Re: [PATCH v2] remoteproc: imx_dsp_rproc: Add support for DSP-specific features

[Mathieu Poirier <mathieu.poirier@linaro.org>]

Good day Iuliana,

On Tue, Mar 18, 2025 at 11:50:07PM +0200, Iuliana Prodan (OSS) wrote:
> From: Iuliana Prodan <iuliana.prodan@nxp.com>
> 
> Some DSP firmware requires a FW_READY signal before proceeding, while
> others do not.
> Therefore, add support to handle i.MX DSP-specific features.
> 
> Implement handle_rsc callback to handle resource table parsing and to
> process DSP-specific resource, to determine if waiting is needed.
> 
> Update imx_dsp_rproc_start() to handle this condition accordingly.
> 
> Signed-off-by: Iuliana Prodan <iuliana.prodan@nxp.com>
> ---
> Changes in v2:
> - Reviews from Mathieu Poirier:
>   - Use vendor-specific resource table entry.
>   - Implement resource handler specific to the i.MX DSP.
> - Revise commit message to include recent updates.
> - Link to v1: https://lore.kernel.org/all/20250305123923.514386-1-iuliana.prodan@oss.nxp.com/
> 
>  drivers/remoteproc/imx_dsp_rproc.c | 59 +++++++++++++++++++++++++++++-
>  1 file changed, 57 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/remoteproc/imx_dsp_rproc.c b/drivers/remoteproc/imx_dsp_rproc.c
> index b9bb15970966..1729bfbc602e 100644
> --- a/drivers/remoteproc/imx_dsp_rproc.c
> +++ b/drivers/remoteproc/imx_dsp_rproc.c
> @@ -35,9 +35,17 @@ module_param_named(no_mailboxes, no_mailboxes, int, 0644);
>  MODULE_PARM_DESC(no_mailboxes,
>  		 "There is no mailbox between cores, so ignore remote proc reply after start, default is 0 (off).");
>  
> +/* Flag indicating that the remote is up and running */
>  #define REMOTE_IS_READY				BIT(0)
> +/* Flag indicating that the host should wait for a firmware-ready response */
> +#define HOST_WAIT_FW_READY			BIT(1)

Right now the default behavior is to wait for a ready message from the remote
processor.  We need to keep this implementation in order to remain backward
compatible with older firwmare images.  As such the above should be:

#define HOST_DONT_WAIT_FW_READY			BIT(1)

>  #define REMOTE_READY_WAIT_MAX_RETRIES		500
>  
> +/* This flag is set in the DSP resource table's features field to indicate
> + * that the firmware requires the host to wait for a FW_READY response.
> + */
> +#define WAIT_FW_READY				BIT(0)
> +
>  /* att flags */
>  /* DSP own area */
>  #define ATT_OWN					BIT(31)
> @@ -136,6 +144,19 @@ struct imx_dsp_rproc_dcfg {
>  	int (*reset)(struct imx_dsp_rproc *priv);
>  };
>  
> +/**
> + * struct fw_rsc_imx_dsp - i.MX DSP specific info
> + * @len: length of the resource entry
> + * @features: feature flags supported by the i.MX DSP firmware
> + *
> + * This represents a DSP-specific resource in the firmware's
> + * resource table, providing information on supported features.
> + */
> +struct fw_rsc_imx_dsp {
> +	uint32_t len;
> +	uint32_t features;
> +};
> +

This needs a version field and magic number.  Have a look at what Xilinx did for
inspiration [1] (both magic and complement of the magic number).  You will also
need a "__packed" attribute for the structure.

[1].
https://elixir.bootlin.com/linux/v6.14-rc6/source/drivers/remoteproc/xlnx_r5_remoteproc.c#L103

>  static const struct imx_rproc_att imx_dsp_rproc_att_imx8qm[] = {
>  	/* dev addr , sys addr  , size	    , flags */
>  	{ 0x596e8000, 0x556e8000, 0x00008000, ATT_OWN },
> @@ -300,6 +321,39 @@ static int imx_dsp_rproc_ready(struct rproc *rproc)
>  	return -ETIMEDOUT;
>  }
>  
> +/**
> + * imx_dsp_rproc_handle_rsc() - Handle DSP-specific resource table entries
> + * @rproc: remote processor instance
> + * @rsc_type: resource type identifier
> + * @rsc: pointer to the resource entry
> + * @offset: offset of the resource entry
> + * @avail: available space in the resource table
> + *
> + * Parse the DSP-specific resource entry and update flags accordingly.
> + * If the WAIT_FW_READY feature is set, the host must wait for the firmware
> + * to signal readiness before proceeding with execution.
> + *
> + * Return: RSC_HANDLED if processed successfully, RSC_IGNORED otherwise.
> + */
> +static int imx_dsp_rproc_handle_rsc(struct rproc *rproc, u32 rsc_type,
> +				    void *rsc, int offset, int avail)
> +{
> +	struct imx_dsp_rproc *priv = rproc->priv;
> +	struct fw_rsc_imx_dsp *imx_dsp_rsc = rsc;
> +
> +	if (!imx_dsp_rsc || imx_dsp_rsc->len != sizeof(imx_dsp_rsc->features)) {
> +		priv->flags |= HOST_WAIT_FW_READY;

Unless I am missing something, this way of enforcing the default behavior won't
work because in older images, function imx_dsp_rproc_handle_rsc() will never be
called.

To fix this I suggest renaming HOST_WAIT_FW_READY to
HOST_DONT_WAIT_FOR_FW_READY.  That way the host can skip waiting for a FW ready
message only when it was specifically told to do so.

Thanks,
Mathieu

> +		return RSC_IGNORED;
> +	}
> +
> +	if (imx_dsp_rsc->features & WAIT_FW_READY)
> +		priv->flags |= HOST_WAIT_FW_READY;
> +	else
> +		priv->flags &= ~HOST_WAIT_FW_READY;
> +
> +	return RSC_HANDLED;
> +}
> +
>  /*
>   * Start function for rproc_ops
>   *
> @@ -335,8 +389,8 @@ static int imx_dsp_rproc_start(struct rproc *rproc)
>  
>  	if (ret)
>  		dev_err(dev, "Failed to enable remote core!\n");
> -	else
> -		ret = imx_dsp_rproc_ready(rproc);
> +	else if (priv->flags & HOST_WAIT_FW_READY)
> +		return imx_dsp_rproc_ready(rproc);
>  
>  	return ret;
>  }
> @@ -936,6 +990,7 @@ static const struct rproc_ops imx_dsp_rproc_ops = {
>  	.kick		= imx_dsp_rproc_kick,
>  	.load		= imx_dsp_rproc_elf_load_segments,
>  	.parse_fw	= imx_dsp_rproc_parse_fw,
> +	.handle_rsc	= imx_dsp_rproc_handle_rsc,
>  	.find_loaded_rsc_table = rproc_elf_find_loaded_rsc_table,
>  	.sanity_check	= rproc_elf_sanity_check,
>  	.get_boot_addr	= rproc_elf_get_boot_addr,
> -- 
> 2.25.1
>