root/sys/dev/virtio/mmio/virtio_mmio.c
/*-
 * Copyright (c) 2014 Ruslan Bukin <br@bsdpad.com>
 * Copyright (c) 2014 The FreeBSD Foundation
 * All rights reserved.
 *
 * This software was developed by SRI International and the University of
 * Cambridge Computer Laboratory under DARPA/AFRL contract (FA8750-10-C-0237)
 * ("CTSRD"), as part of the DARPA CRASH research programme.
 *
 * Portions of this software were developed by Andrew Turner
 * under sponsorship from the FreeBSD Foundation.
 *
 * 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 AUTHOR 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 AUTHOR 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.
 */

/*
 * VirtIO MMIO interface.
 * This driver is heavily based on VirtIO PCI interface driver.
 */

#include <sys/param.h>
#include <sys/systm.h>
#include <sys/bus.h>
#include <sys/kernel.h>
#include <sys/module.h>
#include <sys/malloc.h>
#include <sys/rman.h>
#include <sys/endian.h>

#include <machine/bus.h>
#include <machine/resource.h>

#include <dev/virtio/virtio.h>
#include <dev/virtio/virtqueue.h>
#include <dev/virtio/mmio/virtio_mmio.h>

#include "virtio_bus_if.h"
#include "virtio_if.h"

struct vtmmio_virtqueue {
        struct virtqueue        *vtv_vq;
        int                      vtv_no_intr;
};

static int      vtmmio_detach(device_t);
static int      vtmmio_suspend(device_t);
static int      vtmmio_resume(device_t);
static int      vtmmio_shutdown(device_t);
static void     vtmmio_driver_added(device_t, driver_t *);
static void     vtmmio_child_detached(device_t, device_t);
static int      vtmmio_read_ivar(device_t, device_t, int, uintptr_t *);
static int      vtmmio_write_ivar(device_t, device_t, int, uintptr_t);
static uint64_t vtmmio_negotiate_features(device_t, uint64_t);
static int      vtmmio_finalize_features(device_t);
static bool     vtmmio_with_feature(device_t, uint64_t);
static void     vtmmio_set_virtqueue(struct vtmmio_softc *sc,
                    struct virtqueue *vq, uint32_t size);
static int      vtmmio_alloc_virtqueues(device_t, int,
                    struct vq_alloc_info *);
static int      vtmmio_setup_intr(device_t, enum intr_type);
static void     vtmmio_stop(device_t);
static int      vtmmio_reinit(device_t, uint64_t);
static void     vtmmio_reinit_complete(device_t);
static void     vtmmio_notify_virtqueue(device_t, uint16_t, bus_size_t);
static int      vtmmio_config_generation(device_t);
static uint8_t  vtmmio_get_status(device_t);
static void     vtmmio_set_status(device_t, uint8_t);
static void     vtmmio_read_dev_config(device_t, bus_size_t, void *, int);
static uint64_t vtmmio_read_dev_config_8(struct vtmmio_softc *, bus_size_t);
static void     vtmmio_write_dev_config(device_t, bus_size_t, const void *, int);
static void     vtmmio_describe_features(struct vtmmio_softc *, const char *,
                    uint64_t);
static void     vtmmio_probe_and_attach_child(struct vtmmio_softc *);
static int      vtmmio_reinit_virtqueue(struct vtmmio_softc *, int);
static void     vtmmio_free_interrupts(struct vtmmio_softc *);
static void     vtmmio_free_virtqueues(struct vtmmio_softc *);
static void     vtmmio_release_child_resources(struct vtmmio_softc *);
static void     vtmmio_reset(struct vtmmio_softc *);
static void     vtmmio_select_virtqueue(struct vtmmio_softc *, int);
static void     vtmmio_vq_intr(void *);

/*
 * I/O port read/write wrappers.
 */
#define vtmmio_write_config_1(sc, o, v)                         \
        bus_write_1((sc)->res[0], (o), (v))
#define vtmmio_write_config_2(sc, o, v)                         \
        bus_write_2((sc)->res[0], (o), (v))
#define vtmmio_write_config_4(sc, o, v)                         \
        bus_write_4((sc)->res[0], (o), (v))

