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_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 btl $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 * This is the only entry point used for 64-bit system calls. The
107 * hardware interface is reasonably well designed and the register to
108 * argument mapping Linux uses fits well with the registers that are
109 * available when SYSCALL is used.
111 * SYSCALL instructions can be found inlined in libc implementations as
112 * well as some other programs and libraries. There are also a handful
113 * of SYSCALL instructions in the vDSO used, for example, as a
114 * clock_gettimeofday fallback.
116 * 64-bit SYSCALL saves rip to rcx, clears rflags.RF, then saves rflags to r11,
117 * then loads new ss, cs, and rip from previously programmed MSRs.
118 * rflags gets masked by a value from another MSR (so CLD and CLAC
119 * are not needed). SYSCALL does not save anything on the stack
120 * and does not change rsp.
122 * Registers on entry:
123 * rax system call number
125 * r11 saved rflags (note: r11 is callee-clobbered register in C ABI)
129 * r10 arg3 (needs to be moved to rcx to conform to C ABI)
132 * (note: r12-r15, rbp, rbx are callee-preserved in C ABI)
134 * Only called from user space.
136 * When user can change pt_regs->foo always force IRET. That is because
137 * it deals with uncanonical addresses better. SYSRET has trouble
138 * with them due to bugs in both AMD and Intel CPUs.
141 .pushsection .entry_trampoline, "ax"
144 * The code in here gets remapped into cpu_entry_area's trampoline. This means
145 * that the assembler and linker have the wrong idea as to where this code
146 * lives (and, in fact, it's mapped more than once, so it's not even at a
147 * fixed address). So we can't reference any symbols outside the entry
148 * trampoline and expect it to work.
150 * Instead, we carefully abuse %rip-relative addressing.
151 * _entry_trampoline(%rip) refers to the start of the remapped) entry
152 * trampoline. We can thus find cpu_entry_area with this macro:
155 #define CPU_ENTRY_AREA \
156 _entry_trampoline - CPU_ENTRY_AREA_entry_trampoline(%rip)
158 /* The top word of the SYSENTER stack is hot and is usable as scratch space. */
159 #define RSP_SCRATCH CPU_ENTRY_AREA_entry_stack + \
160 SIZEOF_entry_stack - 8 + CPU_ENTRY_AREA
162 ENTRY(entry_SYSCALL_64_trampoline)
166 /* Stash the user RSP. */
167 movq %rsp, RSP_SCRATCH
169 /* Note: using %rsp as a scratch reg. */
170 SWITCH_TO_KERNEL_CR3 scratch_reg=%rsp
172 /* Load the top of the task stack into RSP */
173 movq CPU_ENTRY_AREA_tss + TSS_sp1 + CPU_ENTRY_AREA, %rsp
175 /* Start building the simulated IRET frame. */
176 pushq $__USER_DS /* pt_regs->ss */
177 pushq RSP_SCRATCH /* pt_regs->sp */
178 pushq %r11 /* pt_regs->flags */
179 pushq $__USER_CS /* pt_regs->cs */
180 pushq %rcx /* pt_regs->ip */
183 * x86 lacks a near absolute jump, and we can't jump to the real
184 * entry text with a relative jump. We could push the target
185 * address and then use retq, but this destroys the pipeline on
186 * many CPUs (wasting over 20 cycles on Sandy Bridge). Instead,
187 * spill RDI and restore it in a second-stage trampoline.
190 movq $entry_SYSCALL_64_stage2, %rdi
192 END(entry_SYSCALL_64_trampoline)
196 ENTRY(entry_SYSCALL_64_stage2)
199 jmp entry_SYSCALL_64_after_hwframe
200 END(entry_SYSCALL_64_stage2)
202 ENTRY(entry_SYSCALL_64)
205 * Interrupts are off on entry.
206 * We do not frame this tiny irq-off block with TRACE_IRQS_OFF/ON,
207 * it is too small to ever cause noticeable irq latency.
212 * This path is only taken when PAGE_TABLE_ISOLATION is disabled so it
213 * is not required to switch CR3.
215 movq %rsp, PER_CPU_VAR(rsp_scratch)
216 movq PER_CPU_VAR(cpu_current_top_of_stack), %rsp
218 /* Construct struct pt_regs on stack */
219 pushq $__USER_DS /* pt_regs->ss */
220 pushq PER_CPU_VAR(rsp_scratch) /* pt_regs->sp */
221 pushq %r11 /* pt_regs->flags */
222 pushq $__USER_CS /* pt_regs->cs */
223 pushq %rcx /* pt_regs->ip */
224 GLOBAL(entry_SYSCALL_64_after_hwframe)
225 pushq %rax /* pt_regs->orig_ax */
227 PUSH_AND_CLEAR_REGS rax=$-ENOSYS
233 call do_syscall_64 /* returns with IRQs disabled */
235 TRACE_IRQS_IRETQ /* we're about to change IF */
238 * Try to use SYSRET instead of IRET if we're returning to
239 * a completely clean 64-bit userspace context. If we're not,
240 * go to the slow exit path.
245 cmpq %rcx, %r11 /* SYSRET requires RCX == RIP */
246 jne swapgs_restore_regs_and_return_to_usermode
249 * On Intel CPUs, SYSRET with non-canonical RCX/RIP will #GP
250 * in kernel space. This essentially lets the user take over
251 * the kernel, since userspace controls RSP.
253 * If width of "canonical tail" ever becomes variable, this will need
254 * to be updated to remain correct on both old and new CPUs.
256 * Change top bits to match most significant bit (47th or 56th bit
257 * depending on paging mode) in the address.
