Nouveau - Oct 2021 - [PATCH 01/15] gpu/drm: make drm_add_edid_modes() consistent when updating connector->display_info

On Fri, Oct 15, 2021 at 03:03:13PM +0300, Ville Syrj?l?
wrote:
> On Fri, Oct 15, 2021 at 01:36:59PM +0200, Claudio Suarez wrote:
> > According to the documentation, drm_add_edid_modes
> > "... Also fills out the &drm_display_info structure and ELD
in @connector
> > with any information which can be derived from the edid."
> > 
> > drm_add_edid_modes accepts a struct edid *edid parameter which may
have a
> > value or may be null. When it is not null, connector->display_info
and
> > connector->eld are updated according to the edid. When edid=NULL,
only
> > connector->eld is reset. Reset connector->display_info to be
consistent
> > and accurate.
> > 
> > Signed-off-by: Claudio Suarez <cssk at net-c.es>
> > ---
> >  drivers/gpu/drm/drm_edid.c | 2 ++
> >  1 file changed, 2 insertions(+)
> > 
> > diff --git a/drivers/gpu/drm/drm_edid.c b/drivers/gpu/drm/drm_edid.c
> > index 6325877c5fd6..6cbe09b2357c 100644
> > --- a/drivers/gpu/drm/drm_edid.c
> > +++ b/drivers/gpu/drm/drm_edid.c
> > @@ -5358,10 +5358,12 @@ int drm_add_edid_modes(struct drm_connector
*connector, struct edid *edid)
> >  
> >  	if (edid == NULL) {
> >  		clear_eld(connector);
> > +		drm_reset_display_info(connector);
> >  		return 0;
> >  	}
> >  	if (!drm_edid_is_valid(edid)) {
> >  		clear_eld(connector);
> > +		drm_reset_display_info(connector);
> 
> Looks easier if you pull both of those out from these branches and
> just call them unconditionally at the start.
After looking at the full code, I am not sure. This is the code:
=================int drm_add_edid_modes(struct drm_connector *connector, struct
edid *edid)
{
        int num_modes = 0;
        u32 quirks;

        if (edid == NULL) {
                clear_eld(connector);
                drm_reset_display_info(connector); <--- added by me
                return 0;
        }
        if (!drm_edid_is_valid(edid)) {
                clear_eld(connector);
                drm_reset_display_info(connector); <--- added by me
                drm_warn(connector->dev, "%s: EDID invalid.\n",
                         connector->name);
                return 0;
        }

        drm_edid_to_eld(connector, edid);

        quirks = drm_add_display_info(connector, edid);
	etc...
================
If we move those out of these branches and edid != NULL, we are executing an
unnecessary clear_eld(connector) and an unnecessary
drm_reset_display_info(connector)
because the fields will be set in the next drm_edid_to_eld(connector, edid) and
drm_add_display_info(connector, edid)

Do we want this ?

BR
Claudio Suarez

On Fri, Oct 15, 2021 at 09:24:06PM +0200, Claudio Suarez
wrote:
> On Fri, Oct 15, 2021 at 03:03:13PM +0300, Ville Syrj?l? wrote:
> > On Fri, Oct 15, 2021 at 01:36:59PM +0200, Claudio Suarez wrote:
> > > According to the documentation, drm_add_edid_modes
> > > "... Also fills out the &drm_display_info structure and
ELD in @connector
> > > with any information which can be derived from the edid."
> > > 
> > > drm_add_edid_modes accepts a struct edid *edid parameter which
may have a
> > > value or may be null. When it is not null,
connector->display_info and
> > > connector->eld are updated according to the edid. When
edid=NULL, only
> > > connector->eld is reset. Reset connector->display_info to
be consistent
> > > and accurate.
> > > 
> > > Signed-off-by: Claudio Suarez <cssk at net-c.es>
> > > ---
> > >  drivers/gpu/drm/drm_edid.c | 2 ++
> > >  1 file changed, 2 insertions(+)
> > > 
> > > diff --git a/drivers/gpu/drm/drm_edid.c
b/drivers/gpu/drm/drm_edid.c
> > > index 6325877c5fd6..6cbe09b2357c 100644
> > > --- a/drivers/gpu/drm/drm_edid.c
> > > +++ b/drivers/gpu/drm/drm_edid.c
> > > @@ -5358,10 +5358,12 @@ int drm_add_edid_modes(struct
drm_connector *connector, struct edid *edid)
> > >  
> > >  	if (edid == NULL) {
> > >  		clear_eld(connector);
> > > +		drm_reset_display_info(connector);
> > >  		return 0;
> > >  	}
> > >  	if (!drm_edid_is_valid(edid)) {
> > >  		clear_eld(connector);
> > > +		drm_reset_display_info(connector);
> > 
> > Looks easier if you pull both of those out from these branches and
> > just call them unconditionally at the start.
> 
> After looking at the full code, I am not sure. This is the code:
> =================> int drm_add_edid_modes(struct drm_connector
*connector, struct edid *edid)
> {
>         int num_modes = 0;
>         u32 quirks;
> 
>         if (edid == NULL) {
>                 clear_eld(connector);
>                 drm_reset_display_info(connector); <--- added by me
>                 return 0;
>         }
>         if (!drm_edid_is_valid(edid)) {
>                 clear_eld(connector);
>                 drm_reset_display_info(connector); <--- added by me
>                 drm_warn(connector->dev, "%s: EDID
invalid.\n",
>                          connector->name);
>                 return 0;
>         }
> 
>         drm_edid_to_eld(connector, edid);
> 
>         quirks = drm_add_display_info(connector, edid);
> 	etc...
> ================> 
> If we move those out of these branches and edid != NULL, we are executing
an
> unnecessary clear_eld(connector) and an unnecessary
drm_reset_display_info(connector)
> because the fields will be set in the next drm_edid_to_eld(connector, edid)
and
> drm_add_display_info(connector, edid)
> 
> Do we want this ?
Seems fine by me. And maybe we could nuke the second
drm_reset_display_info() from deeper inside drm_add_display_info()?
Not sure if drm_add_display_info() still has to be able to operate
standalone or not.

Hmm. Another option is to just move all these NULL/invalid edid
checks into drm_edid_to_eld() and drm_add_display_info().

-- 
Ville Syrj?l?
Intel