Hello Jerome et al.,
on Amlogic Meson SoCs which use the AIU audio controller we have the
so-called "machine gun noise" (MGN) issue, reproducible for example in
Kodi. So far nobody was able to identify the cause of this issue. My
idea was to at least narrow down the issue so we can categorize it. For
that I wanted to get the SPDIF output from AIU to the HDMI controller
working.
On Amlogic Meson GXBB/GXL/GXM SoCs a DesignWare HDMI TX controller is
used. This has an SPDIF input but there's currently not driver for it.
On Meson8/8b/8m2 SoCs I am working on a HDMI driver for the TransSwitch
HDMI controller which - just like DesignWare HDMI TX - supports SPDIF
and I2S inputs.
I decided to add SPDIF support to the latter since the code from the
vendor driver is much easier.
It took me a while to figure out why I would not get any audio output
from AIU SPDIF to the HDMI controller - or from there to the sink.
The "fix" for this issue is the RFC patch which is part of this series.
Any feedback would be great as I am still new to the ASoC subsystem.
Another part I am still struggling with is the audio "routing" (due to
lack of a better term - please correct me if this is not the right word
to use for this case). I have the following description in my .dts:
sound {
compatible = "amlogic,gx-sound-card";
model = "M8B-ODROID-C1";
assigned-clocks = <&clkc CLKID_MPLL0>,
<&clkc CLKID_MPLL1>;
assigned-clock-rates = <294912000>,
<270950400>;
dai-link-0 {
sound-dai = <&aiu AIU_CPU CPU_I2S_FIFO>;
};
dai-link-1 {
sound-dai = <&aiu AIU_CPU CPU_SPDIF_FIFO>;
};
dai-link-2 {
sound-dai = <&aiu AIU_CPU CPU_I2S_ENCODER>;
dai-format = "i2s";
mclk-fs = <256>;
codec-0 {
sound-dai = <&aiu AIU_HDMI CTRL_I2S>;
};
};
dai-link-3 {
sound-dai = <&aiu AIU_CPU CPU_SPDIF_ENCODER>;
codec-0 {
sound-dai = <&aiu AIU_HDMI CTRL_PCM>;
};
};
dai-link-4 {
sound-dai = <&aiu AIU_HDMI CTRL_OUT>;
codec-0 {
sound-dai = <&hdmi_tx>;
};
};
};
The driver for &hdmi_tx defines:
struct hdmi_codec_pdata pdata = {
.ops = &txc_48352_hdmi_codec_ops,
.i2s = 1,
.spdif = 1,
.max_i2s_channels = 8,
.data = priv,
};
In hdmi_codec_ops.hw_params I always get fmt->fmt HDMI_I2S unless I
remove all I2S references from the .dts snipped above (only then
HDMI_SPDIF is detected). Based on the selection of the "HDMI Source"
enum in aiu-codec-ctrl I was expecting the format to update as well.
That unfortunately doesn't happen and I don't know how that can be
achieved.
Best regards,
Martin
Martin Blumenstingl (1):
ASoC: meson: aiu: Fix HDMI codec control selection
sound/soc/meson/aiu-codec-ctrl.c | 108 ++++++++++++++++++++++--------
sound/soc/meson/aiu-encoder-i2s.c | 6 --
2 files changed, 80 insertions(+), 34 deletions(-)
--
2.33.0
On Sun, Oct 03, 2021 at 01:43:12AM +0200, Martin Blumenstingl wrote:
> Hello Jerome et al.,
>
> on Amlogic Meson SoCs which use the AIU audio controller we have the
> so-called "machine gun noise" (MGN) issue, reproducible for example in
> Kodi. So far nobody was able to identify the cause of this issue. My
> idea was to at least narrow down the issue so we can categorize it. For
> that I wanted to get the SPDIF output from AIU to the HDMI controller
> working.
>
Hi, Martin,
The MGN issue surely must be that one when using ALSA directly
(no dmix, no JACK, no Pulseaudio, no Pipewire, no nothing) and
when the track changes or we pause it, when it resumes we get
noise instead of sound, correct?
If so, thanks for trying to track this down!
I bring it up because Googling it seems to head straight to a
Neil Armstrong post to linux-amlogic about what works and what
doesn't on linux-meson currently, and only if you dig deeper
you find a couple of reports apparently by the same person
about the "machine gun noise" thing.
It doesn't sound like a machine gun noise to me :)
More like brown noise, but anyway what I'd like to note to
everyone involved is that this one is a major dealbreaker.
Unless using JACK, Pulseaudio, Pipewire (?) or at least dmix
audio will degrade at the second song or when the user first
pauses and then resumes a stream, completely ruining the
experience.
> On Amlogic Meson GXBB/GXL/GXM SoCs a DesignWare HDMI TX controller is
> used. This has an SPDIF input but there's currently not driver for it.
> On Meson8/8b/8m2 SoCs I am working on a HDMI driver for the TransSwitch
> HDMI controller which - just like DesignWare HDMI TX - supports SPDIF
> and I2S inputs.
> I decided to add SPDIF support to the latter since the code from the
> vendor driver is much easier.
>
> It took me a while to figure out why I would not get any audio output
> from AIU SPDIF to the HDMI controller - or from there to the sink.
> The "fix" for this issue is the RFC patch which is part of this series.
> Any feedback would be great as I am still new to the ASoC subsystem.
>
> Another part I am still struggling with is the audio "routing" (due to
> lack of a better term - please correct me if this is not the right word
> to use for this case). I have the following description in my .dts:
> sound {
> compatible = "amlogic,gx-sound-card";
> model = "M8B-ODROID-C1";
>
> assigned-clocks = <&clkc CLKID_MPLL0>,
> <&clkc CLKID_MPLL1>;
> assigned-clock-rates = <294912000>,
> <270950400>;
> dai-link-0 {
> sound-dai = <&aiu AIU_CPU CPU_I2S_FIFO>;
> };
>
> dai-link-1 {
> sound-dai = <&aiu AIU_CPU CPU_SPDIF_FIFO>;
> };
>
> dai-link-2 {
> sound-dai = <&aiu AIU_CPU CPU_I2S_ENCODER>;
> dai-format = "i2s";
> mclk-fs = <256>;
>
> codec-0 {
> sound-dai = <&aiu AIU_HDMI CTRL_I2S>;
> };
> };
>
> dai-link-3 {
> sound-dai = <&aiu AIU_CPU CPU_SPDIF_ENCODER>;
>
> codec-0 {
> sound-dai = <&aiu AIU_HDMI CTRL_PCM>;
> };
> };
>
> dai-link-4 {
> sound-dai = <&aiu AIU_HDMI CTRL_OUT>;
>
> codec-0 {
> sound-dai = <&hdmi_tx>;
> };
> };
> };
> The driver for &hdmi_tx defines:
> struct hdmi_codec_pdata pdata = {
> .ops = &txc_48352_hdmi_codec_ops,
> .i2s = 1,
> .spdif = 1,
> .max_i2s_channels = 8,
> .data = priv,
> };
> In hdmi_codec_ops.hw_params I always get fmt->fmt HDMI_I2S unless I
> remove all I2S references from the .dts snipped above (only then
> HDMI_SPDIF is detected). Based on the selection of the "HDMI Source"
> enum in aiu-codec-ctrl I was expecting the format to update as well.
