Re: [Qemu-devel] [RFC PATCH 1/3] virtio-bus : Introduce VirtioBus.

[KONRAD Frédéric]

On 20/11/2012 15:12, Cornelia Huck wrote:
> On Mon, 19 Nov 2012 17:33:01 +0000
> Peter Maydell <address@hidden> wrote:
>
> > On 16 November 2012 15:35,  <address@hidden> wrote:
> > > From: KONRAD Frederic <address@hidden>
> > You forgot to CC enough people...
> >
> > > This patch create a new VirtioBus, which can be added to Virtio transports like
> > > virtio-pci, virtio-mmio,...
> > >
> > > One VirtIODevice can be connected to this device, like virtio-blk in the 3rd
> > > patch.
> > >
> > > The VirtioBus shares through a VirtioBusInfo structure :
> > >
> > >      * two callbacks with the transport : init_cb and exit_cb, which must be
> > >        called by the VirtIODevice, after the initialization and before the
> > >        destruction, to put the right PCI IDs and/or stop the event fd.
> Can we specify the general purpose of the {init,exit}_cb a bit better?
> Is it analogous to any of the functions performed by the transport
> busses today?
For example, look at the function static int 
virtio_blk_init_pci(PCIDevice *pci_dev) in virtio-pci.c,

it creates the VirtIODevice and then call virtio_init_pci(proxy, vdev);

It is what I tryed to reproduce, and abstract it from the new device.

eg : for the new virtio-blk I add, in the function static int 
virtio_blk_qdev_init(DeviceState *qdev) in virtio-blk.c :

it creates the VirtIODevice, create the virtiobus and call the 
virtio_bus_init_cb(bus, vdev)
 which call the virtio_init_pci callback when virtio-pci bridge is used.

it is the same thing for the exit.

Is it making sense ?