#define vtmmio_read_config_1(sc, o) \
        bus_read_1((sc)->res[0], (o))
#define vtmmio_read_config_2(sc, o) \
        bus_read_2((sc)->res[0], (o))
#define vtmmio_read_config_4(sc, o) \
        bus_read_4((sc)->res[0], (o))

static device_method_t vtmmio_methods[] = {
        /* Device interface. */
        DEVMETHOD(device_attach,                  vtmmio_attach),
        DEVMETHOD(device_detach,                  vtmmio_detach),
        DEVMETHOD(device_suspend,                 vtmmio_suspend),
        DEVMETHOD(device_resume,                  vtmmio_resume),
        DEVMETHOD(device_shutdown,                vtmmio_shutdown),

        /* Bus interface. */
        DEVMETHOD(bus_driver_added,               vtmmio_driver_added),
        DEVMETHOD(bus_child_detached,             vtmmio_child_detached),
        DEVMETHOD(bus_child_pnpinfo,              virtio_child_pnpinfo),
        DEVMETHOD(bus_read_ivar,                  vtmmio_read_ivar),
        DEVMETHOD(bus_write_ivar,                 vtmmio_write_ivar),

        /* VirtIO bus interface. */
        DEVMETHOD(virtio_bus_negotiate_features,  vtmmio_negotiate_features),
        DEVMETHOD(virtio_bus_finalize_features,   vtmmio_finalize_features),
        DEVMETHOD(virtio_bus_with_feature,        vtmmio_with_feature),
        DEVMETHOD(virtio_bus_alloc_virtqueues,    vtmmio_alloc_virtqueues),
        DEVMETHOD(virtio_bus_setup_intr,          vtmmio_setup_intr),
        DEVMETHOD(virtio_bus_stop,                vtmmio_stop),
        DEVMETHOD(virtio_bus_reinit,              vtmmio_reinit),
        DEVMETHOD(virtio_bus_reinit_complete,     vtmmio_reinit_complete),
        DEVMETHOD(virtio_bus_notify_vq,           vtmmio_notify_virtqueue),
        DEVMETHOD(virtio_bus_config_generation,   vtmmio_config_generation),
        DEVMETHOD(virtio_bus_read_device_config,  vtmmio_read_dev_config),
        DEVMETHOD(virtio_bus_write_device_config, vtmmio_write_dev_config),

        DEVMETHOD_END
};

DEFINE_CLASS_0(virtio_mmio, vtmmio_driver, vtmmio_methods,
    sizeof(struct vtmmio_softc));

MODULE_VERSION(virtio_mmio, 1);

int
vtmmio_probe(device_t dev)
{
        struct vtmmio_softc *sc;
        int rid;
        uint32_t magic, version;

        sc = device_get_softc(dev);

        rid = 0;
        sc->res[0] = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &rid,
            RF_ACTIVE);
        if (sc->res[0] == NULL) {
                device_printf(dev, "Cannot allocate memory window.\n");
                return (ENXIO);
        }

        magic = vtmmio_read_config_4(sc, VIRTIO_MMIO_MAGIC_VALUE);
        if (magic != VIRTIO_MMIO_MAGIC_VIRT) {
                device_printf(dev, "Bad magic value %#x\n", magic);
                bus_release_resource(dev, SYS_RES_MEMORY, rid, sc->res[0]);
                return (ENXIO);
        }

        version = vtmmio_read_config_4(sc, VIRTIO_MMIO_VERSION);
        if (version < 1 || version > 2) {
                device_printf(dev, "Unsupported version: %#x\n", version);
                bus_release_resource(dev, SYS_RES_MEMORY, rid, sc->res[0]);
                return (ENXIO);
        }

        if (vtmmio_read_config_4(sc, VIRTIO_MMIO_DEVICE_ID) == 0) {
                bus_release_resource(dev, SYS_RES_MEMORY, rid, sc->res[0]);
                return (ENXIO);
        }

