kernel - Implement spectre mitigations part 1

[dragonfly.git] / sys / platform / pc64 / x86_64 / vm_machdep.c

/*-
 * Copyright (c) 1982, 1986 The Regents of the University of California.
 * Copyright (c) 1989, 1990 William Jolitz
 * Copyright (c) 1994 John Dyson
 * Copyright (c) 2008-2018 The DragonFly Project.
 * All rights reserved.
 *
 * This code is derived from software contributed to Berkeley by
 * the Systems Programming Group of the University of Utah Computer
 * Science Department, and William Jolitz.
 *
 * 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 by the University of
 *      California, Berkeley and its contributors.
 * 4. Neither the name of the University nor the names of its contributors
 *    may be used to endorse or promote products derived from this software
 *    without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE REGENTS 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 REGENTS 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.
 *
 *      from: @(#)vm_machdep.c  7.3 (Berkeley) 5/13/91
 *      Utah $Hdr: vm_machdep.c 1.16.1.1 89/06/23$
 * $FreeBSD: src/sys/i386/i386/vm_machdep.c,v 1.132.2.9 2003/01/25 19:02:23 dillon Exp $
 */

#include <sys/param.h>
#include <sys/systm.h>
#include <sys/malloc.h>
#include <sys/proc.h>
#include <sys/buf.h>
#include <sys/interrupt.h>
#include <sys/vnode.h>
#include <sys/vmmeter.h>
#include <sys/kernel.h>
#include <sys/sysctl.h>
#include <sys/unistd.h>
#include <sys/lwp.h>

#include <machine/clock.h>
#include <machine/cpu.h>
#include <machine/md_var.h>
#include <machine/smp.h>
#include <machine/pcb.h>
#include <machine/pcb_ext.h>
#include <machine/segments.h>
#include <machine/globaldata.h> /* npxthread */
#include <machine/vmm.h>

#include <vm/vm.h>
#include <vm/vm_param.h>
#include <sys/lock.h>
#include <vm/vm_kern.h>
#include <vm/vm_page.h>
#include <vm/vm_map.h>
#include <vm/vm_extern.h>

#include <sys/thread2.h>
#include <sys/mplock2.h>

#include <bus/isa/isa.h>

static void     cpu_reset_real (void);

int spectre_mitigation = -1;

/*
 * Finish a fork operation, with lwp lp2 nearly set up.
 * Copy and update the pcb, set up the stack so that the child
 * ready to run and return to user mode.
 */
void
cpu_fork(struct lwp *lp1, struct lwp *lp2, int flags)
{
        struct pcb *pcb2;
        struct pmap *pmap2;

        if ((flags & RFPROC) == 0) {
                if ((flags & RFMEM) == 0) {
                        /*
                         * Unshare user LDT.  > 1 test is MPSAFE.  While
                         * it can potentially race a 2->1 transition, the
                         * worst that happens is that we do an unnecessary
                         * ldt replacement.
                         */
                        struct pcb *pcb1 = lp1->lwp_thread->td_pcb;
                        struct pcb_ldt *pcb_ldt = pcb1->pcb_ldt;

                        if (pcb_ldt && pcb_ldt->ldt_refcnt > 1) {
                                pcb_ldt = user_ldt_alloc(pcb1,pcb_ldt->ldt_len);
                                user_ldt_free(pcb1);
                                pcb1->pcb_ldt = pcb_ldt;
                                set_user_ldt(pcb1);
                        }
                }
                return;
        }

        /* Ensure that lp1's pcb is up to date. */
        if (mdcpu->gd_npxthread == lp1->lwp_thread)
                npxsave(lp1->lwp_thread->td_savefpu);

        /*
         * Copy lp1's PCB.  This really only applies to the
         * debug registers and FP state, but its faster to just copy the
         * whole thing.  Because we only save the PCB at switchout time,
         * the register state may not be current.
         */
        pcb2 = lp2->lwp_thread->td_pcb;
        *pcb2 = *lp1->lwp_thread->td_pcb;

        /*
         * Create a new fresh stack for the new process.
         * Copy the trap frame for the return to user mode as if from a
         * syscall.  This copies the user mode register values.
         *
         * pcb_rsp must allocate an additional call-return pointer below
         * the trap frame which will be restored by cpu_heavy_restore from
         * PCB_RIP, and the thread's td_sp pointer must allocate an
         * additonal two quadwords below the pcb_rsp call-return pointer to
         * hold the LWKT restore function pointer and rflags.
         *
         * The LWKT restore function pointer must be set to cpu_heavy_restore,
         * which is our standard heavy-weight process switch-in function.
         * YYY eventually we should shortcut fork_return and fork_trampoline
         * to use the LWKT restore function directly so we can get rid of
         * all the extra crap we are setting up.
         */
        lp2->lwp_md.md_regs = (struct trapframe *)pcb2 - 1;
        bcopy(lp1->lwp_md.md_regs, lp2->lwp_md.md_regs, sizeof(*lp2->lwp_md.md_regs));

