Re: [PATCH 2/2] drm: sun4i: hdmi: Add support for sun4i HDMI encoder audio

From: Maxime Ripard
Date: Wed Jan 15 2020 - 03:32:38 EST


Hi Stefan,

On Tue, Jan 14, 2020 at 11:04:55AM +0200, Stefan Mavrodiev wrote:
> On 1/10/20 6:26 PM, Maxime Ripard wrote:
> > Hi,
> >
> > On Fri, Jan 10, 2020 at 04:11:40PM +0200, Stefan Mavrodiev wrote:
> > > Add HDMI audio support for the sun4i-hdmi encoder, used on
> > > the older Allwinner chips - A10, A20, A31.
> > >
> > > Most of the code is based on the BSP implementation. In it
> > > dditional formats are supported (S20_3LE and S24_LE), however
> > > there where some problems with them and only S16_LE is left.
> > >
> > > Signed-off-by: Stefan Mavrodiev <stefan@xxxxxxxxxx>
> > > ---
> > > drivers/gpu/drm/sun4i/Kconfig | 1 +
> > > drivers/gpu/drm/sun4i/Makefile | 1 +
> > > drivers/gpu/drm/sun4i/sun4i_hdmi.h | 30 ++
> > > drivers/gpu/drm/sun4i/sun4i_hdmi_audio.c | 375 +++++++++++++++++++++++
> > > drivers/gpu/drm/sun4i/sun4i_hdmi_enc.c | 4 +
> > > 5 files changed, 411 insertions(+)
> > > create mode 100644 drivers/gpu/drm/sun4i/sun4i_hdmi_audio.c
> > >
> > > diff --git a/drivers/gpu/drm/sun4i/Kconfig b/drivers/gpu/drm/sun4i/Kconfig
> > > index 37e90e42943f..192b732b10cd 100644
> > > --- a/drivers/gpu/drm/sun4i/Kconfig
> > > +++ b/drivers/gpu/drm/sun4i/Kconfig
> > > @@ -19,6 +19,7 @@ if DRM_SUN4I
> > > config DRM_SUN4I_HDMI
> > > tristate "Allwinner A10 HDMI Controller Support"
> > > default DRM_SUN4I
> > > + select SND_PCM_ELD
> > > help
> > > Choose this option if you have an Allwinner SoC with an HDMI
> > > controller.
> > > diff --git a/drivers/gpu/drm/sun4i/Makefile b/drivers/gpu/drm/sun4i/Makefile
> > > index 0d04f2447b01..e2d82b451c36 100644
> > > --- a/drivers/gpu/drm/sun4i/Makefile
> > > +++ b/drivers/gpu/drm/sun4i/Makefile
> > > @@ -5,6 +5,7 @@ sun4i-frontend-y += sun4i_frontend.o
> > > sun4i-drm-y += sun4i_drv.o
> > > sun4i-drm-y += sun4i_framebuffer.o
> > >
> > > +sun4i-drm-hdmi-y += sun4i_hdmi_audio.o
> > > sun4i-drm-hdmi-y += sun4i_hdmi_ddc_clk.o
> > > sun4i-drm-hdmi-y += sun4i_hdmi_enc.o
> > > sun4i-drm-hdmi-y += sun4i_hdmi_i2c.o
> > > diff --git a/drivers/gpu/drm/sun4i/sun4i_hdmi.h b/drivers/gpu/drm/sun4i/sun4i_hdmi.h
> > > index 7ad3f06c127e..456964e681b0 100644
> > > --- a/drivers/gpu/drm/sun4i/sun4i_hdmi.h
> > > +++ b/drivers/gpu/drm/sun4i/sun4i_hdmi.h
> > > @@ -42,7 +42,32 @@
> > > #define SUN4I_HDMI_VID_TIMING_POL_VSYNC BIT(1)
> > > #define SUN4I_HDMI_VID_TIMING_POL_HSYNC BIT(0)
> > >
> > > +#define SUN4I_HDMI_AUDIO_CTRL_REG 0x040
> > > +#define SUN4I_HDMI_AUDIO_CTRL_ENABLE BIT(31)
> > > +#define SUN4I_HDMI_AUDIO_CTRL_RESET BIT(30)
> > > +
> > > +#define SUN4I_HDMI_AUDIO_FMT_REG 0x048
> > > +#define SUN4I_HDMI_AUDIO_FMT_SRC BIT(31)
> > > +#define SUN4I_HDMI_AUDIO_FMT_LAYOUT BIT(3)
> > > +#define SUN4I_HDMI_AUDIO_FMT_CH_CFG(n) (n - 1)
> > There's the issue multiple times in the headers, but you should wrap n
> > in parentheses to make sure we have no issue with precedence when
> > calling the macro.
> >
> > > +int sun4i_hdmi_audio_create(struct sun4i_hdmi *hdmi)
> > > +{
> > > + struct snd_soc_card *card = &sun4i_hdmi_audio_card;
> > > + struct snd_soc_dai_link_component *comp;
> > > + struct snd_soc_dai_link *link;
> > > + int ret;
> > > +
> > > + ret = devm_snd_dmaengine_pcm_register(hdmi->dev,
> > > + &sun4i_hdmi_audio_pcm_config, 0);
> > > + if (ret) {
> > > + DRM_ERROR("Could not register PCM\n");
> > > + return ret;
> > > + }
> > > +
> > > + ret = devm_snd_soc_register_component(hdmi->dev,
> > > + &sun4i_hdmi_audio_component,
> > > + &sun4i_hdmi_audio_dai, 1);
> > > + if (ret) {
> > > + DRM_ERROR("Could not register DAI\n");
> > > + return ret;
> > > + }
> > > +
> > > + link = devm_kzalloc(hdmi->dev, sizeof(*link), GFP_KERNEL);
> > > + if (!link)
> > > + return -ENOMEM;
> > > +
> > > + comp = devm_kzalloc(hdmi->dev, sizeof(*comp) * 3, GFP_KERNEL);
> > > + if (!comp)
> > > + return -ENOMEM;
> > > +
> > > + link->cpus = &comp[0];
> > > + link->codecs = &comp[1];
> > > + link->platforms = &comp[2];
> > > +
> > > + link->num_cpus = 1;
> > > + link->num_codecs = 1;
> > > + link->num_platforms = 1;
> > > +
> > > + link->playback_only = 1;
> > > +
> > > + link->name = "SUN4I-HDMI";
> > > + link->stream_name = "SUN4I-HDMI PCM";
> > > +
> > > + link->codecs->name = dev_name(hdmi->dev);
> > > + link->codecs->dai_name = sun4i_hdmi_audio_dai.name;
> > > +
> > > + link->cpus->dai_name = dev_name(hdmi->dev);
> > > +
> > > + link->platforms->name = dev_name(hdmi->dev);
> > > +
> > > + link->dai_fmt = SND_SOC_DAIFMT_I2S;
> > > +
> > > + card->dai_link = link;
> > > + card->num_links = 1;
> > > + card->dev = hdmi->dev;
> > > +
> > > + snd_soc_card_set_drvdata(card, hdmi);
> > > + return devm_snd_soc_register_card(hdmi->dev, card);
> >
> > Out of curiosity, did you try to remove the module with that patch
> > applied? IIRC, these functions will overwrite the device drvdata, and
> > we will try to access them in unbind / remove.
>
> Actually I did not. Just tried that and you're right. The module
> crashes at the unbind call. I use sun4i_hdmi struct only for
> regmap. Maybe create separate private structure and copy only
> regmap?

