Re: [PATCH v3 11/15] drm/tidss: oldi: Fix OLDI signal polarities

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:

Pre-existing issues:
- [High] The `bus_format` field in `struct tidss_oldi` caches transient atomic state in a persistent device structure, causing state leakage that can lead to incorrect hardware programming.
--

commit aa70ea0cc740dc55d3518c6263c986fe44dc96b7
Author: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>

drm/tidss: oldi: Fix OLDI signal polarities

This commit fixes OLDI signal polarities by removing drm_bridge_timings,
setting the correct input_bus_cfg.flags in tidss_oldi_atomic_check(),
and setting OLDI_DEPOL based on the DE bus flag.

> diff --git a/drivers/gpu/drm/tidss/tidss_oldi.c b/drivers/gpu/drm/tidss/tidss_oldi.c
> index 17c535bfa0576..e925ddaa4fd6a 100644
> --- a/drivers/gpu/drm/tidss/tidss_oldi.c
> +++ b/drivers/gpu/drm/tidss/tidss_oldi.c
> @@ -183,7 +184,8 @@ static int tidss_oldi_config(struct tidss_oldi *oldi)
>  			 "OLDI%u: DSS port width %d not supported\n",
>  			 oldi->oldi_instance, bus_fmt->data_width);
>  
> -	oldi_cfg |= OLDI_DEPOL;
> +	if (bridge_state->input_bus_cfg.flags & DRM_BUS_FLAG_DE_LOW)
> +		oldi_cfg |= OLDI_DEPOL; /* 1 = active low */
>  
>  	oldi_cfg = (oldi_cfg & (~OLDI_MAP)) | (bus_fmt->oldi_mode_reg_val << 1);

[Severity: High]
This isn't a bug introduced by this patch, but does the use of bus_fmt
(which is assigned from the persistent oldi->bus_format earlier in this
function) cache transient atomic state in the persistent device structure?

During format negotiation in tidss_oldi_atomic_get_input_bus_fmts (called
during the atomic check phase), the negotiated bus format is cached in the
persistent oldi->bus_format pointer. If user space performs a
DRM_MODE_ATOMIC_TEST_ONLY commit that requests a different format,
oldi->bus_format is overwritten.

Since it is a test-only commit, the hardware state remains unchanged.
However, if the system later suspends and resumes, drm_atomic_helper_resume
commits the saved active state directly without re-running the atomic_check
hooks.

When tidss_oldi_atomic_pre_enable executes during resume, tidss_oldi_config
reads the corrupted oldi->bus_format from the aborted test commit instead
of deriving it from the active bridge_state. This results in the OLDI
hardware being programmed with the wrong data width and register values,
distorting or blanking the display.

Since this patch already passed bridge_state down to tidss_oldi_config(),
could the format derivation be updated to use
bridge_state->input_bus_cfg.format (or output_bus_cfg.format) instead of
the cached oldi->bus_format?

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260529-beagley-ai-display-v3-0-7fefdc5d1adf@ideasonboard.com?part=11