Il 07/03/24 15:03, Rob Herring ha scritto:
On Tue, Mar 5, 2024 at 5:20 AM AngeloGioacchino Del Regno
<angelogioacchino.delregno@xxxxxxxxxxxxx> wrote:
Il 04/03/24 15:23, Rob Herring ha scritto:
On Tue, Feb 27, 2024 at 01:09:36PM +0100, AngeloGioacchino Del Regno wrote:
Document the dai-link subnodes and the audio-routing property, allowing
to describe machine specific audio hardware and links in device tree.
While at it, also deprecate the old properties which were previously
used with the driver's partially hardcoded configuration.
Signed-off-by: AngeloGioacchino Del Regno <angelogioacchino.delregno@xxxxxxxxxxxxx>
---
.../sound/mt8192-mt6359-rt1015-rt5682.yaml | 129 ++++++++++++++++--
1 file changed, 121 insertions(+), 8 deletions(-)
diff --git a/Documentation/devicetree/bindings/sound/mt8192-mt6359-rt1015-rt5682.yaml b/Documentation/devicetree/bindings/sound/mt8192-mt6359-rt1015-rt5682.yaml
index 7e50f5d65c8f..78e221003750 100644
--- a/Documentation/devicetree/bindings/sound/mt8192-mt6359-rt1015-rt5682.yaml
+++ b/Documentation/devicetree/bindings/sound/mt8192-mt6359-rt1015-rt5682.yaml
@@ -20,6 +20,15 @@ properties:
- mediatek,mt8192_mt6359_rt1015p_rt5682
- mediatek,mt8192_mt6359_rt1015p_rt5682s
+ audio-routing:
+ $ref: /schemas/types.yaml#/definitions/non-unique-string-array
Already defined in sound-card-common.yaml. Add a $ref.
Right. Done for v2.
+ description:
+ A list of the connections between audio components. Each entry is a
+ pair of strings, the first being the connection's sink, the second
+ being the connection's source.
+ Valid names could be the input or output widgets of audio components,
+ power supplies, MicBias of codec and the software switch.
Generally the names are defined here.
...but those drivers want to support multiple codecs and multiple boards, so
for each board we would maybe have to add (software defined) names in here
which don't always correspond to a HW pin name (but that's not really a problem).
Sure a subset of the names can't change but, on the other hand, some others
can (as in, may be added).
Hence the question:
Is it mandatory to define the names in an enum here, or can that be avoided?
If it is, I can add them no problem.
Does the OS depend on what the names are? As-in if a name was "bar"
and it changed to "baz" in either the DT or the kernel, would that
break things? If yes, then yes, you need them defined here.
Yes, I need them defined here, definitely.
+
mediatek,platform:
$ref: /schemas/types.yaml#/definitions/phandle
description: The phandle of MT8192 ASoC platform.
@@ -27,10 +36,12 @@ properties:
mediatek,hdmi-codec:
$ref: /schemas/types.yaml#/definitions/phandle
description: The phandle of HDMI codec.
+ deprecated: true
headset-codec:
type: object
additionalProperties: false
+ deprecated: true
properties:
sound-dai:
@@ -41,6 +52,7 @@ properties:
speaker-codecs:
type: object
additionalProperties: false
+ deprecated: true
properties:
sound-dai:
@@ -51,13 +63,83 @@ properties:
required:
- sound-dai
+patternProperties:
+ ".*-dai-link$":
+ type: object
+ description:
+ Container for dai-link level properties and CODEC sub-nodes.
+
+ properties:
+ link-name:
+ description: Indicates dai-link name and PCM stream name
+ items:
+ enum:
+ - I2S0
+ - I2S1
+ - I2S2
+ - I2S3
+ - I2S4
+ - I2S5
+ - I2S6
+ - I2S7
+ - I2S8
+ - I2S9
+ - TDM
+
+ codec:
+ description: Holds subnode which indicates codec dai.
+ type: object
+ additionalProperties: false
+ properties:
+ sound-dai:
+ minItems: 1
+ maxItems: 2
+ required:
+ - sound-dai
+
+ dai-format:
+ description: audio format
+ items:
+ enum:
+ - i2s
+ - right_j
+ - left_j
+ - dsp_a
+ - dsp_b
+
+ mediatek,clk-provider:
+ $ref: /schemas/types.yaml#/definitions/string
+ description: Indicates dai-link clock master.
+ items:
+ enum:
+ - cpu
+ - codec
+
+ additionalProperties: false
Move this before properties.
Done for v2.
+
+ required:
+ - link-name
+
additionalProperties: false
required:
- compatible
- mediatek,platform
- - headset-codec
- - speaker-codecs
+
+allOf:
+ # Disallow dai-link-xxx nodes if the legacy properties are specified
xxx-dai-link?
Oh! Yes, thanks for catching this.
That's what I initially wanted to do, but then I opted for xxx-dai-link and
forgot to update this comment.
Fixed for v2.
+ - if:
+ patternProperties:
+ ".*-dai-link$": false
+ then:
+ required:
+ - headset-codec
+ - speaker-codecs
+ else:
+ properties:
+ headset-codec: false
+ speaker-codecs: false
+ mediatek,hdmi-codec: false
Allowing both would preserve compatibility. That's not needed? If so,
say why in the commit msg.
I'm thinking of writing:
"Since describing machine specific audio hardware and links replaces the
now deprecated old logic doing the same in a driver hardcoded fashion,
it is not allowed to have both the old and new properties together."
What happened to that. Instead you just sent a new version with
nothing about this.
The same thing that happened to that card "model" error that shouldn't have
been there because I catched it before sending and fixed, then...
...I have ultimately sent the wrong changeset. My bad.
Anyway - since that's a bigger series, I'll wait for a few days and will
send the v3 with the names added to the audio-routing and this mentioned
in the commit description (so, that's happening next week).
...but in short - both the old and the new can do exactly the same, but
imo it doesn't make any sense to actually rely on both as:
1. It's redundant (and one set of them makes the other useless);
2. I want to avoid confusion (as the other set won't be parsed);
3. I'm trying to *enforce* consistency as MTK cards have different
bindings for .. really, no good reason;
4. I want to see custom stuff disappear completely (and/or as much as
possible anyway) and use something that is (at least somewhat) common
between all MTK and non-MTK or anyway as a start at least consistent
between MTK cards.
In theory, though, speaking of the driver side, there's nothing preventing
you from specifying both audio-routing xxx-dai-link and mediatek,hdmi-codec,
as the drivers' action will be, in short
if (new_bindings)
forget_about_old_bindings_use_the_new_ones();
else
use_old_hardcoded_stuff(); /* and be sad */
That works for newer kernels with this change, but existing kernels
will only have:
use_old_hardcoded_stuff(); /* and not know it's sad */
If you want to support a new DT and old kernel, you need to populate
both properties.
For that, I really don't want to allow both sets of properties - please, please,
tell me that I don't *have to* remove this block :-)
Ultimately it is your decision as Mediatek maintainer, not mine. My
only requirement is the commit message explain why the above
combination is not important for these platforms.
You could leave it, but keep both in the dts files for some time
period. That will cause warnings, but what's a few more. The ABI
doesn't have to be a forever thing. Things evolve and there will be
other reasons to upgrade.
Thanks for explaining.
Cheers,
Angelo