> > >      * a VirtIOBindings structure.
> > >
> > > Signed-off-by: KONRAD Frederic <address@hidden>
> > > ---
> > >   hw/Makefile.objs |   1 +
> > >   hw/virtio-bus.c  | 111 +++++++++++++++++++++++++++++++++++++++++++++++++++++++
> > >   hw/virtio-bus.h  |  49 ++++++++++++++++++++++++
> > >   3 files changed, 161 insertions(+)
> > >   create mode 100644 hw/virtio-bus.c
> > >   create mode 100644 hw/virtio-bus.h
> > >
> > > diff --git a/hw/Makefile.objs b/hw/Makefile.objs
> > > index af4ab0c..1b25d77 100644
> > > --- a/hw/Makefile.objs
> > > +++ b/hw/Makefile.objs
> > > @@ -1,6 +1,7 @@
> > >   common-obj-y = usb/ ide/
> > >   common-obj-y += loader.o
> > >   common-obj-$(CONFIG_VIRTIO) += virtio-console.o
> > > +common-obj-$(CONFIG_VIRTIO) += virtio-bus.o
> > >   common-obj-$(CONFIG_VIRTIO_PCI) += virtio-pci.o
> > >   common-obj-y += fw_cfg.o
> > >   common-obj-$(CONFIG_PCI) += pci.o pci_bridge.o pci_bridge_dev.o
> > > diff --git a/hw/virtio-bus.c b/hw/virtio-bus.c
> > > new file mode 100644
> > > index 0000000..b2e7e9c
> > > --- /dev/null
> > > +++ b/hw/virtio-bus.c
> > > @@ -0,0 +1,111 @@
> > > +/*
> > > + * VirtioBus
> > > + *
> > > + *  Copyright (C) 2012 : GreenSocs Ltd
> > > + *      http://www.greensocs.com/ , email: address@hidden
> > > + *
> > > + *  Developed by :
> > > + *  Frederic Konrad   <address@hidden>
> > > + *
> > > + * This work is licensed under the terms of the GNU GPL, version 2.  See
> > > + * the COPYING file in the top-level directory.
> > Unless you copied this code from an existing v2-only file, preferred
> > license for new code is v2-or-any-later-version.
> >
> > > + */
> > > +
> > > +#include "hw.h"
> > > +#include "qemu-error.h"
> > > +#include "qdev.h"
> > > +#include "virtio-bus.h"
> > > +#include "virtio.h"
> > > +
> > > +#define DEBUG_VIRTIO_BUS
> > > +
> > > +#ifdef DEBUG_VIRTIO_BUS
> > > +
> > > +#define DPRINTF(fmt, ...) \
> > > +do { printf("virtio_bus: " fmt , ## __VA_ARGS__); } while (0)
> > > +#else
> > > +#define DPRINTF(fmt, ...) do {} while (0)
> > > +#endif
> > > +
> > > +static char *virtio_bus_get_fw_dev_path(DeviceState *dev);
> > Is this function necessary? I think your implementation
> > is doing the same as the default.
> >
> > > +static void virtio_bus_class_init(ObjectClass *klass, void *data)
> > > +{
> > > +    BusClass *k = BUS_CLASS(klass);
> > > +    k->get_fw_dev_path = virtio_bus_get_fw_dev_path;
> > > +}
> > > +
> > > +static const TypeInfo virtio_bus_info = {
> > > +    .name = TYPE_VIRTIO_BUS,
> > > +    .parent = TYPE_BUS,
> > > +    .instance_size = sizeof(VirtioBus),
> > > +    .class_init = virtio_bus_class_init,
> > > +};
> > > +
> > > +/* VirtioBus */
> > > +
> > > +static int next_virtio_bus;
> > > +
> > > +/* Create a virtio bus, and attach to transport.  */
> > > +void virtio_bus_new(VirtioBus *bus, DeviceState *host,
> > > +                    const VirtioBusInfo *info)
> > > +{
> > > +    /*
> > > +     * Setting name to NULL return always "virtio.0"
> > > +     * as bus name in info qtree.
> > > +     */
> > > +    char *bus_name = g_strdup_printf("%s.%d", BUS_NAME, next_virtio_bus);
> > > +    qbus_create_inplace(&bus->qbus, TYPE_VIRTIO_BUS, host, bus_name);
> > > +    bus->busnr = next_virtio_bus++;
> > This looks suspicious to me -- is keeping a count of the global
> > bus number really the right way to do this?
> If we want an unique number, we need to track usage of numbers, else we
> end up with duplicates on wraparound.
I put that because the number was all identical in "info qtree", is it 
the right thing to do ?

> > > +    bus->info = info;
> > > +    /* no hotplug for the moment ? */
> > > +    bus->qbus.allow_hotplug = 0;
> > Correct -- we don't need to hotplug the virtio backend.
> >
> > > +    bus->bus_in_use = false;
> > > +    DPRINTF("bus %s created\n", bus_name);
> > > +}
> > > +
> > > +/* Bind the VirtIODevice to the VirtioBus. */
> > > +void virtio_bus_bind_device(VirtioBus *bus)
> > > +{
> > > +    assert(bus != NULL);
> > > +    assert(bus->vdev != NULL);
> > > +    virtio_bind_device(bus->vdev, &(bus->info->virtio_bindings),
> > > +                       bus->qbus.parent);
> > This looks wrong -- virtio_bind_device() is part of the
> > old-style transport API.


