Re: [PATCH v5 2/2] drm/msm/dpu: fix WD timer handling on DPU 8.x

[Marijn Suijten <marijn.suijten@somainline.org>]

On 2025-12-28 05:57:12, Dmitry Baryshkov wrote:
> Since DPU 8.x Watchdog timer settings were moved from the TOP to the
> INTF block. Support programming the timer in the INTF block. Fixes tag
> points to the commit which removed register access to thos registers on

thos -> those

> DPU 8.x+ (and which also should have added proper support for WD timer
> on those devices).

Right, yes.  Commit 2f69e5458447 ("drm/msm/dpu: skip watchdog timer programming
through TOP on >= SM8450") was already a fixup of that (though marked as fixing
the followup commit 100d7ef ("drm/msm/dpu: add support for SM8450") for being
the first to use the new DPU_MDP_PERIPH_0_REMOVED flag).

> 
> Fixes: 43e3293fc614 ("drm/msm/dpu: add support for MDP_TOP blackhole")
> Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>

Reviewed-by: Marijn Suijten <marijn.suijten@somainline.org>

...
> @@ -791,7 +791,7 @@ static void _dpu_encoder_update_vsync_source(struct dpu_encoder_virt *dpu_enc,
>  
>  		if (phys_enc->has_intf_te && phys_enc->hw_intf->ops.vsync_sel)
>  			phys_enc->hw_intf->ops.vsync_sel(phys_enc->hw_intf,
> -							 vsync_cfg.vsync_source);
> +							 &vsync_cfg);

In some way this makes me wonder if we simply need another struct, in favour
of not missing fields that are never / not-yet read, although resulting in more
clutter.

(Just a nit / question, not a request)

...
> +	if (cfg->vsync_source == DPU_VSYNC_SOURCE_WD_TIMER_0) {
> +		u32 reg;
> +
> +		DPU_REG_WRITE(c, INTF_WD_TIMER_0_LOAD_VALUE,
> +			      CALCULATE_WD_LOAD_VALUE(cfg->frame_rate));
> +
> +		DPU_REG_WRITE(c, INTF_WD_TIMER_0_CTL, BIT(0)); /* clear timer */
> +		reg = DPU_REG_READ(c, INTF_WD_TIMER_0_CTL2);
> +		reg |= BIT(8);		/* enable heartbeat timer */
> +		reg |= BIT(0);		/* enable WD timer */

My downstream also sets BIT(1) for "select default 16 clock ticks":

https://github.com/sonyxperiadev/kernel-techpack-display-driver/blob/61a727e1ce1fda617a73793b2cbb76b5ca846ea2/msm/sde/sde_hw_intf.c#L511

Although it doesn't read back the current register value.  Do we need that; or
maybe you are inferring this "missing" BIT(1) via this readback?

After all downstream removed the readback exactly in favour of setting BIT(1)
though because of a "default value change" since MDSS 9.x.x:

https://github.com/sonyxperiadev/kernel-techpack-display-driver/commit/e55c68138b04770d51067c158f92de526e0c926e

- Marijn