Re: [PATCH v6 4/8] drm: tda998x: use cec_notifier_conn_(un)register

Re: [PATCH v6 4/8] drm: tda998x: use cec_notifier_conn_(un)register

[Russell King - ARM Linux admin]

On Tue, Aug 13, 2019 at 01:02:36PM +0200, Dariusz Marcinkiewicz wrote:
> Use the new cec_notifier_conn_(un)register() functions to
> (un)register the notifier for the HDMI connector, and fill
> in the cec_connector_info.
> 
> Changes since v2:
> 	- cec_notifier_phys_addr_invalidate where appropriate,
> 	- don't check for NULL notifier before calling
> 	cec_notifier_conn_unregister.
> Changes since v1:
> 	Add memory barrier to make sure that the notifier
> 	becomes visible to the irq thread once it is
> 	fully constructed.
> 
> Signed-off-by: Dariusz Marcinkiewicz <darekm@google.com>
> Tested-by: Hans Verkuil <hverkuil-cisco@xs4all.nl>
> ---
>  drivers/gpu/drm/i2c/tda998x_drv.c | 33 +++++++++++++++++++++----------
>  1 file changed, 23 insertions(+), 10 deletions(-)
> 
> diff --git a/drivers/gpu/drm/i2c/tda998x_drv.c b/drivers/gpu/drm/i2c/tda998x_drv.c
> index 61e042918a7fc..19a63ee1b3f53 100644
> --- a/drivers/gpu/drm/i2c/tda998x_drv.c
> +++ b/drivers/gpu/drm/i2c/tda998x_drv.c
> @@ -804,9 +804,14 @@ static irqreturn_t tda998x_irq_thread(int irq, void *data)
>  			if (lvl & CEC_RXSHPDLEV_HPD) {
>  				tda998x_edid_delay_start(priv);
>  			} else {
> +				struct cec_notifier *notify;
> +
>  				schedule_work(&priv->detect_work);
> -				cec_notifier_set_phys_addr(priv->cec_notify,
> -						   CEC_PHYS_ADDR_INVALID);
> +
> +				notify = READ_ONCE(priv->cec_notify);
> +				if (notify)
> +					cec_notifier_phys_addr_invalidate(
> +							notify);
>  			}
>  
>  			handled = true;
> @@ -1331,6 +1336,8 @@ static int tda998x_connector_init(struct tda998x_priv *priv,
>  				  struct drm_device *drm)
>  {
>  	struct drm_connector *connector = &priv->connector;
> +	struct cec_connector_info conn_info;
> +	struct cec_notifier *notifier;
>  	int ret;
>  
>  	connector->interlace_allowed = 1;
> @@ -1347,6 +1354,19 @@ static int tda998x_connector_init(struct tda998x_priv *priv,
>  	if (ret)
>  		return ret;
>  
> +	cec_fill_conn_info_from_drm(&conn_info, connector);
> +
> +	notifier = cec_notifier_conn_register(priv->cec_glue.parent,
> +					      NULL, &conn_info);
> +	if (!notifier)
> +		return -ENOMEM;
> +	/*
> +	 * Make sure that tda998x_irq_thread does see the notifier
> +	 * when it fully constructed.
> +	 */
> +	smp_wmb();
> +	priv->cec_notify = notifier;

To nitpick, this comment and the following code do not go together.

I think what you actually mean is:

	 * Make sure that tda998x_irq_thread sees the notifier
	 * only after it is fully constructed.

> +
>  	drm_connector_attach_encoder(&priv->connector,
>  				     priv->bridge.encoder);
>  
> @@ -1790,8 +1810,7 @@ static void tda998x_destroy(struct device *dev)
>  
>  	i2c_unregister_device(priv->cec);
>  
> -	if (priv->cec_notify)
> -		cec_notifier_put(priv->cec_notify);
> +	cec_notifier_conn_unregister(priv->cec_notify);

This also doesn't make sense: tda998x_destroy() is the opposite of
tda998x_create().  However, tda998x_connector_destroy() is the
opposite of tda998x_connector_create().

