machine driver pdata handling is broken

[plbossart]

in #164 the patch 'ASoC: SOF: Refine registeration of platform device' modified the way the machine driver is registered, and modified the arguments that are passed as pdata.

In the acpi case, we passed the 'mach' structure as argument because we need to check the actual device name. This is now broken, so this patch needs to be redone - with probably different arguments passed depending on the machine driver (e.g. mach pointer for i2s machine drivers and codec-mask pointer for hdaudio)

This change was also made only for ACPI and PCI but not for SPI, so @lyakh and @lgirdwood would need to chime in on this.

And last there is no consistent handling of the platform_device_unregister(), this was not removed for ACPI, removed for ACPI and removed in the next patch for SPI

To me this was a wild merge that hasn't been tested on all platforms...

[RanderWang]

@plbossart I can't test it with SPI, does it work at that time? I will check the code again although I don't get the point.

[plbossart]

@RanderWang forget SPI, I am more worried about all the other machine drivers that use the pdata.

/data/pbossart/ktest/sof-dev-rebase/sound/soc/intel/boards$ git grep platform_data
bxt_pcm512x.c:  mach = (&pdev->dev)->platform_data;
bxt_wm8804.c:   mach = (&pdev->dev)->platform_data;
bytcht_da7213.c:        mach = (&pdev->dev)->platform_data;
bytcht_es8316.c:        mach = (&pdev->dev)->platform_data;
bytcr_rt5640.c: mach = byt_rt5640_card.dev->platform_data;
bytcr_rt5651.c: mach = byt_rt5651_card.dev->platform_data;
cht_bsw_rt5645.c:       mach = (&pdev->dev)->platform_data;
cht_bsw_rt5672.c:       struct snd_soc_acpi_mach *mach = pdev->dev.platform_data;

if you changed the parameter passed as pdata then you broke all those platforms.

[RanderWang]

@plbossart my idea is: my fault in this PR is registering wrong platform_data to BYT, right ?
and this commit fix this:99d5562.
with this, mach is registered for BYT

If I am wrong, please notify me and I will refine later

[plbossart]

The current code expects struct snd_soc_acpi_mach pointer to be passed as pdata to BYT/CHT platforms.

In the PR, the new registration passes a different structure:

struct snd_sof_pdata *plat_data;

/* register machine driver */
	plat_data->pdev_mach =
		platform_device_register_data(sdev->dev,
					      plat_data->machine->drv_name,
					      -1, plat_data,
					      sizeof(*plat_data));

As I mentioned it, we can change the parameter to something more organized (https://github.com/plbossart/sound/commits/topic/mach-data). Most likely the code should be something like

/* register machine driver */
	plat_data->pdev_mach =
		platform_device_register_data(sdev->dev,
					      plat_data->machine->drv_name,
					      -1, plat_data->machine,
					      sizeof(*plat_data->machine));

makes sense?

[RanderWang]

Yes, you have prepared patches and want to PR by yourself ?

[RanderWang]

I study the PRs, they are done in common code. So I need to refine SOF side, right ?
If so, I will wait for these PRs merged

[plbossart]

@RanderWang I submitted the patches upstream, can you try and suggest an update for the SOF side? I am running out of time and you are probably the best person to make the changes. Thanks for your help.

[RanderWang]

No problem. I will work on it

[plbossart]

@RanderWang kind reminder. The code is now upstream so before I add the patches i'd need to make sure your changes are compatible and merge in one shot.