259 shl $(64 - (__VIRTUAL_MASK_SHIFT+1)), %rcx
260 sar $(64 - (__VIRTUAL_MASK_SHIFT+1)), %rcx
262 /* If this changed %rcx, it was not canonical */
264 jne swapgs_restore_regs_and_return_to_usermode
266 cmpq $__USER_CS, CS(%rsp) /* CS must match SYSRET */
267 jne swapgs_restore_regs_and_return_to_usermode
270 cmpq %r11, EFLAGS(%rsp) /* R11 == RFLAGS */
271 jne swapgs_restore_regs_and_return_to_usermode
274 * SYSCALL clears RF when it saves RFLAGS in R11 and SYSRET cannot
275 * restore RF properly. If the slowpath sets it for whatever reason, we
276 * need to restore it correctly.
278 * SYSRET can restore TF, but unlike IRET, restoring TF results in a
279 * trap from userspace immediately after SYSRET. This would cause an
280 * infinite loop whenever #DB happens with register state that satisfies
281 * the opportunistic SYSRET conditions. For example, single-stepping
284 * movq $stuck_here, %rcx
289 * would never get past 'stuck_here'.
291 testq $(X86_EFLAGS_RF|X86_EFLAGS_TF), %r11
292 jnz swapgs_restore_regs_and_return_to_usermode
294 /* nothing to check for RSP */
296 cmpq $__USER_DS, SS(%rsp) /* SS must match SYSRET */
297 jne swapgs_restore_regs_and_return_to_usermode
300 * We win! This label is here just for ease of understanding
301 * perf profiles. Nothing jumps here.
303 syscall_return_via_sysret:
304 /* rcx and r11 are already restored (see code above) */
305 POP_REGS pop_rdi=0 skip_r11rcx=1
308 * Now all regs are restored except RSP and RDI.
309 * Save old stack pointer and switch to trampoline stack.
312 movq PER_CPU_VAR(cpu_tss_rw + TSS_sp0), %rsp
315 pushq RSP-RDI(%rdi) /* RSP */
316 pushq (%rdi) /* RDI */
319 * We are on the trampoline stack. All regs except RDI are live.
320 * We can do future final exit work right here.
322 SWITCH_TO_USER_CR3_STACK scratch_reg=%rdi
327 END(entry_SYSCALL_64)
333 ENTRY(__switch_to_asm)
336 * Save callee-saved registers
337 * This must match the order in inactive_task_frame
348 movq %rsp, TASK_threadsp(%rdi)
349 movq TASK_threadsp(%rsi), %rsp
351 #ifdef CONFIG_CC_STACKPROTECTOR
352 movq TASK_stack_canary(%rsi), %rbx
353 movq %rbx, PER_CPU_VAR(irq_stack_union)+stack_canary_offset
356 #ifdef CONFIG_RETPOLINE
358 * When switching from a shallower to a deeper call stack
359 * the RSB may either underflow or use entries populated
360 * with userspace addresses. On CPUs where those concerns
361 * exist, overwrite the RSB with entries which capture
362 * speculative execution to prevent attack.
364 FILL_RETURN_BUFFER %r12, RSB_CLEAR_LOOPS, X86_FEATURE_RSB_CTXSW
367 /* restore callee-saved registers */
380 * A newly forked process directly context switches into this address.
382 * rax: prev task we switched from
383 * rbx: kernel thread func (NULL for user thread)
384 * r12: kernel thread arg
389 call schedule_tail /* rdi: 'prev' task parameter */
391 testq %rbx, %rbx /* from kernel_thread? */
392 jnz 1f /* kernel threads are uncommon */
397 call syscall_return_slowpath /* returns with IRQs disabled */
398 TRACE_IRQS_ON /* user mode is traced as IRQS on */
399 jmp swapgs_restore_regs_and_return_to_usermode
406 * A kernel thread is allowed to return here after successfully
407 * calling do_execve(). Exit to userspace to complete the execve()
415 * Build the entry stubs with some assembler magic.
416 * We pack 1 stub into every 8-byte block.
419 ENTRY(irq_entries_start)
420 vector=FIRST_EXTERNAL_VECTOR
421 .rept (FIRST_SYSTEM_VECTOR - FIRST_EXTERNAL_VECTOR)
422 UNWIND_HINT_IRET_REGS
423 pushq $(~vector+0x80) /* Note: always in signed byte range */
428 END(irq_entries_start)
430 .macro DEBUG_ENTRY_ASSERT_IRQS_OFF
431 #ifdef CONFIG_DEBUG_ENTRY
434 testl $X86_EFLAGS_IF, %eax
443 * Enters the IRQ stack if we're not already using it. NMI-safe. Clobbers
444 * flags and puts old RSP into old_rsp, and leaves all other GPRs alone.
445 * Requires kernel GSBASE.
447 * The invariant is that, if irq_count != -1, then the IRQ stack is in use.
449 .macro ENTER_IRQ_STACK regs=1 old_rsp
450 DEBUG_ENTRY_ASSERT_IRQS_OFF
454 UNWIND_HINT_REGS base=\old_rsp
457 incl PER_CPU_VAR(irq_count)
458 jnz .Lirq_stack_push_old_rsp_\@
461 * Right now, if we just incremented irq_count to zero, we've
462 * claimed the IRQ stack but we haven't switched to it yet.
464 * If anything is added that can interrupt us here without using IST,
465 * it must be *extremely* careful to limit its stack usage. This
466 * could include kprobes and a hypothetical future IST-less #DB
469 * The OOPS unwinder relies on the word at the top of the IRQ
470 * stack linking back to the previous RSP for the entire time we're
471 * on the IRQ stack. For this to work reliably, we need to write
472 * it before we actually move ourselves to the IRQ stack.
475 movq \old_rsp, PER_CPU_VAR(irq_stack_union + IRQ_STACK_SIZE - 8)
476 movq PER_CPU_VAR(irq_stack_ptr), %rsp
478 #ifdef CONFIG_DEBUG_ENTRY
480 * If the first movq above becomes wrong due to IRQ stack layout
481 * changes, the only way we'll notice is if we try to unwind right
482 * here. Assert that we set up the stack right to catch this type
485 cmpq -8(%rsp), \old_rsp
486 je .Lirq_stack_okay\@
491 .Lirq_stack_push_old_rsp_\@:
495 UNWIND_HINT_REGS indirect=1
500 * Undoes ENTER_IRQ_STACK.