> That unfortunately doesn't happen and I don't know how that can be
> achieved.
>
Hm, that sounds weird. I hope you get the answers you are looking for.
Thanks,
Geraldo Nascimento
>
> Best regards,
> Martin
>
>
> Martin Blumenstingl (1):
> ASoC: meson: aiu: Fix HDMI codec control selection
>
> sound/soc/meson/aiu-codec-ctrl.c | 108 ++++++++++++++++++++++--------
> sound/soc/meson/aiu-encoder-i2s.c | 6 --
> 2 files changed, 80 insertions(+), 34 deletions(-)
>
> --
> 2.33.0
>
> On 3 Oct 2021, at 9:13 am, Geraldo Nascimento <[email protected]> wrote:
>
> On Sun, Oct 03, 2021 at 01:43:12AM +0200, Martin Blumenstingl wrote:
>> Hello Jerome et al.,
>>
>> on Amlogic Meson SoCs which use the AIU audio controller we have the
>> so-called "machine gun noise" (MGN) issue, reproducible for example in
>> Kodi. So far nobody was able to identify the cause of this issue. My
>> idea was to at least narrow down the issue so we can categorize it. For
>> that I wanted to get the SPDIF output from AIU to the HDMI controller
>> working.
>>
>
> Hi, Martin,
>
> The MGN issue surely must be that one when using ALSA directly
> (no dmix, no JACK, no Pulseaudio, no Pipewire, no nothing) and
> when the track changes or we pause it, when it resumes we get
> noise instead of sound, correct?
>
> If so, thanks for trying to track this down!
>
> I bring it up because Googling it seems to head straight to a
> Neil Armstrong post to linux-amlogic about what works and what
> doesn't on linux-meson currently, and only if you dig deeper
> you find a couple of reports apparently by the same person
> about the "machine gun noise" thing.
>
> It doesn't sound like a machine gun noise to me :)
>
> More like brown noise, but anyway what I'd like to note to
> everyone involved is that this one is a major dealbreaker.
> Unless using JACK, Pulseaudio, Pipewire (?) or at least dmix
> audio will degrade at the second song or when the user first
> pauses and then resumes a stream, completely ruining the
> experience.
I’m the origin of most reports that Google will eventually discover. I was told elsewhere that this is known as MGN so I’ve described it as such, but that could be wrong. For clarity this is the noise (triggered, and with AVR volume deliberately right down) followed by some navigation in the Kodi GUI to change output:
https://chewitt.libreelec.tv/testing/wp2_audio_noise.mov
In the LibreELEC images I work on where this occurs it’s not something I associate with play/pause (although I can’t definitively rule it out) or second or third songs (not something that happens that often or predictably) but more when there’s a hard transition like starting or stopping playback. I hit it most with simple scrolling around in the Kodi GUI, although that could be because (due to this issue) I limit media playback testing on AIU devices. It’s particularly simple to trigger from the Kodi Settings > System > Audio page, just scroll around it for 10-30 seconds, which is handy because the only way to stop the noise is switching audio source (from anything to anything) on the same page.
CH.
>> On Amlogic Meson GXBB/GXL/GXM SoCs a DesignWare HDMI TX controller is
>> used. This has an SPDIF input but there's currently not driver for it.
>> On Meson8/8b/8m2 SoCs I am working on a HDMI driver for the TransSwitch
>> HDMI controller which - just like DesignWare HDMI TX - supports SPDIF
>> and I2S inputs.
>> I decided to add SPDIF support to the latter since the code from the
>> vendor driver is much easier.
>>
>> It took me a while to figure out why I would not get any audio output
>> from AIU SPDIF to the HDMI controller - or from there to the sink.
>> The "fix" for this issue is the RFC patch which is part of this series.
>> Any feedback would be great as I am still new to the ASoC subsystem.
>>
>> Another part I am still struggling with is the audio "routing" (due to
>> lack of a better term - please correct me if this is not the right word
>> to use for this case). I have the following description in my .dts:
>> sound {
>> compatible = "amlogic,gx-sound-card";
>> model = "M8B-ODROID-C1";
>>
>> assigned-clocks = <&clkc CLKID_MPLL0>,
>> <&clkc CLKID_MPLL1>;
>> assigned-clock-rates = <294912000>,
>> <270950400>;
>> dai-link-0 {
>> sound-dai = <&aiu AIU_CPU CPU_I2S_FIFO>;
>> };
>>
>> dai-link-1 {
>> sound-dai = <&aiu AIU_CPU CPU_SPDIF_FIFO>;
>> };
>>
>> dai-link-2 {
>> sound-dai = <&aiu AIU_CPU CPU_I2S_ENCODER>;
>> dai-format = "i2s";
>> mclk-fs = <256>;
>>
>> codec-0 {
>> sound-dai = <&aiu AIU_HDMI CTRL_I2S>;
>> };
>> };
>>
>> dai-link-3 {
>> sound-dai = <&aiu AIU_CPU CPU_SPDIF_ENCODER>;
>>
>> codec-0 {
>> sound-dai = <&aiu AIU_HDMI CTRL_PCM>;
>> };
>> };
>>
>> dai-link-4 {
>> sound-dai = <&aiu AIU_HDMI CTRL_OUT>;
>>
>> codec-0 {
>> sound-dai = <&hdmi_tx>;
>> };
>> };
>> };
>> The driver for &hdmi_tx defines:
>> struct hdmi_codec_pdata pdata = {
>> .ops = &txc_48352_hdmi_codec_ops,
>> .i2s = 1,
>> .spdif = 1,
>> .max_i2s_channels = 8,
>> .data = priv,
>> };
>> In hdmi_codec_ops.hw_params I always get fmt->fmt HDMI_I2S unless I
>> remove all I2S references from the .dts snipped above (only then
>> HDMI_SPDIF is detected). Based on the selection of the "HDMI Source"
>> enum in aiu-codec-ctrl I was expecting the format to update as well.
>> That unfortunately doesn't happen and I don't know how that can be
>> achieved.
>>
>
> Hm, that sounds weird. I hope you get the answers you are looking for.
>
> Thanks,
> Geraldo Nascimento
>
>>
>> Best regards,
>> Martin
>>
>>
>> Martin Blumenstingl (1):
>> ASoC: meson: aiu: Fix HDMI codec control selection
>>
>> sound/soc/meson/aiu-codec-ctrl.c | 108 ++++++++++++++++++++++--------
>> sound/soc/meson/aiu-encoder-i2s.c | 6 --
>> 2 files changed, 80 insertions(+), 34 deletions(-)
>>
>> --
>> 2.33.0
>>
>
On Sun 03 Oct 2021 at 01:43, Martin Blumenstingl <[email protected]> wrote:
> Hello Jerome et al.,
>
> on Amlogic Meson SoCs which use the AIU audio controller we have the
> so-called "machine gun noise" (MGN) issue, reproducible for example in
> Kodi. So far nobody was able to identify the cause of this issue. My
> idea was to at least narrow down the issue so we can categorize it. For
> that I wanted to get the SPDIF output from AIU to the HDMI controller
> working.
Ok, I'll clarify (again) about this issue.
This issue is on AIU only (not the AXG series), on the i2s encoder
regardless of the output (HDMI, internal and external codec)
This encoder support either 2ch or 8ch mode, nothing else.
