Hi Maxime, On Fri Nov 6, 2026 at 2:14 PM CET, Maxime Ripard wrote: > The ingenic driver allocates a structure holding both the DPI encoder > and the source bridge of its pipeline, and allocates it as a whole > through drmm_encoder_alloc(). The bridge is thus not known to > devm_drm_bridge_alloc(), which drm_bridge_attach() already warns > about. > > The bridge private object also gets its name from the devm bridge > allocation, and drm_bridge_attach() will soon report a failure of > that initialization instead of ignoring it, so a bridge allocated any > other way would prevent the driver from probing. > > Allocate the structure through devm_drm_bridge_alloc(), making the > bridge the primary object of the allocation, and initialise the > embedded encoder in place with drmm_encoder_init(). Both objects > stay in the same structure, so the container_of() conversions keep > working, and the bridge funcs assignment moves into the allocator > call. > > Signed-off-by: Maxime Ripard <[email protected]>
Seems there are now 3 competing patches to fix the same issue! :) This, [0] and [1]. [0] https://lore.kernel.org/all/4f30341735d77c77a9a4952abd91879007bc00fd.1787604223.git....@goldelico.com/ [1] https://lore.kernel.org/r/[email protected] However yours does not add a call to [devm_]drm_bridge_add(), so I'd rather prefer applying [1]. Also [1] removes the encoder pointer as it is not really necessary. Luca -- Luca Ceresoli, Bootlin Embedded Linux and Kernel engineering https://bootlin.com