        /*
         * Set registers for trampoline to user mode.  Leave space for the
         * return address on stack.  These are the kernel mode register values.
         *
         * Set the new pmap CR3.  If the new process uses isolated VM spaces,
         * also set the isolated CR3.
         */
        pmap2 = vmspace_pmap(lp2->lwp_proc->p_vmspace);
        pcb2->pcb_cr3 = vtophys(pmap2->pm_pml4);
        if ((pcb2->pcb_flags & PCB_ISOMMU) && pmap2->pm_pmlpv_iso) {
                pcb2->pcb_cr3_iso = vtophys(pmap2->pm_pml4_iso);
        } else {
                pcb2->pcb_flags &= ~PCB_ISOMMU;
                pcb2->pcb_cr3_iso = 0;
        }

#if 0
        /*
         * Per-process spectre mitigation (future)
         */
        pcb2->pcb_flags &= ~(PCB_IBRS1 | PCB_IBRS2);
        switch (spectre_mitigation) {
        case 1:
                pcb2->pcb_flags |= PCB_IBRS1;
                break;
        case 2:
                pcb2->pcb_flags |= PCB_IBRS2;
                break;
        default:
                break;
        }
#endif

        pcb2->pcb_rbx = (unsigned long)fork_return;     /* fork_trampoline argument */
        pcb2->pcb_rbp = 0;
        pcb2->pcb_rsp = (unsigned long)lp2->lwp_md.md_regs - sizeof(void *);
        pcb2->pcb_r12 = (unsigned long)lp2;             /* fork_trampoline argument */
        pcb2->pcb_r13 = 0;
        pcb2->pcb_r14 = 0;
        pcb2->pcb_r15 = 0;
        pcb2->pcb_rip = (unsigned long)fork_trampoline;
        lp2->lwp_thread->td_sp = (char *)(pcb2->pcb_rsp - sizeof(void *));
        *(u_int64_t *)lp2->lwp_thread->td_sp = PSL_USER;
        lp2->lwp_thread->td_sp -= sizeof(void *);
        *(void **)lp2->lwp_thread->td_sp = (void *)cpu_heavy_restore;

        /*
         * pcb2->pcb_ldt:       duplicated below, if necessary.
         * pcb2->pcb_savefpu:   cloned above.
         * pcb2->pcb_flags:     cloned above
         * pcb2->pcb_onfault:   cloned above (always NULL here).
         * pcb2->pcb_onfault_sp:cloned above (dont care)
         */

        /*
         * XXX don't copy the i/o pages.  this should probably be fixed.
         */
        pcb2->pcb_ext = NULL;

        /* Copy the LDT, if necessary. */
        if (pcb2->pcb_ldt != NULL) {
                if (flags & RFMEM) {
                        atomic_add_int(&pcb2->pcb_ldt->ldt_refcnt, 1);
                } else {
                        pcb2->pcb_ldt = user_ldt_alloc(pcb2,
                                                       pcb2->pcb_ldt->ldt_len);
                }
        }
        bcopy(&lp1->lwp_thread->td_tls, &lp2->lwp_thread->td_tls,
              sizeof(lp2->lwp_thread->td_tls));
        /*
         * Now, cpu_switch() can schedule the new lwp.
         * pcb_rsp is loaded pointing to the cpu_switch() stack frame
         * containing the return address when exiting cpu_switch.
         * This will normally be to fork_trampoline(), which will have
         * %rbx loaded with the new lwp's pointer.  fork_trampoline()
         * will set up a stack to call fork_return(lp, frame); to complete
         * the return to user-mode.
         */
}

/*
 * Prepare new lwp to return to the address specified in params.
 */
int
cpu_prepare_lwp(struct lwp *lp, struct lwp_params *params)
{
        struct trapframe *regs = lp->lwp_md.md_regs;
        void *bad_return = NULL;
        int error;

        regs->tf_rip = (long)params->lwp_func;
        regs->tf_rsp = (long)params->lwp_stack;
        /* Set up argument for function call */
        regs->tf_rdi = (long)params->lwp_arg;