        bus_release_resource(dev, SYS_RES_MEMORY, rid, sc->res[0]);

        device_set_desc(dev, "VirtIO MMIO adapter");
        return (BUS_PROBE_DEFAULT);
}

static int
vtmmio_setup_intr(device_t dev, enum intr_type type)
{
        struct vtmmio_softc *sc;
        int rid;

        sc = device_get_softc(dev);

        rid = 0;
        sc->res[1] = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
                RF_ACTIVE);
        if (!sc->res[1]) {
                device_printf(dev, "Can't allocate interrupt\n");
                return (ENXIO);
        }

        if (bus_setup_intr(dev, sc->res[1], type | INTR_MPSAFE,
                NULL, vtmmio_vq_intr, sc, &sc->ih)) {
                device_printf(dev, "Can't setup the interrupt\n");
                return (ENXIO);
        }

        return (0);
}

int
vtmmio_attach(device_t dev)
{
        struct vtmmio_softc *sc;
        device_t child;
        int rid;

        sc = device_get_softc(dev);
        sc->dev = dev;

        rid = 0;
        sc->res[0] = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &rid,
                        RF_ACTIVE);
        if (sc->res[0] == NULL) {
                device_printf(dev, "Cannot allocate memory window.\n");
                return (ENXIO);
        }

        sc->vtmmio_version = vtmmio_read_config_4(sc, VIRTIO_MMIO_VERSION);

        vtmmio_reset(sc);

        /* Tell the host we've noticed this device. */
        vtmmio_set_status(dev, VIRTIO_CONFIG_STATUS_ACK);

        if ((child = device_add_child(dev, NULL, DEVICE_UNIT_ANY)) == NULL) {
                device_printf(dev, "Cannot create child device.\n");
                vtmmio_set_status(dev, VIRTIO_CONFIG_STATUS_FAILED);
                vtmmio_detach(dev);
                return (ENOMEM);
        }

        sc->vtmmio_child_dev = child;
        vtmmio_probe_and_attach_child(sc);

        return (0);
}

static int
vtmmio_detach(device_t dev)
{
        struct vtmmio_softc *sc;
        int error;

        sc = device_get_softc(dev);

        error = bus_generic_detach(dev);
        if (error)
                return (error);

        vtmmio_reset(sc);

        if (sc->res[0] != NULL) {
                bus_release_resource(dev, SYS_RES_MEMORY, 0,
                    sc->res[0]);
                sc->res[0] = NULL;
        }

        return (0);
}

static int
vtmmio_suspend(device_t dev)
{

        return (bus_generic_suspend(dev));
}

static int
vtmmio_resume(device_t dev)
{

        return (bus_generic_resume(dev));
}

static int
vtmmio_shutdown(device_t dev)
{

        (void) bus_generic_shutdown(dev);

        /* Forcibly stop the host device. */
        vtmmio_stop(dev);

        return (0);
}

static void
vtmmio_driver_added(device_t dev, driver_t *driver)
{
        struct vtmmio_softc *sc;

        sc = device_get_softc(dev);

        vtmmio_probe_and_attach_child(sc);
}

static void
vtmmio_child_detached(device_t dev, device_t child)
{
        struct vtmmio_softc *sc;

        sc = device_get_softc(dev);

        vtmmio_reset(sc);
        vtmmio_release_child_resources(sc);
}

static int
vtmmio_read_ivar(device_t dev, device_t child, int index, uintptr_t *result)
{
        struct vtmmio_softc *sc;

        sc = device_get_softc(dev);

        if (sc->vtmmio_child_dev != child)
                return (ENOENT);

