ANY)
1205 TRACE_IRQS_OFF_DEBUG
1206 testl %ebx, %ebx /* swapgs needed? */
1207 jnz .Lparanoid_exit_no_swapgs
1209 RESTORE_CR3 scratch_reg=%rbx save_reg=%r14
1211 jmp .Lparanoid_exit_restore
1212 .Lparanoid_exit_no_swapgs:
1213 TRACE_IRQS_IRETQ_DEBUG
1214 .Lparanoid_exit_restore:
1215 jmp restore_regs_and_return_to_kernel
1219 * Save all registers in pt_regs, and switch gs if needed.
1220 * Return: EBX=0: came from user mode; EBX=1: otherwise
1227 ENCODE_FRAME_POINTER 8
1229 testb $3, CS+8(%rsp)
1230 jz .Lerror_kernelspace
1233 * We entered from user mode or we're pretending to have entered
1234 * from user mode due to an IRET fault.
1237 /* We have user CR3. Change to kernel CR3. */
1238 SWITCH_TO_KERNEL_CR3 scratch_reg=%rax
1240 .Lerror_entry_from_usermode_after_swapgs:
1241 /* Put us onto the real thread stack. */
1242 popq %r12 /* save return addr in %12 */
1243 movq %rsp, %rdi /* arg0 = pt_regs pointer */
1245 movq %rax, %rsp /* switch stack */
1246 ENCODE_FRAME_POINTER
1250 * We need to tell lockdep that IRQs are off. We can't do this until
1251 * we fix gsbase, and we should do it before enter_from_user_mode
1252 * (which can take locks).
1255 CALL_enter_from_user_mode
1263 * There are two places in the kernel that can potentially fault with
1264 * usergs. Handle them here. B stepping K8s sometimes report a
1265 * truncated RIP for IRET exceptions returning to compat mode. Check
1266 * for these here too.
1268 .Lerror_kernelspace:
1270 leaq native_irq_return_iret(%rip), %rcx
1271 cmpq %rcx, RIP+8(%rsp)
1273 movl %ecx, %eax /* zero extend */
1274 cmpq %rax, RIP+8(%rsp)
1276 cmpq $.Lgs_change, RIP+8(%rsp)
1277 jne .Lerror_entry_done
1280 * hack: .Lgs_change can fail with user gsbase. If this happens, fix up
1281 * gsbase and proceed. We'll fix up the exception and land in
1282 * .Lgs_change's error handler with kernel gsbase.
1285 SWITCH_TO_KERNEL_CR3 scratch_reg=%rax
1286 jmp .Lerror_entry_done
1289 /* Fix truncated RIP */
1290 movq %rcx, RIP+8(%rsp)
1295 * We came from an IRET to user mode, so we have user
1296 * gsbase and CR3. Switch to kernel gsbase and CR3:
1299 SWITCH_TO_KERNEL_CR3 scratch_reg=%rax
1302 * Pretend that the exception came from user mode: set up pt_regs
1303 * as if we faulted immediately after IRET and clear EBX so that
1304 * error_exit knows that we will be returning to user mode.
1310 jmp .Lerror_entry_from_usermode_after_swapgs
1315 * On entry, EBX is a "return to kernel mode" flag:
1316 * 1: already in kernel mode, don't need SWAPGS
1317 * 0: user gsbase is loaded, we need SWAPGS and standard preparation for return to usermode
1321 DISABLE_INTERRUPTS(CLBR_ANY)
1329 * Runs on exception stack. Xen PV does not go through this path at all,
1330 * so we can use real assembly here.
1333 * %r14: Used to save/restore the CR3 of the interrupted context
1334 * when PAGE_TABLE_ISOLATION is in use. Do not clobber.
1337 UNWIND_HINT_IRET_REGS
1340 * We allow breakpoints in NMIs. If a breakpoint occurs, then
1341 * the iretq it performs will take us out of NMI context.
1342 * This means that we can have nested NMIs where the next
1343 * NMI is using the top of the stack of the previous NMI. We
1344 * can't let it execute because the nested NMI will corrupt the
1345 * stack of the previous NMI. NMI handlers are not re-entrant
1348 * To handle this case we do the following:
1349 * Check the a special location on the stack that contains
1350 * a variable that is set when NMIs are executing.
1351 * The interrupted task's stack is also checked to see if it
1353 * If the variable is not set and the stack is not the NMI
1355 * o Set the special variable on the stack
1356 * o Copy the interrupt frame into an "outermost" location on the
1358 * o Copy the interrupt frame into an "iret" location on the stack
1359 * o Continue processing the NMI
1360 * If the variable is set or the previous stack is the NMI stack:
1361 * o Modify the "iret" location to jump to the repeat_nmi
1362 * o return back to the first NMI
1364 * Now on exit of the first NMI, we first clear the stack variable
1365 * The NMI stack will tell any nested NMIs at that point that it is
1366 * nested. Then we pop the stack normally with iret, and if there was
1367 * a nested NMI that updated the copy interrupt stack frame, a
1368 * jump will be made to the repeat_nmi code that will handle the second
1371 * However, espfix prevents us from directly returning to userspace
1372 * with a single IRET instruction. Similarly, IRET to user mode
1373 * can fault. We therefore handle NMIs from user space like
1374 * other IST entries.
