ASoC: SOF: Intel: hda-mlink: add support for Extended Multi Links #4158
Conversation
plbossart
requested review from
ajitkupandey,
bardliao,
dbaluta,
kuanhsuncheng,
kv2019i,
lgirdwood,
libinyang,
lyakh,
ranj063,
ujfalusi and
yaochunhung
as code owners
plbossart
force-pushed
the
hdaudio/extended-multi-link
branch
from
d3d234d to
442f1f3
Compare
|
2 checkpatch warnings missed, oops. fixed now. |
|
@ujfalusi @RanderWang @ranj063 @bardliao can you please look into this PR? it's rather important and gates the rest of the SoundWire patches. |
ranj063
left a comment
ranj063
left a comment
There was a problem hiding this comment.
looks great. just minor comments
|
v2: addressed comments, moved interrupt code in hdaml section. |
plbossart
force-pushed
the
hdaudio/extended-multi-link
branch
from
442f1f3 to
db258a6
Compare
plbossart
force-pushed
the
hdaudio/extended-multi-link
branch
from
db258a6 to
df257de
Compare
|
v3: modified hardware poll delay to 10us and max 1ms |
These helpers configure the ratio between the base clock and the hardware signal used for link synchronization. The SYNCPRD is written before the first sublink is powered-up. The SYNCPU bit is set, but it will only be cleared after the link is powered-up, hence the implementation with a set/wait pattern. These helpers are currently only needed by SoundWire support, where the lock is taken at a higher level, so only the _unlocked versions are exposed for now. Note that the _wait_bit() implementation is similar to previous helpers in drivers/soundwire, but with sleep duration and timeout aligned with hardware recommendations. If desired, this helper could be modified in a second step with e.g. readl_poll_timeout(). Signed-off-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>
The multi-link synchronization uses the same concept and registers, but moved to the HDAudio extended links. Add helpers for sync_arm and sync_go which are the basic for the bus reset, bank switch and clock stop. Since SoundWire is the only user of those helpers, only expose the _unlocked versions for now. Note that SYNCGO is a write-only bit, so no error can be reported. We still return 0 for compatibility with the SoundWire stream management headers. Signed-off-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>
This helper is an optimization where sync_go is only called when the cmdsync field is actually set to a non-zero value. Since this is also only used by SoundWire for now, only expose the _unlocked version. Signed-off-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>
Each SoundWire peripheral can be programmed from the manager side either with a regular command FIFO, or with the HDaudio CORB/RIRB DMA-based mechanism. The mapping between SoundWire peripheral and SDI address is handled with the LSDIID register. This mapping only works of course if each peripheral has a unique address across all links. This has already been enforced in previous Intel contributions allowing for an IDA-based solution for the device number allocation. The checks on the dev_num are handled at the SoundWire level, but the locking is handled at the hda-mlink level. Signed-off-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>
Small helpers to make DAI ops simpler. Signed-off-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>
For DMIC and SSP, the DSP will be responsible for programming the blobs and link registers. Signed-off-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>
For SoundWire usages, we need to use the global eml_lock to serialize/protect all accesses to shared registers. Due to the split implementation across two subsystems, we need to pass a pointer around. Signed-off-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>
plbossart
force-pushed
the
hdaudio/extended-multi-link
branch
from
2c13ad1 to
f423930
Compare
|
v4: fixed nocodec issue (wrong Kconfig). sof-test will still fail on sof-logger test until thesofproject/sof-test#1003 is merged. |
|
The only issue in the tests is the hda-mlink module issue sof-test was updated, will re-run the tests. |
|
SOFCI TEST |
|
I don't understand the CI failures, I tested manually on the very same device that fails https://sof-ci.01.org/linuxpr/PR4158/build3223/devicetest/index.html?model=ADLP_RVP_SDW_IPC4ZPH&testcase=check-kmod-load-unload and it's working fine. Probably some caching issue? |
|
SOFCI_TEST |
|
SOFCI TEST |
|
GitHub I hate you, why can't I rebase this branch? "This branch cannot be rebased due to too many changes" What a useless tool. |
|
I'll use a merge commit for now, I don't have time to deal with stupid tools today. |
bardliao
left a comment
bardliao
left a comment
There was a problem hiding this comment.
I am late, but it looks good to me.
5 participants
The LunarLake hardware moved a set of DMIC/SSP/SoundWire registers in the HDaudio multi-link areas. This PR adds a set of helpers to configure the DMIC/SSP/SoundWire. In the case of SoundWire, the configurations will be made both from the sound/soc/sof/intel and drivers/soundwire/intel locations, so the hda-mlink is now a separate module to avoid circular dependencies.
The actual use of these helpers will be provided when this PR #4158 and PR #4157 are merged. There will be cross-dependencies between ALSA, ASoC and SoundWire so we have to progress in steps.
One possible pushback is that this adds yet another layer on top of hdac_link and hdac_ext_link. That's a valid point. The reason why this was added as a separate extension is that we don't have the ability to validate the Skylake and AVS drivers, and there's no reason why these drivers would need to know about a hardware extension added 8 years after the introduction of SKL/KBL.
One could also argue that this extension is not SOF-specific. That's correct from a hardware perspective, however since there are no other users it's not clear where else than sound/soc/sof/intel this module should be located.
@tiwai FYI