diff --git a/src/run/runner/wall_time/perf/mod.rs b/src/run/runner/wall_time/perf/mod.rs
index e0db17e1..5210cd2d 100644
--- a/src/run/runner/wall_time/perf/mod.rs
+++ b/src/run/runner/wall_time/perf/mod.rs
@@ -125,7 +125,7 @@ impl PerfRunner {
             "sh",
             "-c",
             &format!(
-                "perf record {quiet_flag} --user-callchains --freq=999 --switch-output --control=fifo:{},{} --delay=-1 -g --call-graph={cg_mode} --output={} -- {bench_cmd}",
+                "perf record {quiet_flag} --timestamp -k CLOCK_MONOTONIC --user-callchains --freq=999 --switch-output --control=fifo:{},{} --delay=-1 -g --call-graph={cg_mode} --output={} -- {bench_cmd}",
                 perf_fifo.ctl_fifo_path.to_string_lossy(),
                 perf_fifo.ack_fifo_path.to_string_lossy(),
                 perf_file.path().to_string_lossy(),
@@ -167,9 +167,11 @@ impl PerfRunner {
                     .unwrap_or(false)
             })
             .sorted_by_key(|path| path.file_name().unwrap().to_string_lossy().to_string())
-            // The first perf.data will only contain metadata that is not relevant to the benchmarks. We
-            // capture the symbols and unwind data separately.
+            // We split the perf.data in SetExecutedBenchmark which will also split the last
+            // entry, which will add an extra perf.data file that is empty. We can just skip it.
+            .rev()
             .skip(1)
+            .rev()
             .map(|src_path| {
                 let profile_folder = profile_folder.clone();
                 tokio::task::spawn(async move {
@@ -330,19 +332,7 @@ impl PerfRunner {
                 FifoCommand::CurrentBenchmark { pid, uri } => {
                     bench_order_by_pid.entry(pid).or_default().push(uri);
 
-                    #[cfg(target_os = "linux")]
-                    if !symbols_by_pid.contains_key(&pid) && !unwind_data_by_pid.contains_key(&pid)
-                    {
-                        Self::process_memory_mappings(
-                            pid,
-                            &mut symbols_by_pid,
-                            &mut unwind_data_by_pid,
-                        )?;
-                    }
-
-                    runner_fifo.send_cmd(FifoCommand::Ack).await?;
-                }
-                FifoCommand::StartBenchmark => {
+                    // Split the perf.data file
                     let perf_pid = perf_pid.get_or_init(|| {
                         let output = std::process::Command::new("sh")
                             .arg("-c")
@@ -355,10 +345,21 @@ impl PerfRunner {
                             .parse::<u32>()
                             .expect("Failed to parse perf pid")
                     });
-
-                    // Split the perf.data file
                     run_with_sudo(&["kill", "-USR2", &perf_pid.to_string()])?;
 
+                    #[cfg(target_os = "linux")]
+                    if !symbols_by_pid.contains_key(&pid) && !unwind_data_by_pid.contains_key(&pid)
+                    {
+                        Self::process_memory_mappings(
+                            pid,
+                            &mut symbols_by_pid,
+                            &mut unwind_data_by_pid,
+                        )?;
+                    }
+
+                    runner_fifo.send_cmd(FifoCommand::Ack).await?;
+                }
+                FifoCommand::StartBenchmark => {
                     perf_fifo.start_events().await?;
                     runner_fifo.send_cmd(FifoCommand::Ack).await?;
                 }