Hi Arnaud,
On Tue, Apr 4, 2017 at 5:59 AM, Arnaud Mouiche arnaud.mouiche@invoxia.com wrote:
So to summarize:
- Caleb and I don't see the issue without the patch, but we are working on
DSP mode @ 48K (mostly as master of the bus). But the patch break none trivial "playback only" cases. platforms: imx6 quad for Caleb, imx6sl for me. We are working without codec, checking the bit stream generated by one SSI by recording and checking the content using another SSI.
- Fabio and Max experience the issue very easily in I2S mode, acting as
slave (I guess otherwise generating precise 44.1Khz would be hard), connecting to a STGL5000 codec.
That's correct. The mode is I2S slave in our case.
When you are master of the bus, it is important to start EN before TE for the FIFO pre-fill reasons. The samples need to be ready as soon as TE starts. I also guess that ENGcm06222 doesn't affect us when the SSI is master (since the SSI is govern only by its own timing)
As slave, this is less important to start EN before TE because you have little chance to receive the SYNC trigger as soon as EN+TE starts => the DMA did get time to fill the FIFO. Yet, as slave, ENGcm06222 affect the order of channels, as experienced by Fabio.
So I switch on I2S mode for my SSI => SSI tests and, sadly, I didn't experience issues without the patch. I did the test on vanilla 4.11.0-rc5.
which branch/repository are Fabio using for his tests ?
Right now I am using 4.11-rc5 + the pinctrl patch https://patchwork.ozlabs.org/patch/745349/ .
The swap also happens with 3.14, 4.1.15 and 4.10.8.
The way clocks are configured may explain the difference: Dumping /sys/kernel/debug/clk/clk_summary and checking the differences can give some clues. In my case, I have, for the slave SSI #2, while the PCM bus is running at 48khz+I2Smode+2 channels.
pll4 0 0 7864320000 0 pll4_bypass 0 0 786432000 0 0 pll4_audio 0 0 786432000 0 0 pll4_post_div 0 0 786432000 0 0 pll4_audio_div 0 0 786432000 0 0 ssi2_sel 0 0 786432000 0 0 ssi2_pred 0 0 196608000 0 0 ssi2_podf 0 0 98304000 0 0 ssi2 0 0 98304000 0 0
In the slave mode case while the wav is playing:
ipg 5 6 66000000 0 0 usboh3 0 0 66000000 0 0 uart_ipg 1 2 66000000 0 0 ssi3_ipg 0 0 66000000 0 0 ssi2_ipg 0 0 66000000 0 0 ssi1_ipg 1 2 66000000 0 0
Strangely, the 'enable_cnt' is kept equal to zero while the SSI transmit frames correctly... Is there a patch or a branch somewhere that fix this issue ?
Also, here is a dump of SSI registers. /var/root # cat /sys/kernel/debug/regmap/202c000.ssi/registers 00: 00000000 04: 00000000 10: 0000105b 18: 009031a3 1c: 00000285 20: 00000205 24: 0004e100 28: 00040100 2c: 00880888 30: 00000000 34: 00000000 38: 00000000 48: fffffffc 4c: fffffffc 50: 00000000 54: 00000000 58: 00000000
I have the following SSI1 values (with the original 4.1-rc5 + pictrl patch):
# cat /sys/kernel/debug/regmap/2028000.ssi/registers 00: 00000200 04: 00000000 10: 0000105b 18: 009031a3 1c: 0000028d 20: 0000020d 24: 0004e100 28: 00040100 2c: 00880d88 30: 00000000 34: 00000000 38: 00000000 48: 00000000 4c: 00000000 50: 00000000 54: 00000000 58: 00000000