1379 /* Use %rdx as our temp variable throughout */
1382 testb $3, CS-RIP+8(%rsp)
1383 jz .Lnmi_from_kernel
1386 * NMI from user mode. We need to run on the thread stack, but we
1387 * can't go through the normal entry paths: NMIs are masked, and
1388 * we don't want to enable interrupts, because then we'll end
1389 * up in an awkward situation in which IRQs are on but NMIs
1392 * We also must not push anything to the stack before switching
1393 * stacks lest we corrupt the "NMI executing" variable.
1398 SWITCH_TO_KERNEL_CR3 scratch_reg=%rdx
1400 movq PER_CPU_VAR(cpu_current_top_of_stack), %rsp
1401 UNWIND_HINT_IRET_REGS base=%rdx offset=8
1402 pushq 5*8(%rdx) /* pt_regs->ss */
1403 pushq 4*8(%rdx) /* pt_regs->rsp */
1404 pushq 3*8(%rdx) /* pt_regs->flags */
1405 pushq 2*8(%rdx) /* pt_regs->cs */
1406 pushq 1*8(%rdx) /* pt_regs->rip */
1407 UNWIND_HINT_IRET_REGS
1408 pushq $-1 /* pt_regs->orig_ax */
1409 pushq %rdi /* pt_regs->di */
1410 pushq %rsi /* pt_regs->si */
1411 pushq (%rdx) /* pt_regs->dx */
1412 pushq %rcx /* pt_regs->cx */
1413 pushq %rax /* pt_regs->ax */
1414 pushq %r8 /* pt_regs->r8 */
1415 pushq %r9 /* pt_regs->r9 */
1416 pushq %r10 /* pt_regs->r10 */
1417 pushq %r11 /* pt_regs->r11 */
1418 pushq %rbx /* pt_regs->rbx */
1419 pushq %rbp /* pt_regs->rbp */
1420 pushq %r12 /* pt_regs->r12 */
1421 pushq %r13 /* pt_regs->r13 */
1422 pushq %r14 /* pt_regs->r14 */
1423 pushq %r15 /* pt_regs->r15 */
1425 ENCODE_FRAME_POINTER
1428 * At this point we no longer need to worry about stack damage
1429 * due to nesting -- we're on the normal thread stack and we're
1430 * done with the NMI stack.
1438 * Return back to user mode. We must *not* do the normal exit
1439 * work, because we don't want to enable interrupts.
1441 jmp swapgs_restore_regs_and_return_to_usermode
1445 * Here's what our stack frame will look like:
1446 * +---------------------------------------------------------+
1448 * | original Return RSP |
1449 * | original RFLAGS |
1452 * +---------------------------------------------------------+
1453 * | temp storage for rdx |
1454 * +---------------------------------------------------------+
1455 * | "NMI executing" variable |
1456 * +---------------------------------------------------------+
1457 * | iret SS } Copied from "outermost" frame |
1458 * | iret Return RSP } on each loop iteration; overwritten |
1459 * | iret RFLAGS } by a nested NMI to force another |
1460 * | iret CS } iteration if needed. |
1462 * +---------------------------------------------------------+
1463 * | outermost SS } initialized in first_nmi; |
1464 * | outermost Return RSP } will not be changed before |
1465 * | outermost RFLAGS } NMI processing is done. |
1466 * | outermost CS } Copied to "iret" frame on each |
1467 * | outermost RIP } iteration. |
1468 * +---------------------------------------------------------+
1470 * +---------------------------------------------------------+
1472 * The "original" frame is used by hardware. Before re-enabling
1473 * NMIs, we need to be done with it, and we need to leave enough
1474 * space for the asm code here.
1476 * We return by executing IRET while RSP points to the "iret" frame.
1477 * That will either return for real or it will loop back into NMI
1480 * The "outermost" frame is copied to the "iret" frame on each
1481 * iteration of the loop, so each iteration starts with the "iret"
1482 * frame pointing to the final return target.
1486 * Determine whether we're a nested NMI.
1488 * If we interrupted kernel code between repeat_nmi and
1489 * end_repeat_nmi, then we are a nested NMI. We must not
1490 * modify the "iret" frame because it's being written by
1491 * the outer NMI. That's okay; the outer NMI handler is
1492 * about to about to call do_nmi anyway, so we can just
1493 * resume the outer NMI.
1496 movq $repeat_nmi, %rdx
1499 movq $end_repeat_nmi, %rdx
1505 * Now check "NMI executing". If it's set, then we're nested.