        /*
         * Set up fake return address.  As the lwp function may never return,
         * we simply copy out a NULL pointer and force the lwp to receive
         * a SIGSEGV if it returns anyways.
         */
        regs->tf_rsp -= sizeof(void *);
        error = copyout(&bad_return, (void *)regs->tf_rsp, sizeof(bad_return));
        if (error)
                return (error);

        if (lp->lwp_proc->p_vmm) {
                lp->lwp_thread->td_pcb->pcb_cr3 = KPML4phys;
                cpu_set_fork_handler(lp,
                    (void (*)(void *, struct trapframe *))vmm_lwp_return, lp);
        } else {
                cpu_set_fork_handler(lp,
                    (void (*)(void *, struct trapframe *))generic_lwp_return, lp);
        }
        return (0);
}

/*
 * Intercept the return address from a freshly forked process that has NOT
 * been scheduled yet.
 *
 * This is needed to make kernel threads stay in kernel mode.
 */
void
cpu_set_fork_handler(struct lwp *lp, void (*func)(void *, struct trapframe *),
                     void *arg)
{
        /*
         * Note that the trap frame follows the args, so the function
         * is really called like this:  func(arg, frame);
         */
        lp->lwp_thread->td_pcb->pcb_rbx = (long)func;   /* function */
        lp->lwp_thread->td_pcb->pcb_r12 = (long)arg;    /* first arg */
}

void
cpu_set_thread_handler(thread_t td, void (*rfunc)(void), void *func, void *arg)
{
        td->td_pcb->pcb_rbx = (long)func;
        td->td_pcb->pcb_r12 = (long)arg;
        td->td_switch = cpu_lwkt_switch;
        td->td_sp -= sizeof(void *);
        *(void **)td->td_sp = rfunc;    /* exit function on return */
        td->td_sp -= sizeof(void *);
        *(void **)td->td_sp = cpu_kthread_restore;
}

void
cpu_lwp_exit(void)
{
        struct thread *td = curthread;
        struct pcb *pcb;

        pcb = td->td_pcb;

        /* Some x86 functionality was dropped */
        KKASSERT(pcb->pcb_ext == NULL);

        /*
         * disable all hardware breakpoints
         */
        if (pcb->pcb_flags & PCB_DBREGS) {
                reset_dbregs();
                pcb->pcb_flags &= ~PCB_DBREGS;
        }
        td->td_gd->gd_cnt.v_swtch++;

        crit_enter_quick(td);
        if (td->td_flags & TDF_TSLEEPQ)
                tsleep_remove(td);
        lwkt_deschedule_self(td);
        lwkt_remove_tdallq(td);
        cpu_thread_exit();
}

/*
 * Terminate the current thread.  The caller must have already acquired
 * the thread's rwlock and placed it on a reap list or otherwise notified
 * a reaper of its existance.  We set a special assembly switch function which
 * releases td_rwlock after it has cleaned up the MMU state and switched
 * out the stack.
 *
 * Must be caller from a critical section and with the thread descheduled.
 */
void
cpu_thread_exit(void)
{
        npxexit();
        curthread->td_switch = cpu_exit_switch;
        curthread->td_flags |= TDF_EXITING;
        lwkt_switch();
        panic("cpu_thread_exit: lwkt_switch() unexpectedly returned");
}

void
cpu_reset(void)
{
        cpu_reset_real();
}

static void
cpu_reset_real(void)
{
        /*
         * Attempt to do a CPU reset via the keyboard controller,
         * do not turn off the GateA20, as any machine that fails
         * to do the reset here would then end up in no man's land.
         */

#if !defined(BROKEN_KEYBOARD_RESET)
        outb(IO_KBD + 4, 0xFE);
        DELAY(500000);  /* wait 0.5 sec to see if that did it */
        kprintf("Keyboard reset did not work, attempting CPU shutdown\n");
        DELAY(1000000); /* wait 1 sec for kprintf to complete */
#endif
#if 0 /* JG */
        /* force a shutdown by unmapping entire address space ! */
        bzero((caddr_t) PTD, PAGE_SIZE);
#endif

        /* "good night, sweet prince .... <THUNK!>" */
        cpu_invltlb();
        /* NOTREACHED */
        while(1);
}

/*
 * Convert kernel VA to physical address
 */
vm_paddr_t
kvtop(void *addr)
{
        vm_paddr_t pa;

        pa = pmap_kextract((vm_offset_t)addr);
        if (pa == 0)
                panic("kvtop: zero page frame");
        return (pa);
}

static void
swi_vm(void *arg, void *frame)
{
        if (busdma_swi_pending != 0)
                busdma_swi();
}

static void
swi_vm_setup(void *arg)
{
        register_swi_mp(SWI_VM, swi_vm, NULL, "swi_vm", NULL, 0);
}