        switch (index) {
        case VIRTIO_IVAR_DEVTYPE:
        case VIRTIO_IVAR_SUBDEVICE:
                *result = vtmmio_read_config_4(sc, VIRTIO_MMIO_DEVICE_ID);
                break;
        case VIRTIO_IVAR_VENDOR:
                *result = vtmmio_read_config_4(sc, VIRTIO_MMIO_VENDOR_ID);
                break;
        case VIRTIO_IVAR_SUBVENDOR:
        case VIRTIO_IVAR_DEVICE:
                /*
                 * Dummy value for fields not present in this bus.  Used by
                 * bus-agnostic virtio_child_pnpinfo.
                 */
                *result = 0;
                break;
        case VIRTIO_IVAR_MODERN:
                /*
                 * There are several modern (aka MMIO v2) spec compliance
                 * issues with this driver, but keep the status quo.
                 */
                *result = sc->vtmmio_version > 1;
                break;
        default:
                return (ENOENT);
        }

        return (0);
}

static int
vtmmio_write_ivar(device_t dev, device_t child, int index, uintptr_t value)
{
        struct vtmmio_softc *sc;

        sc = device_get_softc(dev);

        if (sc->vtmmio_child_dev != child)
                return (ENOENT);

        switch (index) {
        case VIRTIO_IVAR_FEATURE_DESC:
                sc->vtmmio_child_feat_desc = (void *) value;
                break;
        default:
                return (ENOENT);
        }

        return (0);
}

static uint64_t
vtmmio_negotiate_features(device_t dev, uint64_t child_features)
{
        struct vtmmio_softc *sc;
        uint64_t host_features, features;

        sc = device_get_softc(dev);

        if (sc->vtmmio_version > 1) {
                child_features |= VIRTIO_F_VERSION_1;
        }

        vtmmio_write_config_4(sc, VIRTIO_MMIO_HOST_FEATURES_SEL, 1);
        host_features = vtmmio_read_config_4(sc, VIRTIO_MMIO_HOST_FEATURES);
        host_features <<= 32;

        vtmmio_write_config_4(sc, VIRTIO_MMIO_HOST_FEATURES_SEL, 0);
        host_features |= vtmmio_read_config_4(sc, VIRTIO_MMIO_HOST_FEATURES);

        vtmmio_describe_features(sc, "host", host_features);

        /*
         * Limit negotiated features to what the driver, virtqueue, and
         * host all support.
         */
        features = host_features & child_features;
        features = virtio_filter_transport_features(features);
        sc->vtmmio_features = features;

        vtmmio_describe_features(sc, "negotiated", features);

        vtmmio_write_config_4(sc, VIRTIO_MMIO_GUEST_FEATURES_SEL, 1);
        vtmmio_write_config_4(sc, VIRTIO_MMIO_GUEST_FEATURES, features >> 32);

        vtmmio_write_config_4(sc, VIRTIO_MMIO_GUEST_FEATURES_SEL, 0);
        vtmmio_write_config_4(sc, VIRTIO_MMIO_GUEST_FEATURES, features);

        return (features);
}

static int
vtmmio_finalize_features(device_t dev)
{
        struct vtmmio_softc *sc;
        uint8_t status;

        sc = device_get_softc(dev);

        if (sc->vtmmio_version > 1) {
                /*
                 * Must re-read the status after setting it to verify the
                 * negotiated features were accepted by the device.
                 */
                vtmmio_set_status(dev, VIRTIO_CONFIG_S_FEATURES_OK);

                status = vtmmio_get_status(dev);
                if ((status & VIRTIO_CONFIG_S_FEATURES_OK) == 0) {
                        device_printf(dev, "desired features were not accepted\n");
                        return (ENOTSUP);
                }
        }

        return (0);
}

static bool
vtmmio_with_feature(device_t dev, uint64_t feature)
{
        struct vtmmio_softc *sc;

        sc = device_get_softc(dev);

        return ((sc->vtmmio_features & feature) != 0);
}

static void
vtmmio_set_virtqueue(struct vtmmio_softc *sc, struct virtqueue *vq,
    uint32_t size)
{
        vm_paddr_t paddr;

        vtmmio_write_config_4(sc, VIRTIO_MMIO_QUEUE_NUM, size);

