On 28/3/20 10:59 pm, Matthias Reichl wrote:
On Sat, Mar 28, 2020 at 08:50:52AM +1100, Matt Flax wrote:On 28/3/20 12:23 am, Matthias Reichl wrote:On Fri, Mar 27, 2020 at 11:30:50AM +1100, Matt Flax wrote:On 27/3/20 10:56 am, Matt Flax wrote:Should this patch be handled through the ALSA team the R. Pi team or the BCM team ?Resending again with reduced recipients.thanks Matt On 24/3/20 8:08 pm, Matt Flax wrote:Substream sample alignment was being set in hwparams for both substreams at the same time. This became a problem when the Audio Injector isolated sound card needed to offset sample alignment for high sample rates. The latency difference between playback and capture occurs because of the digital isolation chip propagation time, particularly when the codec is master and the DAC return is twice delayed. This patch sets sample alignment registers based on the substream direction in hwparams. This gives the machine driver more control over sample alignment in the bcm2835 i2s driver. Signed-off-by: Matt Flax <flatmax@xxxxxxxxxxx> --- sound/soc/bcm/bcm2835-i2s.c | 36 +++++++++++++++++++----------------- 1 file changed, 19 insertions(+), 17 deletions(-) diff --git a/sound/soc/bcm/bcm2835-i2s.c b/sound/soc/bcm/bcm2835-i2s.c index e6a12e271b07..9db542699a13 100644 --- a/sound/soc/bcm/bcm2835-i2s.c +++ b/sound/soc/bcm/bcm2835-i2s.c @@ -493,11 +493,6 @@ static int bcm2835_i2s_hw_params(struct snd_pcm_substream *substream, return -EINVAL; } - bcm2835_i2s_calc_channel_pos(&rx_ch1_pos, &rx_ch2_pos, - rx_mask, slot_width, data_delay, odd_slot_offset); - bcm2835_i2s_calc_channel_pos(&tx_ch1_pos, &tx_ch2_pos, - tx_mask, slot_width, data_delay, odd_slot_offset); - /* * Transmitting data immediately after frame start, eg * in left-justified or DSP mode A, only works stable @@ -508,19 +503,26 @@ static int bcm2835_i2s_hw_params(struct snd_pcm_substream *substream, "Unstable slave config detected, L/R may be swapped"); /* - * Set format for both streams. - * We cannot set another frame length - * (and therefore word length) anyway, - * so the format will be the same. + * Set format on a per stream basis. + * The alignment format can be different depending on direction. */ - regmap_write(dev->i2s_regmap, BCM2835_I2S_RXC_A_REG, - format - | BCM2835_I2S_CH1_POS(rx_ch1_pos) - | BCM2835_I2S_CH2_POS(rx_ch2_pos)); - regmap_write(dev->i2s_regmap, BCM2835_I2S_TXC_A_REG, - format - | BCM2835_I2S_CH1_POS(tx_ch1_pos) - | BCM2835_I2S_CH2_POS(tx_ch2_pos)); + if (substream->stream == SNDRV_PCM_STREAM_CAPTURE) { + bcm2835_i2s_calc_channel_pos(&rx_ch1_pos, &rx_ch2_pos, + rx_mask, slot_width, data_delay, odd_slot_offset); + regmap_write(dev->i2s_regmap, BCM2835_I2S_RXC_A_REG, + format + | BCM2835_I2S_CH1_POS(rx_ch1_pos) + | BCM2835_I2S_CH2_POS(rx_ch2_pos)); + } + + if (substream->stream == SNDRV_PCM_STREAM_PLAYBACK) { + bcm2835_i2s_calc_channel_pos(&tx_ch1_pos, &tx_ch2_pos, + tx_mask, slot_width, data_delay, odd_slot_offset); + regmap_write(dev->i2s_regmap, BCM2835_I2S_TXC_A_REG, + format + | BCM2835_I2S_CH1_POS(tx_ch1_pos) + | BCM2835_I2S_CH2_POS(tx_ch2_pos)); + } /* Setup the I2S mode */This will break duplex operation if a second stream is opened when a stream is already running as the channel position registers for the second stream haven't been set up. Note this code at the very beginning of hw_params: /* * If a stream is already enabled, * the registers are already set properly. */ regmap_read(dev->i2s_regmap, BCM2835_I2S_CS_A_REG, &csreg); if (csreg & (BCM2835_I2S_TXON | BCM2835_I2S_RXON)) return 0; The reason for this check is that we can't change bcm2835 I2S registers after I2S RX/TX has been enabled - the reason why is explained in the datasheet:The PCM interface runs asynchronously at the PCM_CLK rate and automatically transfers transmit and receive data across to the internal APB clock domain. The control registers are NOT synchronised and should be programmed before the device is enabled and should NOT be changed whilst the interface is running. Only the EN, RXON and TXON bits of the PCMCS register are synchronised across the PCM - APB clock domain and are allowed to be changed whilst the interface is running.Therefore we need to set up channel masks for both RX and TX before any stream is started.I see what you mean. We can't change the registers once the system has started half duplex and then subsequently changed to full duplex. There are cases however where playback and capture need to be set independently. In these cases the machine driver requires different format settings based on the stream direction. What if we make a check for whether the system is already running and in that case return an error - forcing the user to use specify the same dai_fmt which is already in use before continuing ?I'm not sure if I can follow you. dai_fmt, as the name implies, sets the format of the DAI - you can't have different DAI formats for playback/capture active at the same time. This sounds a bit like you may be trying to work around some hardware or codec configuration issue by creative use of the API.
It is the nature of digital isolation chips. They have very significant latencies. If the codec is master, then the round trip latency which effects the DAC's I2S line is even more significant.
Would there be a better way to achieve different hwparams based on stream direction ?If you really need different DAI formats for playback/capture it's best to disallow full-duplex mode and set the DAI format based on stream direction in the machine driver.
If we were to disallow full duplex mode in the machine driver, how would it look in the machine driver ? Would the user still be able to do full duplex capture/playback ?
We could take that approach if the user could still perform full duplex operations. However it doesn't represent the hardware. The hardware is in full duplex mode and requires different word offsets for ADC and DAC I2S lines.
It seems that the ALSA core system can't handle this because the set_fmt functions don't specify the stream direction. This happens both for the CPU and the Codec drivers.
Matt
