Race in coverage recording?

[Porges]

I noticed while trying to repro #1781 locally that sometimes the recording would not work properly:

Commandline:

$ RUST_LOG=trace cargo run --example coverage -- /workspaces/onefuzz/src/integration-tests/libfuzzer/fuzz.exe @@ --inputs  /workspaces/onefuzz/src/integration-tests/libfuzzer/input.txt

Coverage as expected:

[2022-10-16T22:16:40Z INFO  coverage::block::linux] module load: /lib/x86_64-linux-gnu/librt-2.27.so
[2022-10-16T22:16:40Z INFO  coverage::block::linux] module load: /lib/x86_64-linux-gnu/libm-2.27.so
[2022-10-16T22:16:40Z INFO  coverage::block::linux] module load: /lib/x86_64-linux-gnu/libdl-2.27.so
[2022-10-16T22:16:40Z INFO  coverage::block::linux] module load: /lib/x86_64-linux-gnu/libgcc_s.so.1
[2022-10-16T22:16:40Z INFO  coverage::block::linux] module load: /lib/x86_64-linux-gnu/libc-2.27.so
[2022-10-16T22:16:41Z INFO  coverage::block::linux] new thread: 28928
[2022-10-16T22:16:41Z INFO  coverage] recorded in 2.480324274s
[2022-10-16T22:16:41Z INFO  coverage] input = /workspaces/onefuzz/src/integration-tests/libfuzzer/input.txt
[2022-10-16T22:16:41Z INFO  coverage] ld-2.27.so = 0 / 0 (NaN%)
[2022-10-16T22:16:41Z INFO  coverage] libc-2.27.so = 0 / 0 (NaN%)
[2022-10-16T22:16:41Z INFO  coverage] libdl-2.27.so = 0 / 0 (NaN%)
[2022-10-16T22:16:41Z INFO  coverage] libgcc_s.so.1 = 0 / 0 (NaN%)
[2022-10-16T22:16:41Z INFO  coverage] libm-2.27.so = 0 / 0 (NaN%)
[2022-10-16T22:16:41Z INFO  coverage] libpthread-2.27.so = 185 / 3192 (5.80%)
[2022-10-16T22:16:41Z INFO  coverage] librt-2.27.so = 0 / 0 (NaN%)
[2022-10-16T22:16:41Z INFO  coverage] libstdc++.so.6.0.29 = 0 / 0 (NaN%)
[2022-10-16T22:16:41Z INFO  coverage] fuzz.exe = 5 / 22 (22.73%)
[2022-10-16T22:16:41Z INFO  coverage] ld-2.27.so = 0 / 0 (NaN%)
[2022-10-16T22:16:41Z INFO  coverage] libc-2.27.so = 0 / 0 (NaN%)
[2022-10-16T22:16:41Z INFO  coverage] libdl-2.27.so = 0 / 0 (NaN%)
[2022-10-16T22:16:41Z INFO  coverage] libgcc_s.so.1 = 0 / 0 (NaN%)
[2022-10-16T22:16:41Z INFO  coverage] libm-2.27.so = 0 / 0 (NaN%)
[2022-10-16T22:16:41Z INFO  coverage] libpthread-2.27.so = 185 / 3192 (5.80%)
[2022-10-16T22:16:41Z INFO  coverage] librt-2.27.so = 0 / 0 (NaN%)
[2022-10-16T22:16:41Z INFO  coverage] libstdc++.so.6.0.29 = 0 / 0 (NaN%)
[2022-10-16T22:16:41Z INFO  coverage] fuzz.exe = 5 / 22 (22.73%)

Coverage not as expected:

[2022-10-16T22:19:43Z INFO  coverage::block::linux] module load: /lib/x86_64-linux-gnu/ld-2.27.so
[2022-10-16T22:19:43Z INFO  coverage] recorded in 11.445283278s
[2022-10-16T22:19:43Z INFO  coverage] input = /workspaces/onefuzz/src/integration-tests/libfuzzer/input.txt
[2022-10-16T22:19:43Z INFO  coverage] ld-2.27.so = 0 / 0 (NaN%)
[2022-10-16T22:19:43Z INFO  coverage] fuzz.exe = 0 / 22 (0.00%)
[2022-10-16T22:19:43Z INFO  coverage] ld-2.27.so = 0 / 0 (NaN%)
[2022-10-16T22:19:43Z INFO  coverage] fuzz.exe = 0 / 22 (0.00%)

In the latter case it only receives "module load" events for fuzz.exe & ld. It seems that pete/Ptracer is missing some events?

Traces:

• ok.txt
• bad.txt

AB#41811543