        if (sc->vtmmio_version == 1) {
                vtmmio_write_config_4(sc, VIRTIO_MMIO_QUEUE_ALIGN,
                    VIRTIO_MMIO_VRING_ALIGN);
                paddr = virtqueue_paddr(vq);
                vtmmio_write_config_4(sc, VIRTIO_MMIO_QUEUE_PFN,
                    paddr >> PAGE_SHIFT);
        } else {
                paddr = virtqueue_desc_paddr(vq);
                vtmmio_write_config_4(sc, VIRTIO_MMIO_QUEUE_DESC_LOW,
                    paddr);
                vtmmio_write_config_4(sc, VIRTIO_MMIO_QUEUE_DESC_HIGH,
                    ((uint64_t)paddr) >> 32);

                paddr = virtqueue_avail_paddr(vq);
                vtmmio_write_config_4(sc, VIRTIO_MMIO_QUEUE_AVAIL_LOW,
                    paddr);
                vtmmio_write_config_4(sc, VIRTIO_MMIO_QUEUE_AVAIL_HIGH,
                    ((uint64_t)paddr) >> 32);

                paddr = virtqueue_used_paddr(vq);
                vtmmio_write_config_4(sc, VIRTIO_MMIO_QUEUE_USED_LOW,
                    paddr);
                vtmmio_write_config_4(sc, VIRTIO_MMIO_QUEUE_USED_HIGH,
                    ((uint64_t)paddr) >> 32);

                vtmmio_write_config_4(sc, VIRTIO_MMIO_QUEUE_READY, 1);
        }
}

static int
vtmmio_alloc_virtqueues(device_t dev, int nvqs,
    struct vq_alloc_info *vq_info)
{
        struct vtmmio_virtqueue *vqx;
        struct vq_alloc_info *info;
        struct vtmmio_softc *sc;
        struct virtqueue *vq;
        uint32_t size;
        int idx, error;

        sc = device_get_softc(dev);

        if (sc->vtmmio_nvqs != 0)
                return (EALREADY);
        if (nvqs <= 0)
                return (EINVAL);

        sc->vtmmio_vqs = malloc(nvqs * sizeof(struct vtmmio_virtqueue),
            M_DEVBUF, M_NOWAIT | M_ZERO);
        if (sc->vtmmio_vqs == NULL)
                return (ENOMEM);

        if (sc->vtmmio_version == 1) {
                vtmmio_write_config_4(sc, VIRTIO_MMIO_GUEST_PAGE_SIZE,
                    (1 << PAGE_SHIFT));
        }

        for (idx = 0; idx < nvqs; idx++) {
                vqx = &sc->vtmmio_vqs[idx];
                info = &vq_info[idx];

                vtmmio_select_virtqueue(sc, idx);
                size = vtmmio_read_config_4(sc, VIRTIO_MMIO_QUEUE_NUM_MAX);

                error = virtqueue_alloc(dev, idx, size,
                    VIRTIO_MMIO_QUEUE_NOTIFY, VIRTIO_MMIO_VRING_ALIGN,
                    ~(vm_paddr_t)0, info, &vq);
                if (error) {
                        device_printf(dev,
                            "cannot allocate virtqueue %d: %d\n",
                            idx, error);
                        break;
                }

                vtmmio_set_virtqueue(sc, vq, size);

                vqx->vtv_vq = *info->vqai_vq = vq;
                vqx->vtv_no_intr = info->vqai_intr == NULL;

                sc->vtmmio_nvqs++;
        }

        if (error)
                vtmmio_free_virtqueues(sc);

        return (error);
}

static void
vtmmio_stop(device_t dev)
{

        vtmmio_reset(device_get_softc(dev));
}

static int
vtmmio_reinit(device_t dev, uint64_t features)
{
        struct vtmmio_softc *sc;
        int idx, error;

        sc = device_get_softc(dev);

        if (vtmmio_get_status(dev) != VIRTIO_CONFIG_STATUS_RESET)
                vtmmio_stop(dev);

        /*
         * Quickly drive the status through ACK and DRIVER. The device
         * does not become usable again until vtmmio_reinit_complete().
         */
        vtmmio_set_status(dev, VIRTIO_CONFIG_STATUS_ACK);
        vtmmio_set_status(dev, VIRTIO_CONFIG_STATUS_DRIVER);

