Message-ID: <jymgfmn5pje4qaslyaivzfm4m7ho774pid3uenj6yfrlrdk3ss@2nqojclepgsu>
Date: Thu, 23 Oct 2025 14:43:12 +0300
From: Dmitry Baryshkov <dmitry.baryshkov@....qualcomm.com>
To: Srinivas Kandagatla <srini@...nel.org>
Cc: Srinivas Kandagatla <srinivas.kandagatla@....qualcomm.com>,
        Jianfeng Liu <liujianfeng1994@...il.com>,
        Liam Girdwood <lgirdwood@...il.com>, Mark Brown <broonie@...nel.org>,
        Jaroslav Kysela <perex@...ex.cz>, Takashi Iwai <tiwai@...e.com>,
        linux-arm-msm@...r.kernel.org, Xilin Wu <sophon@...xa.com>,
        Abhinav Kumar <quic_abhinavk@...cinc.com>,
        David Airlie <airlied@...il.com>, Dmitry Baryshkov <lumag@...nel.org>,
        Maarten Lankhorst <maarten.lankhorst@...ux.intel.com>,
        Maxime Ripard <mripard@...nel.org>, Simona Vetter <simona@...ll.ch>,
        Thomas Zimmermann <tzimmermann@...e.de>,
        dri-devel@...ts.freedesktop.org, linux-kernel@...r.kernel.org,
        linux-sound@...r.kernel.org
Subject: Re: [PATCH v2] drm/display: add hw_params callback function to
 drm_connector_hdmi_audio_ops

