Re: [PATCH] drm/display/dsc: Add YCbCr 4:2:2 and 4:2:0 computation for DSC 1.2 RC parameters

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [High] drm/dsc: incorrect second line BPG offset accumulation
--

commit b83356cde3094c64861089db4359f0b624aad6bd
Author: Chaoyi Chen <kernel@airkyi.com>

drm/display/dsc: Add YCbCr 4:2:2 and 4:2:0 computation for DSC 1.2 RC parameters

This commit updates drm_dsc_compute_rc_parameters() to compute rate control
parameters for DSC 1.2 in native 4:2:0 and 4:2:2 modes. It also introduces
__drm_dsc_compute_buffer_offset() to optimize the calculation of rbsMin
using the DSC 1.2 reference C model.

> diff --git a/drivers/gpu/drm/display/drm_dsc_helper.c b/drivers/gpu/drm/display/drm_dsc_helper.c
> --- a/drivers/gpu/drm/display/drm_dsc_helper.c
> +++ b/drivers/gpu/drm/display/drm_dsc_helper.c
> @@ -1300,6 +1300,59 @@ int drm_dsc_setup_rc_params(struct drm_dsc_config *vdsc_cfg, enum drm_dsc_params
[ ... ]
> +static long
> +__drm_dsc_compute_buffer_offset(const struct drm_dsc_config *vdsc_cfg,
> +				unsigned long groups_per_line,
> +				unsigned long grpcnt)
> +{
[ ... ]
> +	if (vdsc_cfg->native_420) {
> +		if (grpcnt <= groups_per_line)
> +			offset -= (grpcnt * vdsc_cfg->nsl_bpg_offset) >> 11;
> +		else if (grpcnt <= 2 * groups_per_line)
> +			offset += (grpcnt - groups_per_line) *
> +				   vdsc_cfg->second_line_bpg_offset -
> +				 ((groups_per_line *
> +				   vdsc_cfg->nsl_bpg_offset) >> 11);
> +		else
> +			offset += (grpcnt - groups_per_line) *
> +				   vdsc_cfg->second_line_bpg_offset -
> +				 (((grpcnt - groups_per_line) *
> +				   vdsc_cfg->nsl_bpg_offset) >> 11);

[Severity: High]
Does this code inflate the accumulated offset for groups beyond the second
line?

In the branch where grpcnt > 2 * groups_per_line, the calculation multiplies
(grpcnt - groups_per_line) by vdsc_cfg->second_line_bpg_offset. Should this
be capped to groups_per_line * vdsc_cfg->second_line_bpg_offset instead?

When grpcnt_id is large (which is common for narrow slices), this unbounded
linear growth could significantly inflate the rate buffer size minimum
(rbs_min). This might result in invalid rate control parameters (rc_bits,
initial_dec_delay) and cause display configuration failures or corrupted
output when a 4:2:0 DSC-capable sink is connected.

> +	}
> +
> +	return offset;
> +}

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260923102616.215-1-kernel@airkyi.com?part=1