[Intel-gfx] [PATCH v3 10/10] drm/i915: uevent for HDCP status change

[Daniel Vetter]

On Wed, Mar 27, 2019 at 08:07:39PM +0530, Ramalingam C wrote:
> On 2019-03-27 at 12:06:39 +0100, Daniel Vetter wrote:
> > On Fri, Mar 22, 2019 at 06:14:48AM +0530, Ramalingam C wrote:
> > > Invoking the uevent generator for the content protection property state
> > > change of a connector. This helps the userspace to detect the new state
> > > change without polling the property val.
> > > 
> > > Signed-off-by: Ramalingam C <ramalingam.c at intel.com>
> > > ---
> > >  drivers/gpu/drm/i915/intel_hdcp.c | 11 +++++++++++
> > >  1 file changed, 11 insertions(+)
> > > 
> > > diff --git a/drivers/gpu/drm/i915/intel_hdcp.c b/drivers/gpu/drm/i915/intel_hdcp.c
> > > index 1eea553cdf81..1c38026f4de7 100644
> > > --- a/drivers/gpu/drm/i915/intel_hdcp.c
> > > +++ b/drivers/gpu/drm/i915/intel_hdcp.c
> > > @@ -8,6 +8,7 @@
> > >  
> > >  #include <drm/drm_hdcp.h>
> > >  #include <drm/i915_component.h>
> > > +#include <drm/drm_sysfs.h>
> > >  #include <linux/i2c.h>
> > >  #include <linux/random.h>
> > >  #include <linux/component.h>
> > > @@ -19,6 +20,14 @@
> > >  #define ENCRYPT_STATUS_CHANGE_TIMEOUT_MS	50
> > >  #define HDCP2_LC_RETRY_CNT			3
> > >  
> > > +static inline
> > > +void intel_hdcp_state_change_event(struct drm_connector *connector)
> > > +{
> > > +	struct drm_property *property = connector->content_protection_property;
> > > +
> > > +	drm_sysfs_connector_status_event(connector, property);
> > > +}
> > > +
> > >  static
> > >  bool intel_hdcp_is_ksv_valid(u8 *ksv)
> > >  {
> > > @@ -943,6 +952,7 @@ static void intel_hdcp_prop_work(struct work_struct *work)
> > >  	if (hdcp->value != DRM_MODE_CONTENT_PROTECTION_UNDESIRED) {
> > >  		state = connector->base.state;
> > >  		state->content_protection = hdcp->value;
> > > +		intel_hdcp_state_change_event(&connector->base);
> > 
> > I think it'd be good to have a helper to update both
> > state->content_protection and send out the event.
> but adding this in intel_hdcp_prop_work also does the same. Do you
> suggest we need to move the intel_hdcp_state_change out of prop_work and
> put both of them in a wrapper?

Yeah something like:

drm_hdcp_update_content_protecion()
{
	assert_lock_held(dev->mode_config.connnection_mutex);
	state->content_protection = hdcp->value;
	drm_sysfs_connector_status_event(&connector->base,
					dev->mode_config.content_protection_prop);
}

And then use that instead of both the assignement and the call to send out
the event in intel_hdcp_prop_work().

> 
> > Locking is a bit
> > complicated, so we also need a lockdep assert to make sure
> > dev->mode_config.connection_mutex is held.
> in intel_hdcp_state_change ?
> > 
> > That way I hope that any update in the property will actually result in
> > the event being sent out, and not accidentally forgotten.
> > 
> > >  	}
> > >  
> > >  	mutex_unlock(&hdcp->mutex);
> > > @@ -2206,6 +2216,7 @@ int intel_hdcp_disable(struct intel_connector *connector)
> > >  			ret = _intel_hdcp2_disable(connector);
> > >  		else if (hdcp->hdcp_encrypted)
> > >  			ret = _intel_hdcp_disable(connector);
> > > +		intel_hdcp_state_change_event(&connector->base);
> > 
> > Why do we need this here? We don't update the property here ...
> Change is requested from the userspace. So we dont update the property
> state in conn_state. But in rethinking over it, only when kernel
> triggers the change of property state we need the uevent. So we dont
> need it here as it is triggered from the userspace.

Yeah, I think that makes sense. Otherwise we might end up in some funny
loop:
1. userspace updates property
2. kernel sends out uevent
3. userspace handles uevent, again resets property, goto 2

Or something silly like that :-)
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch