Re: [Qemu-devel] [PATCH v1 13/16] qdev: gpio: Define qdev_pass_gpios()

[Alexander Graf]

On 12.08.14 12:48, Peter Crosthwaite wrote:
> On Tue, Aug 12, 2014 at 7:24 PM, Alexander Graf <address@hidden> wrote:
> > On 04.08.14 03:58, Peter Crosthwaite wrote:
> > > Allows a container to take ownership of GPIOs in a contained
> > > device and automatically connect them as GPIOs to the container.
> > >
> > > This prepares for deprecation of the SYSBUS IRQ functionality, which
> > > has this feature. We push it up to the device level instead of sysbus
> > > level. There's nothing sysbus specific about passing GPIOs to
> > > containers so its a legitimate device-level generic feature.
> > >
> > > Signed-off-by: Peter Crosthwaite <address@hidden>
> > > ---
> > >
> > >    hw/core/qdev.c         | 28 ++++++++++++++++++++++++++++
> > >    include/hw/qdev-core.h |  3 +++
> > >    2 files changed, 31 insertions(+)
> > >
> > > diff --git a/hw/core/qdev.c b/hw/core/qdev.c
> > > index bf2c227..708363f 100644
> > > --- a/hw/core/qdev.c
> > > +++ b/hw/core/qdev.c
> > > @@ -440,6 +440,34 @@ void qdev_connect_gpio_out(DeviceState * dev, int n,
> > > qemu_irq pin)
> > >        qdev_connect_gpio_out_named(dev, NULL, n, pin);
> > >    }
> > >    +void qdev_pass_gpios(DeviceState *dev, DeviceState *container,
> > > +                     const char *name)
> > > +{
> > > +    int i;
> > > +    NamedGPIOList *ngl = qdev_get_named_gpio_list(dev, name);
> > > +
> > > +    for (i = 0; i < ngl->num_in; i++) {
> > > +        char *propname = g_strdup_printf("%s[%d]",
> > > +                                         ngl->name ? ngl->name :
> > > +                                                   "unnamed-gpio-in",
> >
> > Really just a minor nit, but I think the code flow would look a lot nicer if
> > you did the name check in a separate variable.
> >
> >    const char *name = ngl->name ? ngl->name : "unnamed-gpio-in";
> I think I may even go an extra step and get it macroified. How about:
>
> #define QDEV_GPIO_IN_NAME(a) ((a) ? (a) : "unnamed-gpio-io")
>
> and then you can continue to use it inline without extra variables or
> nasty "?:"?

The variable will get optimized out, so for the sake of readability I 
would still vote to have it around. Whether you declare it via a macro 
or with an a ? a : b macro I don't really care much about :).

In fact, it might make more sense to literally have the typical

  (a ? a : b)

flow macroified in a generic place somewhere.

> >    for (i = 0; ...) {
> >      char *propname = g_strdup_printf("%s[%d]", name, i);
> >      ....
> >    }
> >
> > Also I don't fully grasp what the naming scheme is supposed to be here. Who
> > sets the name and why is there only a single global name for all GPIOs?
> Ideally, the instantiating device sets the names. There's not a global
> name for all GPIOs, just an over-used default. The intention is that
> qdev_init_gpio_in is phased out in favor of qdev_init_gpio_in_named.
> If NULL name is given or qdev_init_gpio_in is used, it defaults to
> this single global name here. All sysbus IRQs also share a  single
> name (seperate from the qdev default) but that sharing is implemented
> on the sysbus level and transparent to qdev. The need for a default
> name is to appease QOM, which needs a valid string for canonical path.

Ah, I missed the fact that qdev_get_named_gpio_list() returns a list of 
elements with exactly the name you searched for in the first place.


Alex