linux-kernel - Re: [RFC PATCH 01/13] misc: atmel_ssc: add device tree DMA support

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:	Thu, 4 Jul 2013 15:44:21 +0200
From:	Richard Genoud <richard.genoud@...il.com>
To:	Bo Shen <voice.shen@...el.com>
Cc:	devicetree-discuss@...ts.ozlabs.org,
	Nicolas Ferre <nicolas.ferre@...el.com>,
	linux-kernel@...r.kernel.org, linux-arm-kernel@...ts.infradead.org,
	Uwe Kleine-König 
	<u.kleine-koenig@...gutronix.de>
Subject: Re: [RFC PATCH 01/13] misc: atmel_ssc: add device tree DMA support

2013/7/4 Bo Shen <voice.shen@...el.com>:
> Hi Richard,
>
>
> On 7/3/2013 23:51, Richard Genoud wrote:
>>>
>>> >but there's a violent hang (kernel stops, no trace) when I try the
>>> > record :
>>> >arecord -v -V stereo -Dplug:default -f cd -t wav -c 2 /tmp/toto.wav
>>> >last thing I see is :
>>> >dma dma0chan3: atc_control (3)
>
>
> I don't meet this issue. Playback and recording works well on my side on
> at91sam9g35ek board.
>
>
>>> >I'll try to trace that.
>>
>> I think it's DMA related.
>> the last thing done by the kernel is:
>>   i2c i2c-0: i2c_outb: 0x34 A
>>   i2c i2c-0: i2c_outb: 0x0c A
>>   i2c i2c-0: i2c_outb: 0x5a A
>> meaning: enable power on, LINE IN, ADC, OSC, on the WM8731
>> so, after that, data is comming from the codec to the SSC and then is
>> handled by the DMA.
>> there must be something nasty on the DMA bus to hang everything like
>> that...
>
>
> Will you try i2c without DMA support to test this issue?

Ok, I nailed it !

To be sure we are on the same base, here is what I have done:
onto next-20130704:
- your 5 patches:
ASoC: atmel_ssc_dai: move set dma data to startup callback
ASoC: atmel_ssc_dai: add error mask define
ASoC: atmel-pcm-dma: move prepare for dma to dai prepare
ARM: atmel-ssc: change phybase type to dma_addr_t
ASoC: atmel-pcm: use generic dmaengine framework
- my patches 4-8:
ARM: at91: DTS: sam9x5: add SSC DMA parameters
ARM: AT91: DTS: sam9x5ek: add WM8731 codec
ARM: AT91: DTS: sam9x5ek: add sound configuration
ARM: AT91: DTS: sam9x5ek: enable SSC
sound: sam9x5_wm8731: machine driver for at91sam9x5 wm8731 boards

To be sure that dma-I2c doesn't disturb something, I use i2c gpio bitbang:
diff --git a/arch/arm/boot/dts/at91sam9x5ek.dtsi
b/arch/arm/boot/dts/at91sam9x5ek.dtsi
index 6684d4b..53a991e 100644
--- a/arch/arm/boot/dts/at91sam9x5ek.dtsi
+++ b/arch/arm/boot/dts/at91sam9x5ek.dtsi
@@ -57,15 +57,6 @@
 				status = "okay";
 			};

-			i2c0: i2c@...10000 {
-				status = "okay";
-
-				wm8731: wm8731@1a {
-					compatible = "wm8731";
-					reg = <0x1a>;
-				};
-			};
-
 			pinctrl@...ff400 {
 				mmc0 {
 					pinctrl_board_mmc0: mmc0-board {
@@ -114,6 +105,15 @@
 		};
 	};

+	i2c@0 {
+		status = "okay";
+
+		wm8731: wm8731@1a {
+			compatible = "wm8731";
+			reg = <0x1a>;
+		};
+	};
+
 	sound {
 		compatible = "atmel,sam9x5-audio-wm8731";

with that configuration, it hangs when I do a:
arecord -v -V stereo -Dplug:default -f cd -t wav -c 2 /tmp/toto.wav

Now, if I remove the overrun error on ssc in rx_mask:
diff --git a/sound/soc/atmel/atmel_ssc_dai.c b/sound/soc/atmel/atmel_ssc_dai.c
index 0ecf356..c04e825 100644
--- a/sound/soc/atmel/atmel_ssc_dai.c
+++ b/sound/soc/atmel/atmel_ssc_dai.c
@@ -83,7 +83,6 @@ static struct atmel_ssc_mask ssc_rx_mask = {
 	.ssc_disable	= SSC_BIT(CR_RXDIS),
 	.ssc_endx	= SSC_BIT(SR_ENDRX),
 	.ssc_endbuf	= SSC_BIT(SR_RXBUFF),
-	.ssc_error	= SSC_BIT(SR_OVRUN),
 	.pdc_enable	= ATMEL_PDC_RXTEN,
 	.pdc_disable	= ATMEL_PDC_RXTDIS,
 };

It doesn't hang any more doing a simple record.
BUT it still hangs if we do record and play at the same time.

Removing the overrun on tx_mask prevent the hang:
diff --git a/sound/soc/atmel/atmel_ssc_dai.c b/sound/soc/atmel/atmel_ssc_dai.c
index 0ecf356..41e15c2 100644
--- a/sound/soc/atmel/atmel_ssc_dai.c
+++ b/sound/soc/atmel/atmel_ssc_dai.c
@@ -73,7 +73,6 @@ static struct atmel_ssc_mask ssc_tx_mask = {
 	.ssc_disable	= SSC_BIT(CR_TXDIS),
 	.ssc_endx	= SSC_BIT(SR_ENDTX),
 	.ssc_endbuf	= SSC_BIT(SR_TXBUFE),
-	.ssc_error	= SSC_BIT(SR_OVRUN),
 	.pdc_enable	= ATMEL_PDC_TXTEN,
 	.pdc_disable	= ATMEL_PDC_TXTDIS,
 };

i.e. when I revert "ASoC: atmel_ssc_dai: add error mask define", I
don't see any hang.

Could you test and confirm that behaviour please ?

I attached a the (simple) .config I used for the tests.

PS: I hope the patches won't be mangled by gmail...

Best regards,
Richard.

Download attachment "snd_9x5_defconfig" of type "application/octet-stream" (3939 bytes)