Re: [PATCH] powerpc/vio: use simple dummy struct device as bus parent

[Greg Kroah-Hartman]

On Thu, Oct 03, 2019 at 03:10:03PM -0400, Dan Streetman wrote:
> On Fri, Sep 27, 2019 at 2:19 PM Greg Kroah-Hartman
> <gregkh@xxxxxxxxxxxxxxxxxxx> wrote:
> >
> > On Fri, Sep 27, 2019 at 09:04:02AM -0400, Dan Streetman wrote:
> > > The dummy vio_bus_device creates the /sys/devices/vio directory, which
> > > contains real vio devices under it; since it represents itself as having
> > > a bus = &vio_bus_type, its /sys/devices/vio/uevent does call the bus's
> > > .uevent function, vio_hotplug(), and as that function won't find a real
> > > device for the dummy vio_dev, it will return -ENODEV.
> > >
> > > One of the main users of the uevent node is udevadm, e.g. when it is called
> > > with 'udevadm trigger --devices'.  Up until recently, it would ignore any
> > > errors returned when writing to devices' uevent file, but it was recently
> > > changed to start returning error if it gets an error writing to any uevent
> > > file:
> > > https://github.com/systemd/systemd/commit/97afc0351a96e0daa83964df33937967c75c644f
> > >
> > > since the /sys/devices/vio/uevent file has always returned ENODEV from
> > > any write to it, this now causes the udevadm trigger command to return
> > > an error.  This may be fixed in udevadm to ignore ENODEV errors, but the
> > > vio driver should still be fixed.
> > >
> > > This patch changes the arch/powerpc/platform/pseries/vio.c 'dummy'
> > > parent device into a real dummy device with no .bus, so its uevent
> > > file will stop returning ENODEV and simply do nothing and return 0.
> > >
> > > Signed-off-by: Dan Streetman <ddstreet@xxxxxxxxxxxxx>
> > > ---
> > >  arch/powerpc/platforms/pseries/vio.c | 11 ++++-------
> > >  1 file changed, 4 insertions(+), 7 deletions(-)
> > >
> > > diff --git a/arch/powerpc/platforms/pseries/vio.c b/arch/powerpc/platforms/pseries/vio.c
> > > index 79e2287991db..63bc16631680 100644
> > > --- a/arch/powerpc/platforms/pseries/vio.c
> > > +++ b/arch/powerpc/platforms/pseries/vio.c
> > > @@ -32,11 +32,8 @@
> > >  #include <asm/page.h>
> > >  #include <asm/hvcall.h>
> > >
> > > -static struct vio_dev vio_bus_device  = { /* fake "parent" device */
> > > -     .name = "vio",
> > > -     .type = "",
> > > -     .dev.init_name = "vio",
> > > -     .dev.bus = &vio_bus_type,
> > > +static struct device vio_bus = {
> > > +     .init_name      = "vio",
> >
> > Eeek, no!  Why are you creating a static device that will then be
> > reference counted?  Not nice :(
> 
> so, I looked again and it seems quite a few places appear to do
> exactly this, is it something that should be fixed?

Yes.