By moving the CEC creation code into tda998x_connector_create(), you
are creating the possibility for the following sequence to mess up
CEC and leak:

	tda998x_create()
	tda998x_connector_create()
	tda998x_connector_destroy()
	tda998x_connector_create()
	tda998x_connector_destroy()
	tda998x_destroy()

Anything you create in tda998x_connector_create() must be cleaned up
by tda998x_connector_destroy().

>  }
>  
>  static int tda998x_create(struct device *dev)
> @@ -1916,12 +1935,6 @@ static int tda998x_create(struct device *dev)
>  		cec_write(priv, REG_CEC_RXSHPDINTENA, CEC_RXSHPDLEV_HPD);
>  	}
>  
> -	priv->cec_notify = cec_notifier_get(dev);
> -	if (!priv->cec_notify) {
> -		ret = -ENOMEM;
> -		goto fail;
> -	}
> -
>  	priv->cec_glue.parent = dev;
>  	priv->cec_glue.data = priv;
>  	priv->cec_glue.init = tda998x_cec_hook_init;
> -- 
> 2.23.0.rc1.153.gdeed80330f-goog
> 
> 

-- 
RMK's Patch system: https://www.armlinux.org.uk/developer/patches/
FTTC broadband for 0.8mile line in suburbia: sync at 12.1Mbps down 622kbps up
According to speedtest.net: 11.9Mbps down 500kbps up

Re: [PATCH v6 4/8] drm: tda998x: use cec_notifier_conn_(un)register

[Dariusz Marcinkiewicz]

Hello.

On Tue, Aug 13, 2019 at 1:20 PM Russell King - ARM Linux admin
<linux@armlinux.org.uk> wrote:
>
> This also doesn't make sense: tda998x_destroy() is the opposite of
> tda998x_create().  However, tda998x_connector_destroy() is the
> opposite of tda998x_connector_create().
>
> By moving the CEC creation code into tda998x_connector_create(), you
> are creating the possibility for the following sequence to mess up
> CEC and leak:
>
>         tda998x_create()
>         tda998x_connector_create()
>         tda998x_connector_destroy()
>         tda998x_connector_create()
>         tda998x_connector_destroy()
>         tda998x_destroy()
>
> Anything you create in tda998x_connector_create() must be cleaned up
> by tda998x_connector_destroy().
>
Thank you.

I've just sent out another revision of the patch, where registration
and deregistration is symmetric. Please take a look.

Best regards.

Re: [PATCH v6 3/8] tda9950: use cec_notifier_cec_adap_(un)register

[Russell King - ARM Linux admin]

On Tue, Aug 13, 2019 at 01:02:35PM +0200, Dariusz Marcinkiewicz wrote:
> Use the new cec_notifier_cec_adap_(un)register() functions to
> (un)register the notifier for the CEC adapter.
> 
> Signed-off-by: Dariusz Marcinkiewicz <darekm@google.com>
> Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl>
> Tested-by: Hans Verkuil <hverkuil-cisco@xs4all.nl>
> ---
>  drivers/gpu/drm/i2c/tda9950.c | 12 ++++++------
>  1 file changed, 6 insertions(+), 6 deletions(-)
> 
> diff --git a/drivers/gpu/drm/i2c/tda9950.c b/drivers/gpu/drm/i2c/tda9950.c
> index 8039fc0d83db4..a5a75bdeb7a5f 100644
> --- a/drivers/gpu/drm/i2c/tda9950.c
> +++ b/drivers/gpu/drm/i2c/tda9950.c
> @@ -420,7 +420,8 @@ static int tda9950_probe(struct i2c_client *client,
>  		priv->hdmi = glue->parent;
>  
>  	priv->adap = cec_allocate_adapter(&tda9950_cec_ops, priv, "tda9950",
> -					  CEC_CAP_DEFAULTS,
> +					  CEC_CAP_DEFAULTS |
> +					  CEC_CAP_CONNECTOR_INFO,
>  					  CEC_MAX_LOG_ADDRS);
>  	if (IS_ERR(priv->adap))
>  		return PTR_ERR(priv->adap);
> @@ -457,13 +458,14 @@ static int tda9950_probe(struct i2c_client *client,
>  	if (ret < 0)
>  		return ret;
>  
> -	priv->notify = cec_notifier_get(priv->hdmi);
> +	priv->notify = cec_notifier_cec_adap_register(priv->hdmi, NULL,
> +						      priv->adap);
>  	if (!priv->notify)
>  		return -ENOMEM;
>  
>  	ret = cec_register_adapter(priv->adap, priv->hdmi);
>  	if (ret < 0) {
> -		cec_notifier_put(priv->notify);
> +		cec_notifier_cec_adap_unregister(priv->notify);
>  		return ret;
>  	}
>  
> @@ -473,8 +475,6 @@ static int tda9950_probe(struct i2c_client *client,
>  	 */
>  	devm_remove_action(dev, tda9950_cec_del, priv);
>  
> -	cec_register_cec_notifier(priv->adap, priv->notify);
> -
>  	return 0;
>  }
>  
> @@ -482,8 +482,8 @@ static int tda9950_remove(struct i2c_client *client)
>  {
>  	struct tda9950_priv *priv = i2c_get_clientdata(client);
>  
> +	cec_notifier_cec_adap_unregister(priv->notify);
>  	cec_unregister_adapter(priv->adap);
> -	cec_notifier_put(priv->notify);

