diff mbox series

[v6,09/10] drm/vc4: hdmi: stop rereading EDID in get_modes()

Message ID 20241206-drm-bridge-hdmi-connector-v6-9-50dc145a9c06@linaro.org (mailing list archive)
State Superseded
Headers show
Series drm: add DRM HDMI Codec framework | expand

Commit Message

Dmitry Baryshkov Dec. 6, 2024, 10:16 a.m. UTC
The vc4_hdmi_connector_detect_ctx() via vc4_hdmi_handle_hotplug()
already reads EDID and propagates it to the drm_connector. Stop
rereading EDID as a part of the .get_modes() callback and just update
the list of modes. This matches the behaviour of the i915 driver.

Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@linaro.org>
---
 drivers/gpu/drm/vc4/vc4_hdmi.c | 21 ---------------------
 1 file changed, 21 deletions(-)

Comments

Maxime Ripard Dec. 6, 2024, 2:34 p.m. UTC | #1
Hi,

On Fri, Dec 06, 2024 at 12:16:03PM +0200, Dmitry Baryshkov wrote:
> The vc4_hdmi_connector_detect_ctx() via vc4_hdmi_handle_hotplug()
> already reads EDID and propagates it to the drm_connector. Stop
> rereading EDID as a part of the .get_modes() callback and just update
> the list of modes. This matches the behaviour of the i915 driver.
> 
> Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@linaro.org>
> ---
>  drivers/gpu/drm/vc4/vc4_hdmi.c | 21 ---------------------
>  1 file changed, 21 deletions(-)
> 
> diff --git a/drivers/gpu/drm/vc4/vc4_hdmi.c b/drivers/gpu/drm/vc4/vc4_hdmi.c
> index e5ab42f72f618b90f956482db6c9c8074c1e3bf1..3364ef90968dad3074800f02926300ffceb75c69 100644
> --- a/drivers/gpu/drm/vc4/vc4_hdmi.c
> +++ b/drivers/gpu/drm/vc4/vc4_hdmi.c
> @@ -470,31 +470,10 @@ static int vc4_hdmi_connector_detect_ctx(struct drm_connector *connector,
>  
>  static int vc4_hdmi_connector_get_modes(struct drm_connector *connector)
>  {
> -	struct vc4_hdmi *vc4_hdmi = connector_to_vc4_hdmi(connector);
>  	struct vc4_dev *vc4 = to_vc4_dev(connector->dev);
> -	const struct drm_edid *drm_edid;
>  	int ret = 0;
>  
> -	/*
> -	 * NOTE: This function should really take vc4_hdmi->mutex, but doing so
> -	 * results in reentrancy issues since cec_s_phys_addr() might call
> -	 * .adap_enable, which leads to that funtion being called with our mutex
> -	 * held.
> -	 *
> -	 * Concurrency isn't an issue at the moment since we don't share
> -	 * any state with any of the other frameworks so we can ignore
> -	 * the lock for now.
> -	 */
> -
> -	drm_edid = drm_edid_read_ddc(connector, vc4_hdmi->ddc);
> -	drm_edid_connector_update(connector, drm_edid);
> -	cec_s_phys_addr(vc4_hdmi->cec_adap,
> -			connector->display_info.source_physical_address, false);
> -	if (!drm_edid)
> -		return 0;
> -
>  	ret = drm_edid_connector_add_modes(connector);
> -	drm_edid_free(drm_edid);

I don't think that works though, for mostly two reasons:

 1) We're not sure we'll even have an hotplug interrupt if the system
    boots with the connector plugged in for example.

 2) afaik, the get_modes ioctl directly calls get_modes too.