502 .macro LEAVE_IRQ_STACK regs=1
503 DEBUG_ENTRY_ASSERT_IRQS_OFF
504 /* We need to be off the IRQ stack before decrementing irq_count. */
512 * As in ENTER_IRQ_STACK, irq_count == 0, we are still claiming
513 * the irq stack but we're not on it.
516 decl PER_CPU_VAR(irq_count)
520 * Interrupt entry/exit.
522 * Interrupt entry points save only callee clobbered registers in fast path.
524 * Entry runs with interrupts off.
527 /* 0(%rsp): ~(interrupt number) */
528 .macro interrupt func
531 testb $3, CS-ORIG_RAX(%rsp)
534 FENCE_SWAPGS_USER_ENTRY
535 call switch_to_thread_stack
538 FENCE_SWAPGS_KERNEL_ENTRY
547 * IRQ from user mode.
549 * We need to tell lockdep that IRQs are off. We can't do this until
550 * we fix gsbase, and we should do it before enter_from_user_mode
551 * (which can take locks). Since TRACE_IRQS_OFF idempotent,
552 * the simplest way to handle it is to just call it twice if
553 * we enter from user mode. There's no reason to optimize this since
554 * TRACE_IRQS_OFF is a no-op if lockdep is off.
558 CALL_enter_from_user_mode
561 ENTER_IRQ_STACK old_rsp=%rdi
562 /* We entered an interrupt context - irqs are off: */
565 call \func /* rdi points to pt_regs */
569 * The interrupt stubs push (~vector+0x80) onto the stack and
570 * then jump to common_interrupt.
572 .p2align CONFIG_X86_L1_CACHE_SHIFT
575 addq $-0x80, (%rsp) /* Adjust vector to [-256, -1] range */
577 /* 0(%rsp): old RSP */
579 DISABLE_INTERRUPTS(CLBR_ANY)
587 /* Interrupt came from user space */
590 call prepare_exit_to_usermode
593 GLOBAL(swapgs_restore_regs_and_return_to_usermode)
594 #ifdef CONFIG_DEBUG_ENTRY
595 /* Assert that pt_regs indicates user mode. */
604 * The stack is now user RDI, orig_ax, RIP, CS, EFLAGS, RSP, SS.
605 * Save old stack pointer and switch to trampoline stack.
608 movq PER_CPU_VAR(cpu_tss_rw + TSS_sp0), %rsp
611 /* Copy the IRET frame to the trampoline stack. */
612 pushq 6*8(%rdi) /* SS */
613 pushq 5*8(%rdi) /* RSP */
614 pushq 4*8(%rdi) /* EFLAGS */
615 pushq 3*8(%rdi) /* CS */
616 pushq 2*8(%rdi) /* RIP */
618 /* Push user RDI on the trampoline stack. */
622 * We are on the trampoline stack. All regs except RDI are live.
623 * We can do future final exit work right here.
626 SWITCH_TO_USER_CR3_STACK scratch_reg=%rdi
634 /* Returning to kernel space */
636 #ifdef CONFIG_PREEMPT
637 /* Interrupts are off */
638 /* Check if we need preemption */
639 btl $9, EFLAGS(%rsp) /* were interrupts off? */
641 0: cmpl $0, PER_CPU_VAR(__preempt_count)
643 call preempt_schedule_irq
648 * The iretq could re-enable interrupts:
652 GLOBAL(restore_regs_and_return_to_kernel)
653 #ifdef CONFIG_DEBUG_ENTRY
654 /* Assert that pt_regs indicates kernel mode. */
661 addq $8, %rsp /* skip regs->orig_ax */
665 UNWIND_HINT_IRET_REGS
667 * Are we returning to a stack segment from the LDT? Note: in
668 * 64-bit mode SS:RSP on the exception stack is always valid.
670 #ifdef CONFIG_X86_ESPFIX64
671 testb $4, (SS-RIP)(%rsp)
672 jnz native_irq_return_ldt
675 .global native_irq_return_iret
676 native_irq_return_iret:
678 * This may fault. Non-paranoid faults on return to userspace are
679 * handled by fixup_bad_iret. These include #SS, #GP, and #NP.
680 * Double-faults due to espfix64 are handled in do_double_fault.
681 * Other faults here are fatal.
685 #ifdef CONFIG_X86_ESPFIX64
686 native_irq_return_ldt:
688 * We are running with user GSBASE. All GPRs contain their user
689 * values. We have a percpu ESPFIX stack that is eight slots
690 * long (see ESPFIX_STACK_SIZE). espfix_waddr points to the bottom
691 * of the ESPFIX stack.
693 * We clobber RAX and RDI in this code. We stash RDI on the
694 * normal stack and RAX on the ESPFIX stack.
696 * The ESPFIX stack layout we set up looks like this:
698 * --- top of ESPFIX stack ---
703 * RIP <-- RSP points here when we're done
704 * RAX <-- espfix_waddr points here
705 * --- bottom of ESPFIX stack ---
708 pushq %rdi /* Stash user RDI */
709 SWAPGS /* to kernel GS */
710 SWITCH_TO_KERNEL_CR3 scratch_reg=%rdi /* to kernel CR3 */
712 movq PER_CPU_VAR(espfix_waddr), %rdi
713 movq %rax, (0*8)(%rdi) /* user RAX */
714 movq (1*8)(%rsp), %rax /* user RIP */
715 movq %rax, (1*8)(%rdi)
716 movq (2*8)(%rsp), %rax /* user CS */
717 movq %rax, (2*8)(%rdi)
718 movq (3*8)(%rsp), %rax /* user RFLAGS */
719 movq %rax, (3*8)(%rdi)
720 movq (5*8)(%rsp), %rax /* user SS */
721 movq %rax, (5*8)(%rdi)
722 movq (4*8)(%rsp), %rax /* user RSP */
723 movq %rax, (4*8)(%rdi)
724 /* Now RAX == RSP. */
726 andl $0xffff0000, %eax /* RAX = (RSP & 0xffff0000) */
729 * espfix_stack[31:16] == 0. The page tables are set up such that
730 * (espfix_stack | (X & 0xffff0000)) points to a read-only alias of
731 * espfix_waddr for any X. That is, there are 65536 RO aliases of
732 * the same page. Set up RSP so that RSP[31:16] contains the
733 * respective 16 bits of the /userspace/ RSP and RSP nonetheless
734 * still points to an RO alias of the ESPFIX stack.