As long as only 2ch was supported, there was no problem.
In order to switch between 2ch and 8ch mode. AIU_RST_SOFT_I2S_FAST and
AIU_I2S_SYNC shall be "poked" (check iu-fifo-i2s.c and
aiu-encoder-i2c.c). Without this, the switch is not done.
What these do, how and when it should be done is unclear. I tried
several combination. At the time I found something stable-ish but it was
not perfect.
At the time, It was completely new driver. Even if was not rock solid,
it was still progress and I opted to upstream it with an imperfect 8ch
support so people could help debug it. This was mentioned in the
original submission.
The other solution is to restrict to 2ch mode and remove
AIU_RST_SOFT_I2S_FAST and AIU_I2S_SYNC pokes. There will be no noise
anymore.
>
> On Amlogic Meson GXBB/GXL/GXM SoCs a DesignWare HDMI TX controller is
> used. This has an SPDIF input but there's currently not driver for it.
> On Meson8/8b/8m2 SoCs I am working on a HDMI driver for the TransSwitch
> HDMI controller which - just like DesignWare HDMI TX - supports SPDIF
> and I2S inputs.
> I decided to add SPDIF support to the latter since the code from the
> vendor driver is much easier.
>
> It took me a while to figure out why I would not get any audio output
> from AIU SPDIF to the HDMI controller - or from there to the sink.
> The "fix" for this issue is the RFC patch which is part of this series.
> Any feedback would be great as I am still new to the ASoC subsystem.
>
> Another part I am still struggling with is the audio "routing" (due to
> lack of a better term - please correct me if this is not the right word
> to use for this case). I have the following description in my .dts:
> sound {
> compatible = "amlogic,gx-sound-card";
> model = "M8B-ODROID-C1";
>
> assigned-clocks = <&clkc CLKID_MPLL0>,
> <&clkc CLKID_MPLL1>;
> assigned-clock-rates = <294912000>,
> <270950400>;
> dai-link-0 {
> sound-dai = <&aiu AIU_CPU CPU_I2S_FIFO>;
> };
>
> dai-link-1 {
> sound-dai = <&aiu AIU_CPU CPU_SPDIF_FIFO>;
> };
>
> dai-link-2 {
> sound-dai = <&aiu AIU_CPU CPU_I2S_ENCODER>;
> dai-format = "i2s";
> mclk-fs = <256>;
>
> codec-0 {
> sound-dai = <&aiu AIU_HDMI CTRL_I2S>;
> };
> };
>
> dai-link-3 {
> sound-dai = <&aiu AIU_CPU CPU_SPDIF_ENCODER>;
>
> codec-0 {
> sound-dai = <&aiu AIU_HDMI CTRL_PCM>;
Ok, the obove sums up the problem around this RFC.
Here you describe a DAI link (think of it as wires) between the SPDIF
encoder (output) and AIU_HDMI input PCM. This is not what the HW is and
it is not possible.
Let's start from the HDMI controller.
The designware (on amlogic SoC) has 2 interface for audio input.
1) PCM/I2S: a classic interface 2 clocks and N data line
2) SPDIF: The Sony Philips 1 wire interface
Whatever comes up on 1) has to be some sort of i2s signal. So SPDIF
won't fly there.
AIU_HDMI output is Hardwired to 1). It is just just a digital mux,
implemented as an ASoC codec which allows to seleted one of 2 audio
sources:
A) the i2s output implemented as part of the AIU
B) the PCM output, part the AUDIN (yes, an output in AUDIN) block. This
is not implemented ATM.
So trying to handle SPDIF through AIU_HDMI is wrong.
> };
> };
>
> dai-link-4 {
> sound-dai = <&aiu AIU_HDMI CTRL_OUT>;
>
> codec-0 {
> sound-dai = <&hdmi_tx>;
your HDMI controller should have an argument to the phandle if it
supported both I2S and SPDIF inputs
> };
> };
> };
> The driver for &hdmi_tx defines:
> struct hdmi_codec_pdata pdata = {
> .ops = &txc_48352_hdmi_codec_ops,
> .i2s = 1,
> .spdif = 1,
> .max_i2s_channels = 8,
> .data = priv,
> };
> In hdmi_codec_ops.hw_params I always get fmt->fmt HDMI_I2S unless I
> remove all I2S references from the .dts snipped above (only then
> HDMI_SPDIF is detected). Based on the selection of the "HDMI Source"
> enum in aiu-codec-ctrl I was expecting the format to update as well.
> That unfortunately doesn't happen and I don't know how that can be
> achieved.
It is not meant to. The dai_link and the endpoint are i2s.
Your HDMI controller should have 2 inputs and should have a way to
select one or the other. The format at each of the (internal) input does
not change
>
>
> Best regards,
> Martin
>
>
> Martin Blumenstingl (1):
> ASoC: meson: aiu: Fix HDMI codec control selection
>
> sound/soc/meson/aiu-codec-ctrl.c | 108 ++++++++++++++++++++++--------
> sound/soc/meson/aiu-encoder-i2s.c | 6 --
> 2 files changed, 80 insertions(+), 34 deletions(-)
Hi Jerome,
first of all: thanks for taking the time to go through all of this!
On Sun, Oct 3, 2021 at 6:34 PM Jerome Brunet <[email protected]> wrote:
>
>
> On Sun 03 Oct 2021 at 01:43, Martin Blumenstingl <[email protected]> wrote:
>
> > Hello Jerome et al.,
> >
> > on Amlogic Meson SoCs which use the AIU audio controller we have the
> > so-called "machine gun noise" (MGN) issue, reproducible for example in
> > Kodi. So far nobody was able to identify the cause of this issue. My
> > idea was to at least narrow down the issue so we can categorize it. For
> > that I wanted to get the SPDIF output from AIU to the HDMI controller
> > working.
>
> Ok, I'll clarify (again) about this issue.
>
> This issue is on AIU only (not the AXG series), on the i2s encoder
> regardless of the output (HDMI, internal and external codec)
Thanks for repeating this. It was clear to me that only AIU is
affected but I didn't know that it only affects I2S
> This encoder support either 2ch or 8ch mode, nothing else.
> As long as only 2ch was supported, there was no problem.
> In order to switch between 2ch and 8ch mode. AIU_RST_SOFT_I2S_FAST and
> AIU_I2S_SYNC shall be "poked" (check iu-fifo-i2s.c and
> aiu-encoder-i2c.c). Without this, the switch is not done.
>
> What these do, how and when it should be done is unclear. I tried
> several combination. At the time I found something stable-ish but it was
> not perfect.
old 32-bit u-boot sources from the Endless Mini do have some
documentation on AIU_I2S_SYNC [0]:
// 8'b0_000_1_1_10
// A write to this register will cause the interface to repeat the current
// sample. Can be used to regain synchronization.
// A read from this register indicates that the next sample to be sent
// out of the interface should go into the _left_ channel of the dac.
There's also a note about AIU_I2S_MISC stating:
// Bit 4 if true, force each audio data to left or right according to
the bit attached with the audio data
// This bit should be used with Register AIU_i2s_sync(0x511) together
To be honest: to me this is not helpful since I don't understand
how/why the left channel is of importance.
> At the time, It was completely new driver. Even if was not rock solid,
> it was still progress and I opted to upstream it with an imperfect 8ch
> support so people could help debug it. This was mentioned in the
> original submission.
>
> The other solution is to restrict to 2ch mode and remove
> AIU_RST_SOFT_I2S_FAST and AIU_I2S_SYNC pokes. There will be no noise
> anymore.
I think Christian (Hewitt) agrees on this point as he mentioned that
your earlier versions of the AIU code (before it got upstream) were
not affected by the "machine gun noise" issue.
Does the documentation from above give you any new ideas (assuming
that it's correct since it's the best we have)? Should I try playing
with AIU_RST_SOFT_I2S_FAST and AIU_I2S_SYNC to see if I can make a
difference?
[...]
> Here you describe a DAI link (think of it as wires) between the SPDIF
> encoder (output) and AIU_HDMI input PCM. This is not what the HW is and
> it is not possible.
>
> Let's start from the HDMI controller.
> The designware (on amlogic SoC) has 2 interface for audio input.
> 1) PCM/I2S: a classic interface 2 clocks and N data line
> 2) SPDIF: The Sony Philips 1 wire interface
The Transwitch HDMI TX controller supports these two inputs so even
though the IP is different the basic functionality (which we'll
discuss below) is the same.
> Whatever comes up on 1) has to be some sort of i2s signal. So SPDIF
> won't fly there.
I agree with this
> AIU_HDMI output is Hardwired to 1). It is just just a digital mux,
> implemented as an ASoC codec which allows to seleted one of 2 audio
> sources:
> A) the i2s output implemented as part of the AIU
> B) the PCM output, part the AUDIN (yes, an output in AUDIN) block. This
> is not implemented ATM.
This is some interesting information, I thought that PCM was used
because PCM audio can be transmitted over SPDIF.
For A) my understanding is different though:
- for AIU_HDMI_CLK_DATA_CTRL[5:4] (HDMI_DATA_SEL) your description
matches my understanding. For me it makes sense that SPDIF data cannot
be selected with this register since it's a one-wire protocol (and
doesn't have separate data/clock lines). Value 0x2 selects the I2S
data interface
- for AIU_HDMI_CLK_DATA_CTRL[1:0] (HDMI_DATA_CLK_SEL) however I have a
different understanding: 0x0 disables the clock signal to the HDMI TX
controller, 0x1 selects the PCM clock (which now I have learned is
related to the AUDIN block) and 0x2 selects the "AIU clock" (see
below)
- my understanding is that "AIU clock" comes from AIU_CLK_CTRL_MORE[6]
(HDMITX_SEL_AOCLKX2) where 0x0 selects "cts_i958 as AIU clk to
hdmi_tx_audio_master_clk" (SPDIF clock) and 0x1 selects
"cts_aoclkx2_int as AIU clk to hdmi_tx_audio_master_clk" (I2S clock)
So to me this means that there's three different muxes:
- data mux to choose between 0x0 (all zeros), 0x1 (PCM) and 0x2 (I2S)
- clock mux to choose between 0x0 (disabled), 0x1 (PCM) and 0x2
(hdmi_tx_audio_master_clk)
- hdmi_tx_audio_master_clk clock mux to choose between 0x0 (cts_i958)
and 0x1 (cts_aoclkx2_int)
Based on that I think that it's not possible to have AIU output the
I2S and SPDIF signals at the same time to the HDMI TX controller and
then letting the HDMI TX controller choose which one to use.
Based on whichever signal (I2S or SPDIF) we want to use for HDMI we
need to configure AIU accordingly (data and clock).
[...]
> It is not meant to. The dai_link and the endpoint are i2s.
> Your HDMI controller should have 2 inputs and should have a way to
> select one or the other. The format at each of the (internal) input does
> not change
ah, that makes sense.
Let's say AIU has some internal muxing for the HDMI output then AIU
would have two outputs as well (let's call them HDMI_OUT_I2S and
HDMI_OUT_SPDIF).
Then I'd wire up each of these to their corresponding inputs on the
HDMI TX controller.
Best regards,
Martin
On Sun, Oct 3, 2021 at 9:17 PM Martin Blumenstingl
<[email protected]> wrote:
[...]
> documentation on AIU_I2S_SYNC [0]:
I forgot to add the link to my last email, sorry about that - adding
it here to clarify where I got this info from.
[0] https://github.com/endlessm/u-boot-meson/blob/f1ee03e3f7547d03e1478cc1fc967a9e5a121d92/arch/arm/include/asm/arch-m8b/register.h#L2147:L2152
On Sun, Oct 03, 2021 at 11:00:15AM +0400, Christian Hewitt wrote:
>
> > On 3 Oct 2021, at 9:13 am, Geraldo Nascimento <[email protected]> wrote:
> >
> > On Sun, Oct 03, 2021 at 01:43:12AM +0200, Martin Blumenstingl wrote:
> >> Hello Jerome et al.,
> >>
> >> on Amlogic Meson SoCs which use the AIU audio controller we have the
> >> so-called "machine gun noise" (MGN) issue, reproducible for example in
> >> Kodi. So far nobody was able to identify the cause of this issue. My
> >> idea was to at least narrow down the issue so we can categorize it. For
> >> that I wanted to get the SPDIF output from AIU to the HDMI controller
> >> working.
> >>
> >
> > Hi, Martin,
> >
> > The MGN issue surely must be that one when using ALSA directly
> > (no dmix, no JACK, no Pulseaudio, no Pipewire, no nothing) and
> > when the track changes or we pause it, when it resumes we get
> > noise instead of sound, correct?
> >
> > If so, thanks for trying to track this down!
> >
> > I bring it up because Googling it seems to head straight to a
> > Neil Armstrong post to linux-amlogic about what works and what
> > doesn't on linux-meson currently, and only if you dig deeper
> > you find a couple of reports apparently by the same person
> > about the "machine gun noise" thing.
> >
> > It doesn't sound like a machine gun noise to me :)
> >
> > More like brown noise, but anyway what I'd like to note to
> > everyone involved is that this one is a major dealbreaker.
> > Unless using JACK, Pulseaudio, Pipewire (?) or at least dmix
> > audio will degrade at the second song or when the user first
> > pauses and then resumes a stream, completely ruining the
> > experience.
>
>
> I’m the origin of most reports that Google will eventually
> discover. I was told elsewhere that this is known as MGN so
> I’ve described it as such, but that could be wrong. For
> clarity this is the noise (triggered, and with AVR volume
> deliberately right down) followed by some navigation in the
> Kodi GUI to change output:
>
> https://chewitt.libreelec.tv/testing/wp2_audio_noise.mov
>
Hi, Christian,
Thanks for the video.
> In the LibreELEC images I work on where this occurs it’s not
> something I associate with play/pause (although I can’t
> definitively rule it out) or second or third songs
> (not something that happens that often or predictably) but
> more when there’s a hard transition like starting or
> stopping playback. I hit it most with simple scrolling around
> in the Kodi GUI, although that could be because (due to this issue)
> I limit media playback testing on AIU devices. It’s particularly
> simple to trigger from the Kodi Settings > System > Audio page,
> just scroll around it for 10-30 seconds, which is handy because
> the only way to stop the noise is switching audio source (from
> anything to anything) on the same page.
>
> CH.
>
I wasn't able to reproduce the bug by simply scrolling around. On
the other hand, once I flipped the Send Low Volume Noise switch
to off, all hell broke loose!
This is one show-stopper bug, and I empathize to how particularly
nasty this must be for a "Just enough Linux for Kodi" approach.
But I also see Jerome Brunet has already clarified the subject in
response to Martin Blumenstingl's original post.
Thank you,
Geraldo Nascimento
> >> On Amlogic Meson GXBB/GXL/GXM SoCs a DesignWare HDMI TX controller is
> >> used. This has an SPDIF input but there's currently not driver for it.
> >> On Meson8/8b/8m2 SoCs I am working on a HDMI driver for the TransSwitch
> >> HDMI controller which - just like DesignWare HDMI TX - supports SPDIF
> >> and I2S inputs.
> >> I decided to add SPDIF support to the latter since the code from the
> >> vendor driver is much easier.
> >>
> >> It took me a while to figure out why I would not get any audio output
> >> from AIU SPDIF to the HDMI controller - or from there to the sink.
> >> The "fix" for this issue is the RFC patch which is part of this series.
> >> Any feedback would be great as I am still new to the ASoC subsystem.
> >>
> >> Another part I am still struggling with is the audio "routing" (due to
> >> lack of a better term - please correct me if this is not the right word
> >> to use for this case). I have the following description in my .dts:
> >> sound {
> >> compatible = "amlogic,gx-sound-card";
> >> model = "M8B-ODROID-C1";
> >>
> >> assigned-clocks = <&clkc CLKID_MPLL0>,
> >> <&clkc CLKID_MPLL1>;
> >> assigned-clock-rates = <294912000>,
> >> <270950400>;
> >> dai-link-0 {
> >> sound-dai = <&aiu AIU_CPU CPU_I2S_FIFO>;
> >> };
> >>
> >> dai-link-1 {
> >> sound-dai = <&aiu AIU_CPU CPU_SPDIF_FIFO>;
> >> };
> >>
> >> dai-link-2 {
> >> sound-dai = <&aiu AIU_CPU CPU_I2S_ENCODER>;
> >> dai-format = "i2s";
> >> mclk-fs = <256>;
> >>
> >> codec-0 {
> >> sound-dai = <&aiu AIU_HDMI CTRL_I2S>;
> >> };
> >> };
> >>
> >> dai-link-3 {
> >> sound-dai = <&aiu AIU_CPU CPU_SPDIF_ENCODER>;
> >>
> >> codec-0 {
> >> sound-dai = <&aiu AIU_HDMI CTRL_PCM>;
> >> };
> >> };
> >>
> >> dai-link-4 {
> >> sound-dai = <&aiu AIU_HDMI CTRL_OUT>;
> >>
> >> codec-0 {
> >> sound-dai = <&hdmi_tx>;
> >> };
> >> };
> >> };
> >> The driver for &hdmi_tx defines:
> >> struct hdmi_codec_pdata pdata = {
> >> .ops = &txc_48352_hdmi_codec_ops,
> >> .i2s = 1,
> >> .spdif = 1,
> >> .max_i2s_channels = 8,
> >> .data = priv,
> >> };
> >> In hdmi_codec_ops.hw_params I always get fmt->fmt HDMI_I2S unless I
> >> remove all I2S references from the .dts snipped above (only then
> >> HDMI_SPDIF is detected). Based on the selection of the "HDMI Source"
> >> enum in aiu-codec-ctrl I was expecting the format to update as well.
> >> That unfortunately doesn't happen and I don't know how that can be
> >> achieved.
> >>
> >
> > Hm, that sounds weird. I hope you get the answers you are looking for.
> >
> > Thanks,
> > Geraldo Nascimento
> >
> >>
> >> Best regards,
> >> Martin
> >>
> >>
> >> Martin Blumenstingl (1):
> >> ASoC: meson: aiu: Fix HDMI codec control selection
> >>
> >> sound/soc/meson/aiu-codec-ctrl.c | 108 ++++++++++++++++++++++--------
> >> sound/soc/meson/aiu-encoder-i2s.c | 6 --
> >> 2 files changed, 80 insertions(+), 34 deletions(-)
> >>
> >> --
> >> 2.33.0
> >>
> >
>
On Sun, Oct 03, 2021 at 05:57:34PM +0200, Jerome Brunet wrote:
>
> On Sun 03 Oct 2021 at 01:43, Martin Blumenstingl <[email protected]> wrote:
>
> > Hello Jerome et al.,
> >
> > on Amlogic Meson SoCs which use the AIU audio controller we have the
> > so-called "machine gun noise" (MGN) issue, reproducible for example in
> > Kodi. So far nobody was able to identify the cause of this issue. My
> > idea was to at least narrow down the issue so we can categorize it. For
> > that I wanted to get the SPDIF output from AIU to the HDMI controller
> > working.
>
> Ok, I'll clarify (again) about this issue.
>
> This issue is on AIU only (not the AXG series), on the i2s encoder
> regardless of the output (HDMI, internal and external codec)
>
> This encoder support either 2ch or 8ch mode, nothing else.
> As long as only 2ch was supported, there was no problem.
> In order to switch between 2ch and 8ch mode. AIU_RST_SOFT_I2S_FAST and
> AIU_I2S_SYNC shall be "poked" (check iu-fifo-i2s.c and
> aiu-encoder-i2c.c). Without this, the switch is not done.
>
> What these do, how and when it should be done is unclear. I tried
> several combination. At the time I found something stable-ish but it was
> not perfect.
>
> At the time, It was completely new driver. Even if was not rock solid,
> it was still progress and I opted to upstream it with an imperfect 8ch
> support so people could help debug it. This was mentioned in the
> original submission.
>
> The other solution is to restrict to 2ch mode and remove
> AIU_RST_SOFT_I2S_FAST and AIU_I2S_SYNC pokes. There will be no noise
> anymore.
>
Hi, Jerome,
Thanks for the extensive clarification. I have compiled 5.15-rc3
commenting out the pokes for 8ch as you have suggested, both for
AIU_RST_SOFT_I2S_FAST and AIU_I2S_SYNC in both aiu-fifo-i2s.c
and aiu-encoder-i2c.c.
It works, the bug is gone just like you said it would be.
My suggestion is that, for the time being, until someone can solve
this properly, you restrict enablement of 8ch configuration under
an EXPERIMENTAL kernel option, with a big fat warning in the help
description, saying it will break 2ch.
Last but no least, thanks for helping Martin out.
Thank you,
Geraldo Nascimento
On Sun 03 Oct 2021 at 21:17, Martin Blumenstingl <[email protected]> wrote:
> Hi Jerome,
>
> first of all: thanks for taking the time to go through all of this!
>
> On Sun, Oct 3, 2021 at 6:34 PM Jerome Brunet <[email protected]> wrote:
>>
>>
>> On Sun 03 Oct 2021 at 01:43, Martin Blumenstingl <[email protected]> wrote:
>>
>> > Hello Jerome et al.,
>> >
>> > on Amlogic Meson SoCs which use the AIU audio controller we have the
>> > so-called "machine gun noise" (MGN) issue, reproducible for example in
>> > Kodi. So far nobody was able to identify the cause of this issue. My
>> > idea was to at least narrow down the issue so we can categorize it. For
>> > that I wanted to get the SPDIF output from AIU to the HDMI controller
>> > working.
>>
>> Ok, I'll clarify (again) about this issue.