It looks weird to have an unexpectedly different ordering of
unregistration from the registration path - normally, unregistration
is the reverse order of initialisation.

In the initialisation path, it seems that we register the notifier
and _then_ the adapter.  Here, we unregister the notifier and then
the adapter rather than what would normally be expected.  Why is
this?  I suspect there will be drivers created that do this the
"normal" way round, so if this is a requirement, it needs to be made
plainly obvious.

>  
>  	return 0;
>  }
> -- 
> 2.23.0.rc1.153.gdeed80330f-goog
> 
> 

-- 
RMK's Patch system: https://www.armlinux.org.uk/developer/patches/
FTTC broadband for 0.8mile line in suburbia: sync at 12.1Mbps down 622kbps up
According to speedtest.net: 11.9Mbps down 500kbps up

Re: [PATCH v6 3/8] tda9950: use cec_notifier_cec_adap_(un)register

[Hans Verkuil]

On 8/13/19 1:32 PM, Russell King - ARM Linux admin wrote:
> On Tue, Aug 13, 2019 at 01:02:35PM +0200, Dariusz Marcinkiewicz wrote:
>> Use the new cec_notifier_cec_adap_(un)register() functions to
>> (un)register the notifier for the CEC adapter.
>>
>> Signed-off-by: Dariusz Marcinkiewicz <darekm@google.com>
>> Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl>
>> Tested-by: Hans Verkuil <hverkuil-cisco@xs4all.nl>
>> ---
>>  drivers/gpu/drm/i2c/tda9950.c | 12 ++++++------
>>  1 file changed, 6 insertions(+), 6 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/i2c/tda9950.c b/drivers/gpu/drm/i2c/tda9950.c
>> index 8039fc0d83db4..a5a75bdeb7a5f 100644
>> --- a/drivers/gpu/drm/i2c/tda9950.c
>> +++ b/drivers/gpu/drm/i2c/tda9950.c
>> @@ -420,7 +420,8 @@ static int tda9950_probe(struct i2c_client *client,
>>  		priv->hdmi = glue->parent;
>>  
>>  	priv->adap = cec_allocate_adapter(&tda9950_cec_ops, priv, "tda9950",
>> -					  CEC_CAP_DEFAULTS,
>> +					  CEC_CAP_DEFAULTS |
>> +					  CEC_CAP_CONNECTOR_INFO,
>>  					  CEC_MAX_LOG_ADDRS);
>>  	if (IS_ERR(priv->adap))
>>  		return PTR_ERR(priv->adap);
>> @@ -457,13 +458,14 @@ static int tda9950_probe(struct i2c_client *client,
>>  	if (ret < 0)
>>  		return ret;
>>  
>> -	priv->notify = cec_notifier_get(priv->hdmi);
>> +	priv->notify = cec_notifier_cec_adap_register(priv->hdmi, NULL,
>> +						      priv->adap);
>>  	if (!priv->notify)
>>  		return -ENOMEM;
>>  
>>  	ret = cec_register_adapter(priv->adap, priv->hdmi);
>>  	if (ret < 0) {
>> -		cec_notifier_put(priv->notify);
>> +		cec_notifier_cec_adap_unregister(priv->notify);
>>  		return ret;
>>  	}
>>  
>> @@ -473,8 +475,6 @@ static int tda9950_probe(struct i2c_client *client,
>>  	 */
>>  	devm_remove_action(dev, tda9950_cec_del, priv);
>>  
>> -	cec_register_cec_notifier(priv->adap, priv->notify);
>> -
>>  	return 0;
>>  }
>>  
>> @@ -482,8 +482,8 @@ static int tda9950_remove(struct i2c_client *client)
>>  {
>>  	struct tda9950_priv *priv = i2c_get_clientdata(client);
>>  
>> +	cec_notifier_cec_adap_unregister(priv->notify);
>>  	cec_unregister_adapter(priv->adap);
>> -	cec_notifier_put(priv->notify);
> 
> It looks weird to have an unexpectedly different ordering of
> unregistration from the registration path - normally, unregistration
> is the reverse order of initialisation.
> 
> In the initialisation path, it seems that we register the notifier
> and _then_ the adapter.  Here, we unregister the notifier and then
> the adapter rather than what would normally be expected.  Why is
> this?  I suspect there will be drivers created that do this the
> "normal" way round, so if this is a requirement, it needs to be made
> plainly obvious.

