dai-zephyr: check max block count supported by DMA #6823
Conversation
abonislawski
requested review from
lgirdwood,
plbossart,
mmaka1,
lbetlej and
dbaluta
as code owners
src/audio/dai-zephyr.c
Outdated
| comp_info(dev, "dai_playback_params(): block count = %d not supported by DMA", period_count); | ||
| buf_size = period_count * period_bytes; | ||
| period_count = max_block_count; | ||
| period_bytes = buf_size / period_count; |
There was a problem hiding this comment.
Maybe add an assert for the case where this does not divide evenly? Should never happen of course.
There was a problem hiding this comment.
hmm, maybe there's some implicit rule that I'm not aware of, but by just looking at this, it might well not divide evenly. We enter here when the caller requested more periods than the DMA can handle. We then calculate a buffer size from that larger block count and divide it by the the DMA maximum block count. So if max was 16 and the caller requested 17 and the period size isn't a multiple of 16, this won't divide well. Or is this meant for cases where the requested period count is a multiple of the block count, fitting multiple periods in a single DMA block?
There was a problem hiding this comment.
The only one real use case is limiting period_count to 1 for HDA DAI so more like as you said - fitting multiple period in a single DMA block, I thought we can make this more generic.
Division problem we can quickly solve with assert or ALIGN_DOWN but Im now not sure if this will result in a valid config, lets say 4x128 is still valid if requested 8x64?
There was a problem hiding this comment.
lets not assert() here, just log() return an error up the stack to host.
There was a problem hiding this comment.
Updated with alignment check and period count decreasing, please take a look if it's not over complicated
marc-hb
added
P1
|
Re-running to clear https://sof-ci.01.org/sofpr/PR6823/build2944/boottest https://sof-ci.01.org/sofpr/PR6823/build2944/devicetest/index.html is all green! |
|
SOFCI TEST EDIT: https://sof-ci.01.org/sofpr/PR6823/build3199/boottest/index.html is now clear. |
abonislawski
force-pushed
the
dai_hda_sg
branch
from
b8bda47 to
a379898
Compare
Check max block count supported by DMA using new Zephyr API This will allow to prepare a correct config regardless of specific DMA capabilities Signed-off-by: Adrian Bonislawski <adrian.bonislawski@intel.com>
abonislawski
force-pushed
the
dai_hda_sg
branch
from
a379898 to
94e7541
Compare
|
Test results are good, including alsabat test now passing https://sof-ci.01.org/sofpr/PR6823/build3247/devicetest/index.html?model=TGLU_UP_HDA_IPC4ZPH&testcase=check-alsabat-headset-playback There are multiple checkpatch warnings from long log/trace lines, but I think this is ok as it is not recommended to split string literals across multiple lines. As this is a critical fix, proceeding with merge. |
check max block count supported by DMA to prepare a correct config for DMA
Fixes: #6709
OPEN: currently it is recalculating block count to max supported but it could also return error and recalculate to 1 only for HDA