736 orq PER_CPU_VAR(espfix_stack), %rax
738 SWITCH_TO_USER_CR3_STACK scratch_reg=%rdi
739 SWAPGS /* to user GS */
740 popq %rdi /* Restore user RDI */
743 UNWIND_HINT_IRET_REGS offset=8
746 * At this point, we cannot write to the stack any more, but we can
749 popq %rax /* Restore user RAX */
752 * RSP now points to an ordinary IRET frame, except that the page
753 * is read-only and RSP[31:16] are preloaded with the userspace
754 * values. We can now IRET back to userspace.
756 jmp native_irq_return_iret
758 END(common_interrupt)
763 .macro apicinterrupt3 num sym do_sym
765 UNWIND_HINT_IRET_REGS
774 /* Make sure APIC interrupt handlers end up in the irqentry section: */
775 #define PUSH_SECTION_IRQENTRY .pushsection .irqentry.text, "ax"
776 #define POP_SECTION_IRQENTRY .popsection
778 .macro apicinterrupt num sym do_sym
779 PUSH_SECTION_IRQENTRY
780 apicinterrupt3 \num \sym \do_sym
785 apicinterrupt3 IRQ_MOVE_CLEANUP_VECTOR irq_move_cleanup_interrupt smp_irq_move_cleanup_interrupt
786 apicinterrupt3 REBOOT_VECTOR reboot_interrupt smp_reboot_interrupt
790 apicinterrupt3 UV_BAU_MESSAGE uv_bau_message_intr1 uv_bau_message_interrupt
793 apicinterrupt LOCAL_TIMER_VECTOR apic_timer_interrupt smp_apic_timer_interrupt
794 apicinterrupt X86_PLATFORM_IPI_VECTOR x86_platform_ipi smp_x86_platform_ipi
796 #ifdef CONFIG_HAVE_KVM
797 apicinterrupt3 POSTED_INTR_VECTOR kvm_posted_intr_ipi smp_kvm_posted_intr_ipi
798 apicinterrupt3 POSTED_INTR_WAKEUP_VECTOR kvm_posted_intr_wakeup_ipi smp_kvm_posted_intr_wakeup_ipi
799 apicinterrupt3 POSTED_INTR_NESTED_VECTOR kvm_posted_intr_nested_ipi smp_kvm_posted_intr_nested_ipi
802 #ifdef CONFIG_X86_MCE_THRESHOLD
803 apicinterrupt THRESHOLD_APIC_VECTOR threshold_interrupt smp_threshold_interrupt
806 #ifdef CONFIG_X86_MCE_AMD
807 apicinterrupt DEFERRED_ERROR_VECTOR deferred_error_interrupt smp_deferred_error_interrupt
810 #ifdef CONFIG_X86_THERMAL_VECTOR
811 apicinterrupt THERMAL_APIC_VECTOR thermal_interrupt smp_thermal_interrupt
815 apicinterrupt CALL_FUNCTION_SINGLE_VECTOR call_function_single_interrupt smp_call_function_single_interrupt
816 apicinterrupt CALL_FUNCTION_VECTOR call_function_interrupt smp_call_function_interrupt
817 apicinterrupt RESCHEDULE_VECTOR reschedule_interrupt smp_reschedule_interrupt
820 apicinterrupt ERROR_APIC_VECTOR error_interrupt smp_error_interrupt
821 apicinterrupt SPURIOUS_APIC_VECTOR spurious_interrupt smp_spurious_interrupt
823 #ifdef CONFIG_IRQ_WORK
824 apicinterrupt IRQ_WORK_VECTOR irq_work_interrupt smp_irq_work_interrupt
828 * Exception entry points.
830 #define CPU_TSS_IST(x) PER_CPU_VAR(cpu_tss_rw) + (TSS_ist + ((x) - 1) * 8)
833 * Switch to the thread stack. This is called with the IRET frame and
834 * orig_ax on the stack. (That is, RDI..R12 are not on the stack and
835 * space has not been allocated for them.)
837 ENTRY(switch_to_thread_stack)
841 /* Need to switch before accessing the thread stack. */
842 SWITCH_TO_KERNEL_CR3 scratch_reg=%rdi
844 movq PER_CPU_VAR(cpu_current_top_of_stack), %rsp
845 UNWIND_HINT sp_offset=16 sp_reg=ORC_REG_DI
847 pushq 7*8(%rdi) /* regs->ss */
848 pushq 6*8(%rdi) /* regs->rsp */
849 pushq 5*8(%rdi) /* regs->eflags */
850 pushq 4*8(%rdi) /* regs->cs */
851 pushq 3*8(%rdi) /* regs->ip */
852 pushq 2*8(%rdi) /* regs->orig_ax */
853 pushq 8(%rdi) /* return address */
858 END(switch_to_thread_stack)
860 .macro idtentry sym do_sym has_error_code:req paranoid=0 shift_ist=-1 create_gap=0
862 UNWIND_HINT_IRET_REGS offset=\has_error_code*8
865 .if \shift_ist != -1 && \paranoid == 0
866 .error "using shift_ist requires paranoid=1"
871 .if \has_error_code == 0
872 pushq $-1 /* ORIG_RAX: no syscall to restart */
876 testb $3, CS-ORIG_RAX(%rsp) /* If coming from userspace, switch stacks */
877 jnz .Lfrom_usermode_switch_stack_\@
882 * If coming from kernel space, create a 6-word gap to allow the
883 * int3 handler to emulate a call instruction.
