Hi,
On Thu, Jan 16, 2025 at 03:04:19AM +0200, Dmitry Baryshkov wrote:
> On Wed, Jan 15, 2025 at 10:05:32PM +0100, Maxime Ripard wrote:
> > Now that connectors are no longer necessarily created by the bridges
> > drivers themselves but might be created by drm_bridge_connector, it's
> > pretty hard for bridge drivers to retrieve pointers to the connector and
> > CRTC they are attached to.
> >
> > Indeed, the only way to retrieve the CRTC is to follow the drm_bridge
> > encoder field, and then the drm_encoder crtc field, both of them being
> > deprecated.
> >
> > And for the connector, since we can have multiple connectors attached to
> > a CRTC, we don't really have a reliable way to get it.
> >
> > Let's provide both pointers in the drm_bridge_state structure so we
> > don't have to follow deprecated, non-atomic, pointers, and be more
> > consistent with the other KMS entities.
> >
> > Signed-off-by: Maxime Ripard <mripard@xxxxxxxxxx>
> > ---
> > drivers/gpu/drm/drm_atomic_state_helper.c | 5 +++++
> > drivers/gpu/drm/drm_bridge.c | 21 +++++++++++++--------
> > include/drm/drm_atomic.h | 14 ++++++++++++++
> > 3 files changed, 32 insertions(+), 8 deletions(-)
> >
> > diff --git a/drivers/gpu/drm/drm_atomic_state_helper.c b/drivers/gpu/drm/drm_atomic_state_helper.c
> > index 519228eb109533d2596e899a57b571fa0995824f..66661dca077215b78dffca7bc1712f56d35e3918 100644
> > --- a/drivers/gpu/drm/drm_atomic_state_helper.c
> > +++ b/drivers/gpu/drm/drm_atomic_state_helper.c
> > @@ -777,10 +777,15 @@ EXPORT_SYMBOL(drm_atomic_helper_bridge_duplicate_state);
> > * that don't subclass the bridge state.
> > */
> > void drm_atomic_helper_bridge_destroy_state(struct drm_bridge *bridge,
> > struct drm_bridge_state *state)
> > {
> > + if (state->connector) {
> > + drm_connector_put(state->connector);
> > + state->connector = NULL;
> > + }
> > +
> > kfree(state);
> > }
> > EXPORT_SYMBOL(drm_atomic_helper_bridge_destroy_state);
> >
> > /**
> > diff --git a/drivers/gpu/drm/drm_bridge.c b/drivers/gpu/drm/drm_bridge.c
> > index c937980d6591fd98e33e37d799ebf84e7e6c5529..069c105aa59636c64caffbefcf482133b0db97d9 100644
> > --- a/drivers/gpu/drm/drm_bridge.c
> > +++ b/drivers/gpu/drm/drm_bridge.c
> > @@ -829,19 +829,24 @@ EXPORT_SYMBOL(drm_atomic_bridge_chain_enable);
> >
> > static int drm_atomic_bridge_check(struct drm_bridge *bridge,
> > struct drm_crtc_state *crtc_state,
> > struct drm_connector_state *conn_state)
> > {
> > + struct drm_bridge_state *bridge_state;
> > + int ret;
> > +
> > + bridge_state = drm_atomic_get_new_bridge_state(crtc_state->state,
> > + bridge);
>
> It felt like an error to me to call this function for a non-atomic
> bridges, until I fully followed the code path to find that it will
> return NULL if the bridge isn't registered as a private object.
Yeah.. I wasn't too sure what to do about this one either. I think it
would be more consistent to always have a state properly filled, even if
we have !atomic drivers. It's what happens with the rest of the
framework.
But also, I have no idea what the side-effects might be.
One thing though: a driver having an atomic_check callback is not an
indication of whether it supports atomic mode-setting or not.
atomic_check is optional, so we can have atomic drivers without
atomic_check.
> BTW: if my grep-foo isn't deceiving me, we currently have 34 non-atomic
> bridges out of 90. Should we start forcebly updating them to use atomic
> interface in attempt to drop the mode_fixup() and other non-atomic
> callbacks?
Maybe? I'm not sure forcing anyone to anything really helps. sii8620 for
example is going to be a fun one, and I'd rather stay away from it :)
Maxime
Attachment:
signature.asc
Description: PGP signature
