root/sys/arch/amd64/amd64/spl.S
/*      $NetBSD: spl.S,v 1.50 2026/01/17 10:53:19 bouyer Exp $  */

/*
 * Copyright (c) 2003 Wasabi Systems, Inc.
 * All rights reserved.
 *
 * Written by Frank van der Linden for Wasabi Systems, Inc.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 * 3. All advertising materials mentioning features or use of this software
 *    must display the following acknowledgement:
 *      This product includes software developed for the NetBSD Project by
 *      Wasabi Systems, Inc.
 * 4. The name of Wasabi Systems, Inc. may not be used to endorse
 *    or promote products derived from this software without specific prior
 *    written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY WASABI SYSTEMS, INC. ``AS IS'' AND
 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
 * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL WASABI SYSTEMS, INC
 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
 * POSSIBILITY OF SUCH DAMAGE.
 */

/*
 * Copyright (c) 1998, 2007, 2008 The NetBSD Foundation, Inc.
 * All rights reserved.
 *
 * This code is derived from software contributed to The NetBSD Foundation
 * by Charles M. Hannum and Andrew Doran.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 *
 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
 * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
 * POSSIBILITY OF SUCH DAMAGE.
 */

#include "opt_ddb.h"
#include "opt_kasan.h"
#include "opt_kmsan.h"

#define ALIGN_TEXT      .align 16,0x90

#include <machine/asm.h>
#include <machine/trap.h>
#include <machine/segments.h>
#include <machine/frameasm.h>

#include "assym.h"

        .text

/*
 * int splraise(int s);
 */
ENTRY(splraise)
        movzbl  CPUVAR(ILEVEL),%eax
        cmpl    %edi,%eax
        cmoval  %eax,%edi
        movb    %dil,CPUVAR(ILEVEL)
        KMSAN_INIT_RET(4)
        ret
END(splraise)

/*
 * Xsoftintr()
 *
 * Switch to the LWP assigned to handle interrupts from the given
 * source.  We borrow the VM context from the interrupted LWP.
 *
 * On entry:
 *
 *      %rax            intrsource
 *      %r13            address to return to
 */
IDTVEC(softintr)
        /* set up struct switchframe */
        pushq   $_C_LABEL(softintr_ret)
        pushq   %rbx
        pushq   %r12
        pushq   %r13
        pushq   %r14
        pushq   %r15

        movb    $IPL_HIGH,CPUVAR(ILEVEL)
        movq    CPUVAR(CURLWP),%r15
        movq    IS_LWP(%rax),%rdi       /* switch to handler LWP */
        movq    L_PCB(%rdi),%rdx
        movq    L_PCB(%r15),%rcx
        /*
         * Simple MOV to set curlwp to softlwp.  See below on ordering
         * required to restore softlwp like cpu_switchto.
         *
         * 1. Don't need store-before-store barrier because x86 is TSO.
         *
         * 2. Don't need store-before-load barrier because when we
         *    enter a softint lwp, it can't be holding any mutexes, so
         *    it can't release any until after it has acquired them, so
         *    we need not participate in the protocol with
         *    mutex_vector_enter barriers here.
         *
         * Hence no need for XCHG or barriers around MOV.
         */
        movq    %rdi,CPUVAR(CURLWP)

#ifdef KASAN
        /* clear the new stack */
        pushq   %rax
        pushq   %rdx
        pushq   %rcx
        callq   _C_LABEL(kasan_softint)
        popq    %rcx
        popq    %rdx
        popq    %rax
#endif

#ifdef KMSAN
        pushq   %rax
        pushq   %rdx
        pushq   %rcx
        callq   _C_LABEL(kmsan_softint)
        popq    %rcx
        popq    %rdx
        popq    %rax
#endif

        /* save old context */
        movq    %rsp,PCB_RSP(%rcx)
        movq    %rbp,PCB_RBP(%rcx)

        /* switch to the new stack */
        movq    PCB_RSP0(%rdx),%rsp

        /* dispatch */
        STI(di)
        movq    %r15,%rdi               /* interrupted LWP */
        movl    IS_MAXLEVEL(%rax),%esi  /* ipl to run at */
        call    _C_LABEL(softint_dispatch)/* run handlers */
        CLI(di)

