Re: [PATCH v2 03/12] drm/drv: DOC: Add driver example code

From: "Noralf Trønnes" <noralf@tronnes.org>
To: Sam Ravnborg <sam@ravnborg.org>
Cc: daniel.vetter@ffwll.ch, david@lechnology.com,
	dri-devel@lists.freedesktop.org
Subject: Re: [PATCH v2 03/12] drm/drv: DOC: Add driver example code
Date: Tue, 12 Feb 2019 14:20:17 +0100	[thread overview]
Message-ID: <beba0a8b-8860-cb6e-e4fd-b136edd4c894@tronnes.org> (raw)
In-Reply-To: <20190210210352.GA2983@ravnborg.org>

Den 10.02.2019 22.03, skrev Sam Ravnborg:
> Hi Noralf
> 
> On Sun, Feb 10, 2019 at 02:10:30PM +0100, Noralf Trønnes wrote:
>> Add driver example that shows how devm_drm_dev_init() can be used.
>>
>> Signed-off-by: Noralf Trønnes <noralf@tronnes.org>
> 
> Always good with examples!
> 
> 
>> ---
>>
>> I'm not sure how detailed such an example such be and a description of 
>> some kind is also required. Help is needed :-)
>>
>> Noralf.
>>
>>  drivers/gpu/drm/drm_drv.c | 118 ++++++++++++++++++++++++++++++++++++++
>>  1 file changed, 118 insertions(+)
>>
>> diff --git a/drivers/gpu/drm/drm_drv.c b/drivers/gpu/drm/drm_drv.c
>> index 351f128ec4b7..99ca3551688f 100644
>> --- a/drivers/gpu/drm/drm_drv.c
>> +++ b/drivers/gpu/drm/drm_drv.c
>> @@ -286,6 +286,124 @@ void drm_minor_release(struct drm_minor *minor)
>>   * Note that the lifetime rules for &drm_device instance has still a lot of
>>   * historical baggage. Hence use the reference counting provided by
>>   * drm_dev_get() and drm_dev_put() only carefully.
>> + *
>> + * Display driver example
>> + * ~~~~~~~~~~~~~~~~~~~~~~
> 
> Maybe add something like this:
> 
> The following example shows a typical structure of a DRM display driver.
> The example focus on the probe() function and the other functions that is
> almost always present.
> The example is used to demonstrate the use of devm_drm_dev_init(), and
> shows how the use of a device managed drm_device allows for simpler
> remove() and shuttdown() functions.

Using devm_drm_dev_init() doesn't affect the use of shutdown(), and for
remove() it only saves one drm_dev_put(). The main benefit that I see is
for a simpler probe() function.

Maybe the last part can be:

The example is used to demonstrate the use of devm_drm_dev_init() and
its accompanying drm_driver->release callback.

Thanks for helping me with this. Daniel is big on documentation so I try
to follow up, but words for humans isn't my strong side :-)

> 
> 
>> + *
>> + * .. code-block:: c
>> + *
>> + *	struct driver_device {
>> + *		struct drm_device drm;

Maybe I should add a comment for userspace_facing:

     *		[ Represents a resource that is accessed by userspace ]
>> + *		void *userspace_facing;
>> + *		struct clk *pclk;
>> + *	};
>> + *
>> + *	static inline struct driver_device *drm_to_priv(struct drm_device *drm)
>> + *	{
>> + *		return container_of(drm, struct driver_device, drm);
>> + *	}
>> + *
>> + *	static void driver_drm_release(struct drm_device *drm)
>> + *	{
>> + *		struct driver_device *priv = drm_to_priv(drm);
>> + *
>> + *		drm_mode_config_cleanup(drm);
>> + *		drm_dev_fini(drm);
>> + *		kfree(priv->userspace_facing);
>> + *		kfree(priv);
>> + *	}
>> + *
>> + *	static struct drm_driver driver_drm_driver = {
>> + *		[...]
>> + *		.release = driver_drm_release,
>> + *	};
>> + *
>> + *	static int driver_probe(struct platform_device *pdev)
>> + *	{
>> + *		struct driver_device *priv;
>> + *		struct drm_device *drm;
>> + *		int ret;
>> + *
>> + *		priv = kzalloc(sizeof(*priv), GFP_KERNEL);
> Add a comment that explains why we cannot use devm_kzalloc()?
> It is used in a lot of places today...

That makes sense:

devm_kzalloc() can't be used here because the drm_device lifetime can
exceed the device lifetime if driver unbind happens when userspace still
has open file descriptors.