SYSINIT(swi_vm_setup, SI_BOOT2_MACHDEP, SI_ORDER_ANY, swi_vm_setup, NULL);

/*
 * NOTE: This routine is also called after a successful microcode
 *       reload on cpu 0.
 */
void spectre_vm_setup(void *arg);

/*
 * Check for IBRS support
 */
static
int
spectre_check_support(void)
{
        uint32_t p[4];

        p[0] = 0;
        p[1] = 0;
        p[2] = 0;
        p[3] = 0;
        cpuid_count(7, 0, p);
        if ((p[3] & 0x0C000000U) == 0x0C000000U) {

                /*
                 * SPEC_CTRL (bit 26) and STIBP support (bit 27)
                 *
                 * 0x80000008 p[0] bit 12 indicates IBPB support
                 */
                p[0] = 0;
                p[1] = 0;
                p[2] = 0;
                p[3] = 0;
                do_cpuid(0x80000008U, p);
                if (p[0] & 0x00001000)
                        return 1;
        }
        return 0;
}

/*
 * Iterate CPUs and adjust MSR for global operations, since
 * the KMMU* code won't do it if spectre_mitigation is 0 or 2.
 */
static
void
spectre_sysctl_changed(int old_value)
{
        globaldata_t save_gd;
        int n;

        save_gd = mycpu;
        for (n = 0; n < ncpus; ++n) {
                lwkt_setcpu_self(globaldata_find(n));

                pscpu->trampoline.tr_pcb_gflags &= ~(PCB_IBRS1 | PCB_IBRS2);

                switch(spectre_mitigation) {
                case 0:
                        if (old_value >= 0)
                                wrmsr(0x48, 0);
                        break;
                case 1:
                        pscpu->trampoline.tr_pcb_gflags |= PCB_IBRS1;
                        wrmsr(0x48, 1);
                        break;
                case 2:
                        pscpu->trampoline.tr_pcb_gflags |= PCB_IBRS2;
                        wrmsr(0x48, 1);
                        break;
                }
        }
        if (save_gd != mycpu)
                lwkt_setcpu_self(save_gd);
}

/*
 * User changes sysctl value
 */
static int
sysctl_spectre_mitigation(SYSCTL_HANDLER_ARGS)
{
        int new_spectre;
        int old_spectre;
        int error;

        old_spectre = spectre_mitigation;
        new_spectre = old_spectre;
        error = sysctl_handle_int(oidp, &new_spectre, 0, req);
        if (error || req->newptr == NULL)
                return error;
        spectre_mitigation = new_spectre;
        spectre_sysctl_changed(old_spectre);

        return 0;
}

SYSCTL_PROC(_machdep, OID_AUTO, spectre_mitigation, CTLTYPE_INT | CTLFLAG_RW,
        0, 0, sysctl_spectre_mitigation, "I", "Spectre exploit mitigation");

void
spectre_vm_setup(void *arg)
{
        int inconsistent = 0;
        int old_value = spectre_mitigation;

        if (spectre_mitigation < 0) {
                TUNABLE_INT_FETCH("machdep.spectre_mitigation",
                                  &spectre_mitigation);
        }

        if (cpu_vendor_id == CPU_VENDOR_INTEL) {
                if (spectre_check_support()) {
                        /*
                         * Must be supported on all cpus before we
                         * can enable it.  Returns silently if it
                         * isn't.
                         *
                         * NOTE! arg != NULL indicates we were called
                         *       from cpuctl after a successful microcode
                         *       update.
                         */
                        if (arg != NULL) {
                                globaldata_t save_gd;
                                int n;

                                save_gd = mycpu;
                                for (n = 0; n < ncpus; ++n) {
                                        lwkt_setcpu_self(globaldata_find(n));
                                        if (spectre_check_support() == 0) {
                                                inconsistent = 1;
                                                break;
                                        }
                                }
                                if (save_gd != mycpu)
                                        lwkt_setcpu_self(save_gd);
                        }
                        if (inconsistent == 0) {
                                if (spectre_mitigation < 0)
                                        spectre_mitigation = 1;
                        } else {
                                spectre_mitigation = -1;
                        }
                } else {
                        spectre_mitigation = -1;
                }
        } else {
                spectre_mitigation = -1;                /* no support */
        }

        /*
         * Be silent while microcode is being loaded on various CPUs,
         * until all done.
         */
        if (inconsistent)
                return;

        /*
         * Disallow sysctl changes when there is no support (otherwise
         * the wrmsr will cause a protection fault).
         */
        switch(spectre_mitigation) {
        case 0:
                sysctl___machdep_spectre_mitigation.oid_kind |= CTLFLAG_WR;
                kprintf("machdep.spectre_mitigation available but disabled\n");
                break;
        case 1:
                sysctl___machdep_spectre_mitigation.oid_kind |= CTLFLAG_WR;
                kprintf("machdep.spectre_mitigation available, system call\n"
                        "performance and kernel operation will be impacted\n");
                break;
        case 2:
                sysctl___machdep_spectre_mitigation.oid_kind |= CTLFLAG_WR;
                kprintf("machdep.spectre_mitigation available, whole machine\n"
                        "performance will be impacted\n");
                break;
        default:
                sysctl___machdep_spectre_mitigation.oid_kind &= ~CTLFLAG_WR;
                if (cpu_vendor_id == CPU_VENDOR_INTEL)
                        kprintf("no microcode spectre mitigation available\n");
                break;
        }
        spectre_sysctl_changed(old_value);
}

SYSINIT(spectre_vm_setup, SI_BOOT2_MACHDEP, SI_ORDER_ANY,
        spectre_vm_setup, NULL);

/*
 * platform-specific vmspace initialization (nothing for x86_64)
 */
void
cpu_vmspace_alloc(struct vmspace *vm __unused)
{
}

void
cpu_vmspace_free(struct vmspace *vm __unused)
{
}

int
kvm_access_check(vm_offset_t saddr, vm_offset_t eaddr, int prot)
{
        vm_offset_t addr;

        if (saddr < KvaStart)
                return EFAULT;
        if (eaddr >= KvaEnd)
                return EFAULT;
        for (addr = saddr; addr < eaddr; addr += PAGE_SIZE)  {
                if (pmap_kextract(addr) == 0)
                        return EFAULT;
        }
        if (!kernacc((caddr_t)saddr, eaddr - saddr, prot))
                return EFAULT;
        return 0;
}

#if 0

void _test_frame_enter(struct trapframe *frame);
void _test_frame_exit(struct trapframe *frame);

void
_test_frame_enter(struct trapframe *frame)
{
        thread_t td = curthread;

        if (ISPL(frame->tf_cs) == SEL_UPL) {
                KKASSERT(td->td_lwp);
                KASSERT(td->td_lwp->lwp_md.md_regs == frame,
                        ("_test_frame_exit: Frame mismatch %p %p",
                        td->td_lwp->lwp_md.md_regs, frame));
            td->td_lwp->lwp_saveusp = (void *)frame->tf_rsp;
            td->td_lwp->lwp_saveupc = (void *)frame->tf_rip;
        }
        if ((char *)frame < td->td_kstack ||
            (char *)frame > td->td_kstack + td->td_kstack_size) {
                panic("_test_frame_exit: frame not on kstack %p kstack=%p",
                        frame, td->td_kstack);
        }
}

void
_test_frame_exit(struct trapframe *frame)
{
        thread_t td = curthread;

        if (ISPL(frame->tf_cs) == SEL_UPL) {
                KKASSERT(td->td_lwp);
                KASSERT(td->td_lwp->lwp_md.md_regs == frame,
                        ("_test_frame_exit: Frame mismatch %p %p",
                        td->td_lwp->lwp_md.md_regs, frame));
                if (td->td_lwp->lwp_saveusp != (void *)frame->tf_rsp) {
                        kprintf("_test_frame_exit: %s:%d usp mismatch %p/%p\n",
                                td->td_comm, td->td_proc->p_pid,
                                td->td_lwp->lwp_saveusp,
                                (void *)frame->tf_rsp);
                }
                if (td->td_lwp->lwp_saveupc != (void *)frame->tf_rip) {
                        kprintf("_test_frame_exit: %s:%d upc mismatch %p/%p\n",
                                td->td_comm, td->td_proc->p_pid,
                                td->td_lwp->lwp_saveupc,
                                (void *)frame->tf_rip);
                }

                /*
                 * adulterate the fields to catch entries that
                 * don't run through test_frame_enter
                 */
                td->td_lwp->lwp_saveusp =
                        (void *)~(intptr_t)td->td_lwp->lwp_saveusp;
                td->td_lwp->lwp_saveupc =
                        (void *)~(intptr_t)td->td_lwp->lwp_saveupc;
        }
        if ((char *)frame < td->td_kstack ||
            (char *)frame > td->td_kstack + td->td_kstack_size) {
                panic("_test_frame_exit: frame not on kstack %p kstack=%p",
                        frame, td->td_kstack);
        }
}

#endif