>>
>> This issue is on AIU only (not the AXG series), on the i2s encoder
>> regardless of the output (HDMI, internal and external codec)
> Thanks for repeating this. It was clear to me that only AIU is
> affected but I didn't know that it only affects I2S
>
>> This encoder support either 2ch or 8ch mode, nothing else.
>> As long as only 2ch was supported, there was no problem.
>> In order to switch between 2ch and 8ch mode. AIU_RST_SOFT_I2S_FAST and
>> AIU_I2S_SYNC shall be "poked" (check iu-fifo-i2s.c and
>> aiu-encoder-i2c.c). Without this, the switch is not done.
>>
>> What these do, how and when it should be done is unclear. I tried
>> several combination. At the time I found something stable-ish but it was
>> not perfect.
> old 32-bit u-boot sources from the Endless Mini do have some
> documentation on AIU_I2S_SYNC [0]:
> // 8'b0_000_1_1_10
> // A write to this register will cause the interface to repeat the current
> // sample. Can be used to regain synchronization.
> // A read from this register indicates that the next sample to be sent
> // out of the interface should go into the _left_ channel of the dac.
>
> There's also a note about AIU_I2S_MISC stating:
> // Bit 4 if true, force each audio data to left or right according to
> the bit attached with the audio data
> // This bit should be used with Register AIU_i2s_sync(0x511) together
>
> To be honest: to me this is not helpful since I don't understand
> how/why the left channel is of importance.
Same here, I already had this description and it is not really that helpful.
>
>> At the time, It was completely new driver. Even if was not rock solid,
>> it was still progress and I opted to upstream it with an imperfect 8ch
>> support so people could help debug it. This was mentioned in the
>> original submission.
>>
>> The other solution is to restrict to 2ch mode and remove
>> AIU_RST_SOFT_I2S_FAST and AIU_I2S_SYNC pokes. There will be no noise
>> anymore.
> I think Christian (Hewitt) agrees on this point as he mentioned that
> your earlier versions of the AIU code (before it got upstream) were
> not affected by the "machine gun noise" issue.
> Does the documentation from above give you any new ideas (assuming
> that it's correct since it's the best we have)? Should I try playing
> with AIU_RST_SOFT_I2S_FAST and AIU_I2S_SYNC to see if I can make a
> difference?
Nothing new here.
I suspect that the encoder restart out of sync with the FIFO, picking
mis-aligned samples but I have not confirmed that yet.
>
> [...]
>> Here you describe a DAI link (think of it as wires) between the SPDIF
>> encoder (output) and AIU_HDMI input PCM. This is not what the HW is and
>> it is not possible.
>>
>> Let's start from the HDMI controller.
>> The designware (on amlogic SoC) has 2 interface for audio input.
>> 1) PCM/I2S: a classic interface 2 clocks and N data line
>> 2) SPDIF: The Sony Philips 1 wire interface
> The Transwitch HDMI TX controller supports these two inputs so even
> though the IP is different the basic functionality (which we'll
> discuss below) is the same.
>
>> Whatever comes up on 1) has to be some sort of i2s signal. So SPDIF
>> won't fly there.
> I agree with this
>
>> AIU_HDMI output is Hardwired to 1). It is just just a digital mux,
>> implemented as an ASoC codec which allows to seleted one of 2 audio
>> sources:
>> A) the i2s output implemented as part of the AIU
>> B) the PCM output, part the AUDIN (yes, an output in AUDIN) block. This
>> is not implemented ATM.
> This is some interesting information, I thought that PCM was used
> because PCM audio can be transmitted over SPDIF.
>
> For A) my understanding is different though:
> - for AIU_HDMI_CLK_DATA_CTRL[5:4] (HDMI_DATA_SEL) your description
> matches my understanding. For me it makes sense that SPDIF data cannot
> be selected with this register since it's a one-wire protocol (and
> doesn't have separate data/clock lines). Value 0x2 selects the I2S
> data interface
> - for AIU_HDMI_CLK_DATA_CTRL[1:0] (HDMI_DATA_CLK_SEL) however I have a
> different understanding: 0x0 disables the clock signal to the HDMI TX
> controller, 0x1 selects the PCM clock (which now I have learned is
> related to the AUDIN block) and 0x2 selects the "AIU clock" (see
> below)
> - my understanding is that "AIU clock" comes from AIU_CLK_CTRL_MORE[6]
> (HDMITX_SEL_AOCLKX2) where 0x0 selects "cts_i958 as AIU clk to
> hdmi_tx_audio_master_clk" (SPDIF clock) and 0x1 selects
> "cts_aoclkx2_int as AIU clk to hdmi_tx_audio_master_clk" (I2S clock)
Not quite sure it works that way.
Each FIFO (i2s and spdif) is mainly target at the related encoder BUT
there is way to feed to both encoder from a single FIFO. In such case,
the proper must selected for the encoder and the downstream element to
keep things in sync
This at least with i2s FIFO feeding both the i2s and spdif encoders.
I suspect the same is possible with the spdif FIFO. Maybe the bit is
needed to also switch HDMI part to the spdif clock in this case.
This bit could also be a remain of an older design, not really connected
to anything meaningful. It would not be the first time.
The AIU looks like an IP that has evolved a lot over the years, not always
in a coordinated fashion. Some scenario are well supported and easy,
others seem to require a magic spell.
Last (but not least), in AML vendor kernel, the only time this bit poked
is around 8ch support (1 for 8ch, 0 otherwise) ... I have no idea why.
>
> So to me this means that there's three different muxes:
> - data mux to choose between 0x0 (all zeros), 0x1 (PCM) and 0x2 (I2S)
> - clock mux to choose between 0x0 (disabled), 0x1 (PCM) and 0x2
> (hdmi_tx_audio_master_clk)
> - hdmi_tx_audio_master_clk clock mux to choose between 0x0 (cts_i958)
> and 0x1 (cts_aoclkx2_int)
>
> Based on that I think that it's not possible to have AIU output the
> I2S and SPDIF signals at the same time to the HDMI TX controller and
> then letting the HDMI TX controller choose which one to use.
i2s and spdif can be 2 independent paths (with different clock sources).
I have already made them work concurrently. I believe (but have no
proof) that the same should be possible with the HDMI.
Until the hdmi controller has support for both inputs, we don't need to
worry about it.
> Based on whichever signal (I2S or SPDIF) we want to use for HDMI we
> need to configure AIU accordingly (data and clock).
TBH, I'm not really sure what this bit does. From the description, it
seems it allows to select an HDMI clock (not idea which one actually)
for either the i2s or the spdif path.
Nothing says it is related to the HDMI_CTRL (codec), maybe it is not.
Maybe (MAYBE) it defines what will become "the AIU clk" in
AIU_HDMI_CLK_DATA_CTRL[1:0]. Until we can feed the i2s encoder from the
spdif FIFO, I'm not sure we need to play with it.
>
> [...]
>> It is not meant to. The dai_link and the endpoint are i2s.
>> Your HDMI controller should have 2 inputs and should have a way to
>> select one or the other. The format at each of the (internal) input does
>> not change
> ah, that makes sense.
> Let's say AIU has some internal muxing for the HDMI output then AIU
> would have two outputs as well (let's call them HDMI_OUT_I2S and
> HDMI_OUT_SPDIF).
