iMX6 ESAI TDM with underrun and overrun support

[aurelienbouin]

I am trying to use the ESAI of the iMX6...

I need to use 2x 32 slots of 8 bits at 8Khz, and I am in slave mode since i am listenning Frame Synchro and Tx Clock

I am testing the interface simply sending a word of 0xAA on slot 0 and slot 15, the others are 0x00

I am probably facing the underrun and overrun chip errata (ERR008000) on ESAI since my channels are switching randomly ...

I already tested the kernel from :

-> git://git.freescale.com/imx/linux-2.6-imx.git --- branch : imx_3.10.31_1.1.0_alpha

-> git://git.freescale.com/imx/linux-2.6-imx.git --- branch : imx_3.10.31_1.1.0_beta2

And now I am trying to use the main kernel stable git :

-> git://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git --- branch master

But I have some trouble actually with the mailine kernel :

[  745.648380] fsl-esai-dai 2024000.esai: Missing dma channel for stream: 0

[  745.655198] fsl-esai-dai 2024000.esai: ASoC: pcm constructor failed: -22

I figured out that the imx6qdl.dtsi was not giving enough information to use the ESAI so I add to my own dts :

pinctrl_esai_gip2: esaigrp-gip2 {

                fsl,pins = <

                        MX6QDL_PAD_GPIO_2__ESAI_TX_FS           0x1b030

                        MX6QDL_PAD_GPIO_3__XTALOSC_REF_CLK_24M 0x1b030

                        MX6QDL_PAD_GPIO_1__ESAI_RX_CLK       0x1b030

                        MX6QDL_PAD_GPIO_6__ESAI_TX_CLK         0x1b030

                        MX6QDL_PAD_GPIO_17__ESAI_TX0            0x1b030

                        MX6QDL_PAD_GPIO_8__ESAI_TX5_RX0       0x1b030

                        MX6QDL_PAD_GPIO_9__ESAI_RX_FS           0x1b030

                        MX6QDL_PAD_GPIO_18__ESAI_TX1 0x1b030

                        MX6QDL_PAD_GPIO_7__ESAI_TX4_RX1 0x1b030

                >;

&esai {

    dma-names = "rx", "tx";

    clocks = <&clks 228>, <&clks 229>, <&clks 118>, <&clks 228>, <&clks 156>;

    clock-names = "core", "mem", "extal", "fsys", "dma";

    compatible = "fsl,imx35-esai";

    fsl,esai-dma-events = <24 23>;

    pinctrl-names = "default";

    pinctrl-0 = <&pinctrl_esai_gip2>;

    fsl,esai-synchronous;

    fsl,fifo-depth = <64>; /* Transmit FIFO watermark => the Transmit FIFO Empty flag will set => 32 slots * 2 bus of TX the same for RX */

    status = "okay";

};

sound {

  compatible = "fsl,imx6q-gipc-generic-esai",

      "fsl,imx-audio-generic-esai";

  model = "generic-esai-audio";

  esai-controller = <&esai>;

  };

I do have in my .config :

#

# SoC Audio for Freescale CPUs

#

#

# Common SoC Audio options for Freescale CPUs:

#

# CONFIG_SND_SOC_FSL_ASRC is not set

CONFIG_SND_SOC_FSL_SAI=y

CONFIG_SND_SOC_FSL_SSI=y

CONFIG_SND_SOC_FSL_SPDIF=y

CONFIG_SND_SOC_FSL_ESAI=y

CONFIG_SND_SOC_FSL_UTILS=y

CONFIG_SND_SOC_IMX_PCM_DMA=y

CONFIG_SND_SOC_IMX_AUDMUX=y

CONFIG_SND_IMX_SOC=y

CONFIG_SND_SOC_IMX_PCM_FIQ=y

I simply make a module that connect to a dummy codec the dai interface ESAI to use it in a generic way : http://pastebin.com/CupYva7Z

thank you for your return

Aurelien BOUIN

[catalinpetrescu]

Hi Aurelien,

I assume that you have this working now. Will you please share the device tree configuration? I'm working on a similar project and I hope to start from a working configuration.

Thanks,

Catalin Petrescu.

[aurelienbouin]

Hi,

Unfortunately I have never had the chance to make it work ... not enough support of the ESAI iMX6 black box ...

