diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index 12ed355..e174912 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -85,9 +85,8 @@ jobs:
       - name: Run the benchmarks
         uses: CodSpeedHQ/action@main
         if: matrix.codspeed-mode != 'off'
-        env:
-          CODSPEED_PERF_ENABLED: true
         with:
+          mode: ${{ matrix.codspeed-mode }}
           run: examples/google_benchmark_cmake/build/benchmark_example
           token: ${{ secrets.CODSPEED_TOKEN }}
 
@@ -124,9 +123,8 @@ jobs:
       - name: Run the benchmarks
         uses: CodSpeedHQ/action@main
         if: matrix.codspeed-mode != 'off'
-        env:
-          CODSPEED_PERF_ENABLED: true
         with:
+          mode: ${{ matrix.codspeed-mode }}
           run: bazel run //examples/google_benchmark_bazel:my_benchmark --//core:codspeed_mode=${{ matrix.codspeed-mode }}
           token: ${{ secrets.CODSPEED_TOKEN }}
 
diff --git a/core/BUILD b/core/BUILD
index 75c1d8f..d1cdef2 100644
--- a/core/BUILD
+++ b/core/BUILD
@@ -29,6 +29,8 @@ cc_library(
             "-Wno-unused-parameter",
             "-Wno-unused-but-set-variable",
             "-Wno-type-limits",
+            "-Wno-format",
+            "-Wno-format-security",
         ],
     }),
     visibility = ["//visibility:public"],
diff --git a/core/CMakeLists.txt b/core/CMakeLists.txt
index 6a2bea3..21b2639 100644
--- a/core/CMakeLists.txt
+++ b/core/CMakeLists.txt
@@ -35,6 +35,8 @@ if(CMAKE_CXX_COMPILER_ID MATCHES "GNU|Clang")
         -Wno-unused-parameter
         -Wno-unused-but-set-variable
         -Wno-type-limits
+        -Wno-format
+        -Wno-format-security
     )
 elseif(CMAKE_CXX_COMPILER_ID STREQUAL "MSVC")
     target_compile_options(
diff --git a/core/include/measurement.hpp b/core/include/measurement.hpp
index 89de7be..1644415 100644
--- a/core/include/measurement.hpp
+++ b/core/include/measurement.hpp
@@ -1,6 +1,7 @@
 #ifndef MEASUREMENT_H
 #define MEASUREMENT_H
 
+#include <cassert>
 #include <string>
 #ifdef _WIN32
 #include <process.h>
@@ -49,4 +50,23 @@ ALWAYS_INLINE void measurement_set_executed_benchmark(const std::string& name) {
   instrument_hooks_executed_benchmark(g_hooks, current_pid, name.c_str());
 }
 
+ALWAYS_INLINE uint64_t measurement_current_timestamp() {
+  return instrument_hooks_current_timestamp();
+}
+
+ALWAYS_INLINE int8_t measurement_add_marker(uint8_t marker_type,
+                                            uint64_t timestamp) {
+  auto pid = getpid();
+  return instrument_hooks_add_marker(g_hooks, pid, marker_type, timestamp);
+}
+
+ALWAYS_INLINE void measurement_add_benchmark_timestamps(uint64_t start,
+                                                        uint64_t end) {
+  assert(start <= end);
+  assert(start != 0 && end != 0);
+
+  measurement_add_marker(MARKER_TYPE_BENCHMARK_START, start);
+  measurement_add_marker(MARKER_TYPE_BENCHMARK_END, end);
+}
+
 #endif  // MEASUREMENT_H
diff --git a/core/instrument-hooks b/core/instrument-hooks
index a9ae7a4..b1e401a 160000
--- a/core/instrument-hooks
+++ b/core/instrument-hooks
@@ -1 +1 @@
-Subproject commit a9ae7a4d897dbb694cb0355c9994141a7c4a1ab9
+Subproject commit b1e401a4d031ad308edb22ed59a52253a1ebe924
diff --git a/examples/google_benchmark_bazel/helper.hpp b/examples/google_benchmark_bazel/helper.hpp
new file mode 120000
index 0000000..bc2f02c
--- /dev/null
+++ b/examples/google_benchmark_bazel/helper.hpp
@@ -0,0 +1 @@
+../google_benchmark_cmake/helper.hpp
\ No newline at end of file
diff --git a/examples/google_benchmark_bazel/multithread_bench.hpp b/examples/google_benchmark_bazel/multithread_bench.hpp
new file mode 120000
index 0000000..5709a0d
--- /dev/null
+++ b/examples/google_benchmark_bazel/multithread_bench.hpp
@@ -0,0 +1 @@
+../google_benchmark_cmake/multithread_bench.hpp
\ No newline at end of file
diff --git a/examples/google_benchmark_bazel/pause_timing_bench.hpp b/examples/google_benchmark_bazel/pause_timing_bench.hpp
new file mode 120000
index 0000000..24e68c9
--- /dev/null
+++ b/examples/google_benchmark_bazel/pause_timing_bench.hpp
@@ -0,0 +1 @@
+../google_benchmark_cmake/pause_timing_bench.hpp
\ No newline at end of file
diff --git a/examples/google_benchmark_cmake/helper.hpp b/examples/google_benchmark_cmake/helper.hpp
new file mode 100644
index 0000000..2be319c
--- /dev/null
+++ b/examples/google_benchmark_cmake/helper.hpp
@@ -0,0 +1,26 @@
+#ifndef HELPER_HPP
+#define HELPER_HPP
+
+#ifdef _MSC_VER
+#define NOINLINE __declspec(noinline)
+#else
+#define NOINLINE __attribute__((noinline))
+#endif
+
+#include <benchmark/benchmark.h>
+
+#include <cstdint>
+#include <numeric>
+#include <vector>
+
+NOINLINE static uint64_t recursive_fib(int n) {
+  if (n <= 1) return n;
+  return recursive_fib(n - 1) + recursive_fib(n - 2);
+}
+
+NOINLINE static uint64_t expensive_operation() {
+  // Expensive recursive compuation that will dominate flamegraph
+  return 42 + recursive_fib(30);
+}
+
+#endif  // HELPER_HPP
diff --git a/examples/google_benchmark_cmake/main.cpp b/examples/google_benchmark_cmake/main.cpp
index 9dd39dd..525cebd 100644
--- a/examples/google_benchmark_cmake/main.cpp
+++ b/examples/google_benchmark_cmake/main.cpp
@@ -4,6 +4,8 @@
 
