early-access version 2824

README.md

@@ -1,7 +1,7 @@
 yuzu emulator early access
 =============
 
-This is the source code for early-access 2823.
+This is the source code for early-access 2824.
 
 ## Legal Notice
 

externals/cubeb/.github/workflows/build.yml

@@ -9,11 +9,11 @@ jobs:
       BUILD_TYPE: ${{ matrix.type }}
     strategy:
       matrix:
-        os: [ubuntu-20.04, windows-2019, macos-10.15]
+        os: [ubuntu-20.04, windows-2019, macos-11]
         type: [Release, Debug]
 
     steps:
-    - uses: actions/checkout@v2
+    - uses: actions/checkout@v3
       with:
         submodules: true
 
@@ -46,13 +46,13 @@ jobs:
       matrix:
         type: [Release, Debug]
     steps:
-    - uses: actions/checkout@v2
+    - uses: actions/checkout@v3
       with:
         submodules: true
 
     - name: Configure CMake
       shell: bash
-      run: cmake -S . -B build -DCMAKE_BUILD_TYPE=$BUILD_TYPE -DCMAKE_TOOLCHAIN_FILE=$ANDROID_NDK_HOME/build/cmake/android.toolchain.cmake -DANDROID_NATIVE_API_LEVEL=android-26
+      run: cmake -S . -B build -DCMAKE_BUILD_TYPE=$BUILD_TYPE -DCMAKE_TOOLCHAIN_FILE=$ANDROID_NDK_HOME/build/cmake/android.toolchain.cmake -DANDROID_NATIVE_API_LEVEL=android-28
 
     - name: Build
       shell: bash
@@ -61,7 +61,7 @@ jobs:
   check_format:
     runs-on: ubuntu-20.04
     steps:
-    - uses: actions/checkout@v2
+    - uses: actions/checkout@v3
       with:
         submodules: true
 

externals/cubeb/INSTALL.md

@@ -5,9 +5,9 @@ You must have CMake v3.1 or later installed.
 1. `git clone --recursive https://github.com/mozilla/cubeb.git`
 2. `mkdir cubeb-build`
 3. `cd cubeb-build`
-3. `cmake ../cubeb`
+4. `cmake ../cubeb`
-4. `cmake --build .`
+5. `cmake --build .`
-5. `ctest`
+6. `ctest`
 
 # Windows build notes
 
@@ -41,6 +41,6 @@ To build with MinGW-w64, install the following items:
 - Download and install MinGW-w64 with Win32 threads.
 - Download and install CMake.
 - Run MinGW-w64 Terminal from the Start Menu.
-- Follow the build steps at the top of this file, but at step 3 run:
+- Follow the build steps at the top of this file, but at step 4 run:
-  `cmake -G "MinGW Makefiles" ..`
+  `cmake -G "MinGW Makefiles" ../cubeb`
 - Continue the build steps at the top of this file.

externals/cubeb/README.md

@@ -2,6 +2,6 @@
 
 See INSTALL.md for build instructions.
 
-See [Backend Support](https://github.com/kinetiknz/cubeb/wiki/Backend-Support) in the wiki for the support level of each backend.
+See [Backend Support](https://github.com/mozilla/cubeb/wiki/Backend-Support) in the wiki for the support level of each backend.
 
 Licensed under an ISC-style license.  See LICENSE for details.

externals/cubeb/src/cubeb_audiounit.cpp

@@ -541,6 +541,13 @@ audiounit_input_callback(void * user_ptr, AudioUnitRenderActionFlags * flags,
   long outframes = cubeb_resampler_fill(stm->resampler.get(),
                                         stm->input_linear_buffer->data(),
                                         &total_input_frames, NULL, 0);
+  if (outframes < 0) {
+    stm->shutdown = true;
+    OSStatus r = AudioOutputUnitStop(stm->input_unit);
+    assert(r == 0);
+    stm->state_callback(stm, stm->user_ptr, CUBEB_STATE_ERROR);
+    return noErr;
+  }
   stm->draining = outframes < total_input_frames;
 
   // Reset input buffer

externals/cubeb/src/cubeb_log.cpp

@@ -123,7 +123,7 @@ cubeb_async_log(char const * fmt, ...)
 }
 
 void
-cubeb_async_log_reset_threads()
+cubeb_async_log_reset_threads(void)
 {
   if (!g_cubeb_log_callback) {
     return;

externals/cubeb/src/cubeb_log.h

@@ -35,7 +35,7 @@ extern cubeb_log_callback g_cubeb_log_callback PRINTF_FORMAT(1, 2);
 void
 cubeb_async_log(const char * fmt, ...);
 void
-cubeb_async_log_reset_threads();
+cubeb_async_log_reset_threads(void);
 