Sorry

Regards,

[aurelienbouin]

Hello,

I get back to linux branch imx_3.10.31_1.1.0_beta2 at git://git.freescale.com/imx/linux-2.6-imx.git

I figured out that the TIE and TEIE bits in the TCR register are not set at any moment ...

So I add in the fsl_esai_set_dai_fmt

#define MAKE_TCR_INTERRUPTIBLE

#ifdef MAKE_TCR_INTERRUPTIBLE

  //make it interruptible

  printk(KERN_ERR"%s(%d)\n",__func__,__LINE__);

  mask |= ESAI_xCR_xEIE_MASK | ESAI_xCR_xIE_MASK;

  xcr |= ESAI_xCR_xEIE | ESAI_xCR_xIE;

  //make it interruptible

#endif /* MAKE_TCR_INTERRUPTIBLE */

I also add a dump function that show the register at  fsl_esai_hw_params : ESAI_TCR   0x00728100

I also add debug in the function fsl_esai_check_xrun that show the saisr register

When underrun occured (when I see that with an oscilloscope)

There is no interrupt generated ...

FYI here is my fsl_esai.c modified : fsl_esai.c - Pastebin.com

Thank you for your help,

Sincerely,

Aurelien BOUIN

[aurelienbouin]

For your information the problem occured only in slave mode ... (Clock and Frame Synchro is in input)

It is working correctly apparently in master mode ... but the clock and the frame synchro are not correct ... When I ask 2048000Hz I get 2207000 hz

If you have any suggestion ...

Thank you for your help

[aurelienbouin]

Why the problem is marked as solved ??? it is definitely not the case ...

[ZeeFrench]

Bonjour Aurélien,

Your problem seems to have multiple aspects, setting dma was one of it and the oscilloscope above looks like a channel swapping issue (in case of overrun) that is supposed to be fixed by an other group of patches in Yocto http://git.freescale.com/git/cgit.cgi/imx/linux-2.6-imx.git/log/sound/soc/fsl/fsl_esai.c?h=imx_3.10.... (apparently not in the mainline kernel). This is becoming difficult to follow (do you still have swapping now you are back on beta2 ?), especially if you add a new issue below that is not your use case (looking at our alsa driver documentation, 22 kHz is a standard frequency but no trace of 20 kHz).

I think we need to dig the root cause of underun (and / or overrun ?), not necessary in the esai driver but maybe in dma, since you open a new thread, i propose to follow on that one if that ofk with you ?

Best regards, Philippe.

[fabio_estevam]

Yes, Philippe's suggestion is a good one. Could you please try 3.10.53 GA kernel?

http://git.freescale.com/git/cgit.cgi/imx/linux-2.6-imx.git/log/?h=imx_3.10.53_1.1.0_ga

Regards,

Fabio Estevam

[aurelienbouin]

Hello Philippe and Fabio,

I keep getting swapping channels with the linux branch imx_3.10.31_1.1.0_beta2 at git://git.freescale.com/imx/linux-2.6-imx.git and it is the same implementation that the one you give me (http://git.freescale.com/git/cgit.cgi/imx/linux-2.6-imx.git/log/sound/soc/fsl/fsl_esai.c?h=imx_3.10....

Again in this driver the implementation is limited to 16 channels (why ?)

FYI here is my fsl_esai.c modified :[C] fsl_esai_updated.c - Pastebin.com

I don't understand 20KHz trace ?

I am using 64 slots of 8 bits word at 8000Hz => mclk=2048000Hz

Thank you for your interest in my problem, I really appreciate it

Aurelien BOUIN

[fabio_estevam]

Hi Aurelien,

3.10.53 has some additional esai patches that seem to be important for your case.

Please see this one about channel swap:

linux-2.6-imx.git - Freescale i.MX Linux Tree

I would suggest you to try 3.10.53.

Regards,

Fabio Estevam

[aurelienbouin]

Hi again,

I add the patch (http://git.freescale.com/git/cgit.cgi/imx/linux-2.6-imx.git/commit/sound/soc/fsl/fsl_esai.c?h=imx_3....) but i keep getting swapping

My new code is now http://pastebin.com/66GkXME2

Regards,

Aurelien BOUIN

[ZeeFrench]

The patch set is not only affecting esai driver, I haven't locate something (need to dig in the sdma) that could explain this phenomenum but it would still be interresting to test a complete kernel. The workaround to the hardware errata is to manage startup condition and reset esai in case of overrun but I don't see any fix in case of underrun : your message below seems to indicate you are experiencing both over and underruns ?