        /*
         * TODO: Check that features are not added as to what was
         * originally negotiated.
         */
        vtmmio_negotiate_features(dev, features);
        error = vtmmio_finalize_features(dev);
        if (error) {
                device_printf(dev, "cannot finalize features during reinit\n");
                return (error);
        }

        if (sc->vtmmio_version == 1) {
                vtmmio_write_config_4(sc, VIRTIO_MMIO_GUEST_PAGE_SIZE,
                    (1 << PAGE_SHIFT));
        }

        for (idx = 0; idx < sc->vtmmio_nvqs; idx++) {
                error = vtmmio_reinit_virtqueue(sc, idx);
                if (error)
                        return (error);
        }

        return (0);
}

static void
vtmmio_reinit_complete(device_t dev)
{

        vtmmio_set_status(dev, VIRTIO_CONFIG_STATUS_DRIVER_OK);
}

static void
vtmmio_notify_virtqueue(device_t dev, uint16_t queue, bus_size_t offset)
{
        struct vtmmio_softc *sc;

        sc = device_get_softc(dev);
        MPASS(offset == VIRTIO_MMIO_QUEUE_NOTIFY);

        vtmmio_write_config_4(sc, offset, queue);
}

static int
vtmmio_config_generation(device_t dev)
{
        struct vtmmio_softc *sc;
        uint32_t gen;

        sc = device_get_softc(dev);

        if (sc->vtmmio_version > 1)
                gen = vtmmio_read_config_4(sc, VIRTIO_MMIO_CONFIG_GENERATION);
        else
                gen = 0;

        return (gen);
}

static uint8_t
vtmmio_get_status(device_t dev)
{
        struct vtmmio_softc *sc;

        sc = device_get_softc(dev);

        return (vtmmio_read_config_4(sc, VIRTIO_MMIO_STATUS));
}

static void
vtmmio_set_status(device_t dev, uint8_t status)
{
        struct vtmmio_softc *sc;

        sc = device_get_softc(dev);

        if (status != VIRTIO_CONFIG_STATUS_RESET)
                status |= vtmmio_get_status(dev);

        vtmmio_write_config_4(sc, VIRTIO_MMIO_STATUS, status);
}

static void
vtmmio_read_dev_config(device_t dev, bus_size_t offset,
    void *dst, int length)
{
        struct vtmmio_softc *sc;
        bus_size_t off;
        uint8_t *d;
        int size;

        sc = device_get_softc(dev);
        off = VIRTIO_MMIO_CONFIG + offset;

        /*
         * The non-legacy MMIO specification adds the following restriction:
         *
         *   4.2.2.2: For the device-specific configuration space, the driver
         *   MUST use 8 bit wide accesses for 8 bit wide fields, 16 bit wide
         *   and aligned accesses for 16 bit wide fields and 32 bit wide and
         *   aligned accesses for 32 and 64 bit wide fields.
         *
         * The endianness also varies between non-legacy and legacy:
         *
         *   2.4: Note: The device configuration space uses the little-endian
         *   format for multi-byte fields.
         *
         *   2.4.3: Note that for legacy interfaces, device configuration space
         *   is generally the guest’s native endian, rather than PCI’s
         *   little-endian. The correct endian-ness is documented for each
         *   device.
         */
        if (sc->vtmmio_version > 1) {
                switch (length) {
                case 1:
                        *(uint8_t *)dst = vtmmio_read_config_1(sc, off);
                        break;
                case 2:
                        *(uint16_t *)dst =
                            le16toh(vtmmio_read_config_2(sc, off));
                        break;
                case 4:
                        *(uint32_t *)dst =
                            le32toh(vtmmio_read_config_4(sc, off));
                        break;
                case 8:
                        *(uint64_t *)dst = vtmmio_read_dev_config_8(sc, off);
                        break;
                default:
                        panic("%s: invalid length %d\n", __func__, length);
                }

