Re: [PATCH v3] drm/tilcdc: Fix removal actions in case of failed probe

[Kory Maincent <kory.maincent@bootlin.com>]

On Wed, 19 Nov 2025 18:12:40 +0100
"Luca Ceresoli" <luca.ceresoli@bootlin.com> wrote:

> Hello Köry,
> 
> On Tue Nov 18, 2025 at 2:38 PM CET, Kory Maincent wrote:
> > From: "Kory Maincent (TI.com)" <kory.maincent@bootlin.com>
> >
> > The drm_kms_helper_poll_fini() and drm_atomic_helper_shutdown() helpers
> > should only be called when the device has been successfully registered.
> > Currently, these functions are called unconditionally in tilcdc_fini(),
> > which causes warnings during probe deferral scenarios.
> >
> > [    7.972317] WARNING: CPU: 0 PID: 23 at
> > drivers/gpu/drm/drm_atomic_state_helper.c:175
> > drm_atomic_helper_crtc_duplicate_state+0x60/0x68 ... [    8.005820]
> > drm_atomic_helper_crtc_duplicate_state from
> > drm_atomic_get_crtc_state+0x68/0x108 [    8.005858]
> > drm_atomic_get_crtc_state from drm_atomic_helper_disable_all+0x90/0x1c8 [
> >  8.005885]  drm_atomic_helper_disable_all from
> > drm_atomic_helper_shutdown+0x90/0x144 [    8.005911]
> > drm_atomic_helper_shutdown from tilcdc_fini+0x68/0xf8 [tilcdc] [
> > 8.005957]  tilcdc_fini [tilcdc] from tilcdc_pdev_probe+0xb0/0x6d4 [tilcdc]
> >
> > Fix this by rewriting the failed probe cleanup path using the standard
> > goto error handling pattern, which ensures that cleanup functions are
> > only called on successfully initialized resources. Additionally, remove
> > the now-unnecessary is_registered flag.
> >
> > Cc: stable@vger.kernel.org
> > Fixes: 3c4babae3c4a ("drm: Call drm_atomic_helper_shutdown() at
> > shutdown/remove time for misc drivers") Signed-off-by: Kory Maincent
> > (TI.com) <kory.maincent@bootlin.com>  
> 
> Except for the bug reported by the kernel test robot, this patch looks
> good to me. Just a couple thoughts, below.
> 
> > @@ -372,16 +371,34 @@ static int tilcdc_init(const struct drm_driver *ddrv,
> > struct device *dev)
> >
> >  	ret = drm_dev_register(ddev, 0);
> >  	if (ret)
> > -		goto init_failed;
> > -	priv->is_registered = true;
> > +		goto stop_poll;
> >
> >  	drm_client_setup_with_color_mode(ddev, bpp);
> >
> >  	return 0;
> >
> > -init_failed:
> > -	tilcdc_fini(ddev);
> > +stop_poll:
> > +	drm_kms_helper_poll_fini(ddev);
> > +	tilcdc_irq_uninstall(ddev);
> > +unbind_component:
> > +	if (priv->is_componentized)
> > +		component_unbind_all(dev, ddev);
> > +unregister_cpufreq_notif:
> > +#ifdef CONFIG_CPU_FREQ
> > +	cpufreq_unregister_notifier(&priv->freq_transition,
> > +				    CPUFREQ_TRANSITION_NOTIFIER);
> > +#endif
> > +destroy_crtc:
> > +	tilcdc_crtc_destroy(priv->crtc);
> > +disable_pm:
> > +	pm_runtime_disable(dev);
> > +	clk_put(priv->clk);
> > +free_wq:
> > +	destroy_workqueue(priv->wq);
> > +put_drm:
> >  	platform_set_drvdata(pdev, NULL);  
> 
> I'm not 100% sure this is needed, but perhaps it is because of the
> component framework being used.

Yes not sure either but as it was already present I let it here.
Do you think I should remove it?

> 
> If it is needed, then shouldn't it be present in tilcdc_fini() as well?
> 
> > +	ddev->dev_private = NULL;
> > +	drm_dev_put(ddev);
> >
> >  	return ret;
> >  }  
> 
> About tilcdc_fini(), I think it can be itself cleaned up a lot (in another
> patch). Basically it should do the same thing (almost) that are here below
> the 'return 0' line, and in the same order. Now the list of actions is auite
> different and the order is very different.

Yes indeed, but this won't be a fix as there is no real issue in the remove
AFAIK.

Regards,
-- 
Köry Maincent, Bootlin
Embedded Linux and kernel engineering
https://bootlin.com