On 09/03/2014 05:37 AM, Li.Xiubo@freescale.com wrote:
Subject: Re: [PATCHv2 1/4] ASoC: simple-card: add asoc_simple_card_fmt_master()
...
This won't work. The logic for cpu node needs to be negated for codec node.
Yes, actually it should be.
As my previous patches about this:
Since from the DAI format micro SND_SOC_DAIFMT_CBx_CFx, the 'CBx' mean Codec's bit clock is as master/slave and the 'CFx' mean Codec's frame clock is as master/slave.
So these same DAI formats should be informed to CPU and CODE DAIs at the same time. For the Codec driver will set the bit clock and frame clock as the DAI formats said, but for the CPU driver, if the the bit clock or frame clock is as Codec master, so it should be set CPU DAI device as bit clock or frame clock as slave, and vice versa.
The old code will cause confusion, and we should be clear that the letter 'C' here mean to Codec.
For the master format, no matter for CPU or CODEC, it always means Codec is master or slave for bit/frame clock, not means the local DAI device's bit/frame clock as master or slave.
So your CPU DAI device driver should negate this locally as the existed Ones do.
Yes, but there is double negation in this patch. The switch-case assignments depend on whether the bitclkmaster and framemaster DT-node pointers are compared to a cpu-dai-node or codec-dai-node. When your patch compares the codec-node, it does the decisions like it was a cpu-node, which produces inverted CBM and CFM setting.
However, Kurinori-san's patch fixes this problem because it just uses the daifmt generated by comparing to codec node for both cpu and codec nodes.
The reason why I did the comparison per node basis, was to make the code more ready for tdm setups with multiple codecs on a same wire. But writing code for something that is not really needed yet is usually a bad idea, like it was this time too.
Kurinori-san's version of the fix should be fine and it cleans up the code quite nicely.
Best regards, Jyri