> 
>> + *		if (!priv)
>> + *			return -ENOMEM;
>> + *
>> + *		drm = &priv->drm;
>> + *
>> + *		ret = devm_drm_dev_init(&pdev->dev, drm, &driver_drm_driver);
>> + *		if (ret) {
>> + *			kfree(drm);
> This should be kfree(priv);
> 
>> + *			return ret;
>> + *		}
>> + *
>> + *		drm_mode_config_init(drm);
>> + *
>> + *		priv->userspace_facing = kzalloc(..., GFP_KERNEL);
>> + *		if (!priv->userspace_facing)
>> + *			return -ENOMEM;
> If we fail here we should also free priv?
> 

This is where one benfit of device managed resources kicks in.
If probing fails, happens in drivers/base/dd.c:really_probe(),
devres_release_all() is called which results in:
devm_drm_dev_init_release() -> drm_dev_put() -> driver_drm_release().

If the driver only uses devm_ managed resources or releases the
resource(s) in drm_driver->release, there's no need for an error path in
the probe function.

The ->userspace_facing resource is an attempt to showcase a resource
that is released in drm_driver->release, but I'm unsure if it's really a
good example. It would have been better to have a real world example.

>> + *
>> + *		priv->pclk = devm_clk_get(dev, "PCLK");
>> + *		if (IS_ERR(priv->pclk))
>> + *			return PTR_ERR(priv->pclk);
>> + *
>> + *		[ Further setup, display pipeline etc ]
>> + *
>> + *		drm_mode_config_reset(drm);
>> + *
>> + *		ret = drm_dev_register(drm);
>> + *		if (ret)
>> + *			return ret;
>> + *
>> + *		platform_set_drvdata(pdev, drm);
>> + *
>> + *		drm_fbdev_generic_setup(drm, 32);
>> + *
>> + *		return 0;
>> + *	}
>> + *
>> + *	[ This function is called before the devm_ resources are released ]
>> + *	static int driver_remove(struct platform_device *pdev)
>> + *	{
>> + *		struct drm_device *drm = platform_get_drvdata(pdev);
>> + *
>> + *		drm_dev_unregister(drm);

You asked about drm_dev_unplug() in the other patch. It is used if the
driver supports device unplug, like for USB or unloading a Device Tree
overlay (not sure if all the pieces exists in mainline yet for that).

Maybe I can add this section after the example:

Drivers that want to support device unplugging (USB, DT overlay unload)
should use drm_dev_unplug() instead of drm_dev_unregister(). The driver
must protect regions that is accessing device resources to avoid use
after they're released. This is done using drm_dev_enter() and
drm_dev_exit(). There is one shortcoming however, drm_dev_unplug() marks
the drm_device as unplugged before drm_atomic_helper_shutdown() is
called. This means that if the disable code paths are protected, they
will not run on regular driver module unload, possibily leaving the
hardware enabled.

>> + *		drm_atomic_helper_shutdown(drm)
>> + *
>> + *		return 0;
>> + *	}
>> + *

I can add a comment for shutdown:

This function is called on kernel restart and shutdown

>> + *	static void driver_shutdown(struct platform_device *pdev)
>> + *	{
>> + *		drm_atomic_helper_shutdown(platform_get_drvdata(pdev));
>> + *	}
>> + *
>> + *	static int __maybe_unused driver_pm_suspend(struct device *dev)
>> + *	{
>> + *		return drm_mode_config_helper_suspend(dev_get_drvdata(dev));
>> + *	}
>> + *
>> + *	static int __maybe_unused driver_pm_resume(struct device *dev)
>> + *	{
>> + *		drm_mode_config_helper_resume(dev_get_drvdata(dev));
>> + *
>> + *		return 0;
>> + *	}
>> + *
>> + *	static const struct dev_pm_ops driver_pm_ops = {
>> + *		SET_SYSTEM_SLEEP_PM_OPS(driver_pm_suspend, driver_pm_resume)
>> + *	};
>> + *
>> + *	static struct platform_driver driver_driver = {
>> + *		.driver = {
>> + *			[...]
>> + *			.pm = &driver_pm_ops,
>> + *		},
>> + *		.probe = driver_probe,
>> + *		.remove = driver_remove,
>> + *		.shutdown = driver_shutdown,
>> + *	};
>> + *	module_platform_driver(driver_driver);
>> + *
>>   */
>>  
>>  /**
>> -- 
>> 2.20.1
>>
>> _______________________________________________
>> dri-devel mailing list
>> dri-devel@lists.freedesktop.org
>> https://lists.freedesktop.org/mailman/listinfo/dri-devel
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel