[Bug] Deadlocks when instrumenting tracing

[lisasgoh]

Describe the bug

When instrumenting tracing in Firecracker, there are at least two sources of deadlocks using the default clippy-tracing command.

1. Firecracker process hangs while starting up
   a. main_exec() in main.rs → LOGGER.update(config)
   b. Logger::update() at logging.rs → acquires LOGGER mutex → calls open_file_nonblock() when log-path is configured.
   c. open_file_nonblock() is instrumented → __Instrument::new() → log::trace!()
   d. Logger::log() tries to acquire LOGGER mutex → deadlock

2. When resuming from snapshot:
   a. Main thread calls resume_vm() → send_event() → sends Resume on channel, sets immediate_exit = 1 → sends RT signal to fc_vcpu
   b. fc_vcpu is in paused(), wakes up from recv(), receives Resume, checks immediate_exit = 1 and calls warn!() → Logger::log() → holds LOGGER mutex.
   c. RT signal arrives and interrupts fc_vcpu, the signal handler handle_signal is instrumented so it tries to acquire the LOGGER mutex as well but deadlocks.

To Reproduce

As above.

Expected behaviour

No deadlocks. I had to exclude utils/ and vpu.rs from the tracing instrumentation.

Environment

• Firecracker version: 1.15.0
• Host and guest kernel versions:
• Rootfs used:
• Architecture:
• Any other relevant software versions:

Checks

• Have you searched the Firecracker Issues database for similar problems?
• Have you read the existing relevant Firecracker documentation?
• Are you certain the bug being reported is a Firecracker issue?