885 testb $3, CS-ORIG_RAX(%rsp)
886 jnz .Lfrom_usermode_no_gap_\@
890 UNWIND_HINT_IRET_REGS offset=8
891 .Lfrom_usermode_no_gap_\@:
900 /* returned flag: ebx=0: need swapgs on exit, ebx=1: don't need it */
904 TRACE_IRQS_OFF_DEBUG /* reload IDT in case of recursion */
910 movq %rsp, %rdi /* pt_regs pointer */
913 movq ORIG_RAX(%rsp), %rsi /* get error code */
914 movq $-1, ORIG_RAX(%rsp) /* no syscall to restart */
916 xorl %esi, %esi /* no error code */
920 subq $EXCEPTION_STKSZ, CPU_TSS_IST(\shift_ist)
926 addq $EXCEPTION_STKSZ, CPU_TSS_IST(\shift_ist)
929 /* these procedures expect "no swapgs" flag in ebx */
938 * Entry from userspace. Switch stacks and treat it
939 * as a normal entry. This means that paranoid handlers
940 * run in real process context if user_mode(regs).
942 .Lfrom_usermode_switch_stack_\@:
945 movq %rsp, %rdi /* pt_regs pointer */
948 movq ORIG_RAX(%rsp), %rsi /* get error code */
949 movq $-1, ORIG_RAX(%rsp) /* no syscall to restart */
951 xorl %esi, %esi /* no error code */
961 idtentry divide_error do_divide_error has_error_code=0
962 idtentry overflow do_overflow has_error_code=0
963 idtentry bounds do_bounds has_error_code=0
964 idtentry invalid_op do_invalid_op has_error_code=0
965 idtentry device_not_available do_device_not_available has_error_code=0
966 idtentry double_fault do_double_fault has_error_code=1 paranoid=2
967 idtentry coprocessor_segment_overrun do_coprocessor_segment_overrun has_error_code=0
968 idtentry invalid_TSS do_invalid_TSS has_error_code=1
969 idtentry segment_not_present do_segment_not_present has_error_code=1
970 idtentry spurious_interrupt_bug do_spurious_interrupt_bug has_error_code=0
971 idtentry coprocessor_error do_coprocessor_error has_error_code=0
972 idtentry alignment_check do_alignment_check has_error_code=1
973 idtentry simd_coprocessor_error do_simd_coprocessor_error has_error_code=0
977 * Reload gs selector with exception handling
980 ENTRY(native_load_gs_index)
983 DISABLE_INTERRUPTS(CLBR_ANY & ~CLBR_RDI)
987 2: ALTERNATIVE "", "mfence", X86_BUG_SWAPGS_FENCE
992 ENDPROC(native_load_gs_index)
993 EXPORT_SYMBOL(native_load_gs_index)
995 _ASM_EXTABLE(.Lgs_change, bad_gs)
996 .section .fixup, "ax"
997 /* running with kernelgs */
999 SWAPGS /* switch back to user gs */
1001 /* This can't be a string because the preprocessor needs to see it. */
1002 movl $__USER_DS, %eax
1005 ALTERNATIVE "", "ZAP_GS", X86_BUG_NULL_SEG
1011 /* Call softirq on interrupt stack. Interrupts are off. */
1012 ENTRY(do_softirq_own_stack)
1015 ENTER_IRQ_STACK regs=0 old_rsp=%r11
1017 LEAVE_IRQ_STACK regs=0
1020 ENDPROC(do_softirq_own_stack)
1023 idtentry hypervisor_callback xen_do_hypervisor_callback has_error_code=0
1026 * A note on the "critical region" in our callback handler.
1027 * We want to avoid stacking callback handlers due to events occurring
1028 * during handling of the last event. To do this, we keep events disabled
1029 * until we've done all processing. HOWEVER, we must enable events before
1030 * popping the stack frame (can't be done atomically) and so it would still
1031 * be possible to get enough handler activations to overflow the stack.
1032 * Although unlikely, bugs of that kind are hard to track down, so we'd
1033 * like to avoid the possibility.
1034 * So, on entry to the handler we detect whether we interrupted an
1035 * existing activation in its critical region -- if so, we pop the current
1036 * activation and restart the handler using the previous one.
1038 ENTRY(xen_do_hypervisor_callback) /* do_hypervisor_callback(struct *pt_regs) */
1041 * Since we don't modify %rdi, evtchn_do_upall(struct *pt_regs) will
1042 * see the correct pointer to the pt_regs
1045 movq %rdi, %rsp /* we don't return, adjust the stack frame */
1048 ENTER_IRQ_STACK old_rsp=%r10
1049 call xen_evtchn_do_upcall
1052 #ifndef CONFIG_PREEMPT
1053 call xen_maybe_preempt_hcall
1056 END(xen_do_hypervisor_callback)
1059 * Hypervisor uses this for application faults while it executes.
1060 * We get here for two reasons:
1061 * 1. Fault while reloading DS, ES, FS or GS
1062 * 2. Fault while executing IRET
1063 * Category 1 we do not need to fix up as Xen has already reloaded all segment
1064 * registers that could be reloaded and zeroed the others.
1065 * Category 2 we fix up by killing the current process. We cannot use the
1066 * normal Linux return path in this case because if we use the IRET hypercall
1067 * to pop the stack frame we end up in an infinite loop of failsafe callbacks.
1068 * We distinguish between categories by comparing each saved segment register
1069 * with its current contents: any discrepancy means we in category 1.
