topology: sof-apl-pcm512x: add option for codec master mode and 44.1kHz

[plbossart]

This changes the SSP configuration and moves the scheduling to DMA instead of TIMER. 44.1 kHz is also supported, a first in the history of SOF I guess :-)

[singalsu]

@plbossart There's no ASRC in the media pipeline so I'd assume it should be DMA like it was before the topologies were switched to timer based scheduling.

[lgirdwood]

Both should be DMA, IIRC media is not using ASRC and codec is master.

[paulstelian97]

If it's DMA you should also disable the agent as it only works correctly with timer driven pipelines. If not done already.

[plbossart]

If it's DMA you should also disable the agent as it only works correctly with timer driven pipelines. If not done already.

?? Are we saying that we need a new firmware with a different config when the topology changes from SOC master to SOC slave?
@tlauda can you please comment?

[plbossart]

If the Kconfig documentation is right, then this AGENT is really wrong:

config HAVE_AGENT
bool "Enable system agent"
default y
help
Enables system agent. It can be disabled on systems
which are still unstable and cannot assure that
system agent will always execute on time or systems
with DMA based scheduling, where asynchronous interrupts
can potentially starve the agent.

If we provide a means to change the behavior with the topology, then somehow the agent should be disabled automatically for pipelines/DAIs that rely on DMA. It doesn't seem right to me to require a new binary be generated.

[paulstelian97]

If the Kconfig documentation is right, then this AGENT is really wrong:

config HAVE_AGENT
bool "Enable system agent"
default y
help
Enables system agent. It can be disabled on systems
which are still unstable and cannot assure that
system agent will always execute on time or systems
with DMA based scheduling, where asynchronous interrupts
can potentially starve the agent.

If we provide a means to change the behavior with the topology, then somehow the agent should be disabled automatically for pipelines/DAIs that rely on DMA. It doesn't seem right to me to require a new binary be generated.

@tlauda Any comments on what should be done with this?

[tlauda]

If the Kconfig documentation is right, then this AGENT is really wrong:

config HAVE_AGENT
bool "Enable system agent"
default y
help
Enables system agent. It can be disabled on systems
which are still unstable and cannot assure that
system agent will always execute on time or systems
with DMA based scheduling, where asynchronous interrupts
can potentially starve the agent.

If we provide a means to change the behavior with the topology, then somehow the agent should be disabled automatically for pipelines/DAIs that rely on DMA. It doesn't seem right to me to require a new binary be generated.

@tlauda Any comments on what should be done with this?

We can easily dynamically turn off the agent as soon as there is any pipeline scheduled on DMA created. It will give us much more flexibility in terms of topology creation. Platforms which are not capable at all to schedule pipelines on timer should have agent disabled in kConfig (which I think is currently done).

[plbossart]

If the Kconfig documentation is right, then this AGENT is really wrong:

config HAVE_AGENT
bool "Enable system agent"
default y
help
Enables system agent. It can be disabled on systems
which are still unstable and cannot assure that
system agent will always execute on time or systems
with DMA based scheduling, where asynchronous interrupts
can potentially starve the agent.

If we provide a means to change the behavior with the topology, then somehow the agent should be disabled automatically for pipelines/DAIs that rely on DMA. It doesn't seem right to me to require a new binary be generated.

@tlauda Any comments on what should be done with this?

We can easily dynamically turn off the agent as soon as there is any pipeline scheduled on DMA created. It will give us much more flexibility in terms of topology creation. Platforms which are not capable at all to schedule pipelines on timer should have agent disabled in kConfig (which I think is currently done).

For my education, what does this 'system agent' do, and could it still be used for paths that will always be timer-based (HDA/HDMI + SoundWire + DMIC)?

[lgirdwood]

For my education, what does this 'system agent' do, and could it still be used for paths that will always be timer-based (HDA/HDMI + SoundWire + DMIC)?

It's primary purpose it to detect deadlock or anything that blocks main loop re-entry. i.e. we expect to enter the main loop but FW is stuck somewhere else for too long. It will cause a panic so drivers can attempt an recovery.

It really needs a little more intelligence in programming the new timeout to deal with sync and async conditions rather than a disable for use case X.

[singalsu]

Cool, the 44100 Hz version works too! Though only with 16 bit content. With 24 bit and 32 bit it triggers an issue with mixer buffer sizes. This is seen in trace:

[ 86533.020833] ( 34716.824219) c0 IPC src/ipc/handler.c:239 ipc: comp 0 -> params
[ 86536.770833] ( 3.750000) c0 PIPE 1.9 src/audio/pipeline.c:419 pipe params dir 0 frame_fmt 2 buffer_fmt 0 rate 44100
[ 86539.531250] ( 2.760417) c0 PIPE 1.9 src/audio/pipeline.c:423 pipe params stream_tag 1 channels 2 sample_valid_bytes 4 sample_container_bytes 4
[ 86573.020833] ( 33.489582) c0 DMA .../intel/cavs/hda-dma.c:405 hda-dmac: 5 channel 0 -> get
[ 86577.552083] ( 4.531250) c0 DMA .../intel/cavs/hda-dma.c:587 hda-dmac: 5 channel 0 -> config
[ 86590.937500] ( 13.385417) c0 mixer 1.3 src/audio/mixer.c:204 ERROR mixer_params(): sink buffer size is insufficient
[ 86595.572917] ( 4.635417) c0 PIPE src/audio/pipeline.c:443 ERROR pipeline_params(): ret = -12, host->comp.id = 0
[ 86599.739583] ( 4.166667) c0 IPC src/ipc/handler.c:302 ERROR ipc: pipe 1 comp 0 params failed -12

The media PCM has the same issue. It can playback 16 kHz 48 kHz content via SRC so I think configuration should be OK.

The SRC likely suffers from insufficient buffers since there's audio discontinuity (few timer per second). I can fix later as well as the mixer buffer so you don't need to deep dive into them.

[singalsu]

Unrelated issues but I can fix them later.

[plbossart]

please don't merge, I have a better one coming

[lgirdwood]

CI known issues