ci: Use fp-based perf profiling to reduce unknown symbols (#1718)

.github/workflows/bench.yml

@@ -8,8 +8,8 @@ env:
   CARGO_TERM_COLOR: always
   RUST_BACKTRACE: 1
   TOOLCHAIN: nightly
-  RUSTFLAGS: -C link-arg=-fuse-ld=lld -C link-arg=-Wl,--no-rosegment
-  PERF_CMD: record -o perf.data -F997 --call-graph lbr -g
+  RUSTFLAGS: -C link-arg=-fuse-ld=lld -C link-arg=-Wl,--no-rosegment, -Cforce-frame-pointers=yes
+  PERF_CMD: record -o perf.data -F997 --call-graph fp -g
 
 jobs:
   bench:
@@ -81,8 +81,6 @@ jobs:
             taskset -c 0 nice -n -20 \
             cargo +$TOOLCHAIN flamegraph -v -c "$PERF_CMD" \
               --bin neqo-server -- --db ../test-fixture/db $HOST:4433 || true; } &
-          # Give the server time to start.
-          sleep 3
           mkdir client; \
             cd client; \
             time taskset -c 1 nice -n -20 \