Re: [PATCH] platform_driver_register: warn if probe is in .init.text

From: Dmitry Torokhov
Date: Mon Jan 25 2010 - 12:15:10 EST


On Sat, Jan 23, 2010 at 10:49:16AM +0100, Uwe Kleine-König wrote:
> Helle Dmitry,
>
> On Fri, Jan 22, 2010 at 03:49:37PM -0800, Dmitry Torokhov wrote:
> > On Fri, Jan 22, 2010 at 08:49:18PM +0100, Uwe Kleine-König wrote:
> > > +int platform_driver_register(struct platform_driver *drv)
> > > +{
> > > + int ret = __platform_driver_register(drv);
> > > +
> > > +#if defined(CONFIG_HOTPLUG)
> > > + /*
> > > + * drivers that are registered by platform_driver_register
> > > + * should not have their probe function in .init.text. The
> > > + * reason is that a probe can happen after .init.text is
> > > + * discarded which then results in an oops. The alternatives
> > > + * are using .devinit.text for the probe function or "register"
> > > + * with platform_driver_probe.
> > > + */
> >
> >
> > What about CONFIG_MODULE case (without hotplug)? The driver may be
> > compiled in (and thus probe will be discarded) but device might be
> > created by a loadable module which is loaded later.
> I don't see the problem here.

You are right, without hotplug we don't try to bind to devices that
come in later so we are safe to discard probe.

> If the driver is registered using
> platform_driver_register and the probe function lives in .init.text it
> doesn't matter where the device comes from. If there is one you can
> unbind and bind at a later time when .init is already discarded. No?
>

Yes.

> More general, using platform_driver_register the pointer to the probe
> function have to stay valid as it is saved for later use, so it must not
> be an .init pointer.


>
> > > + if (drv->probe && kernel_init_text_address((unsigned long)drv->probe))
> > > + pr_warning("oops-warning: probe function of platform driver %s"
> > > + " lives in .init.text\n", drv->driver.name);
> > > +#else
> > > + /*
> > > + * without HOTPLUG probe functions can be discarded after the driver is
> > > + * loaded.
> > > + * There is a little chance for false positives, namely if the driver is
> > > + * registered after the .init sections are discarded.
> > > + */
> > > + if (drv->probe && !kernel_init_text_address((unsigned long)drv->probe))
> > > + pr_info("probably the probe function of platform driver %s can"
> > > + " be moved to .init.text\n", drv->driver.name);
> >
> > It really depends on the user's (or rather builder's) local
> > configuration so how can we make this suggestion?
> Can you point out a situation that makes the suggestion wrong? AFAIK
> there only the possibility that the driver is loaded after .init is
> discarded (as written in the comment).
>

The presence of CONFIG_HOTPLUG depends on particular .config, so, for
example, you may not have it set while I have. So you will boot, get the
suggestion, submit a patch, then I will boot and will get a warning that
the probe function is in wrong section.

So in general I wonder if it is possible to request that all probe
functions are marked as __devinit or not marked at all, regardless of
CONFIG_HOTPLUG and CONFIG_MODULE settings.

--
Dmitry
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/