I think the issue is that:

- In bind, we first call dev_set_drvdata on the bound device, with a
pointer to struct sun4i_hdmi as the value. The driver_data field
in struct device is now a pointer to our instance of struct
sun4i_hdmi.

- In audio create, you then call snd_soc_card_set_drvdata with a
pointer to struct sun4i_hdmi as the value. The drvdata field in
the struct snd_soc_card is now a pointer to our instance of struct
sun4i_hdmi (so far so good).

- Then you call (devm_)snd_soc_register_card. One of the thing that
it will do is call drv_set_drvdata on the card->dev device,
setting it to our pointer to the struct snd_soc_card we provided.
However, since you set card->dev to the same device than the one
initially bound, this means that you just overwrote the struct
sun4i_hdmi pointer with a pointer to struct snd_soc_card.

- The driver will operate properly, since we never really use the
driver_data field, in the HDMI driver, except when...

- At unbind, you retrieve the driver_data field, expecting a struct
sun4i_hdmi pointer, except you have a pointer to struct
snd_soc_card, and everything explodes.

I think the way to work around that would be to create a new
(platform_)device for the HDMI audio component, so that ASoC can work
on that device instead.

This seems to be what dw-hdmi is doing here:
https://elixir.bootlin.com/linux/latest/source/drivers/gpu/drm/bridge/synopsys/dw-hdmi.c#L2812

(Except that they are also using platform_data, since they have
multiple drivers, we wouldn't, so we can just lookup sun4i_hdmi using
the parent's device driver_data).

> > > +}
> > > diff --git a/drivers/gpu/drm/sun4i/sun4i_hdmi_enc.c b/drivers/gpu/drm/sun4i/sun4i_hdmi_enc.c
> > > index a7c4654445c7..79ecd89fb705 100644
> > > --- a/drivers/gpu/drm/sun4i/sun4i_hdmi_enc.c
> > > +++ b/drivers/gpu/drm/sun4i/sun4i_hdmi_enc.c
> > > @@ -114,6 +114,9 @@ static void sun4i_hdmi_enable(struct drm_encoder *encoder)
> > > val |= SUN4I_HDMI_VID_CTRL_HDMI_MODE;
> > >
> > > writel(val, hdmi->base + SUN4I_HDMI_VID_CTRL_REG);
> > > +
> > > + if (hdmi->hdmi_audio && sun4i_hdmi_audio_create(hdmi))
> > > + DRM_ERROR("Couldn't create the HDMI audio adapter\n");
> >
> > So you create the audio card each time the display is enabled? I guess
> > this is to deal with the hotplug?
>
> Yes. See below.
>
> > I'm not sure this is the right thing to do. If I remember well, the
> > ELD are here precisely to let userspace know that the display is
> > plugged (and audio-capable) or not.
> >
> > Also, you don't remove that card in the disable, which mean that if
> > you end up in a situation where you would enable the display, disable
> > it and then enable it again, you have two audio cards now.
>
> There is issue with the hotplug. When inserting the cable, the event
> is detected and the hdmi encoder is enabled. Thus the card is
> created. However further removal and insertions are not
> detected.

I guess we would need to fix that then?

> This is why I don't remove the card.
>
> Also I count on devm_snd_soc_register_card() to release the card.

I think you should really create the card all the time, and just
update the ELD to let the userspace know when something has been
created.

And yeah, we should have a working hotplug, but that's a separate
story :)

Maxime

Attachment: signature.asc
Description: PGP signature