 #include "fibonacci_bench.hpp"
 #include "fixture_bench.hpp"
+#include "multithread_bench.hpp"
+#include "pause_timing_bench.hpp"
 #include "sleep_bench.hpp"
 #include "template_bench.hpp"
 
diff --git a/examples/google_benchmark_cmake/multithread_bench.hpp b/examples/google_benchmark_cmake/multithread_bench.hpp
new file mode 100644
index 0000000..a67b769
--- /dev/null
+++ b/examples/google_benchmark_cmake/multithread_bench.hpp
@@ -0,0 +1,41 @@
+#pragma once
+
+#include <benchmark/benchmark.h>
+
+#include <atomic>
+#include <thread>
+
+#include "helper.hpp"
+
+// Simple multithreaded benchmark: spawn thread, do work, join
+static void BM_SimpleMultithread(benchmark::State& state) {
+  for (auto _ : state) {
+    state.PauseTiming();
+    std::atomic<bool> work_start{false};
+    std::atomic<bool> work_done{false};
+
+    std::thread worker([&]() {
+      // Wait for the signal to start
+      while (!work_start.load()) {
+        std::this_thread::yield();
+      }
+
+      // Do some CPU work
+      benchmark::DoNotOptimize(expensive_operation());
+      work_done.store(true);
+    });
+    state.ResumeTiming();
+
+    // Signal the worker to start
+    work_start.store(true);
+
+    // Wait for worker to complete
+    while (!work_done.load()) {
+      std::this_thread::yield();
+    }
+
+    worker.join();
+  }
+}
+
+BENCHMARK(BM_SimpleMultithread);
diff --git a/examples/google_benchmark_cmake/pause_timing_bench.hpp b/examples/google_benchmark_cmake/pause_timing_bench.hpp
new file mode 100644
index 0000000..5687470
--- /dev/null
+++ b/examples/google_benchmark_cmake/pause_timing_bench.hpp
@@ -0,0 +1,40 @@
+#pragma once
+
+#include <benchmark/benchmark.h>
+
+#include <chrono>
+#include <thread>
+
+#include "helper.hpp"
+
+NOINLINE static uint64_t actual_work() {
+  std::this_thread::sleep_for(std::chrono::milliseconds(1));
+  return 42;
+}
+
+static void BM_large_setup_teardown(benchmark::State& state) {
+  for (auto _ : state) {
+    state.PauseTiming();
+    benchmark::DoNotOptimize(expensive_operation());
+    state.ResumeTiming();
+
+    benchmark::DoNotOptimize(actual_work());
+
+    state.PauseTiming();
+    benchmark::DoNotOptimize(expensive_operation());
+    state.ResumeTiming();
+  }
+}
+// IMPORTANT: Use fixed iterations, otherwise we'll run for 10+ minutes
+BENCHMARK(BM_large_setup_teardown)->Iterations(100);
+
+static void BM_large_setup(benchmark::State& state) {
+  for (auto _ : state) {
+    state.PauseTiming();
+    benchmark::DoNotOptimize(expensive_operation());
+    state.ResumeTiming();
+
+    benchmark::DoNotOptimize(actual_work());
+  }
+}
+BENCHMARK(BM_large_setup)->Iterations(100);
diff --git a/google_benchmark/include/benchmark/benchmark.h b/google_benchmark/include/benchmark/benchmark.h
index f30977b..0e7d96d 100644
--- a/google_benchmark/include/benchmark/benchmark.h
+++ b/google_benchmark/include/benchmark/benchmark.h
@@ -949,6 +949,9 @@ class BENCHMARK_EXPORT BENCHMARK_INTERNAL_CACHELINE_ALIGNED State {
 #if defined(CODSPEED_INSTRUMENTATION) || defined(CODSPEED_WALLTIME)
   codspeed::CodSpeed* codspeed_;
 #endif
+#ifdef CODSPEED_WALLTIME
+  uint64_t resume_timestamp_;
+#endif
 
