[PATCH 2/2] drm: bridge: ti-sn65dsi83: Add error recovery mechanism

[Maxime Ripard]

On Mon, Oct 28, 2024 at 03:28:58PM +0200, Laurent Pinchart wrote:
> On Mon, Oct 28, 2024 at 01:21:45PM +0100, Maxime Ripard wrote:
> > On Mon, Oct 28, 2024 at 01:28:57PM +0200, Laurent Pinchart wrote:
> > > On Mon, Oct 28, 2024 at 09:13:31AM +0100, Herve Codina wrote:
> > > > On Sun, 27 Oct 2024 18:23:50 +0200 Laurent Pinchart wrote:
> > > > 
> > > > [...]
> > > > > > +static int sn65dsi83_reset_pipeline(struct sn65dsi83 *sn65dsi83)
> > > > > > +{
> > > > > > +	struct drm_device *dev = sn65dsi83->bridge.dev;
> > > > > > +	struct drm_modeset_acquire_ctx ctx;
> > > > > > +	struct drm_atomic_state *state;
> > > > > > +	int err;
> > > > > > +
> > > > > > +	/* Use operation done in drm_atomic_helper_suspend() followed by
> > > > > > +	 * operation done in drm_atomic_helper_resume() but without releasing
> > > > > > +	 * the lock between suspend()/resume()
> > > > > > +	 */
> > > > > > +
> > > > > > +	DRM_MODESET_LOCK_ALL_BEGIN(dev, ctx, 0, err);
> > > > > > +
> > > > > > +	state = drm_atomic_helper_duplicate_state(dev, &ctx);
> > > > > > +	if (IS_ERR(state)) {
> > > > > > +		err = PTR_ERR(state);
> > > > > > +		goto unlock;
> > > > > > +	}
> > > > > > +
> > > > > > +	err = drm_atomic_helper_disable_all(dev, &ctx);
> > > > > > +	if (err < 0)
> > > > > > +		goto unlock;
> > > > > > +
> > > > > > +	drm_mode_config_reset(dev);
> > > > > > +
> > > > > > +	err = drm_atomic_helper_commit_duplicated_state(state, &ctx);  
> > > > > 
> > > > > Committing a full atomic state from a bridge driver in an asynchronous
> > > > > way seems quite uncharted territory, and it worries me. It's also a very
> > > > > heavyweight, you disable all outputs here, instead of focussing on the
> > > > > output connected to the bridge. Can you either implement something more
> > > > > local, resetting the bridge only, or create a core helper to handle this
> > > > > kind of situation, on a per-output basis ?
> > > > 
> > > > A full restart of the bridge (power off/on) is needed and so we need to
> > > > redo the initialization sequence. This initialization sequence has to be
> > > > done with the DSI data lanes (bridge inputs) driven in LP11 state and so
> > > > without any video stream. Only focussing on bridge outputs will not be
> > > > sufficient. That's why I brought the pipeline down and restarted it.
> > > 
> > > Fair point.
> > > 
> > > > Of course, I can copy/paste sn65dsi83_reset_pipeline() to a core helper
> > > > function. Is drm_atomic_helper_reset_all() could be a good candidate?
> > > 
> > > The helper should operate on a single output, unrelated outputs should
> > > not be affected.
> > 
> > Also, you don't want to reset anything, you just want the last commit to
> > be replayed.
> 
> I'm not sure about that. If the last commit is just a page flip, that
> won't help, will it ?

The alternative would be that you start anew with a blank state, which
effectively drops every configuration that has been done by userspace.
This is terrible.

And a page flip wouldn't have affected the connector and
connector->state would still be to the last state that affected it, so
it would work.

Maxime
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 273 bytes
Desc: not available
URL: <https://lists.freedesktop.org/archives/dri-devel/attachments/20241028/1e940200/attachment.sig>