Maxime
Dmitry Baryshkov Dec. 8, 2024, 11:06 a.m. UTC | #2
On Fri, Dec 06, 2024 at 03:34:52PM +0100, Maxime Ripard wrote:
> Hi,
> 
> On Fri, Dec 06, 2024 at 12:16:03PM +0200, Dmitry Baryshkov wrote:
> > The vc4_hdmi_connector_detect_ctx() via vc4_hdmi_handle_hotplug()
> > already reads EDID and propagates it to the drm_connector. Stop
> > rereading EDID as a part of the .get_modes() callback and just update
> > the list of modes. This matches the behaviour of the i915 driver.
> > 
> > Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@linaro.org>
> > ---
> >  drivers/gpu/drm/vc4/vc4_hdmi.c | 21 ---------------------
> >  1 file changed, 21 deletions(-)
> > 
> > diff --git a/drivers/gpu/drm/vc4/vc4_hdmi.c b/drivers/gpu/drm/vc4/vc4_hdmi.c
> > index e5ab42f72f618b90f956482db6c9c8074c1e3bf1..3364ef90968dad3074800f02926300ffceb75c69 100644
> > --- a/drivers/gpu/drm/vc4/vc4_hdmi.c
> > +++ b/drivers/gpu/drm/vc4/vc4_hdmi.c
> > @@ -470,31 +470,10 @@ static int vc4_hdmi_connector_detect_ctx(struct drm_connector *connector,
> >  
> >  static int vc4_hdmi_connector_get_modes(struct drm_connector *connector)
> >  {
> > -	struct vc4_hdmi *vc4_hdmi = connector_to_vc4_hdmi(connector);
> >  	struct vc4_dev *vc4 = to_vc4_dev(connector->dev);
> > -	const struct drm_edid *drm_edid;
> >  	int ret = 0;
> >  
> > -	/*
> > -	 * NOTE: This function should really take vc4_hdmi->mutex, but doing so
> > -	 * results in reentrancy issues since cec_s_phys_addr() might call
> > -	 * .adap_enable, which leads to that funtion being called with our mutex
> > -	 * held.
> > -	 *
> > -	 * Concurrency isn't an issue at the moment since we don't share
> > -	 * any state with any of the other frameworks so we can ignore
> > -	 * the lock for now.
> > -	 */
> > -
> > -	drm_edid = drm_edid_read_ddc(connector, vc4_hdmi->ddc);
> > -	drm_edid_connector_update(connector, drm_edid);
> > -	cec_s_phys_addr(vc4_hdmi->cec_adap,
> > -			connector->display_info.source_physical_address, false);
> > -	if (!drm_edid)
> > -		return 0;
> > -
> >  	ret = drm_edid_connector_add_modes(connector);
> > -	drm_edid_free(drm_edid);
> 
> I don't think that works though, for mostly two reasons:
> 
>  1) We're not sure we'll even have an hotplug interrupt if the system
>     boots with the connector plugged in for example.
> 
>  2) afaik, the get_modes ioctl directly calls get_modes too.

