This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

Linux, sound on McASP : external MCLK, CPU generates WCLK and BCLK, slave codec(s), full duplex issues

Other Parts Discussed in Thread: OMAP-L138, TLV320AIC3256

Hi, as we will be using multiple codecs on McASP, we have put the codecs as slaves.

We are also using a 8.192MHz external master clock.

Linux is v3.3; at the moment, we have patched davinci-mcasp.c to handle this mode ("SND_SOC_DAIFMT_EM_CBS_CFS"), and refer to it for the CPU DAI configuration in the machine driver:

diff --git a/sound/soc/davinci/davinci-mcasp.c b/sound/soc/davinci/davinci-mcasp.c
index 95441bf..167d5cf 100644
--- a/sound/soc/davinci/davinci-mcasp.c
+++ b/sound/soc/davinci/davinci-mcasp.c
@@ -461,6 +496,21 @@ static int davinci_mcasp_set_dai_fmt(struct snd_soc_dai *cpu_dai,
                mcasp_clr_bits(base + DAVINCI_MCASP_PDIR_REG,
                                ACLKX | AHCLKX | AFSX | ACLKR | AHCLKR | AFSR);
                break;
+       
+       case SND_SOC_DAIFMT_EM_CBS_CFS:
+               /* external master clock, cpu generates bit and frame clocks */
+               mcasp_clr_bits(base + DAVINCI_MCASP_AHCLKXCTL_REG, AHCLKXE);
+               mcasp_set_bits(base + DAVINCI_MCASP_ACLKXCTL_REG, ACLKXE);
+               mcasp_set_bits(base + DAVINCI_MCASP_TXFMCTL_REG, AFSXE);
+
+               mcasp_clr_bits(base + DAVINCI_MCASP_AHCLKRCTL_REG, AHCLKRE);
+               mcasp_set_bits(base + DAVINCI_MCASP_ACLKRCTL_REG, ACLKRE);
+               mcasp_set_bits(base + DAVINCI_MCASP_RXFMCTL_REG, AFSRE);
+
+               mcasp_clr_bits(base + DAVINCI_MCASP_PDIR_REG, AHCLKX | AHCLKR);
+               mcasp_set_bits(base + DAVINCI_MCASP_PDIR_REG, ACLKX | ACLKR);
+               mcasp_set_bits(base + DAVINCI_MCASP_PDIR_REG, AFSX | AFSR);
+               break;
 
        default:
                return -EINVAL;



Our setup currently has an AIC3X codec, connected to the TX clocks of McASP.

The RX clocks are not used, they are left hanging, so we need to enable the TX clocks even when recording:

diff --git a/sound/soc/davinci/davinci-mcasp.c b/sound/soc/davinci/davinci-mcasp.c
index 95441bf..167d5cf 100644
--- a/sound/soc/davinci/davinci-mcasp.c
+++ b/sound/soc/davinci/davinci-mcasp.c
@@ -291,6 +292,9 @@
 
 #define DAVINCI_MCASP_NUM_SERIALIZER   16
 
+static int playing = 0;
+static int recording = 0;
 static inline void mcasp_set_bits(void __iomem *reg, u32 val)
 {
        __raw_writel(__raw_readl(reg) | val, reg);
@@ -346,6 +355,12 @@ static void mcasp_start_rx(struct davinci_audio_dev *dev)
 
        mcasp_set_ctl_reg(dev->base + DAVINCI_MCASP_GBLCTLR_REG, RXSMRST);
        mcasp_set_ctl_reg(dev->base + DAVINCI_MCASP_GBLCTLR_REG, RXFSRST);
+
+       if (!playing) {
+               pr_debug("%s: also starting tx\n", __func__);
+               mcasp_start_tx(dev);
+       }
+
 }
 
 static void mcasp_start_tx(struct davinci_audio_dev *dev)
@@ -384,23 +406,36 @@ static void davinci_mcasp_start(struct davinci_audio_dev *dev, int stream)
                        mcasp_set_bits(dev->base + DAVINCI_MCASP_WFIFOCTL,
                                                                FIFO_ENABLE);
                mcasp_start_tx(dev);
+               playing = 1;
        } else {
                if (dev->rxnumevt)      /* enable FIFO */
                        mcasp_set_bits(dev->base + DAVINCI_MCASP_RFIFOCTL,
                                                                FIFO_ENABLE);
                mcasp_start_rx(dev);
+               recording = 1;
        }
 }
 
 static void mcasp_stop_rx(struct davinci_audio_dev *dev)
 {
+       recording = 0;
+       pr_debug("%s\n", __func__);
        mcasp_set_reg(dev->base + DAVINCI_MCASP_GBLCTLR_REG, 0);
        mcasp_set_reg(dev->base + DAVINCI_MCASP_RXSTAT_REG, 0xFFFFFFFF);
+       if (!playing) {
+               pr_debug("%s: not playing, shutting down tx too\n", __func__);
+               mcasp_set_reg(dev->base + DAVINCI_MCASP_GBLCTLX_REG, 0);
+       }
 }
  
 static void mcasp_stop_tx(struct davinci_audio_dev *dev)
 {
-       mcasp_set_reg(dev->base + DAVINCI_MCASP_GBLCTLX_REG, 0);
+       playing = 0;
+       pr_debug("%s\n", __func__);
+       if (!recording) {
+               pr_debug("%s: not recording, shutting down\n", __func__);
+               mcasp_set_reg(dev->base + DAVINCI_MCASP_GBLCTLX_REG, 0);
+       }
        mcasp_set_reg(dev->base + DAVINCI_MCASP_TXSTAT_REG, 0xFFFFFFFF);
 }

@@ -641,11 +691,21 @@ static void davinci_hw_param(struct davinci_audio_dev *dev, int stream) # wrong numbers
 
        mcasp_clr_bits(dev->base + DAVINCI_MCASP_ACLKXCTL_REG, TX_ASYNC);
 
        if (stream == SNDRV_PCM_STREAM_PLAYBACK) {
-               /* bit stream is MSB first  with no delay */
+               /* bit stream is MSB first with no delay */
                /* DSP_B mode */
+               /* HACK: disable master transmit clock generation
                mcasp_set_bits(dev->base + DAVINCI_MCASP_AHCLKXCTL_REG,
                                AHCLKXE);
+               */
                mcasp_set_reg(dev->base + DAVINCI_MCASP_TXTDM_REG, mask);
                mcasp_set_bits(dev->base + DAVINCI_MCASP_TXFMT_REG, TXORD);
 @@ -661,13 +721,22 @@ static void davinci_hw_param(struct davinci_audio_dev *dev, int stream)
                /* bit stream is MSB first with no delay */
                /* DSP_B mode */
                mcasp_set_bits(dev->base + DAVINCI_MCASP_RXFMT_REG, RXORD);
+               /* HACK: disable master receive clock generation
                mcasp_set_bits(dev->base + DAVINCI_MCASP_AHCLKRCTL_REG,
                                AHCLKRE);
+               */
                mcasp_set_reg(dev->base + DAVINCI_MCASP_RXTDM_REG, mask);
 
-               if ((dev->tdm_slots >= 2) && (dev->tdm_slots <= 32))
+               if ((dev->tdm_slots >= 2) && (dev->tdm_slots <= 32)) {
                        mcasp_mod_bits(dev->base + DAVINCI_MCASP_RXFMCTL_REG,
                                        FSRMOD(dev->tdm_slots), FSRMOD(0x1FF));
+
+                       if (!playing) {
+                               /* HACK: generate frame sync for Tx too */
+                               mcasp_mod_bits(dev->base + DAVINCI_MCASP_TXFMCTL_REG,
+                                       FSXMOD(dev->tdm_slots), FSXMOD(0x1FF));
+                       }
+               }
                else
                        printk(KERN_ERR "capture tdm slot %d not supported\n",
                                dev->tdm_slots);

We only plan to support 32kHz, so currently we have left an ugly hack in the code:

diff --git a/sound/soc/davinci/davinci-mcasp.c b/sound/soc/davinci/davinci-mcasp.c
index 95441bf..167d5cf 100644
--- a/sound/soc/davinci/davinci-mcasp.c
+++ b/sound/soc/davinci/davinci-mcasp.c
@@ -641,11 +691,21 @@ static void davinci_hw_param(struct davinci_audio_dev *dev, int stream)
 
        mcasp_clr_bits(dev->base + DAVINCI_MCASP_ACLKXCTL_REG, TX_ASYNC);
 
+       /* UGLY: set clock divider, no support for frequencies != 32k
+        *
+        * We want 32kHz FCLK with 32bits (left, right) transferred for
+        * every frame, so it makes a 1.024MHz BCLK (/8).
+        */
+       mcasp_mod_bits(dev->base + DAVINCI_MCASP_ACLKXCTL_REG, ACLKXDIV(8-1),
+        ACLKXDIV_MASK);
+
        if (stream == SNDRV_PCM_STREAM_PLAYBACK) {

It almost works, but we have a nasty problem:

  • When performing a capture alone, the sound is recorded as expected
  • When performing a playback alone, no problem
  • When starting a playback then a capture, no problem
  • When starting a capture then a playback, the playback will hang, the stream won't accept more data; no perturbation on the recording side

Currently I am reviewing the McASP code with regards to SPRUH77A (OMAP-L138 Technical Reference Manual), but some help would be appreciated.