                return;
        }

        for (d = dst; length > 0; d += size, off += size, length -= size) {
#ifdef ALLOW_WORD_ALIGNED_ACCESS
                if (length >= 4) {
                        size = 4;
                        *(uint32_t *)d = vtmmio_read_config_4(sc, off);
                } else if (length >= 2) {
                        size = 2;
                        *(uint16_t *)d = vtmmio_read_config_2(sc, off);
                } else
#endif
                {
                        size = 1;
                        *d = vtmmio_read_config_1(sc, off);
                }
        }
}

static uint64_t
vtmmio_read_dev_config_8(struct vtmmio_softc *sc, bus_size_t off)
{
        device_t dev;
        int gen;
        uint32_t val0, val1;

        dev = sc->dev;

        do {
                gen = vtmmio_config_generation(dev);
                val0 = le32toh(vtmmio_read_config_4(sc, off));
                val1 = le32toh(vtmmio_read_config_4(sc, off + 4));
        } while (gen != vtmmio_config_generation(dev));

        return (((uint64_t) val1 << 32) | val0);
}

static void
vtmmio_write_dev_config(device_t dev, bus_size_t offset,
    const void *src, int length)
{
        struct vtmmio_softc *sc;
        bus_size_t off;
        const uint8_t *s;
        int size;

        sc = device_get_softc(dev);
        off = VIRTIO_MMIO_CONFIG + offset;

        /*
         * The non-legacy MMIO specification adds size and alignment
         * restrctions. It also changes the endianness from native-endian to
         * little-endian. See vtmmio_read_dev_config.
         */
        if (sc->vtmmio_version > 1) {
                switch (length) {
                case 1:
                        vtmmio_write_config_1(sc, off, *(const uint8_t *)src);
                        break;
                case 2:
                        vtmmio_write_config_2(sc, off,
                            htole16(*(const uint16_t *)src));
                        break;
                case 4:
                        vtmmio_write_config_4(sc, off,
                            htole32(*(const uint32_t *)src));
                        break;
                case 8:
                        vtmmio_write_config_4(sc, off,
                            htole32(*(const uint64_t *)src));
                        vtmmio_write_config_4(sc, off + 4,
                            htole32((*(const uint64_t *)src) >> 32));
                        break;
                default:
                        panic("%s: invalid length %d\n", __func__, length);
                }

                return;
        }

        for (s = src; length > 0; s += size, off += size, length -= size) {
#ifdef ALLOW_WORD_ALIGNED_ACCESS
                if (length >= 4) {
                        size = 4;
                        vtmmio_write_config_4(sc, off, *(uint32_t *)s);
                } else if (length >= 2) {
                        size = 2;
                        vtmmio_write_config_2(sc, off, *(uint16_t *)s);
                } else
#endif
                {
                        size = 1;
                        vtmmio_write_config_1(sc, off, *s);
                }
        }
}

static void
vtmmio_describe_features(struct vtmmio_softc *sc, const char *msg,
    uint64_t features)
{
        device_t dev, child;

        dev = sc->dev;
        child = sc->vtmmio_child_dev;

        if (device_is_attached(child) || bootverbose == 0)
                return;

        virtio_describe(dev, msg, features, sc->vtmmio_child_feat_desc);
}

static void
vtmmio_probe_and_attach_child(struct vtmmio_softc *sc)
{
        device_t dev, child;

        dev = sc->dev;
        child = sc->vtmmio_child_dev;

        if (child == NULL)
                return;

        if (device_get_state(child) != DS_NOTPRESENT) {
                return;
        }

        if (device_probe(child) != 0) {
                return;
        }

        vtmmio_set_status(dev, VIRTIO_CONFIG_STATUS_DRIVER);
        if (device_attach(child) != 0) {
                vtmmio_set_status(dev, VIRTIO_CONFIG_STATUS_FAILED);
                vtmmio_reset(sc);
                vtmmio_release_child_resources(sc);
                /* Reset status for future attempt. */
                vtmmio_set_status(dev, VIRTIO_CONFIG_STATUS_ACK);
        } else {
                vtmmio_set_status(dev, VIRTIO_CONFIG_STATUS_DRIVER_OK);
                VIRTIO_ATTACH_COMPLETED(child);
        }
}

static int
vtmmio_reinit_virtqueue(struct vtmmio_softc *sc, int idx)
{
        struct vtmmio_virtqueue *vqx;
        struct virtqueue *vq;
        int error;
        uint16_t size;

        vqx = &sc->vtmmio_vqs[idx];
        vq = vqx->vtv_vq;