I think both paths use .fill_modes, not get_modes. And fill_modes
explicitly calls .detect / .detect_ctx. It would be nice if somebody can
verify the change on the acual hw. But as I wrote, i915 driver does
exactly the same: the EDID is read / updated in .detect_ctx(), while
.get_modes() only performs drm_edid_connector_add_modes().
Okay, there is one significant difference which we should probably patch
out: the .force() is also implemented to update EDID in the forced case.
I should probably do the same for the drivers being handled in this
patchset.
Dmitry Baryshkov Dec. 16, 2024, 11:26 a.m. UTC | #3
On Sun, Dec 08, 2024 at 01:06:46PM +0200, Dmitry Baryshkov wrote:
> On Fri, Dec 06, 2024 at 03:34:52PM +0100, Maxime Ripard wrote:
> > Hi,
> > 
> > On Fri, Dec 06, 2024 at 12:16:03PM +0200, Dmitry Baryshkov wrote:
> > > The vc4_hdmi_connector_detect_ctx() via vc4_hdmi_handle_hotplug()
> > > already reads EDID and propagates it to the drm_connector. Stop
> > > rereading EDID as a part of the .get_modes() callback and just update
> > > the list of modes. This matches the behaviour of the i915 driver.
> > > 
> > > Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@linaro.org>
> > > ---
> > >  drivers/gpu/drm/vc4/vc4_hdmi.c | 21 ---------------------
> > >  1 file changed, 21 deletions(-)
> > > 
> > > diff --git a/drivers/gpu/drm/vc4/vc4_hdmi.c b/drivers/gpu/drm/vc4/vc4_hdmi.c
> > > index e5ab42f72f618b90f956482db6c9c8074c1e3bf1..3364ef90968dad3074800f02926300ffceb75c69 100644
> > > --- a/drivers/gpu/drm/vc4/vc4_hdmi.c
> > > +++ b/drivers/gpu/drm/vc4/vc4_hdmi.c
> > > @@ -470,31 +470,10 @@ static int vc4_hdmi_connector_detect_ctx(struct drm_connector *connector,
> > >  
> > >  static int vc4_hdmi_connector_get_modes(struct drm_connector *connector)
> > >  {
> > > -	struct vc4_hdmi *vc4_hdmi = connector_to_vc4_hdmi(connector);
> > >  	struct vc4_dev *vc4 = to_vc4_dev(connector->dev);
> > > -	const struct drm_edid *drm_edid;
> > >  	int ret = 0;
> > >  
> > > -	/*
> > > -	 * NOTE: This function should really take vc4_hdmi->mutex, but doing so
> > > -	 * results in reentrancy issues since cec_s_phys_addr() might call
> > > -	 * .adap_enable, which leads to that funtion being called with our mutex
> > > -	 * held.
> > > -	 *
> > > -	 * Concurrency isn't an issue at the moment since we don't share
> > > -	 * any state with any of the other frameworks so we can ignore
> > > -	 * the lock for now.
> > > -	 */
> > > -
> > > -	drm_edid = drm_edid_read_ddc(connector, vc4_hdmi->ddc);
> > > -	drm_edid_connector_update(connector, drm_edid);
> > > -	cec_s_phys_addr(vc4_hdmi->cec_adap,
> > > -			connector->display_info.source_physical_address, false);
> > > -	if (!drm_edid)
> > > -		return 0;
> > > -
> > >  	ret = drm_edid_connector_add_modes(connector);
> > > -	drm_edid_free(drm_edid);
> > 
> > I don't think that works though, for mostly two reasons:
> > 
> >  1) We're not sure we'll even have an hotplug interrupt if the system
> >     boots with the connector plugged in for example.
> > 
> >  2) afaik, the get_modes ioctl directly calls get_modes too.
> 
> I think both paths use .fill_modes, not get_modes. And fill_modes
> explicitly calls .detect / .detect_ctx. It would be nice if somebody can
> verify the change on the acual hw. But as I wrote, i915 driver does
> exactly the same: the EDID is read / updated in .detect_ctx(), while
> .get_modes() only performs drm_edid_connector_add_modes().
> Okay, there is one significant difference which we should probably patch
> out: the .force() is also implemented to update EDID in the forced case.
> I should probably do the same for the drivers being handled in this
> patchset.

Maxime, will .force() cover your concerns? I probably fail to see a path
when .get_mode() is called without calling .detect() / .force() first.
Maxime Ripard Dec. 17, 2024, 5:13 p.m. UTC | #4
On Sun, Dec 08, 2024 at 01:06:46PM +0200, Dmitry Baryshkov wrote:
> On Fri, Dec 06, 2024 at 03:34:52PM +0100, Maxime Ripard wrote:
> > Hi,
> > 
> > On Fri, Dec 06, 2024 at 12:16:03PM +0200, Dmitry Baryshkov wrote:
> > > The vc4_hdmi_connector_detect_ctx() via vc4_hdmi_handle_hotplug()
> > > already reads EDID and propagates it to the drm_connector. Stop
> > > rereading EDID as a part of the .get_modes() callback and just update
> > > the list of modes. This matches the behaviour of the i915 driver.
> > > 
> > > Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@linaro.org>
> > > ---
> > >  drivers/gpu/drm/vc4/vc4_hdmi.c | 21 ---------------------
> > >  1 file changed, 21 deletions(-)
> > > 
> > > diff --git a/drivers/gpu/drm/vc4/vc4_hdmi.c b/drivers/gpu/drm/vc4/vc4_hdmi.c
> > > index e5ab42f72f618b90f956482db6c9c8074c1e3bf1..3364ef90968dad3074800f02926300ffceb75c69 100644
> > > --- a/drivers/gpu/drm/vc4/vc4_hdmi.c
> > > +++ b/drivers/gpu/drm/vc4/vc4_hdmi.c
> > > @@ -470,31 +470,10 @@ static int vc4_hdmi_connector_detect_ctx(struct drm_connector *connector,
> > >  
> > >  static int vc4_hdmi_connector_get_modes(struct drm_connector *connector)
> > >  {
> > > -	struct vc4_hdmi *vc4_hdmi = connector_to_vc4_hdmi(connector);
> > >  	struct vc4_dev *vc4 = to_vc4_dev(connector->dev);
> > > -	const struct drm_edid *drm_edid;
> > >  	int ret = 0;
> > >  
> > > -	/*
> > > -	 * NOTE: This function should really take vc4_hdmi->mutex, but doing so
> > > -	 * results in reentrancy issues since cec_s_phys_addr() might call
> > > -	 * .adap_enable, which leads to that funtion being called with our mutex
> > > -	 * held.
> > > -	 *
> > > -	 * Concurrency isn't an issue at the moment since we don't share
> > > -	 * any state with any of the other frameworks so we can ignore
> > > -	 * the lock for now.
> > > -	 */
> > > -
> > > -	drm_edid = drm_edid_read_ddc(connector, vc4_hdmi->ddc);
> > > -	drm_edid_connector_update(connector, drm_edid);
> > > -	cec_s_phys_addr(vc4_hdmi->cec_adap,
> > > -			connector->display_info.source_physical_address, false);
> > > -	if (!drm_edid)
> > > -		return 0;
> > > -
> > >  	ret = drm_edid_connector_add_modes(connector);
> > > -	drm_edid_free(drm_edid);
> > 
> > I don't think that works though, for mostly two reasons:
> > 
> >  1) We're not sure we'll even have an hotplug interrupt if the system
> >     boots with the connector plugged in for example.
> > 
> >  2) afaik, the get_modes ioctl directly calls get_modes too.
> 
> I think both paths use .fill_modes, not get_modes.

