I also notice that we have other paths like this, where the allocated
pmu (and aliases) are not freed for later error paths, it seems:
parse_events_add_pmu() -> perf_pmu_find() -> pmu_lookup() ->
pmu_add_cpu_aliases().
I had a quick look at the rest of the series, and could not see if we
fix up any of this.
Right, I also found that and wondered why ASAN didn't report it.
I think we should free all pmu instances from the pmus list.
But this can be a later work..