        KASSERT(vq != NULL, ("%s: vq %d not allocated", __func__, idx));

        vtmmio_select_virtqueue(sc, idx);
        size = vtmmio_read_config_4(sc, VIRTIO_MMIO_QUEUE_NUM_MAX);

        error = virtqueue_reinit(vq, size);
        if (error)
                return (error);

        vtmmio_set_virtqueue(sc, vq, size);

        return (0);
}

static void
vtmmio_free_interrupts(struct vtmmio_softc *sc)
{

        if (sc->ih != NULL)
                bus_teardown_intr(sc->dev, sc->res[1], sc->ih);

        if (sc->res[1] != NULL)
                bus_release_resource(sc->dev, SYS_RES_IRQ, 0, sc->res[1]);
}

static void
vtmmio_free_virtqueues(struct vtmmio_softc *sc)
{
        struct vtmmio_virtqueue *vqx;
        int idx;

        for (idx = 0; idx < sc->vtmmio_nvqs; idx++) {
                vqx = &sc->vtmmio_vqs[idx];

                vtmmio_select_virtqueue(sc, idx);
                if (sc->vtmmio_version > 1) {
                        vtmmio_write_config_4(sc, VIRTIO_MMIO_QUEUE_READY, 0);
                        vtmmio_read_config_4(sc, VIRTIO_MMIO_QUEUE_READY);
                } else
                        vtmmio_write_config_4(sc, VIRTIO_MMIO_QUEUE_PFN, 0);

                virtqueue_free(vqx->vtv_vq);
                vqx->vtv_vq = NULL;
        }

        free(sc->vtmmio_vqs, M_DEVBUF);
        sc->vtmmio_vqs = NULL;
        sc->vtmmio_nvqs = 0;
}

static void
vtmmio_release_child_resources(struct vtmmio_softc *sc)
{

        vtmmio_free_interrupts(sc);
        vtmmio_free_virtqueues(sc);
}

static void
vtmmio_reset(struct vtmmio_softc *sc)
{

        /*
         * Setting the status to RESET sets the host device to
         * the original, uninitialized state.
         */
        vtmmio_set_status(sc->dev, VIRTIO_CONFIG_STATUS_RESET);
}

static void
vtmmio_select_virtqueue(struct vtmmio_softc *sc, int idx)
{

        vtmmio_write_config_4(sc, VIRTIO_MMIO_QUEUE_SEL, idx);
}

static void
vtmmio_vq_intr(void *arg)
{
        struct vtmmio_virtqueue *vqx;
        struct vtmmio_softc *sc;
        struct virtqueue *vq;
        uint32_t status;
        int idx;

        sc = arg;

        status = vtmmio_read_config_4(sc, VIRTIO_MMIO_INTERRUPT_STATUS);
        vtmmio_write_config_4(sc, VIRTIO_MMIO_INTERRUPT_ACK, status);

        /* The config changed */
        if (status & VIRTIO_MMIO_INT_CONFIG)
                if (sc->vtmmio_child_dev != NULL)
                        VIRTIO_CONFIG_CHANGE(sc->vtmmio_child_dev);

        /* Notify all virtqueues. */
        if (status & VIRTIO_MMIO_INT_VRING) {
                for (idx = 0; idx < sc->vtmmio_nvqs; idx++) {
                        vqx = &sc->vtmmio_vqs[idx];
                        if (vqx->vtv_no_intr == 0) {
                                vq = vqx->vtv_vq;
                                virtqueue_intr(vq);
                        }
                }
        }
}