It's not a requirement, it just feels better to do it in this order
since cec_unregister_adapter will in general also delete the adapter
(unless an application keeps the cec device open).

So the order is actually: allocate_adapter, then register notifier
and: unregister notifier, then unregister (and typically delete) adapter

Regards,

	Hans

> 
>>  
>>  	return 0;
>>  }
>> -- 
>> 2.23.0.rc1.153.gdeed80330f-goog
>>
>>
> 

Re: [PATCH v6 7/8] drm: dw-hdmi: use cec_notifier_conn_(un)register

[Hans Verkuil]

On 8/13/19 1:02 PM, Dariusz Marcinkiewicz wrote:
> Use the new cec_notifier_conn_(un)register() functions to
> (un)register the notifier for the HDMI connector, and fill in
> the cec_connector_info.
> 
> Changes since v4:
> 	- typo fix
> Changes since v2:
> 	- removed unnecessary NULL check before a call to
> 	cec_notifier_conn_unregister,
> 	- use cec_notifier_phys_addr_invalidate to invalidate physical
> 	address.
> Changes since v1:
> 	Add memory barrier to make sure that the notifier
> 	becomes visible to the irq thread once it is fully
> 	constructed.
> 
> Signed-off-by: Dariusz Marcinkiewicz <darekm@google.com>
> Tested-by: Hans Verkuil <hverkuil-cisco@xs4all.nl>
> ---
>  drivers/gpu/drm/bridge/synopsys/dw-hdmi.c | 36 ++++++++++++++---------
>  1 file changed, 22 insertions(+), 14 deletions(-)
> 
> diff --git a/drivers/gpu/drm/bridge/synopsys/dw-hdmi.c b/drivers/gpu/drm/bridge/synopsys/dw-hdmi.c
> index 83b94b66e464e..c00184700bb9d 100644
> --- a/drivers/gpu/drm/bridge/synopsys/dw-hdmi.c
> +++ b/drivers/gpu/drm/bridge/synopsys/dw-hdmi.c
> @@ -2194,6 +2194,8 @@ static int dw_hdmi_bridge_attach(struct drm_bridge *bridge)
>  	struct dw_hdmi *hdmi = bridge->driver_private;
>  	struct drm_encoder *encoder = bridge->encoder;
>  	struct drm_connector *connector = &hdmi->connector;
> +	struct cec_connector_info conn_info;
> +	struct cec_notifier *notifier;
>  
>  	connector->interlace_allowed = 1;
>  	connector->polled = DRM_CONNECTOR_POLL_HPD;
> @@ -2207,6 +2209,18 @@ static int dw_hdmi_bridge_attach(struct drm_bridge *bridge)
>  
>  	drm_connector_attach_encoder(connector, encoder);
>  
> +	cec_fill_conn_info_from_drm(&conn_info, connector);
> +
> +	notifier = cec_notifier_conn_register(hdmi->dev, NULL, &conn_info);
> +	if (!notifier)
> +		return -ENOMEM;
> +	/*
> +	 * Make sure that dw_hdmi_irq thread does see the notifier
