Re: [PATCH v4 21/37] sm501: make SerialMM a child, export chardev property

[Marc-André Lureau]

On Thu, Nov 21, 2019 at 6:04 PM Peter Maydell <address@hidden> wrote:
>
> On Wed, 20 Nov 2019 at 15:31, Marc-André Lureau
> <address@hidden> wrote:
> >
> > Embed the SerialMM sybus device, and re-export its "chardev" property.
> > That way, we can get rid of PROP_PTR "chr-state" and better track
> > devices relationship.
> >
> > Signed-off-by: Marc-André Lureau <address@hidden>
> > ---
> >  hw/display/sm501.c | 33 ++++++++++++++++++++++++---------
> >  hw/sh4/r2d.c       |  2 +-
> >  2 files changed, 25 insertions(+), 10 deletions(-)
> >
> > diff --git a/hw/display/sm501.c b/hw/display/sm501.c
> > index 1f33c87e65..c4445b28f9 100644
> > --- a/hw/display/sm501.c
> > +++ b/hw/display/sm501.c
> > @@ -1930,13 +1930,14 @@ typedef struct {
> >      SM501State state;
> >      uint32_t vram_size;
> >      uint32_t base;
> > -    void *chr_state;
> > +    SerialMM serial;
> >  } SM501SysBusState;
> >
> >  static void sm501_realize_sysbus(DeviceState *dev, Error **errp)
> >  {
> >      SM501SysBusState *s = SYSBUS_SM501(dev);
> >      SysBusDevice *sbd = SYS_BUS_DEVICE(dev);
> > +    SerialState *ss = &s->serial.serial;
> >      DeviceState *usb_dev;
> >
> >      sm501_init(&s->state, dev, s->vram_size);
> > @@ -1958,17 +1959,19 @@ static void sm501_realize_sysbus(DeviceState *dev, 
> > Error **errp)
> >      sysbus_pass_irq(sbd, SYS_BUS_DEVICE(usb_dev));
> >
> >      /* bridge to serial emulation module */
> > -    if (s->chr_state) {
> > -        serial_mm_init(&s->state.mmio_region, SM501_UART0, 2,
> > -                       NULL, /* TODO : chain irq to IRL */
> > -                       115200, s->chr_state, DEVICE_LITTLE_ENDIAN);
> > +    /* FIXME: SM501_UART0 is always mapped, no need to check for the 
> > backend */
> > +    if (qemu_chr_fe_backend_connected(&ss->chr)) {
> > +        MemoryRegion *mr;
> > +        qdev_init_nofail(DEVICE(&s->serial));
> > +        mr = sysbus_mmio_get_region(SYS_BUS_DEVICE(&s->serial), 0);
> > +        memory_region_add_subregion(&s->state.mmio_region, SM501_UART0, 
> > mr);
> > +        /* TODO : chain irq to IRL */
> >      }
>
> I don't really understand what the FIXME is trying to
> tell me here. If we don't need to check for the backend,
> why is the code checking for it ? It means we have to fish
> around inside the SerialMM's implementation, which seems odd.
> Only mapping the UART registers if there happens to be a backend
> connected also doesn't conceptually seem like the right behaviour,
> because the registers should always exist. Since commit
> 12051d82f004024d5d the chardev mid-layer has correctly handled
> the backend not being connected (ie having a NULL chardev),
> so there's no longer any need for board/device code to special
> case the lack of a chardev.

As you noted, this is addressed by "[PATCH-for-5.0] hw/display/sm501:
Always map the UART0".

It's preferable to keep those 2 seperate. I can include Philippe
change on top, or earlier in my series, is that ok?

>
> >  }
> >
> >  static Property sm501_sysbus_properties[] = {
> >      DEFINE_PROP_UINT32("vram-size", SM501SysBusState, vram_size, 0),
> >      DEFINE_PROP_UINT32("base", SM501SysBusState, base, 0),
> > -    DEFINE_PROP_PTR("chr-state", SM501SysBusState, chr_state),
> >      DEFINE_PROP_END_OF_LIST(),
> >  };
> >
> > @@ -1999,9 +2002,20 @@ static void sm501_sysbus_class_init(ObjectClass 
> > *klass, void *data)
> >      dc->props = sm501_sysbus_properties;
> >      dc->reset = sm501_reset_sysbus;
> >      dc->vmsd = &vmstate_sm501_sysbus;
> > -    /* Note: pointer property "chr-state" may remain null, thus
> > -     * no need for dc->user_creatable = false;
> > -     */
> > +}
> > +
> > +static void sm501_sysbus_init(Object *o)
> > +{
> > +    SM501SysBusState *sm501 = SYSBUS_SM501(o);
> > +    SerialMM *smm = &sm501->serial;
> > +
> > +    sysbus_init_child_obj(o, "serial", smm, sizeof(SerialMM), 
> > TYPE_SERIAL_MM);
> > +    qdev_set_legacy_instance_id(DEVICE(smm), SM501_UART0, 2);
>
> The only board we use the sm501 sysbus device is the sh4 r2d
> board, and we don't care about migration compatibility there
> (indeed I would be unsurprised to find that it doesn't even work ;-))
> So I think we can reasonably not set the legacy-instance-ID
> and just declare in the commit message that this is a migration
> compat break for that board.
>

I am simply adapting serial_mm_init() code from "serial: register vmsd
with DeviceClass" here.

Dropping qdev_set_legacy_instance_id() is proabably a seperate concern.

> > +    qdev_prop_set_uint8(DEVICE(smm), "regshift", 2);
> > +    qdev_prop_set_uint8(DEVICE(smm), "endianness", DEVICE_LITTLE_ENDIAN);
> > +
> > +    object_property_add_alias(o, "chardev",
> > +                              OBJECT(smm), "chardev", &error_abort);
> >  }
> >
> >  static const TypeInfo sm501_sysbus_info = {
> > @@ -2009,6 +2023,7 @@ static const TypeInfo sm501_sysbus_info = {
> >      .parent        = TYPE_SYS_BUS_DEVICE,
> >      .instance_size = sizeof(SM501SysBusState),
> >      .class_init    = sm501_sysbus_class_init,
> > +    .instance_init = sm501_sysbus_init,
> >  };
> >
> >  #define TYPE_PCI_SM501 "sm501"
> > diff --git a/hw/sh4/r2d.c b/hw/sh4/r2d.c
> > index ee0840f380..72bb5285cc 100644
> > --- a/hw/sh4/r2d.c
> > +++ b/hw/sh4/r2d.c
> > @@ -272,7 +272,7 @@ static void r2d_init(MachineState *machine)
> >      busdev = SYS_BUS_DEVICE(dev);
> >      qdev_prop_set_uint32(dev, "vram-size", SM501_VRAM_SIZE);
> >      qdev_prop_set_uint32(dev, "base", 0x10000000);
> > -    qdev_prop_set_ptr(dev, "chr-state", serial_hd(2));
> > +    qdev_prop_set_chr(dev, "chardev", serial_hd(2));
> >      qdev_init_nofail(dev);
> >      sysbus_mmio_map(busdev, 0, 0x10000000);
> >      sysbus_mmio_map(busdev, 1, 0x13e00000);
> > --
> > 2.24.0
>
> thanks
> -- PMM
>


-- 
Marc-André Lureau