I don't think there is another output the HDMI_CTRL mux block. If SPDIF to
HDMITX is supported, I suspect the output of the spdif encoder is
hardwired to the spdif input of the HDMI controller and it is up to the
HDMI controller to take it or not.
That being said, I don't see any hard evidence that AML used SPDIF to
HDMI in their tree, so maybe it does not work at all.
> Then I'd wire up each of these to their corresponding inputs on the
> HDMI TX controller.
>
>
> Best regards,
> Martin
On Sun, Oct 03, 2021 at 11:00:15AM +0400, Christian Hewitt wrote:
> > On 3 Oct 2021, at 9:13 am, Geraldo Nascimento <[email protected]> wrote:
> > It doesn't sound like a machine gun noise to me :)
> > More like brown noise, but anyway what I'd like to note to
> > everyone involved is that this one is a major dealbreaker.
> I’m the origin of most reports that Google will eventually discover. I
> was told elsewhere that this is known as MGN so I’ve described it as
> such, but that could be wrong. For clarity this is the noise
> (triggered, and with AVR volume deliberately right down) followed by
> some navigation in the Kodi GUI to change output:
> https://chewitt.libreelec.tv/testing/wp2_audio_noise.mov
Right, I would just call that static.
Please fix your mail client to word wrap within paragraphs at something
substantially less than 80 columns. Doing this makes your messages much
easier to read and reply to.
On Sun, Oct 03, 2021 at 09:17:39PM +0200, Martin Blumenstingl wrote:
> old 32-bit u-boot sources from the Endless Mini do have some
> documentation on AIU_I2S_SYNC [0]:
> // 8'b0_000_1_1_10
> // A write to this register will cause the interface to repeat the current
> // sample. Can be used to regain synchronization.
> // A read from this register indicates that the next sample to be sent
> // out of the interface should go into the _left_ channel of the dac.
> There's also a note about AIU_I2S_MISC stating:
> // Bit 4 if true, force each audio data to left or right according to
> the bit attached with the audio data
> // This bit should be used with Register AIU_i2s_sync(0x511) together
> To be honest: to me this is not helpful since I don't understand
> how/why the left channel is of importance.
The left channel is important because for most (I think all?) audio
formats the first channel sent after each frame sync is the left
channel, if you're trying to resync it's useful to know when a left
frame is going to be sent.
>
> > At the time, It was completely new driver. Even if was not rock solid,
> > it was still progress and I opted to upstream it with an imperfect 8ch
> > support so people could help debug it. This was mentioned in the
> > original submission.
> >
> > The other solution is to restrict to 2ch mode and remove
> > AIU_RST_SOFT_I2S_FAST and AIU_I2S_SYNC pokes. There will be no noise
> > anymore.
> I think Christian (Hewitt) agrees on this point as he mentioned that
> your earlier versions of the AIU code (before it got upstream) were
> not affected by the "machine gun noise" issue.
> Does the documentation from above give you any new ideas (assuming
> that it's correct since it's the best we have)? Should I try playing
> with AIU_RST_SOFT_I2S_FAST and AIU_I2S_SYNC to see if I can make a
> difference?
>
> [...]
> > Here you describe a DAI link (think of it as wires) between the SPDIF
> > encoder (output) and AIU_HDMI input PCM. This is not what the HW is and
> > it is not possible.
> >
> > Let's start from the HDMI controller.
> > The designware (on amlogic SoC) has 2 interface for audio input.
> > 1) PCM/I2S: a classic interface 2 clocks and N data line
> > 2) SPDIF: The Sony Philips 1 wire interface
> The Transwitch HDMI TX controller supports these two inputs so even
> though the IP is different the basic functionality (which we'll
> discuss below) is the same.
>
> > Whatever comes up on 1) has to be some sort of i2s signal. So SPDIF
> > won't fly there.
> I agree with this
>
> > AIU_HDMI output is Hardwired to 1). It is just just a digital mux,
> > implemented as an ASoC codec which allows to seleted one of 2 audio
> > sources:
> > A) the i2s output implemented as part of the AIU
> > B) the PCM output, part the AUDIN (yes, an output in AUDIN) block. This
> > is not implemented ATM.
> This is some interesting information, I thought that PCM was used
> because PCM audio can be transmitted over SPDIF.
>
> For A) my understanding is different though:
> - for AIU_HDMI_CLK_DATA_CTRL[5:4] (HDMI_DATA_SEL) your description
> matches my understanding. For me it makes sense that SPDIF data cannot
> be selected with this register since it's a one-wire protocol (and
> doesn't have separate data/clock lines). Value 0x2 selects the I2S
> data interface
> - for AIU_HDMI_CLK_DATA_CTRL[1:0] (HDMI_DATA_CLK_SEL) however I have a
> different understanding: 0x0 disables the clock signal to the HDMI TX
> controller, 0x1 selects the PCM clock (which now I have learned is
> related to the AUDIN block) and 0x2 selects the "AIU clock" (see
> below)
> - my understanding is that "AIU clock" comes from AIU_CLK_CTRL_MORE[6]
> (HDMITX_SEL_AOCLKX2) where 0x0 selects "cts_i958 as AIU clk to
> hdmi_tx_audio_master_clk" (SPDIF clock) and 0x1 selects
> "cts_aoclkx2_int as AIU clk to hdmi_tx_audio_master_clk" (I2S clock)
>
> So to me this means that there's three different muxes:
> - data mux to choose between 0x0 (all zeros), 0x1 (PCM) and 0x2 (I2S)
> - clock mux to choose between 0x0 (disabled), 0x1 (PCM) and 0x2
> (hdmi_tx_audio_master_clk)
> - hdmi_tx_audio_master_clk clock mux to choose between 0x0 (cts_i958)
> and 0x1 (cts_aoclkx2_int)
>
> Based on that I think that it's not possible to have AIU output the
> I2S and SPDIF signals at the same time to the HDMI TX controller and
> then letting the HDMI TX controller choose which one to use.
> Based on whichever signal (I2S or SPDIF) we want to use for HDMI we
> need to configure AIU accordingly (data and clock).
>
> [...]
> > It is not meant to. The dai_link and the endpoint are i2s.
> > Your HDMI controller should have 2 inputs and should have a way to
> > select one or the other. The format at each of the (internal) input does
> > not change
> ah, that makes sense.
> Let's say AIU has some internal muxing for the HDMI output then AIU
> would have two outputs as well (let's call them HDMI_OUT_I2S and
> HDMI_OUT_SPDIF).
> Then I'd wire up each of these to their corresponding inputs on the
> HDMI TX controller.
>
>
> Best regards,
> Martin
Hi Jerome,
On Mon, Oct 4, 2021 at 11:02 AM Jerome Brunet <[email protected]> wrote:
[...]
> > old 32-bit u-boot sources from the Endless Mini do have some
> > documentation on AIU_I2S_SYNC [0]:
> > // 8'b0_000_1_1_10
> > // A write to this register will cause the interface to repeat the current
> > // sample. Can be used to regain synchronization.
> > // A read from this register indicates that the next sample to be sent
> > // out of the interface should go into the _left_ channel of the dac.
> >
> > There's also a note about AIU_I2S_MISC stating:
> > // Bit 4 if true, force each audio data to left or right according to
> > the bit attached with the audio data
> > // This bit should be used with Register AIU_i2s_sync(0x511) together
> >
> > To be honest: to me this is not helpful since I don't understand
> > how/why the left channel is of importance.