On Sun, Oct 05, 2025 at 06:29:20PM +0300, Dmitry Baryshkov wrote:
> On Sat, Oct 04, 2025 at 02:16:08PM +0100, Srinivas Kandagatla wrote:
> > 
> > 
> > On 10/3/25 6:02 PM, Dmitry Baryshkov wrote:
> > > On Fri, Oct 03, 2025 at 05:35:16PM +0100, Srinivas Kandagatla wrote:
> > >>
> > >>
> > >> On 9/26/25 4:09 PM, Dmitry Baryshkov wrote:
> > >>> On Fri, Sep 26, 2025 at 11:30:26AM +0100, Srinivas Kandagatla wrote:
> > >>>>
> > >>>>
> > >>>> On 9/25/25 5:28 AM, Dmitry Baryshkov wrote:
> > >>>>> On Thu, Sep 25, 2025 at 12:05:09PM +0800, Jianfeng Liu wrote:
> > >>>>>> After reusing drm_hdmi_audio_* helpers and drm_bridge_connector
> > >>>>>> integration in drm/msm/dp, we have dropped msm_dp_audio_hw_params and
> > >>>>>> use msm_dp_audio_prepare instead. While userspace is still calling
> > >>>>>> hw_params to do audio initialization, and we get the following errors:
> > >>>>>>
> > >>>>>> q6apm-lpass-dais 3700000.remoteproc:glink-edge:gpr:service@1:bedais: q6apm_lpass_dai_prepare() started
> > >>>>>> q6apm-lpass-dais 3700000.remoteproc:glink-edge:gpr:service@1:bedais: q6apm_lpass_dai_prepare() started
> > >>>>>> q6apm-lpass-dais 3700000.remoteproc:glink-edge:gpr:service@1:bedais: q6apm_lpass_dai_prepare() started
> > >>>>>> hdmi-audio-codec hdmi-audio-codec.0.auto: hdmi_codec_hw_params() started
> > >>>>>> q6apm-lpass-dais 3700000.remoteproc:glink-edge:gpr:service@1:bedais: q6apm_lpass_dai_prepare() started
> > >>>>>> qcom-apm gprsvc:service:2:1: Error (1) Processing 0x01001002 cmd
> > >>>>>> qcom-apm gprsvc:service:2:1: DSP returned error[1001002] 1
> > >>>>>> q6apm-lpass-dais 3700000.remoteproc:glink-edge:gpr:service@1:bedais: Failed to start APM port 104
> > >>>>>> q6apm-lpass-dais 3700000.remoteproc:glink-edge:gpr:service@1:bedais: ASoC error (-22): at snd_soc_dai_prepare() on DISPLAY_PORT_RX_0
> > >>>>>> MultiMedia2 Playback: ASoC error (-22): at dpcm_run_update_startup() on MultiMedia2 Playback
> > >>>>>
> > >>>>> And a call to hdmi_codec_prepare() comes only at this place.
> > >>>>>
> > >>>>> Srini, Mark, when selecting to only implement .prepare for codec ops I
> > >>>>> was following the commit 2fef64eec23a ("ASoC: hdmi-codec: Add a prepare
> > >>>>> hook"), which documents that IEC958 status bit is set after
> > >>>>> .hw_params(), so it's only visible during .prepare(). Is it okay to
> > >>>>> implement both callbacks? Or should the audioreach DAI driver be fixed
> > >>>>> somehow instead (I suppose it assumes that the port is available after
> > >>>>> .hw_params(), not sure if that assumption is correct)?
> > >>>>>
> > >>>>>>
> > >>>>>> msm_dp_audio_prepare is not called because hdmi-codec driver only checks
> > >>>>>> and runs hw_params before q6apm_lpass_dai_prepare(). This commit will
> > >>>>>> add hw_params callback same as drm_connector_hdmi_audio_prepare, so that
> > >>>>>> hdmi-codec driver can work with userspace alsa.
> > >>>>>>
> > >>>>>> Tested with Radxa Dragon Q6A.
> > >>>>>>
> > >>>>>> Fixes: 98a8920e7b07 ("drm/msm/dp: reuse generic HDMI codec implementation")
> > >>>>>> Signed-off-by: Jianfeng Liu <liujianfeng1994@...il.com>
> > >>>>>
> > >>>>> The patch LGTM, but I would wait for response from audio maintainers.
> > >>>>>
> > >>>>
> > >>>> The ordering matters in this case as we need clocks and audio
> > >>>> configuration on DP codec side to be setup before we start configuring
> > >>>> the dsp pipeline. Looks like that DSP is trying to setup DP endpoint
> > >>>> even before it is ready.
> > >>>>
> > >>>> q6apm prepare loads the dsp pipeline and starts configuring the
> > >>>> endpoints, if the DP endpoint is not ready dsp would throw an error.
> > >>>>
> > >>>> We might be able to pull in some dsp logs to confirm this, but I dont
> > >>>> have a setup that I can reproduce this issue.
> > >>>
> > >>> What would be your recommendation to proceed? Is it okay for the DAI
> > >>> driver to depend on the .hw_params enabling the clock? Also I see that
> > >>> the error regarding the clocks comes from .prepare callback too. What is
> > >>> the order of .prepare callbacks()? Can we influence it?
> > >>
> > >> prepare follows hw-params, and prepare can be called multiple times
> > >>
> > >> When you mean order of prepare callbacks, you mean w.r.t codec and dsp
> > >> backend dia link drivers ?
> > > 
> > > Yes. Here we got a dependency from the cpu dai to be prepare()'d after
> > > the DP driver performs some actions, which were a part of hw_params()
> > > callback but were moved to be executed during prepare() callback.
> > > 
> > > This leads me to two sets of questions:
> > > - In which order are those callbacks executed? Can we make the ASoC
> > >   enforce some order of DAI's prepare() callbacks?
> > > 
> > > - More improtantly, isn't it a sympthom of DP driver (incorrectly)
> > >   performing too much in the .hw_params() / .prepare() callback? Should
> > >   we move some of the setup to the .audio_startup() instead? What is the
> > >   expected split between those callbacks?
> > 
> > I have not looked at full dp sequences but, if prepare is the only place
> > when it enables the required clocks for audio block, then it is a
> > problem, we should do it early so that DSP can enable the required
> > configuration in prepare.
> 
> As I wrote, previously it was done as a part of hw_params, which is
> also not optimal as hdmi-codec clearly defines both callbacks as
> optional, requiring only one of them. Do you know, what is actually
> required on the DP side for the DSP to be able to preapre()?

Srinivas, did you have a chance to look at this? Or the .hw_prepare
ordering? I really don't want to land this fix only to face the same
issue because on some of the platforms .hw_prepare() callbacks will be
called in a different order.

> 
> > 
> > Its also doable to move out the clock related settings from  prepare to
> > startup which should work aswell.
> > 
> 
> -- 
> With best wishes
> Dmitry

-- 
With best wishes
Dmitry

Powered by blists - more mailing lists