1071 ENTRY(xen_failsafe_callback)
1074 cmpw %cx, 0x10(%rsp)
1077 cmpw %cx, 0x18(%rsp)
1080 cmpw %cx, 0x20(%rsp)
1083 cmpw %cx, 0x28(%rsp)
1085 /* All segments match their saved values => Category 2 (Bad IRET). */
1090 UNWIND_HINT_IRET_REGS offset=8
1091 jmp general_protection
1092 1: /* Segment mismatch => Category 1 (Bad segment). Retry the IRET. */
1096 UNWIND_HINT_IRET_REGS
1097 pushq $-1 /* orig_ax = -1 => not a system call */
1099 ENCODE_FRAME_POINTER
1101 END(xen_failsafe_callback)
1103 apicinterrupt3 HYPERVISOR_CALLBACK_VECTOR \
1104 xen_hvm_callback_vector xen_evtchn_do_upcall
1106 #endif /* CONFIG_XEN */
1108 #if IS_ENABLED(CONFIG_HYPERV)
1109 apicinterrupt3 HYPERVISOR_CALLBACK_VECTOR \
1110 hyperv_callback_vector hyperv_vector_handler
1111 #endif /* CONFIG_HYPERV */
1113 idtentry debug do_debug has_error_code=0 paranoid=1 shift_ist=DEBUG_STACK
1114 idtentry int3 do_int3 has_error_code=0 create_gap=1
1115 idtentry stack_segment do_stack_segment has_error_code=1
1118 idtentry xennmi do_nmi has_error_code=0
1119 idtentry xendebug do_debug has_error_code=0
1122 idtentry general_protection do_general_protection has_error_code=1
1123 idtentry page_fault do_page_fault has_error_code=1
1125 #ifdef CONFIG_KVM_GUEST
1126 idtentry async_page_fault do_async_page_fault has_error_code=1
1129 #ifdef CONFIG_X86_MCE
1130 idtentry machine_check do_mce has_error_code=0 paranoid=1
1134 * Save all registers in pt_regs, and switch gs if needed.
1135 * Use slow, but surefire "are we in kernel?" check.
1136 * Return: ebx=0: need swapgs on exit, ebx=1: otherwise
1138 ENTRY(paranoid_entry)
1141 PUSH_AND_CLEAR_REGS save_ret=1
1142 ENCODE_FRAME_POINTER 8
1144 movl $MSR_GS_BASE, %ecx
1147 js 1f /* negative -> in kernel */
1152 SAVE_AND_SWITCH_TO_KERNEL_CR3 scratch_reg=%rax save_reg=%r14
1154 * The above SAVE_AND_SWITCH_TO_KERNEL_CR3 macro doesn't do an
1155 * unconditional CR3 write, even in the PTI case. So do an lfence
1156 * to prevent GS speculation, regardless of whether PTI is enabled.
1158 FENCE_SWAPGS_KERNEL_ENTRY
1164 * "Paranoid" exit path from exception stack. This is invoked
1165 * only on return from non-NMI IST interrupts that came
1166 * from kernel space.
1168 * We may be returning to very strange contexts (e.g. very early
1169 * in syscall entry), so checking for preemption here would
1170 * be complicated. Fortunately, we there's no good reason
1171 * to try to handle preemption here.
1173 * On entry, ebx is "no swapgs" flag (1: don't need swapgs, 0: need it)
1175 ENTRY(paranoid_exit)
1177 DISABLE_INTERRUPTS(CLBR_ANY)
1178 TRACE_IRQS_OFF_DEBUG
1179 testl %ebx, %ebx /* swapgs needed? */
1180 jnz .Lparanoid_exit_no_swapgs
1182 RESTORE_CR3 scratch_reg=%rbx save_reg=%r14
1184 jmp .Lparanoid_exit_restore
1185 .Lparanoid_exit_no_swapgs:
1186 TRACE_IRQS_IRETQ_DEBUG
1187 RESTORE_CR3 scratch_reg=%rbx save_reg=%r14
1188 .Lparanoid_exit_restore:
1189 jmp restore_regs_and_return_to_kernel
1193 * Save all registers in pt_regs, and switch GS if needed.
1198 PUSH_AND_CLEAR_REGS save_ret=1
1199 ENCODE_FRAME_POINTER 8
1200 testb $3, CS+8(%rsp)
1201 jz .Lerror_kernelspace
1204 * We entered from user mode or we're pretending to have entered
1205 * from user mode due to an IRET fault.
1208 FENCE_SWAPGS_USER_ENTRY
1209 /* We have user CR3. Change to kernel CR3. */
1210 SWITCH_TO_KERNEL_CR3 scratch_reg=%rax
1212 .Lerror_entry_from_usermode_after_swapgs:
1213 /* Put us onto the real thread stack. */
1214 popq %r12 /* save return addr in %12 */
1215 movq %rsp, %rdi /* arg0 = pt_regs pointer */
1217 movq %rax, %rsp /* switch stack */
1218 ENCODE_FRAME_POINTER
1222 * We need to tell lockdep that IRQs are off. We can't do this until
1223 * we fix gsbase, and we should do it before enter_from_user_mode
1224 * (which can take locks).
1227 CALL_enter_from_user_mode
1230 .Lerror_entry_done_lfence:
1231 FENCE_SWAPGS_KERNEL_ENTRY
1237 * There are two places in the kernel that can potentially fault with
1238 * usergs. Handle them here. B stepping K8s sometimes report a
1239 * truncated RIP for IRET exceptions returning to compat mode. Check
1240 * for these here too.
1242 .Lerror_kernelspace:
1243 leaq native_irq_return_iret(%rip), %rcx
1244 cmpq %rcx, RIP+8(%rsp)
1246 movl %ecx, %eax /* zero extend */
1247 cmpq %rax, RIP+8(%rsp)
1249 cmpq $.Lgs_change, RIP+8(%rsp)
1250 jne .Lerror_entry_done_lfence
1253 * hack: .Lgs_change can fail with user gsbase. If this happens, fix up
1254 * gsbase and proceed. We'll fix up the exception and land in
1255 * .Lgs_change's error handler with kernel gsbase.
