hda: separation of l1 settings to power manager

[fkwasowi]

Separating two new functions force and allow l1 to have the current state with separated functions in the power manager so that SOF can call these functions via IPC DMI_FORCE_L1_EXIT. Change related to the addition of a new parameter to force DMI L1 exit on IPC request.

Associated with: https://github.com/thesofproject/sof/pull/8561/files

[teburd]

This appears to be moving some ace specifics into power management for all of Zephyr while not actually changing the logic involved.

If soc specifics are needed in sof then those need to be exposed through the soc headers not through the general pm device header.

[ceolin]

This adsp specific logic and should never be part of Zephyr's device pm interface.

[fkwasowi]

This appears to be moving some ace specifics into power management for all of Zephyr while not actually changing the logic involved.

If soc specifics are needed in sof then those need to be exposed through the soc headers not through the general pm device header.

I changed the location of the function to soc/xtensa/intel_adsp/common/include/intel_adsp_ipc.h

[fkwasowi]

This adsp specific logic and should never be part of Zephyr's device pm interface.

I changed the location of the function to soc/xtensa/intel_adsp/common/include/intel_adsp_ipc.h

[fkwasowi]

Direct Media Interface

Yes. It's about Direct Media Interface

done

[kv2019i]

@marc-hb wrote:

shouldn't these functions be called intel_adsp_ipc_allow_dma_l1_state and intel_adsp_ipc_allow_dma_l0_state ?
Is this maybe Direct Media Interface (PCI) link state?

I don't think it doesn't matter much here. The property we are setting is really tied to DMA (not typo) here. Behind the DMA, there will be DMI in the end, but from the point of view of the DSP, this difference has little practical meaning.

[marc-hb]

When you run west grep -i dmi.*l1, you really don't want to miss one spot.

Names look good to me now.

[iganakov]

Looks good to me

[abonislawski]

Looks fine, I would just remove "ipc" from function names, it seems unrelated

[mwasko]

@nashif @ceolin can we proceed with merge?

[marc-hb]

This fails to compile for LNL:

https://github.com/thesofproject/sof/actions/runs/7196263630/job/19600738125?pr=8618

-o zephyr/drivers/dma/CMakeFiles/drivers__dma.dir/dma_intel_adsp_hda.c.obj -c /zep_workspace/zephyr/drivers/dma/dma_intel_adsp_hda.c
In file included from /zep_workspace/zephyr/drivers/dma/dma_intel_adsp_hda.c:28:
/zep_workspace/zephyr/soc/xtensa/intel_adsp/common/include/intel_adsp_hda.h: In function 'intel_adsp_force_dmi_l0_state':
/zep_workspace/zephyr/soc/xtensa/intel_adsp/common/include/intel_adsp_hda.h:448:30: error: 'ADSP_FORCE_DECOUPLED_HDMA_L1_EXIT_BIT' undeclared (first use in this function)
  448 |         ACE_DfPMCCH.svcfg |= ADSP_FORCE_DECOUPLED_HDMA_L1_EXIT_BIT;
      |                              ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
/zep_workspace/zephyr/soc/xtensa/intel_adsp/common/include/intel_adsp_hda.h:448:30: note: each undeclared identifier is reported only once for each function it appears in
/zep_workspace/zephyr/soc/xtensa/intel_adsp/common/include/intel_adsp_hda.h: In function 'intel_adsp_allow_dmi_l1_state':
/zep_workspace/zephyr/soc/xtensa/intel_adsp/common/include/intel_adsp_hda.h:455:32: error: 'ADSP_FORCE_DECOUPLED_HDMA_L1_EXIT_BIT' undeclared (first use in this function)
  455 |         ACE_DfPMCCH.svcfg &= ~(ADSP_FORCE_DECOUPLED_HDMA_L1_EXIT_BIT);
      |                                ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

[marc-hb]

By using ADSP_FORCE_DECOUPLED_HDMA_L1_EXIT_BIT in a common .h file, this commit
introduces a new requirement:

=> now ADSP_FORCE_DECOUPLED_HDMA_L1_EXIT_BIT must always be defined, even when CONFIG_DMA_INTEL_ADSP_HDA_TIMING_L1_EXIT is false.

Is this new requirement OK?

Of course another question is: why is CONFIG_DMA_INTEL_ADSP_HDA_TIMING_L1_EXIT false on LNL? But that's a different question.

[marc-hb]

Of course another question is: why is CONFIG_DMA_INTEL_ADSP_HDA_TIMING_L1_EXIT false on LNL? But that's a different question.

5 months and probably a lot of wasted time later:

• drivers: dma: intel_adsp_hda: change L1 exit defaults #73733