        /* restore old context */
        movq    L_PCB(%r15),%rcx
        movq    PCB_RSP(%rcx),%rsp

        /*
         * Use XCHG, not MOV, to coordinate mutex_exit on this CPU with
         * mutex_vector_enter on another CPU.
         *
         * 1. Any prior mutex_exit by the softint must be visible to
         *    other CPUs before we restore curlwp on this one,
         *    requiring store-before-store ordering.
         *
         *    (This is always guaranteed by the x86 memory model, TSO,
         *    but other architectures require a explicit barrier before
         *    the store to ci->ci_curlwp.)
         *
         * 2. Restoring curlwp must be visible on all other CPUs before
         *    any subsequent mutex_exit on this one can even test
         *    whether there might be waiters, requiring
         *    store-before-load ordering.
         *
         *    (This is the only ordering x86 TSO ever requires any kind
         *    of barrier for -- in this case, we take advantage of the
         *    sequential consistency implied by XCHG to obviate the
         *    need for MFENCE or something.)
         *
         * See kern_mutex.c for details -- this is necessary for
         * adaptive mutexes to detect whether the lwp is on the CPU in
         * order to safely block without requiring atomic r/m/w in
         * mutex_exit.  See also cpu_switchto.
         */
        xchgq   %r15,CPUVAR(CURLWP)     /* restore curlwp */
        popq    %r15                    /* unwind switchframe */
        addq    $(5 * 8),%rsp
        jmp     *%r13                   /* back to Xspllower/Xdoreti */
IDTVEC_END(softintr)

/*
 * softintr_ret()
 *
 * Trampoline function that gets returned to by cpu_switchto() when
 * an interrupt handler blocks.  On entry:
 *
 *      %rax            prevlwp from cpu_switchto()
 */
ENTRY(softintr_ret)
        incl    CPUVAR(MTX_COUNT)       /* re-adjust after mi_switch */
        CLI(ax)         /* %rax not used by Xspllower/Xdoreti */
        jmp     *%r13                   /* back to Xspllower/Xdoreti */
END(softintr_ret)

/*
 * void softint_trigger(uintptr_t machdep);
 *
 * Software interrupt registration.
 */
ENTRY(softint_trigger)
        shlq    $8,%rdi                 /* clear upper 8 bits */
        shrq    $8,%rdi
        orq     %rdi,CPUVAR(IPENDING)   /* atomic on local cpu */
        ret
END(softint_trigger)

/*
 * Xrecurse_preempt()
 *
 * Handles preemption interrupts via Xspllower().
 */
IDTVEC(recurse_preempt)
        movb    $IPL_PREEMPT,CPUVAR(ILEVEL)
        STI(di)
        xorq    %rdi,%rdi
        KMSAN_INIT_ARG(8)
        call    _C_LABEL(kpreempt)
        CLI(di)
        jmp     *%r13                   /* back to Xspllower */
IDTVEC_END(recurse_preempt)

/*
 * Xresume_preempt()
 *
 * Handles preemption interrupts via Xdoreti().
 */
IDTVEC(resume_preempt)
        movb    $IPL_PREEMPT,CPUVAR(ILEVEL)
        STI(ax)
        testq   $SEL_RPL,TF_CS(%rsp)
        jnz     1f
        movq    TF_RIP(%rsp),%rdi
        KMSAN_INIT_ARG(8)
        call    _C_LABEL(kpreempt)      /* from kernel */
        CLI(ax)
        jmp     *%r13                   /* back to Xdoreti */
1:
        call    _C_LABEL(preempt)       /* from user */
        CLI(ax)
        jmp     *%r13                   /* back to Xdoreti */
IDTVEC_END(resume_preempt)