1258 FENCE_SWAPGS_USER_ENTRY
1259 SWITCH_TO_KERNEL_CR3 scratch_reg=%rax
1260 jmp .Lerror_entry_done
1263 /* Fix truncated RIP */
1264 movq %rcx, RIP+8(%rsp)
1269 * We came from an IRET to user mode, so we have user
1270 * gsbase and CR3. Switch to kernel gsbase and CR3:
1273 FENCE_SWAPGS_USER_ENTRY
1274 SWITCH_TO_KERNEL_CR3 scratch_reg=%rax
1277 * Pretend that the exception came from user mode: set up pt_regs
1278 * as if we faulted immediately after IRET.
1283 jmp .Lerror_entry_from_usermode_after_swapgs
1288 DISABLE_INTERRUPTS(CLBR_ANY)
1296 * Runs on exception stack. Xen PV does not go through this path at all,
1297 * so we can use real assembly here.
1300 * %r14: Used to save/restore the CR3 of the interrupted context
1301 * when PAGE_TABLE_ISOLATION is in use. Do not clobber.
1304 UNWIND_HINT_IRET_REGS
1307 * We allow breakpoints in NMIs. If a breakpoint occurs, then
1308 * the iretq it performs will take us out of NMI context.
1309 * This means that we can have nested NMIs where the next
1310 * NMI is using the top of the stack of the previous NMI. We
1311 * can't let it execute because the nested NMI will corrupt the
1312 * stack of the previous NMI. NMI handlers are not re-entrant
1315 * To handle this case we do the following:
1316 * Check the a special location on the stack that contains
1317 * a variable that is set when NMIs are executing.
1318 * The interrupted task's stack is also checked to see if it
1320 * If the variable is not set and the stack is not the NMI
1322 * o Set the special variable on the stack
1323 * o Copy the interrupt frame into an "outermost" location on the
1325 * o Copy the interrupt frame into an "iret" location on the stack
1326 * o Continue processing the NMI
1327 * If the variable is set or the previous stack is the NMI stack:
1328 * o Modify the "iret" location to jump to the repeat_nmi
1329 * o return back to the first NMI
1331 * Now on exit of the first NMI, we first clear the stack variable
1332 * The NMI stack will tell any nested NMIs at that point that it is
1333 * nested. Then we pop the stack normally with iret, and if there was
1334 * a nested NMI that updated the copy interrupt stack frame, a
1335 * jump will be made to the repeat_nmi code that will handle the second
1338 * However, espfix prevents us from directly returning to userspace
1339 * with a single IRET instruction. Similarly, IRET to user mode
1340 * can fault. We therefore handle NMIs from user space like
1341 * other IST entries.
1346 /* Use %rdx as our temp variable throughout */
1349 testb $3, CS-RIP+8(%rsp)
1350 jz .Lnmi_from_kernel
1353 * NMI from user mode. We need to run on the thread stack, but we
1354 * can't go through the normal entry paths: NMIs are masked, and
1355 * we don't want to enable interrupts, because then we'll end
1356 * up in an awkward situation in which IRQs are on but NMIs
1359 * We also must not push anything to the stack before switching
1360 * stacks lest we corrupt the "NMI executing" variable.
1365 FENCE_SWAPGS_USER_ENTRY
1366 SWITCH_TO_KERNEL_CR3 scratch_reg=%rdx
1368 movq PER_CPU_VAR(cpu_current_top_of_stack), %rsp
1369 UNWIND_HINT_IRET_REGS base=%rdx offset=8
1370 pushq 5*8(%rdx) /* pt_regs->ss */
1371 pushq 4*8(%rdx) /* pt_regs->rsp */
1372 pushq 3*8(%rdx) /* pt_regs->flags */
1373 pushq 2*8(%rdx) /* pt_regs->cs */
1374 pushq 1*8(%rdx) /* pt_regs->rip */
1375 UNWIND_HINT_IRET_REGS
1376 pushq $-1 /* pt_regs->orig_ax */
1377 PUSH_AND_CLEAR_REGS rdx=(%rdx)
1378 ENCODE_FRAME_POINTER
1381 * At this point we no longer need to worry about stack damage
1382 * due to nesting -- we're on the normal thread stack and we're
1383 * done with the NMI stack.
1391 * Return back to user mode. We must *not* do the normal exit
1392 * work, because we don't want to enable interrupts.
1394 jmp swapgs_restore_regs_and_return_to_usermode
1398 * Here's what our stack frame will look like:
1399 * +---------------------------------------------------------+
1401 * | original Return RSP |
1402 * | original RFLAGS |
1405 * +---------------------------------------------------------+
1406 * | temp storage for rdx |
1407 * +---------------------------------------------------------+
1408 * | "NMI executing" variable |
1409 * +---------------------------------------------------------+
1410 * | iret SS } Copied from "outermost" frame |
1411 * | iret Return RSP } on each loop iteration; overwritten |
1412 * | iret RFLAGS } by a nested NMI to force another |
1413 * | iret CS } iteration if needed. |
1415 * +---------------------------------------------------------+
1416 * | outermost SS } initialized in first_nmi; |
1417 * | outermost Return RSP } will not be changed before |
1418 * | outermost RFLAGS } NMI processing is done. |
1419 * | outermost CS } Copied to "iret" frame on each |
1420 * | outermost RIP } iteration. |
1421 * +---------------------------------------------------------+
1423 * +---------------------------------------------------------+
1425 * The "original" frame is used by hardware. Before re-enabling
1426 * NMIs, we need to be done with it, and we need to leave enough
1427 * space for the asm code here.
1429 * We return by executing IRET while RSP points to the "iret" frame.