1506 * This will not detect if we interrupted an outer NMI just
1513 * Now test if the previous stack was an NMI stack. This covers
1514 * the case where we interrupt an outer NMI after it clears
1515 * "NMI executing" but before IRET. We need to be careful, though:
1516 * there is one case in which RSP could point to the NMI stack
1517 * despite there being no NMI active: naughty userspace controls
1518 * RSP at the very beginning of the SYSCALL targets. We can
1519 * pull a fast one on naughty userspace, though: we program
1520 * SYSCALL to mask DF, so userspace cannot cause DF to be set
1521 * if it controls the kernel's RSP. We set DF before we clear
1525 /* Compare the NMI stack (rdx) with the stack we came from (4*8(%rsp)) */
1526 cmpq %rdx, 4*8(%rsp)
1527 /* If the stack pointer is above the NMI stack, this is a normal NMI */
1530 subq $EXCEPTION_STKSZ, %rdx
1531 cmpq %rdx, 4*8(%rsp)
1532 /* If it is below the NMI stack, it is a normal NMI */
1535 /* Ah, it is within the NMI stack. */
1537 testb $(X86_EFLAGS_DF >> 8), (3*8 + 1)(%rsp)
1538 jz first_nmi /* RSP was user controlled. */
1540 /* This is a nested NMI. */
1544 * Modify the "iret" frame to point to repeat_nmi, forcing another
1545 * iteration of NMI handling.
1548 leaq -10*8(%rsp), %rdx
1555 /* Put stack back */
1561 /* We are returning to kernel mode, so this cannot result in a fault. */
1568 /* Make room for "NMI executing". */
1571 /* Leave room for the "iret" frame */
1574 /* Copy the "original" frame to the "outermost" frame */
1578 UNWIND_HINT_IRET_REGS
1580 /* Everything up to here is safe from nested NMIs */
1582 #ifdef CONFIG_DEBUG_ENTRY
1584 * For ease of testing, unmask NMIs right away. Disabled by
1585 * default because IRET is very expensive.
1588 pushq %rsp /* RSP (minus 8 because of the previous push) */
1589 addq $8, (%rsp) /* Fix up RSP */
1591 pushq $__KERNEL_CS /* CS */
1593 iretq /* continues at repeat_nmi below */
1594 UNWIND_HINT_IRET_REGS
1600 * If there was a nested NMI, the first NMI's iret will return
1601 * here. But NMIs are still enabled and we can take another
1602 * nested NMI. The nested NMI checks the interrupted RIP to see
1603 * if it is between repeat_nmi and end_repeat_nmi, and if so
1604 * it will just return, as we are about to repeat an NMI anyway.
1605 * This makes it safe to copy to the stack frame that a nested
1608 * RSP is pointing to "outermost RIP". gsbase is unknown, but, if
1609 * we're repeating an NMI, gsbase has the same value that it had on
1610 * the first iteration. paranoid_entry will load the kernel
1611 * gsbase if needed before we call do_nmi. "NMI executing"
1614 movq $1, 10*8(%rsp) /* Set "NMI executing". */
1617 * Copy the "outermost" frame to the "iret" frame. NMIs that nest
1618 * here must not modify the "iret" frame while we're writing to
1619 * it or it will end up containing garbage.
1629 * Everything below this point can be preempted by a nested NMI.
1630 * If this happens, then the inner NMI will change the "iret"
1631 * frame to point back to repeat_nmi.
1633 pushq $-1 /* ORIG_RAX: no syscall to restart */
1634 ALLOC_PT_GPREGS_ON_STACK
1637 * Use paranoid_entry to handle SWAPGS, but no need to use paranoid_exit
1638 * as we should not be calling schedule in NMI context.
1639 * Even with normal interrupts enabled. An NMI should not be
1640 * setting NEED_RESCHED or anything that normal interrupts and
1641 * exceptions might do.
1646 /* paranoidentry do_nmi, 0; without TRACE_IRQS_OFF */
1651 RESTORE_CR3 scratch_reg=%r15 save_reg=%r14
1653 testl %ebx, %ebx /* swapgs needed? */
1662 * Skip orig_ax and the "outermost" frame to point RSP at the "iret"
1663 * at the "iret" frame.
1668 * Clear "NMI executing". Set DF first so that we can easily
1669 * distinguish the remaining code between here and IRET from
1670 * the SYSCALL entry and exit paths.
1672 * We arguably should just inspect RIP instead, but I (Andy) wrote
1673 * this code when I had the misapprehension that Xen PV supported
1674 * NMIs, and Xen PV would break that approach.
1677 movq $0, 5*8(%rsp) /* clear "NMI executing" */
1680 * iretq reads the "iret" frame and exits the NMI stack in a
1681 * single instruction. We are returning to kernel mode, so this
1682 * cannot result in a fault. Similarly, we don't need to worry
1683 * about espfix64 on the way back to kernel mode.
1688 ENTRY(ignore_sysret)
1694 ENTRY(rewind_stack_do_exit)
1696 /* Prevent any naive code from trying to unwind to our caller. */
1699 movq PER_CPU_VAR(cpu_current_top_of_stack), %rax
1700 leaq -PTREGS_SIZE(%rax), %rsp
1701 UNWIND_HINT_FUNC sp_offset=PTREGS_SIZE
1704 END(rewind_stack_do_exit)