[PATCH v7 3/3] remoteproc: imx_rproc: detect and attach to pre-booted remote cores

[Mathieu Poirier]

On Sun, Jun 29, 2025 at 02:25:12PM -0300, Hiago De Franco wrote:
> From: Hiago De Franco <hiago.franco at toradex.com>
> 
> When the Cortex-M remote core is started and already running before
> Linux boots (typically by the Cortex-A bootloader using a command like
> bootaux), the current driver is unable to attach to it. This is because
> the driver only checks for remote cores running in different SCFW
> partitions. However in this case, the M-core is in the same partition as
> Linux and is already powered up and running by the bootloader.
> 
> This patch adds a check using dev_pm_genpd_is_on() to verify whether the
> M-core's power domains are already on. If all power domain devices are
> on, the driver assumes the M-core is running and proceed to attach to
> it.
> 
> To accomplish this, we need to avoid passing any attach_data or flags to
> dev_pm_domain_attach_list(), allowing the platform device become a
> consumer of the power domain provider without changing its current
> state.
> 
> During probe, also enable and sync the device runtime PM to make sure
> the power domains are correctly managed when the core is controlled by
> the kernel.
> 
> Suggested-by: Ulf Hansson <ulf.hansson at linaro.org>
> Reviewed-by: Ulf Hansson <ulf.hansson at linaro.org>
> Reviewed-by: Peng Fan <peng.fan at nxp.com>
> Signed-off-by: Hiago De Franco <hiago.franco at toradex.com>
> ---
> v6 -> v7:
>  - Added Peng reviewed-by.
> v5 -> v6:
>  - Commit description improved, as suggested. Added Ulf Hansson reviewed
>    by. Comment on imx-rproc.c improved.
> v4 -> v5:
>  - pm_runtime_get_sync() removed in favor of
>    pm_runtime_resume_and_get(). Now it also checks the return value of
>    this function.
>  - Added pm_runtime_disable() and pm_runtime_put() to imx_rproc_remove()
>    function.
> v3 -> v4:
>  - Changed to use the new dev_pm_genpd_is_on() function instead, as
>    suggested by Ulf. This will now get the power status of the two
>    remote cores power domains to decided if imx_rpoc needs to attach or
>    not. In order to do that, pm_runtime_enable() and
>    pm_runtime_get_sync() were introduced and pd_data was removed.
> v2 -> v3:
>  - Unchanged.
> v1 -> v2:
>  - Dropped unecessary include. Removed the imx_rproc_is_on function, as
>    suggested.
> ---
>  drivers/remoteproc/imx_rproc.c | 37 +++++++++++++++++++++++++++++-----
>  1 file changed, 32 insertions(+), 5 deletions(-)
> 
> diff --git a/drivers/remoteproc/imx_rproc.c b/drivers/remoteproc/imx_rproc.c
> index 627e57a88db2..24597b60c5b0 100644
> --- a/drivers/remoteproc/imx_rproc.c
> +++ b/drivers/remoteproc/imx_rproc.c
> @@ -18,6 +18,7 @@
>  #include <linux/of_reserved_mem.h>
>  #include <linux/platform_device.h>
>  #include <linux/pm_domain.h>
> +#include <linux/pm_runtime.h>
>  #include <linux/reboot.h>
>  #include <linux/regmap.h>
>  #include <linux/remoteproc.h>
> @@ -890,10 +891,8 @@ static int imx_rproc_partition_notify(struct notifier_block *nb,
>  static int imx_rproc_attach_pd(struct imx_rproc *priv)
>  {
>  	struct device *dev = priv->dev;
> -	int ret;
> -	struct dev_pm_domain_attach_data pd_data = {
> -		.pd_flags = PD_FLAG_DEV_LINK_ON,
> -	};
> +	int ret, i;
> +	bool detached = true;
>  
>  	/*
>  	 * If there is only one power-domain entry, the platform driver framework
> @@ -902,7 +901,22 @@ static int imx_rproc_attach_pd(struct imx_rproc *priv)
>  	if (dev->pm_domain)
>  		return 0;
>  
> -	ret = dev_pm_domain_attach_list(dev, &pd_data, &priv->pd_list);
> +	ret = dev_pm_domain_attach_list(dev, NULL, &priv->pd_list);
> +	/*
> +	 * If all the power domain devices are already turned on, the remote
> +	 * core is already powered up and running when the kernel booted (e.g.,
> +	 * started by U-Boot's bootaux command). In this case attach to it.
> +	 */
> +	for (i = 0; i < ret; i++) {
> +		if (!dev_pm_genpd_is_on(priv->pd_list->pd_devs[i])) {
> +			detached = false;
> +			break;
> +		}
> +	}

I was doing one final review of this work when I noticed the return code for
dev_pm_domain_attach_list() is never checked for error. 

Thanks,
Mathieu

> +
> +	if (detached)
> +		priv->rproc->state = RPROC_DETACHED;
> +
>  	return ret < 0 ? ret : 0;
>  }
>  
> @@ -1146,6 +1160,15 @@ static int imx_rproc_probe(struct platform_device *pdev)
>  		}
>  	}
>  
> +	if (dcfg->method == IMX_RPROC_SCU_API) {
> +		pm_runtime_enable(dev);
> +		ret = pm_runtime_resume_and_get(dev);
> +		if (ret) {
> +			dev_err(dev, "pm_runtime get failed: %d\n", ret);
> +			goto err_put_clk;
> +		}
> +	}
> +
>  	ret = rproc_add(rproc);
>  	if (ret) {
>  		dev_err(dev, "rproc_add failed\n");
> @@ -1171,6 +1194,10 @@ static void imx_rproc_remove(struct platform_device *pdev)
>  	struct rproc *rproc = platform_get_drvdata(pdev);
>  	struct imx_rproc *priv = rproc->priv;
>  
> +	if (priv->dcfg->method == IMX_RPROC_SCU_API) {
> +		pm_runtime_disable(priv->dev);
> +		pm_runtime_put(priv->dev);
> +	}
>  	clk_disable_unprepare(priv->clk);
>  	rproc_del(rproc);
>  	imx_rproc_put_scu(rproc);
> -- 
> 2.39.5
>