Cheers, Philippe.

[aurelienbouin]

Hello,

So I git clone the linux branch imx_3.10.53_1.1.0_ga on git://git.freescale.com/imx/linux-2.6-imx.git

I patched it with this : http://pastebin.com/RM8pyUSA

It makes 64 slots available and set a correct clock to pll4 to get 2048000Hz

Unfortunetaly I also get swapping without anything un debug that can help me detect it ...

Thank you again for your suggestions,

Regards

Aurelien BOUIN

[aurelienbouin]

Hi,

Like I already told, this swapping happen in slave mode ... not in master mode (apparently)

Also sometimes when I start playing on one of my 64 channels, it is played on a different slot tahn the one usually used ...

i don't know if it can give you some clue ...

Thank you again for your support

Aurélien BOUIN

[ZeeFrench]

Bonjour Aurelien,

Checking the patches, looks like the workaround is checking both ROE and TUE bits to reset ESAI. Other customers are able to use it but not a lot with the full 64 channels, what is the situation if you lower them to 16 or 32 ?

Best regards, Philippe.

[aurelienbouin]

Hello Philippe,

I am still wondering what is the purpose of HCKT in the ESAI module

I tune to get 2048000Hz like is the tx clock for 8Khz Frame Synchro

Maybe there is some recommandation somewhere to put it at a higher level ?

For memory I use the ESAI module with 64 slots (in total) of 8 bits on 2 TX/RX pins at a bandwidth of 8000Hz ... clock bit is at 2048000Hz

I use of course network mode in synchronous mode.

Externally I only use those pins :

MX6QDL_PAD_GPIO_2__ESAI_TX_FS	0x1b030
MX6QDL_PAD_GPIO_6__ESAI_TX_CLK	0x1b030

MX6QDL_PAD_GPIO_17__ESAI_TX0              0x1b030

MX6QDL_PAD_GPIO_8__ESAI_TX5_RX0	0x1b030
MX6QDL_PAD_GPIO_18__ESAI_TX1	0x1b030
MX6QDL_PAD_GPIO_7__ESAI_TX4_RX1	0x1b030

Thank you for any advice ...

Regards,

Aurelien BOUIN

[aurelienbouin]

Hello Philippe,

If I lower to 16 bits the situation stay the same ... there is swapping ...

I can't understand Why I do not get at least a DMA problem ...

Thank you for your support philippe.

Best regards,

Aurelien BOUIN

[aurelienbouin]

So I am using the ESAI with the master branch of the main git git://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git

And I can see that my channels are randomly switching ... seems like the ESAI chip errata is here ...

On the fallowing pictures you can see in blue datas, in yellow frame synchro

You can see on the first attached picture the initial situation : channel[0] et channel[16] have 0xAA, and others 0x0

And after sometimes we get that :

I don't see any debug message concerning xrun or underrrun ...

Thank you for your help,

Regards

Aurelien BOUIN

[aurelienbouin]

Hi again,

When I speak about fake codec I am actually speaking about the snd-soc-dummy that I declared on my module : http://pastebin.com/CupYva7Z :

1. static struct snd_soc_dai_link snd_soc_dapm_route[] = {
2.         {
3.                 .name = "HiFi",
4.                 .stream_name = "HiFi",
5.                 .codec_name = "snd-soc-dummy",
6.                 .codec_dai_name = "snd-soc-dummy-dai",
7.                 .ignore_pmdown_time = 1,
8.                 .ops = &imx_generic_esai_surround_ops,
9.         },
10. };

I feel like I won't be able to do the same, just with the devicetree ?!

Thank you for your reply,

Regards,

Aurelien BOUIN

[fabio_estevam]

why do you need the snd-soc-dummy?

Don't you have a real audio codec connected in your hardware?

[aurelienbouin]

Nope I don't ! I directly connect my ESAI bus to another processor ... They just exchange audio datas directly ...