> +	 * when it fully constructed.
> +	 */
> +	smp_wmb();
> +	hdmi->cec_notifier = notifier;
> +
>  	return 0;
>  }
>  
> @@ -2373,9 +2387,13 @@ static irqreturn_t dw_hdmi_irq(int irq, void *dev_id)
>  				       phy_stat & HDMI_PHY_HPD,
>  				       phy_stat & HDMI_PHY_RX_SENSE);
>  
> -		if ((phy_stat & (HDMI_PHY_RX_SENSE | HDMI_PHY_HPD)) == 0)
> -			cec_notifier_set_phys_addr(hdmi->cec_notifier,
> -						   CEC_PHYS_ADDR_INVALID);
> +		if ((phy_stat & (HDMI_PHY_RX_SENSE | HDMI_PHY_HPD)) == 0) {
> +			struct cec_notifier *notifier;
> +
> +			notifier = READ_ONCE(hdmi->cec_notifier);
> +			if (notifier)
> +				cec_notifier_phys_addr_invalidate(notifier);
> +		}
>  	}
>  
>  	if (intr_stat & HDMI_IH_PHY_STAT0_HPD) {
> @@ -2693,12 +2711,6 @@ __dw_hdmi_probe(struct platform_device *pdev,
>  	if (ret)
>  		goto err_iahb;
>  
> -	hdmi->cec_notifier = cec_notifier_get(dev);
> -	if (!hdmi->cec_notifier) {
> -		ret = -ENOMEM;
> -		goto err_iahb;
> -	}
> -
>  	/*
>  	 * To prevent overflows in HDMI_IH_FC_STAT2, set the clk regenerator
>  	 * N and cts values before enabling phy
> @@ -2796,9 +2808,6 @@ __dw_hdmi_probe(struct platform_device *pdev,
>  		hdmi->ddc = NULL;
>  	}
>  
> -	if (hdmi->cec_notifier)
> -		cec_notifier_put(hdmi->cec_notifier);
> -
>  	clk_disable_unprepare(hdmi->iahb_clk);
>  	if (hdmi->cec_clk)
>  		clk_disable_unprepare(hdmi->cec_clk);
> @@ -2820,8 +2829,7 @@ static void __dw_hdmi_remove(struct dw_hdmi *hdmi)
>  	/* Disable all interrupts */
>  	hdmi_writeb(hdmi, ~0, HDMI_IH_MUTE_PHY_STAT0);
>  
> -	if (hdmi->cec_notifier)
> -		cec_notifier_put(hdmi->cec_notifier);
> +	cec_notifier_conn_unregister(hdmi->cec_notifier);

Russell's review caused me to take another look at this series, and it made
wonder if cec_notifier_conn_unregister() shouldn't be called from bridge_detach?

Regards,

	Hans

>  
>  	clk_disable_unprepare(hdmi->iahb_clk);
>  	clk_disable_unprepare(hdmi->isfr_clk);
> 

Re: [PATCH v6 7/8] drm: dw-hdmi: use cec_notifier_conn_(un)register

[Dariusz Marcinkiewicz]

Hi.

On Tue, Aug 13, 2019 at 1:38 PM Hans Verkuil <hverkuil-cisco@xs4all.nl> wrote:
>
> Russell's review caused me to take another look at this series, and it made
> wonder if cec_notifier_conn_unregister() shouldn't be called from bridge_detach?
>
I've sent out v7 of the series where unregistration is done from bridge detach.

Regards.