Re: [PATCH 02/21] ASoC: fsl: use .auto_selectable_formats

[Hsieh Hung-En <hungen3108@gmail.com>]

Hi,

I tested this on an i.MX95 board with an SAI3 <-> TLV320AIC3101 link,
on top of broonie/sound.git for-7.4 (f42fbd560cd8), which already has
cb125dde4b5d ("ASoC: codecs: tlv320*: use .auto_selectable_formats"),
so both ends of the link advertise their formats.

Two device trees, differing in one property:

  - control:       simple-audio-card,format = "i2s";
  - auto-selected: that property deleted, so the FORMAT bits are left
                   for soc_dai_auto_select_format() to fill in.

bitclock-master and frame-master are present in both, so the
clock-provider relationship stays fixed while format/polarity selection
is exercised.

The card registers and plays in both cases, and the hardware ends up
configured identically (read during a stream):

              control       auto-selected
  SAI TCR4    0x10010F3B    0x10010F3B    FSE|FSP -> I2S, FSD_MSTR,
  SAI RCR4    0x10010F3B    0x10010F3B    SYWD=0x0f, FRSZ=1
  SAI TCR2    0x43000000    0x43000000
  SAI RCR2    0x07000003    0x07000003
  AIC3101 r8  0x00          0x00          codec is clock consumer
  AIC3101 r9  0x00          0x00          bits 7:6 = 00 -> I2S

So auto-selection lands on I2S + NB_NF, matching the effective DAI
format of the control case.

That the card came up at all is informative here, because a wrong pick
is not silent on this codec: aic3x_set_dai_fmt() validates format and
inversion together, and the codec's low-priority entry is
DSP_A|DSP_B|IB_NF while fsl_sai advertises DSP_A, DSP_B and all four
inversion combinations. I checked this with a third device tree using

    simple-audio-card,format = "dsp_a";

which gives DSP_A + NB_NF, a combination the codec refuses:

    tlv320aic3x 3-0018: ASoC error (-22): at snd_soc_dai_set_fmt() on tlv320aic3x-hifi
    asoc-simple-card sound-aic3101: error -EINVAL: parse error
    asoc-simple-card sound-aic3101: probe with driver asoc-simple-card failed with error -22
    No soundcards found.

In other words, an unsupported format/inversion combination fails the
machine driver's probe outright; there is no fallback to another
format.

I also ran an analogue loopback, headphone output cabled back into
LINE1, so the capture path and fsl_sai_pcm_dai_rx_ops are exercised as
well. alsabat plays 1 kHz and FFTs the capture: three runs per device
tree, all six reporting "PASS: Peak detected at target frequency" at
999.76 Hz. alsabat's overall return is non-zero (-1003, "more than one
signal") in every run because the analogue loop also shows a few
spurious peaks, but those are not consistent between runs of the same
device tree, so this is a functional round-trip check and not an
analogue quality measurement.

What this does not cover is the case you asked about, Mark: this board
has no DPCM link, so the dummy DAI is never the far end of the
negotiation here.

Tested-by: Hsieh Hung-En <hungen3108@gmail.com>