The standard fill_modes helper calls get_modes.

> And fill_modes explicitly calls .detect / .detect_ctx. It would be
> nice if somebody can verify the change on the acual hw. But as I
> wrote, i915 driver does exactly the same: the EDID is read / updated
> in .detect_ctx(), while .get_modes() only performs
> drm_edid_connector_add_modes(). Okay, there is one significant
> difference which we should probably patch out: the .force() is also
> implemented to update EDID in the forced case. I should probably do
> the same for the drivers being handled in this patchset.

Ack. Worst case scenario we can fix it later on.

Maxime
diff mbox series

Patch

diff --git a/drivers/gpu/drm/vc4/vc4_hdmi.c b/drivers/gpu/drm/vc4/vc4_hdmi.c
index e5ab42f72f618b90f956482db6c9c8074c1e3bf1..3364ef90968dad3074800f02926300ffceb75c69 100644
--- a/drivers/gpu/drm/vc4/vc4_hdmi.c
+++ b/drivers/gpu/drm/vc4/vc4_hdmi.c
@@ -470,31 +470,10 @@  static int vc4_hdmi_connector_detect_ctx(struct drm_connector *connector,
 
 static int vc4_hdmi_connector_get_modes(struct drm_connector *connector)
 {
-	struct vc4_hdmi *vc4_hdmi = connector_to_vc4_hdmi(connector);
 	struct vc4_dev *vc4 = to_vc4_dev(connector->dev);
-	const struct drm_edid *drm_edid;
 	int ret = 0;
 
-	/*
-	 * NOTE: This function should really take vc4_hdmi->mutex, but doing so
-	 * results in reentrancy issues since cec_s_phys_addr() might call
-	 * .adap_enable, which leads to that funtion being called with our mutex
-	 * held.
-	 *
-	 * Concurrency isn't an issue at the moment since we don't share
-	 * any state with any of the other frameworks so we can ignore
-	 * the lock for now.
-	 */
-
-	drm_edid = drm_edid_read_ddc(connector, vc4_hdmi->ddc);
-	drm_edid_connector_update(connector, drm_edid);
-	cec_s_phys_addr(vc4_hdmi->cec_adap,
-			connector->display_info.source_physical_address, false);
-	if (!drm_edid)
-		return 0;
-
 	ret = drm_edid_connector_add_modes(connector);
-	drm_edid_free(drm_edid);
 
 	if (!vc4->hvs->vc5_hdmi_enable_hdmi_20) {
 		struct drm_device *drm = connector->dev;