Re: [PATCH 07/11] ASoC: SOF: core: Skip firmware test for undefined fw_name

[Cristian Ciocaltea]

On 12/14/23 12:48, Péter Ujfalusi wrote:
> 
> 
> On 09/12/2023 22:53, Cristian Ciocaltea wrote:
>> Some SOF drivers like AMD ACP do not always rely on a single static
>> firmware file, but may require multiple files having their names
>> dynamically computed on probe time, e.g. based on chip name.
> 
> I see, AMD vangogh needs two binary files to be loaded sometimes, it is not using the default name as it hints via the sof_dev_desc ("sof-vangogh.ri").
> 
> The constructed names for the two files are just using different pattern:
> sof-%PLAT%.ri
> vs
> sof-%PLAT%-code.bin
> sof-%PLAT%-data.bin
> 
> iow, instead of the combined .ri file which includes the code and data segment it has 'raw' bin files and cannot use the core for loading the firmware.
> 
> What is the reason for this? an .ri file can have two 'modules' one to be written to IRAM the other to DRAM.
> sof_ipc3_load_fw_to_dsp()
> 
>> In those cases, providing an invalid default_fw_filename in their
>> sof_dev_desc struct will prevent probing due to 'SOF firmware
>> and/or topology file not found' error.
>  
>> Fix the issue by allowing drivers to omit initialization for this member
>> (or alternatively provide a dynamic override via ipc_file_profile_base)
>> and update sof_test_firmware_file() to verify the given profile data and
>> skip firmware testing if either fw_path or fw_name is not defined.
>>
>> Fixes: 6c393ebbd74a ("ASoC: SOF: core: Implement IPC version fallback if firmware files are missing")
>> Signed-off-by: Cristian Ciocaltea <cristian.ciocaltea@xxxxxxxxxxxxx>
>> ---
>>  sound/soc/sof/fw-file-profile.c | 3 +++
>>  1 file changed, 3 insertions(+)
>>
>> diff --git a/sound/soc/sof/fw-file-profile.c b/sound/soc/sof/fw-file-profile.c
>> index 138a1ca2c4a8..e63700234df0 100644
>> --- a/sound/soc/sof/fw-file-profile.c
>> +++ b/sound/soc/sof/fw-file-profile.c
>> @@ -21,6 +21,9 @@ static int sof_test_firmware_file(struct device *dev,
>>  	const u32 *magic;
>>  	int ret;
>>  
>> +	if (!profile->fw_path || !profile->fw_name || !*profile->fw_name)
>> +		return 0;
> 
> I would rather make the firmware file skipping based on custom loader use and print a dev_dbg() alongside:

Agree, that's a better approach. Thanks for the hint!

> diff --git a/sound/soc/sof/fw-file-profile.c b/sound/soc/sof/fw-file-profile.c
> index 138a1ca2c4a8..7b91c9551ada 100644
> --- a/sound/soc/sof/fw-file-profile.c
> +++ b/sound/soc/sof/fw-file-profile.c
> @@ -89,6 +89,15 @@ static int sof_test_topology_file(struct device *dev,
>  	return ret;
>  }
>  
> +static bool sof_platform_uses_generic_loader(struct snd_sof_dev *sdev)
> +{
> +	if (sdev->pdata->desc->ops->load_firmware == snd_sof_load_firmware_raw ||
> +	    sdev->pdata->desc->ops->load_firmware == snd_sof_load_firmware_memcpy)
> +		return true;
> +
> +	return false;
> +}

I would drop the conditional and simply return.

>  static int
>  sof_file_profile_for_ipc_type(struct snd_sof_dev *sdev,
>  			      enum sof_ipc_type ipc_type,
> @@ -130,7 +139,8 @@ sof_file_profile_for_ipc_type(struct snd_sof_dev *sdev,
>  	 * For default path and firmware name do a verification before
>  	 * continuing further.
>  	 */
> -	if (base_profile->fw_path || base_profile->fw_name) {
> +	if ((base_profile->fw_path || base_profile->fw_name) &&
> +	    sof_platform_uses_generic_loader(sdev)) {
>  		ret = sof_test_firmware_file(dev, out_profile, &ipc_type);
>  		if (ret)
>  			return ret;
> @@ -181,7 +191,8 @@ sof_file_profile_for_ipc_type(struct snd_sof_dev *sdev,
>  	out_profile->ipc_type = ipc_type;
>  
>  	/* Test only default firmware file */
> -	if (!base_profile->fw_path && !base_profile->fw_name)
> +	if ((!base_profile->fw_path && !base_profile->fw_name) &&
> +	    sof_platform_uses_generic_loader(sdev))
>  		ret = sof_test_firmware_file(dev, out_profile, NULL);
>  
>  	if (!ret)
> @@ -265,7 +276,9 @@ static void sof_print_profile_info(struct snd_sof_dev *sdev,
>  			 "Using fallback IPC type %d (requested type was %d)\n",
>  			 profile->ipc_type, ipc_type);
>  
> -	dev_info(dev, "Firmware paths/files for ipc type %d:\n", profile->ipc_type);
> +	/* The firmware path only valid when generic loader is used */
> +	if (sof_platform_uses_generic_loader(sdev))
> +		dev_info(dev, "Firmware paths/files for ipc type %d:\n", profile->ipc_type);
>  
>  	dev_info(dev, " Firmware file:     %s/%s\n", profile->fw_path, profile->fw_name);
>  	if (profile->fw_lib_path)
> 
>> +
>>  	fw_filename = kasprintf(GFP_KERNEL, "%s/%s", profile->fw_path,
>>  				profile->fw_name);
>>  	if (!fw_filename)
> 
> checking for custom loader allows you to drop the next patch.

Yes, I will take only the context information and use it to update the
current patch description.

> I would also skip the fw path printing in case of a custom loader.

Sure, makes sense.

Thanks for the review,
Cristian