Version: 6.17.13

In previous versions, I was able to run e.g. `perf record --call-graph
dwarf,16384 -m 64M -p <PID>` and it would exit fairly quickly after
pressing crtl+C. Now, the same command has been running for over an hour
after profiling for about 10 seconds! Every few seconds it prints an error:

addr2line /path/to/symbols/file.debug: could not read first record

The way perf invokes addr2line is known to be very slow, but I have
previously only encountered that when running `perf script`. `perf
record --call-graph dwarf` used to only record stack dumps, not try and
symbolicate them, which other tools, e.g. hotspot or samply, are much
faster at doing. I believe `perf record` *shouldn't* be trying to do
this on its own, at least without being explicitly requested to, and the
manpage makes no mention of it.