The environment we use are as follows: Intel(R) Xeon(R) Platinum 8468VL + Kernel 5.14 + CentOS 7 + perf-5.14
We have the following four different result:
sudo perf -a -e topdown-fe-bound -- sleep 1
<not supported> topdown-fe-bound
sudo perf -a -e slots,topdown-fe-bound -- sleep 1
14,001,930,708 slots
<not supported> topdown-fe-bound
sudo perf -a -e '{topdown-fe-bound}' -- sleep 1
<not supported> topdown-fe-bound
sudo perf -a -e '{slots,topdown-fe-bound}' -- sleep 1
511,115,940 slots
100,354,686 topdown-fe-bound
It seems that topdown-fe-bound works only with both slot and the -e '{eventname}' format.
What is the reason for this dependency?
topdown-fe-bound event needs these two special requirements because:
The Fixed counter 3 must be the leader of the group, that is, the SLOTS event must be the leader of a TopDowngroup, the second event of the group is the sampling event. Please refer to paragraphs from Row52 and Row 296.
Similar interdependence or format requirement for other events and which document should we refer to:
There isn't a uniformed document for this kind of requirements. You need to check such events seperately in kernel code. To simplify your usage, it is recommended to refer to Intel Perfspect events list. This is a more easy and direct way than using kernle wrappered event name.