/*
 * void spllower(int s);
 */
ENTRY(spllower)
        movzbl  CPUVAR(ILEVEL),%edx
        cmpl    %edx,%edi                       /* new level is lower? */
        jae     1f
        xorq    %rcx,%rcx                       /* rcx: ci_ipending mask */
        notq    %rcx
        shrq    $8,%rcx
        movq    %rdi,%r9                        /* r9: shifted new level */
        shlq    $56,%r9
0:
        movq    CPUVAR(IPENDING),%rax
        testq   %rax,CPUVAR(IUNMASK)(,%rdi,8)   /* deferred interrupts? */
        /*
         * On the P4 this jump is cheaper than patching in junk
         * using cmov.  Is cmpxchg expensive if it fails?
         */
        jnz     2f
        movq    %rax,%r8
        andq    %rcx,%r8
        orq     %r9,%r8
        cmpxchgq %r8,CPUVAR(ISTATE)             /* swap in new ilevel */
        jnz     0b
1:
        ret
2:
        jmp     _C_LABEL(Xspllower)
END(spllower)

/*
 * void Xspllower(int s);
 *
 * Process pending interrupts.
 *
 * Important registers:
 *   ebx - cpl
 *   r13 - address to resume loop at
 *
 * It is important that the bit scan instruction is bsr, it will get
 * the highest 2 bits (currently the IPI and clock handlers) first,
 * to avoid deadlocks where one CPU sends an IPI, another one is at
 * splhigh() and defers it, lands in here via splx(), and handles
 * a lower-prio one first, which needs to take the kernel lock -->
 * the sending CPU will never see the that CPU accept the IPI
 * (see pmap_tlb_shootnow).
 * XenPV note: it seems that we can be called at spl0, so need to use CLI2()
 */
IDTVEC(spllower)
        pushq   %rbx
        pushq   %r13
        pushq   %r12
        movl    %edi,%ebx
        leaq    1f(%rip),%r13           /* address to resume loop at */
1:
        movl    %ebx,%eax               /* get cpl */
        movq    CPUVAR(IUNMASK)(,%rax,8),%rax
        CLI2(si, r12)
        andq    CPUVAR(IPENDING),%rax   /* any non-masked bits left? */
        jz      2f
        bsrq    %rax,%rax
        btrq    %rax,CPUVAR(IPENDING)
        movq    CPUVAR(ISOURCES)(,%rax,8),%rax
        jmp     *IS_RECURSE(%rax)
2:
        movb    %bl,CPUVAR(ILEVEL)
        STI(si)
        popq    %r12
        popq    %r13
        popq    %rbx
        ret
IDTVEC_END(spllower)

/*
 * void Xdoreti(void);
 *
 * Handle return from interrupt after device handler finishes.
 *
 * Important registers:
 *   ebx - cpl to restore
 *   r13 - address to resume loop at
 *   rsi, rdi and r14 - free for local use
 */
IDTVEC(doreti)
        popq    %rbx                    /* get previous priority */
        decl    CPUVAR(IDEPTH)
        leaq    1f(%rip),%r13
1:
        movl    %ebx,%eax
        movq    CPUVAR(IUNMASK)(,%rax,8),%rax
        CLI2(si, r14)
        andq    CPUVAR(IPENDING),%rax
        jz      2f
        bsrq    %rax,%rax               /* slow, but not worth optimizing */
        btrq    %rax,CPUVAR(IPENDING)
        movq    CPUVAR(ISOURCES)(,%rax,8),%rax
        jmp     *IS_RESUME(%rax)
2:      /* Check for ASTs on exit to user mode. */
        movb    %bl,CPUVAR(ILEVEL)
5:
        testb   $SEL_RPL,TF_CS(%rsp)
        jz      6f
.Ldoreti_checkast:
        movq    CPUVAR(CURLWP),%r14
        CHECK_ASTPENDING(%r14)
        je      3f
        CLEAR_ASTPENDING(%r14)
        STI(si)
        movl    $T_ASTFLT,TF_TRAPNO(%rsp)       /* XXX undo later.. */
        /* Pushed T_ASTFLT into tf_trapno on entry. */
        movq    %rsp,%rdi
        KMSAN_INIT_ARG(8)
        call    _C_LABEL(trap)
        CLI2(si, rdi)
        jmp     .Ldoreti_checkast
3:
        CHECK_DEFERRED_SWITCH
        jnz     9f
        HANDLE_DEFERRED_FPU
6:
        INTRFASTEXIT
9:
        STI(si)
        call    _C_LABEL(do_pmap_load)
        CLI2(si, rdi)
        jmp     .Ldoreti_checkast               /* recheck ASTs */
IDTVEC_END(doreti)