[keyonjie]

ssp: refine the RX FIFO flushing logic

The SSSR_BSY is not necessary for flushing, we need to read out entries
every time the FIFO is not empty. This flushing may need take place
several rounds, and we need to wait 1 sample time between 2 flushing
rounds.

Fixes #3525

Signed-off-by: Keyon Jie <yang.jie@linux.intel.com>

[keyonjie]

SOFCI TEST

[jdrzaszcz]

Looks like this PR fix this issue: #3525 on external loopack.
I ran tests with trigger pause/stop/start on capture pipeline (100 playback iteration for audio format) on WHL RVP with external loopback and there was no swapped channels.

[keyonjie]

SOFCI TEST

[keyonjie]

Looks like this PR fix this issue: #3525 on external loopack.
I ran tests with trigger pause/stop/start on capture pipeline (100 playback iteration for audio format) on WHL RVP with external loopback and there was no swapped channels.

Thank you so much @jdrzaszcz !

Hi @lgirdwood good news that we finally got the external normal SSP capture channel swap issue fixed with this PR, please help to review, we could add similar fixes for BYT next.

[jdrzaszcz]

@keyonjie I have a TGL RVP with external loopback on SSP.
I'm running tests with pause/stop/start on capture pipeline and on this platform channels are still swapped.

This is some example log with: dai_info on "ssp_empty_rx_fifo(), before flushing, entries %d"

dma_trace_1.log

[keyonjie]

@keyonjie I have a TGL RVP with external loopback on SSP.
I'm running tests with pause/stop/start on capture pipeline and on this platform channels are still swapped.

This is some example log with: dai_info on "ssp_empty_rx_fifo(), before flushing, entries %d"

dma_trace_1.log

Interesting, so looks there is difference between WHL and TGL, let me debug it more.

[lgirdwood]

clock_ticks_per_sample(clock, rate);

[keyonjie]

clock_ticks_per_sample(clock, rate);

good, let me change.

[lgirdwood]

inline comments here explaining usage. I'm assuming we get number of ticks per sample ?

[keyonjie]

yep, let me add inline comments.

[lgirdwood]

Aren't we using rate here ?

[keyonjie]

Aren't we using rate here ?

I wanted to use it as the total timeout control, now I change my mind to use something like this:

#define  SSP_RX_FLUSH_SAMPLE_MAX 16

Will update it in the PR.

[keyonjie]

@lgirdwood FYI that this is to fix normal SSP capture channel swap issue on cAVS (< 2.5) platforms, for cAVS 2.5+, the solution is still open as there is some IP design change and known limitation there. So let's get this merged first before jumping to the TGL issues.

[keyonjie]

SOFCI TEST

[lgirdwood]

sample_rate is mandatory here. If it's not passed in then return 0

[keyonjie]

sample_rate is mandatory here. If it's not passed in then return 0

Fair, let me change.

[lgirdwood]

Btw, why are you committing this ? You are not changing anything in clk. I would have expected to invalidate this before we calculate the ticks ?

[keyonjie]

Btw, why are you committing this ? You are not changing anything in clk. I would have expected to invalidate this before we calculate the ticks ?

Indeed, platform_shared_get() should be used to invalidate the clk_info. It was copy-paste from clock_ms_to_ticks(), let me correct it.