Regards,

  • I also posted this issue on alsa-devel: http://mailman.alsa-project.org/pipermail/alsa-devel/2012-May/051613.html

  • We are using the (Arago) linux kernel 2.6.37 (also labelled v2.6.37_DAVINCIPSP_03.21.00.04). There are several TLV320AIC3256 codec connected to an OMAP-L138 SoM board via I2C and McASP. The board is based on the DA850 EVM.

    To support the codec, I've backported the tlv320aic32x4 driver from the 2.6.39 linux kernel.

    The I2C part is configured correctly, and I can get and set mixer components and other settings via I2C to the codec.

    However, setting up the McASP data path is a complete riddle, and I need help to get this to work.

    In the board files, the following structs "glue" the codec to the audio data transfer unit:

    static struct snd_soc_dai_link da8xx_evm_dai[] = {
            {
                    .name = "TLV320AIC3X-1",
                    .stream_name = "AIC3X-1",
                    .cpu_dai_name= "davinci-mcasp.0",
                    .codec_dai_name = "tlv320aic32x4-hifi",
                    .codec_name = "tlv320aic32x4.11-0018",
                    .platform_name = "davinci-pcm-audio",
                    .init = evm_aic3x_init_analog,
                    .ops = &evm_ops,
            },
            {
                    .name = "TLV320AIC3X-2",
                    .stream_name = "AIC3X-2",
                    .cpu_dai_name= "davinci-mcasp.0",
                    .codec_dai_name = "tlv320aic32x4-hifi",
                    .codec_name = "tlv320aic32x4.12-0018",
                    .platform_name = "davinci-pcm-audio",
                    .init = evm_aic3x_init_analog,
                    .ops = &evm_ops,
            },
    

    The string "davinci-mcasp.0" refers to the struct in sound/soc/davinci/davinci-mcasp.c:

    static struct snd_soc_dai_driver davinci_mcasp_dai[] = {
            {
                    .name           = "davinci-mcasp.0",
                    .playback       = {
                            .channels_min   = 2,
                            .channels_max   = 2,
                            .rates          = DAVINCI_MCASP_RATES,
                            .formats        = SNDRV_PCM_FMTBIT_S8 |
                                                    SNDRV_PCM_FMTBIT_S16_LE |
                                                    SNDRV_PCM_FMTBIT_S32_LE,
                    },
                    .capture        = {
                            .channels_min   = 2,
                            .channels_max   = 2,
                            .rates          = DAVINCI_MCASP_RATES,
                            .formats        = SNDRV_PCM_FMTBIT_S8 |
                                                    SNDRV_PCM_FMTBIT_S16_LE |
                                                    SNDRV_PCM_FMTBIT_S32_LE,
                    },
                    .ops            = &davinci_mcasp_dai_ops,
    
            },

    And in the board config files, there is also a struct that configures the McASP ports:

    static u8 da850_iis_serializer_direction[] = {
            INACTIVE_MODE,  INACTIVE_MODE,  INACTIVE_MODE,  INACTIVE_MODE,
            INACTIVE_MODE,  INACTIVE_MODE,  RX_MODE,        TX_MODE,
            INACTIVE_MODE,  INACTIVE_MODE,  INACTIVE_MODE,  INACTIVE_MODE,
            INACTIVE_MODE,  INACTIVE_MODE,  INACTIVE_MODE,  INACTIVE_MODE,
    };
    
    static struct snd_platform_data da850_evm_snd_data = {
            .tx_dma_offset  = 0x2000,
            .rx_dma_offset  = 0x2000,
            .op_mode        = DAVINCI_MCASP_IIS_MODE,
            .num_serializer = ARRAY_SIZE(da850_iis_serializer_direction),
            .tdm_slots      = 2,
            .serial_dir     = da850_iis_serializer_direction,
            .asp_chan_q     = EVENTQ_1,
            .version        = MCASP_VERSION_2,
            .txnumevt       = 1,
            .rxnumevt       = 1,
    };
    
    ...
    da8xx_register_mcasp(0, &da850_evm_snd_data);
    

    Now this part is a complete mystery. To have multiple CODECs, do i need to register multiple instances of the McASP driver and fill in the serializer structs for each? And if not, how does it know which input is for which codec? And what are all the fields? There's no documentation whatsoever, so the above was my best guess to get at least one codec functioning. It sort of seems to do something, but all samples are zero and I get lots of overrun errors when using arecord to capture data from it.