>
> Same here, I already had this description and it is not really that helpful.
OK, so I guess this leaves experimenting (which requires luck)... :-(
[...]
> This bit could also be a remain of an older design, not really connected
> to anything meaningful. It would not be the first time.
>
> The AIU looks like an IP that has evolved a lot over the years, not always
> in a coordinated fashion. Some scenario are well supported and easy,
> others seem to require a magic spell.
>
> Last (but not least), in AML vendor kernel, the only time this bit poked
> is around 8ch support (1 for 8ch, 0 otherwise) ... I have no idea why.
The 32-bit SoCs use SPDIF to feed 2-channel audio to the HDMI TX
controller and I2S to feed 8-channel audio to the HDMI TX controller.
It seems that Amlogic stopped this for (at least some) 64-bit SoCs.
My testing results indicate that AIU_CLK_CTRL_MORE[6] is still relevant.
I can do another round of testing with various combinations of
AIU_CLK_CTRL_MORE[6] and AIU_HDMI_CLK_DATA_CTRL register values.
If you want me to test any specific combinations then please let me know.
> > So to me this means that there's three different muxes:
> > - data mux to choose between 0x0 (all zeros), 0x1 (PCM) and 0x2 (I2S)
> > - clock mux to choose between 0x0 (disabled), 0x1 (PCM) and 0x2
> > (hdmi_tx_audio_master_clk)
> > - hdmi_tx_audio_master_clk clock mux to choose between 0x0 (cts_i958)
> > and 0x1 (cts_aoclkx2_int)
> >
> > Based on that I think that it's not possible to have AIU output the
> > I2S and SPDIF signals at the same time to the HDMI TX controller and
> > then letting the HDMI TX controller choose which one to use.
>
> i2s and spdif can be 2 independent paths (with different clock sources).
> I have already made them work concurrently. I believe (but have no
> proof) that the same should be possible with the HDMI.
>
> Until the hdmi controller has support for both inputs, we don't need to
> worry about it.
My work-in-progress HDMI TX controller driver for the 32-bit SoCs [0]
does support both (I2S and SPDIF) inputs. That's how I am able to test
all of this.
I wasn't aware of the fact that it's known that the "machine gun
noise" (MGN) issue is related to I2S only. Since I couldn't find a
workaround or fix for the MGN issue I wanted to narrow it down by
adding SPDIF support to the HDMI TX path.
> > Based on whichever signal (I2S or SPDIF) we want to use for HDMI we
> > need to configure AIU accordingly (data and clock).
>
> TBH, I'm not really sure what this bit does. From the description, it
> seems it allows to select an HDMI clock (not idea which one actually)
> for either the i2s or the spdif path.
>
> Nothing says it is related to the HDMI_CTRL (codec), maybe it is not.
> Maybe (MAYBE) it defines what will become "the AIU clk" in
> AIU_HDMI_CLK_DATA_CTRL[1:0].
This is what I am thinking
> Until we can feed the i2s encoder from the spdif FIFO, I'm not sure we
> need to play with it.
I think it's also relevant when we feed HDMI TX from the SPDIF (FIFO +
encoder) which is what I am attempting here.
[...]
> >> It is not meant to. The dai_link and the endpoint are i2s.
> >> Your HDMI controller should have 2 inputs and should have a way to
> >> select one or the other. The format at each of the (internal) input does
> >> not change
> > ah, that makes sense.
> > Let's say AIU has some internal muxing for the HDMI output then AIU
> > would have two outputs as well (let's call them HDMI_OUT_I2S and
> > HDMI_OUT_SPDIF).
>
> I don't think there is another output the HDMI_CTRL mux block. If SPDIF to
> HDMITX is supported, I suspect the output of the spdif encoder is
> hardwired to the spdif input of the HDMI controller and it is up to the
> HDMI controller to take it or not.
I think you are correct with this.
My theory is that AIU_CLK_CTRL_MORE[6] and AIU_HDMI_CLK_DATA_CTRL will
enable one of the outputs (I2S or SPIDF) from AIU to the HDMI TX
controller, meaning that only one signal can be active at a time. The
HDMI TX controller then needs to "select" the correct input based on
the selected output from AIU.
> That being said, I don't see any hard evidence that AML used SPDIF to
> HDMI in their tree, so maybe it does not work at all.
For the 32-bit SoCs the easiest to understand code-snippet is the HDMI
TX audio input configuration [1]
Which basically translates to:
if (more_than_two_channel_audio) {
enable_audio_i2s();
} else {
enable_audio_spdif();
}
The code in the AIU driver just assumes that whatever it's configuring
is what the HDMI TX driver expects - without a clear link between
these two.
Best regards,
Martin
[0] https://github.com/xdarklight/linux/blob/meson-mx-integration-5.15-20211003/drivers/gpu/drm/bridge/transwitch/txccq-txc-48352.c
[1] https://github.com/endlessm/linux-meson/blob/d6e13c220931110fe676ede6da69fc61a7cb04b6/arch/arm/mach-meson8/hdmi_tx_hw/hdmi_tx_hw.c#L3035:L3038
Hi Jerome,
On Mon, Oct 4, 2021 at 11:17 PM Martin Blumenstingl
<[email protected]> wrote:
[...]
> > This bit could also be a remain of an older design, not really connected
> > to anything meaningful. It would not be the first time.
> >
> > The AIU looks like an IP that has evolved a lot over the years, not always
> > in a coordinated fashion. Some scenario are well supported and easy,
> > others seem to require a magic spell.
> >
> > Last (but not least), in AML vendor kernel, the only time this bit poked
> > is around 8ch support (1 for 8ch, 0 otherwise) ... I have no idea why.
> The 32-bit SoCs use SPDIF to feed 2-channel audio to the HDMI TX
> controller and I2S to feed 8-channel audio to the HDMI TX controller.
> It seems that Amlogic stopped this for (at least some) 64-bit SoCs.
>
> My testing results indicate that AIU_CLK_CTRL_MORE[6] is still relevant.
> I can do another round of testing with various combinations of
> AIU_CLK_CTRL_MORE[6] and AIU_HDMI_CLK_DATA_CTRL register values.
> If you want me to test any specific combinations then please let me know.
I have tested various combinations, see the attached result file
(which can be viewed with "column -t /path/to/results.txt").
The short summary is that...
...I2S output requires:
AIU_HDMI_CLK_DATA_CTRL[1:0] = 0x2
AIU_HDMI_CLK_DATA_CTRL[5:4] = 0x2
AIU_CLK_CTRL_MORE[6] = 0x1
...SPDIF output requires:
AIU_HDMI_CLK_DATA_CTRL[1:0] = 0x2
AIU_HDMI_CLK_DATA_CTRL[5:4] = (any)
AIU_CLK_CTRL_MORE[6] = 0x1
My test consisted of running speaker-test -c2 and playing an mp3 with
ffplay on an Odroid-C1.
In other words: this confirms what we have suspected before.
What is your suggestion on how to model these muxes in the driver?
In the meantime I finally understood what #sound-dai-cells = <1>; does
thanks to your previous hints.
With that I can wire up the I2S and SPDIF inputs to the HDMI TX
controller's "HDMI codec".
Many thanks again for this hint!
Best regards,
Martin