Repository navigation
trace output should include component names from topology #508
Description
Currently trace output only contains the component type but not the component name used by topology. This could make trace confusing if a pipeline had multiple components of the same type or multiple pipelines were running simultaneously.
e.g.
0 2 HOST 56308262.447917 2.395833 host.c:650 cpy
0 2 HOST 56308265.312500 2.864583 hda-dma.c:229 GwU
0 2 HOST 56308268.802083 3.489583 host.c:145 irq
0 2 BUFFER 56308272.291667 3.489583 buffer.c:145 pro
0 2 unknown 56308274.843750 2.552083 buffer.c:146 value 25165824
0 2 unknown 56308277.291667 2.447917 buffer.c:147 value 131456
0 2 unknown 56308279.635417 2.343750 buffer.c:148 value 12583104
0 2 POWER 56308282.552083 2.916667 pm_runtime.c:78 put
0 2 PIPE 56308317.135417 34.583332 pipeline.c:914 CD+
0 2 VOLUME 56308319.739583 2.604167 volume.c:445 cpy
0 2 BUFFER 56308350.781250 31.041666 buffer.c:145 pro
0 2 unknown 56308353.281250 2.500000 buffer.c:146 value 50331648
0 2 unknown 56308355.729167 2.447917 buffer.c:147 value 197376
0 2 unknown 56308358.072917 2.343750 buffer.c:148
Output should also contain topology name e.g. "BUFFER 1.1" in the above output.
Activity
bkokoszx commentedon Oct 30, 2018
@lgirdwood What do you mean by topology name? Do you think about pipeline_id.component_id?
Does it not be sufficient if we only print component_id as it is unique?
lgirdwood commentedon Oct 30, 2018
@bkokoszx the topology name for a component or pipeline (they should all be unique). So yes, we will need pipeline->id and component->id for components, pipeline->id for pipelines and also dai->id for SSP, HDA etc.
The intention is to be able to globally identify any part of the trace against topology.
akloniex commentedon Oct 31, 2018
@lgirdwood I don't see an easy way to automate collection of these IDs, and it would be tedious to pass IDs to each and every trace_event macro. And adding separate macro that also accepts IDs will miss the point, as it is suboptimal. Do you have any idea how to implement this feature user-friendly?
lgirdwood commentedon Oct 31, 2018
@akloniex Add pipeline and comp ID fields to the data format then make sure all the trace macros accept these 2 IDs. If a trace user has no IDs then we can just set them as 0s.
logger will then take the pipeline ID and comp ID and then sprintf() it with component type.
akloniex commentedon Nov 7, 2018
FW part done with merged PR #530
bkokoszx commentedon Nov 22, 2018
Logger part was made here: thesofproject/soft#118
I think we can close this issue/enhancement.