> > > +}
> > > +
> > > +/* This must be called to when the VirtIODevice init */
> > > +int virtio_bus_init_cb(VirtioBus *bus, VirtIODevice *vdev)
> > > +{
> > > +    if (bus->bus_in_use == true) {
> > > +        error_report("%s in use.\n", bus->qbus.name);
> > > +        return -1;
> > > +    }
> > > +    assert(bus->info->init_cb != NULL);
> > > +    /* keep the VirtIODevice in the VirtioBus */
> > > +    bus->vdev = vdev;
> > This shouldn't be happening here, it should happen as
> > part of plugging the device into the bus.
> >
> > > +    bus->info->init_cb(bus->qbus.parent);
> > > +
> > > +    bus->bus_in_use = true;
> > > +    return 0;
> > > +}
> > > +
> > > +/* This must be called when the VirtIODevice exit */
> > > +void virtio_bus_exit_cb(VirtioBus *bus)
> > > +{
> > > +    assert(bus->info->exit_cb != NULL);
> > > +    bus->info->exit_cb(bus->qbus.parent);
> > > +    bus->bus_in_use = false;
> > > +}
> > These shouldn't be necessary -- the VirtioDevice should
> > have a pointer to the VirtioBus and can just invoke the
> > init/exit callbacks directly.
> >
> > > +
> > > +static char *virtio_bus_get_fw_dev_path(DeviceState *dev)
> > > +{
> > > +    return g_strdup_printf("%s", qdev_fw_name(dev));
> > > +}
> > > +
> > > +
> > > +static void virtio_register_types(void)
> > > +{
> > > +    type_register_static(&virtio_bus_info);
> > > +}
> > > +
> > > +type_init(virtio_register_types)
> > > diff --git a/hw/virtio-bus.h b/hw/virtio-bus.h
> > > new file mode 100644
> > > index 0000000..6ea5035
> > > --- /dev/null
> > > +++ b/hw/virtio-bus.h
> > > @@ -0,0 +1,49 @@
> > > +/*
> > > + * VirtioBus
> > > + *
> > > + *  Copyright (C) 2012 : GreenSocs Ltd
> > > + *      http://www.greensocs.com/ , email: address@hidden
> > > + *
> > > + *  Developed by :
> > > + *  Frederic Konrad   <address@hidden>
> > > + *
> > > + * This work is licensed under the terms of the GNU GPL, version 2.  See
> > > + * the COPYING file in the top-level directory.
> > > + */
> > > +
> > > +#ifndef _VIRTIO_BUS_H_
> > > +#define _VIRTIO_BUS_H_
> > Leading underscores are reserved by the C standard. VIRTIO_BUS_H will
> > do fine.
> >
> > > +
> > > +#include "qdev.h"
> > > +#include "sysemu.h"
> > > +#include "virtio.h"
> > > +
> > > +#define TYPE_VIRTIO_BUS "VIRTIO"
> > Type names are generally "virtio-bus" by convention.
> >
> > > +#define BUS_NAME "virtio"
> > I don't think you want this.
> >
> >
> > > +typedef struct VirtioBus VirtioBus;
> > > +typedef struct VirtioBusInfo VirtioBusInfo;
> > > +
> > > +struct VirtioBusInfo {
> > > +    void (*init_cb)(DeviceState *dev);
> > > +    void (*exit_cb)(DeviceState *dev);
> > > +    VirtIOBindings virtio_bindings;
> > This doesn't look right -- VirtIOBindings is the
> > old-style way for the transport to specify a bunch
> > of function pointers for its specific implementation.
> > Those function pointers should probably just be in
> > the VirtioBus struct.
> >
> > > +};
> > I was just talking to Anthony on IRC and he said SCSIBusInfo
> > is really kind of a historical accident. So we can just fold
> > this struct into VirtioBus. Sorry for misleading you earlier.
> >
> > > +struct VirtioBus {
> > > +    BusState qbus;
> > > +    int busnr;
> > Why does the bus need to know what number it is?
> >
> > > +    bool bus_in_use;
> > > +    uint16_t pci_device_id;
> > > +    uint16_t pci_class;
> > This shouldn't be here -- VirtioBus should be transport
> > independent, so no PCI related info.
> Agreed - this should be a property of the bridge device.
Yes, So I must add the virtiodevice identifier and put a switch case in 
the transport to set the right PCI IDs?
In that case, the transport won't be device independent.

> > > +    VirtIODevice *vdev;
> > This could use a comment saying that we only ever have one
> > child device on the bus.
> >
> > +    const VirtioBusInfo *info;
> > +};
> > +
> > +void virtio_bus_new(VirtioBus *bus, DeviceState *host,
> > +                    const VirtioBusInfo *info);
> > +void virtio_bus_bind_device(VirtioBus *bus);
> > +int virtio_bus_init_cb(VirtioBus *bus, VirtIODevice *dev);
> > +void virtio_bus_exit_cb(VirtioBus *bus);
> > +
> > +#endif
> > --
> > 1.7.11.7
> > -- PMM