  private:
   bool started_;
@@ -1044,11 +1047,13 @@ struct State::StateIterator {
  private:
   friend class State;
   BENCHMARK_ALWAYS_INLINE
-  StateIterator() : cached_(0), parent_() {}
+  StateIterator() : cached_(0), parent_()
+  {}
 
   BENCHMARK_ALWAYS_INLINE
   explicit StateIterator(State* st)
-      : cached_(st->skipped() ? 0 : st->max_iterations), parent_(st) {}
+      : cached_(st->skipped() ? 0 : st->max_iterations), parent_(st)
+  {}
 
  public:
   BENCHMARK_ALWAYS_INLINE
@@ -1063,7 +1068,9 @@ struct State::StateIterator {
 
   BENCHMARK_ALWAYS_INLINE
   bool operator!=(StateIterator const&) const {
-    if (BENCHMARK_BUILTIN_EXPECT(cached_ != 0, true)) return true;
+    if (BENCHMARK_BUILTIN_EXPECT(cached_ != 0, true)) {
+      return true;
+    }
 #ifdef CODSPEED_INSTRUMENTATION
     measurement_stop();
 #endif
diff --git a/google_benchmark/src/benchmark.cc b/google_benchmark/src/benchmark.cc
index 91c737a..5a7dbb7 100644
--- a/google_benchmark/src/benchmark.cc
+++ b/google_benchmark/src/benchmark.cc
@@ -19,6 +19,10 @@
 #include "codspeed.h"
 #include "internal_macros.h"
 
+#ifdef CODSPEED_WALLTIME
+#include "measurement.hpp"
+#endif
+
 #ifndef BENCHMARK_OS_WINDOWS
 #if !defined(BENCHMARK_OS_FUCHSIA) && !defined(BENCHMARK_OS_QURT)
 #include <sys/resource.h>
@@ -185,6 +189,9 @@ State::State(std::string name, IterationCount max_iters,
       max_iterations(max_iters),
 #if defined(CODSPEED_INSTRUMENTATION) || defined(CODSPEED_WALLTIME)
       codspeed_(codspeed),
+#endif
+#ifdef CODSPEED_WALLTIME
+      resume_timestamp_(0),
 #endif
       started_(false),
       finished_(false),
@@ -252,9 +259,21 @@ State::State(std::string name, IterationCount max_iters,
 }
 
 void State::PauseTiming() {
+#ifdef CODSPEED_WALLTIME
+  uint64_t pause_timestamp = measurement_current_timestamp();
+#endif
+
   // Add in time accumulated so far
   BM_CHECK(started_ && !finished_ && !skipped());
   timer_->StopTimer();
+
+#ifdef CODSPEED_WALLTIME
+  if (resume_timestamp_ != 0) {
+    measurement_add_benchmark_timestamps(resume_timestamp_, pause_timestamp);
+    resume_timestamp_ = 0;
+  }
+#endif
+
   if (perf_counters_measurement_ != nullptr) {
     std::vector<std::pair<std::string, double>> measurements;
     if (!perf_counters_measurement_->Stop(measurements)) {
@@ -276,6 +295,11 @@ void State::ResumeTiming() {
   if (perf_counters_measurement_ != nullptr) {
     perf_counters_measurement_->Start();
   }
+
+#ifdef CODSPEED_WALLTIME
+  BM_CHECK(resume_timestamp_ == 0);
+  resume_timestamp_ = measurement_current_timestamp();
+#endif
 }
 
 void State::SkipWithMessage(const std::string& msg) {