Re: [PATCH v3] drm/panel: novatek-nt36672a: Inline panel init sequences

[Alexey Minnekhanov <alexeymin@minlexx.ru>]

On 23.05.2026 06:57, Chintan Patel wrote:
> Inline the panel initialization command sequences and remove the
> table-based command abstraction used by the NT36672A panel driver.
> 
> Replace the nt36672a_panel_cmd tables and nt36672a_send_cmds()
> helper with explicit initialization functions using
> mipi_dsi_dcs_write_seq_multi() and
> mipi_dsi_dcs_write_var_seq_multi() directly.
> 
> This improves readability by making the panel programming sequence
> explicit in code and allows future sharing of common command
> subsequences between panels. It also removes an unnecessary wrapper
> around the MIPI DSI helpers.
> 
> Additionally, compress repeated register writes into small loops where
> appropriate to reduce duplication in the initialization sequences.
> 
> Add:
> tianma_fhd_video_send_init_cmds_1()
> tianma_fhd_video_send_init_cmds_2()
> tianma_fhd_video_send_deinit_cmds()
> 
> Update nt36672a_panel_desc to use function pointers for panel init
> sequences and invoke them directly from prepare/unprepare paths.
> 
> Signed-off-by: Chintan Patel <chintanlike@gmail.com>
> Reviewed-by: Douglas Anderson <dianders@chromium.org>
> ---
> Changes in v3:
> - Rename on_init_1/on_init_2/off_init to send_init_cmds_1/send_init_cmds_2/send_deinit_cmds
> - Convert all hex values to lowercase (0xFF → 0xff) per kernel coding standards
> 
> Changes in v2:
> - Replace command tables with explicit init functions
> - Remove nt36672a_send_cmds() helper entirely
> - Use mipi_dsi_dcs_write_seq_multi() directly
> - Use mipi_dsi_dcs_write_var_seq_multi() loops for repeated register writes
> - Convert nt36672a_panel_desc command table fields to function pointers
> ---
>   .../gpu/drm/panel/panel-novatek-nt36672a.c    | 522 +++++++-----------
>   1 file changed, 186 insertions(+), 336 deletions(-)
> 
> diff --git a/drivers/gpu/drm/panel/panel-novatek-nt36672a.c b/drivers/gpu/drm/panel/panel-novatek-nt36672a.c
> index 7e8b5e059575..464d9ce47d87 100644
> --- a/drivers/gpu/drm/panel/panel-novatek-nt36672a.c
> +++ b/drivers/gpu/drm/panel/panel-novatek-nt36672a.c

...

> @@ -162,8 +141,8 @@ static int nt36672a_panel_prepare(struct drm_panel *panel)
>   	dsi_ctx.accum_err = nt36672a_panel_power_on(pinfo);
>   
>   	/* send first part of init cmds */
> -	nt36672a_send_cmds(&dsi_ctx, pinfo->desc->on_cmds_1,
> -			   pinfo->desc->num_on_cmds_1);
> +	if (pinfo->desc->send_init_cmds_1)
> +		pinfo->desc->send_init_cmds_1(&dsi_ctx);
>   
>   	mipi_dsi_dcs_exit_sleep_mode_multi(&dsi_ctx);
>   
> @@ -173,8 +152,8 @@ static int nt36672a_panel_prepare(struct drm_panel *panel)
>   	mipi_dsi_dcs_set_display_on_multi(&dsi_ctx);
>   
>   	/* Send rest of the init cmds */
> -	nt36672a_send_cmds(&dsi_ctx, pinfo->desc->on_cmds_2,
> -			   pinfo->desc->num_on_cmds_2);
> +	if (pinfo->desc->send_init_cmds_2)
> +		pinfo->desc->send_init_cmds_2(&dsi_ctx);
>   
>   	mipi_dsi_msleep(&dsi_ctx, 120);
>   

Hi!

Here, the split of panel init sequence into 2 separate functions _1 / _2
is completely artificial and completely unnecessary. For some unknown
reason the initial version of driver had initialization procedure cut in
2 parts on the boundary of exit_sleep_mode + set_display_on commands.

I think the whole init sequence should be glued together into one big
send_init_cmds() callback and have all these calls:

     mipi_dsi_dcs_exit_sleep_mode_multi(&dsi_ctx);
     /* 0x46 = 70 ms delay */
     mipi_dsi_msleep(&dsi_ctx, 70);
     mipi_dsi_dcs_set_display_on_multi(&dsi_ctx);


included in the middle. This opens up a possibility for other panels
based on nt36672a to be supported in this driver, which don't have the
init sequence split this way. The location of exit_sleep_mode and
set_display_on commands there is different. And, importantly, the delay
is different too.

I once did a very similar change to this driver [1] to add support for
more panels (ignore touchscreen changes there, look only at panel).
What is also important, that change was also tested on Xiaomi Poco F1
(xiaomi-beryllium) phone, which is the main user of this panel.
Unfortunately I never got to send it, even though I wanted to..

What do you think?

[1] https://gitlab.com/sdm845-mainline/linux/-/merge_requests/131

--
Regards,
Alexey Minnekhanov