1430 * That will either return for real or it will loop back into NMI
1433 * The "outermost" frame is copied to the "iret" frame on each
1434 * iteration of the loop, so each iteration starts with the "iret"
1435 * frame pointing to the final return target.
1439 * Determine whether we're a nested NMI.
1441 * If we interrupted kernel code between repeat_nmi and
1442 * end_repeat_nmi, then we are a nested NMI. We must not
1443 * modify the "iret" frame because it's being written by
1444 * the outer NMI. That's okay; the outer NMI handler is
1445 * about to about to call do_nmi anyway, so we can just
1446 * resume the outer NMI.
1449 movq $repeat_nmi, %rdx
1452 movq $end_repeat_nmi, %rdx
1458 * Now check "NMI executing". If it's set, then we're nested.
1459 * This will not detect if we interrupted an outer NMI just
1466 * Now test if the previous stack was an NMI stack. This covers
1467 * the case where we interrupt an outer NMI after it clears
1468 * "NMI executing" but before IRET. We need to be careful, though:
1469 * there is one case in which RSP could point to the NMI stack
1470 * despite there being no NMI active: naughty userspace controls
1471 * RSP at the very beginning of the SYSCALL targets. We can
1472 * pull a fast one on naughty userspace, though: we program
1473 * SYSCALL to mask DF, so userspace cannot cause DF to be set
1474 * if it controls the kernel's RSP. We set DF before we clear
1478 /* Compare the NMI stack (rdx) with the stack we came from (4*8(%rsp)) */
1479 cmpq %rdx, 4*8(%rsp)
1480 /* If the stack pointer is above the NMI stack, this is a normal NMI */
1483 subq $EXCEPTION_STKSZ, %rdx
1484 cmpq %rdx, 4*8(%rsp)
1485 /* If it is below the NMI stack, it is a normal NMI */
1488 /* Ah, it is within the NMI stack. */
1490 testb $(X86_EFLAGS_DF >> 8), (3*8 + 1)(%rsp)
1491 jz first_nmi /* RSP was user controlled. */
1493 /* This is a nested NMI. */
1497 * Modify the "iret" frame to point to repeat_nmi, forcing another
1498 * iteration of NMI handling.
1501 leaq -10*8(%rsp), %rdx
1508 /* Put stack back */
1514 /* We are returning to kernel mode, so this cannot result in a fault. */
1521 /* Make room for "NMI executing". */
1524 /* Leave room for the "iret" frame */
1527 /* Copy the "original" frame to the "outermost" frame */
1531 UNWIND_HINT_IRET_REGS
1533 /* Everything up to here is safe from nested NMIs */
1535 #ifdef CONFIG_DEBUG_ENTRY
1537 * For ease of testing, unmask NMIs right away. Disabled by
1538 * default because IRET is very expensive.
1541 pushq %rsp /* RSP (minus 8 because of the previous push) */
1542 addq $8, (%rsp) /* Fix up RSP */
1544 pushq $__KERNEL_CS /* CS */
1546 iretq /* continues at repeat_nmi below */
1547 UNWIND_HINT_IRET_REGS
1553 * If there was a nested NMI, the first NMI's iret will return
1554 * here. But NMIs are still enabled and we can take another
1555 * nested NMI. The nested NMI checks the interrupted RIP to see
1556 * if it is between repeat_nmi and end_repeat_nmi, and if so
1557 * it will just return, as we are about to repeat an NMI anyway.
1558 * This makes it safe to copy to the stack frame that a nested
1561 * RSP is pointing to "outermost RIP". gsbase is unknown, but, if
1562 * we're repeating an NMI, gsbase has the same value that it had on
1563 * the first iteration. paranoid_entry will load the kernel
1564 * gsbase if needed before we call do_nmi. "NMI executing"
1567 movq $1, 10*8(%rsp) /* Set "NMI executing". */
1570 * Copy the "outermost" frame to the "iret" frame. NMIs that nest
1571 * here must not modify the "iret" frame while we're writing to
1572 * it or it will end up containing garbage.
1582 * Everything below this point can be preempted by a nested NMI.
1583 * If this happens, then the inner NMI will change the "iret"
1584 * frame to point back to repeat_nmi.
1586 pushq $-1 /* ORIG_RAX: no syscall to restart */
1589 * Use paranoid_entry to handle SWAPGS, but no need to use paranoid_exit
1590 * as we should not be calling schedule in NMI context.
1591 * Even with normal interrupts enabled. An NMI should not be
1592 * setting NEED_RESCHED or anything that normal interrupts and
1593 * exceptions might do.
1598 /* paranoidentry do_nmi, 0; without TRACE_IRQS_OFF */
1603 RESTORE_CR3 scratch_reg=%r15 save_reg=%r14
1605 testl %ebx, %ebx /* swapgs needed? */
1613 * Skip orig_ax and the "outermost" frame to point RSP at the "iret"
1614 * at the "iret" frame.
1619 * Clear "NMI executing". Set DF first so that we can easily
1620 * distinguish the remaining code between here and IRET from
1621 * the SYSCALL entry and exit paths.
1623 * We arguably should just inspect RIP instead, but I (Andy) wrote
1624 * this code when I had the misapprehension that Xen PV supported
1625 * NMIs, and Xen PV would break that approach.
1628 movq $0, 5*8(%rsp) /* clear "NMI executing" */
1631 * iretq reads the "iret" frame and exits the NMI stack in a
1632 * single instruction. We are returning to kernel mode, so this
1633 * cannot result in a fault. Similarly, we don't need to worry
1634 * about espfix64 on the way back to kernel mode.
1639 ENTRY(ignore_sysret)
1645 ENTRY(rewind_stack_do_exit)
1647 /* Prevent any naive code from trying to unwind to our caller. */
1650 movq PER_CPU_VAR(cpu_current_top_of_stack), %rax
1651 leaq -PTREGS_SIZE(%rax), %rsp
1655 END(rewind_stack_do_exit)