 #ifdef __cplusplus
 }

externals/cubeb/src/cubeb_pulse.c

@@ -280,6 +280,7 @@ trigger_user_callback(pa_stream * s, void const * input_data, size_t nbytes,
     if (got < 0) {
       WRAP(pa_stream_cancel_write)(s);
       stm->shutdown = 1;
+      stm->state_callback(stm, stm->user_ptr, CUBEB_STATE_ERROR);
       return;
     }
     // If more iterations move offset of read buffer
@@ -392,6 +393,9 @@ stream_read_callback(pa_stream * s, size_t nbytes, void * u)
         if (got < 0 || (size_t)got != read_frames) {
           WRAP(pa_stream_cancel_write)(s);
           stm->shutdown = 1;
+          if (got < 0) {
+            stm->state_callback(stm, stm->user_ptr, CUBEB_STATE_ERROR);
+          }
           break;
         }
       }

externals/cubeb/src/cubeb_ringbuffer.h

@@ -110,8 +110,8 @@ public:
     assert_correct_thread(producer_id);
 #endif
 
-    int rd_idx = read_index_.load(std::memory_order_relaxed);
     int wr_idx = write_index_.load(std::memory_order_relaxed);
+    int rd_idx = read_index_.load(std::memory_order_acquire);
 
     if (full_internal(rd_idx, wr_idx)) {
       return 0;
@@ -154,8 +154,8 @@ public:
     assert_correct_thread(consumer_id);
 #endif
 
-    int wr_idx = write_index_.load(std::memory_order_acquire);
     int rd_idx = read_index_.load(std::memory_order_relaxed);
+    int wr_idx = write_index_.load(std::memory_order_acquire);
 
     if (empty_internal(rd_idx, wr_idx)) {
       return 0;
@@ -172,7 +172,7 @@ public:
     }
 
     read_index_.store(increment_index(rd_idx, to_read),
-                      std::memory_order_relaxed);
+                      std::memory_order_release);
 
     return to_read;
   }
@@ -190,7 +190,7 @@ public:
 #endif
     return available_read_internal(
         read_index_.load(std::memory_order_relaxed),
-        write_index_.load(std::memory_order_relaxed));
+        write_index_.load(std::memory_order_acquire));
   }
   /**
    * Get the number of available elements for consuming.
@@ -205,7 +205,7 @@ public:
     assert_correct_thread(producer_id);
 #endif
     return available_write_internal(
-        read_index_.load(std::memory_order_relaxed),
+        read_index_.load(std::memory_order_acquire),
         write_index_.load(std::memory_order_relaxed));
   }
   /**

externals/cubeb/src/cubeb_wasapi.cpp

@@ -240,8 +240,9 @@ wasapi_create_device(cubeb * ctx, cubeb_device_info & ret,
 void
 wasapi_destroy_device(cubeb_device_info * device_info);
 static int
-wasapi_enumerate_devices(cubeb * context, cubeb_device_type type,
+wasapi_enumerate_devices_internal(cubeb * context, cubeb_device_type type,
-                         cubeb_device_collection * out);
+                                  cubeb_device_collection * out,
+                                  DWORD state_mask);
 static int
 wasapi_device_collection_destroy(cubeb * ctx,
                                  cubeb_device_collection * collection);
@@ -409,12 +410,9 @@ struct cubeb_stream {
   float volume = 1.0;
   /* True if the stream is draining. */
   bool draining = false;
-  /* True when we've destroyed the stream. This pointer is leaked on stream
+  /* If the render thread fails to stop, this is set to true and ownership of
-   * destruction if we could not join the thread. */
+   * the stm is "leaked" to the render thread for later cleanup. */
-  std::atomic<std::atomic<bool> *> emergency_bailout{nullptr};
+  std::atomic<bool> emergency_bailout{false};
-  /* Synchronizes render thread start to ensure safe access to
-   * emergency_bailout. */
-  HANDLE thread_ready_event = 0;
   /* This needs an active audio input stream to be known, and is updated in the
    * first audio input callback. */
   std::atomic<int64_t> input_latency_hns{LATENCY_NOT_AVAILABLE_YET};
@@ -753,6 +751,27 @@ private:
 
 namespace {
 
+long
+wasapi_data_callback(cubeb_stream * stm, void * user_ptr,
+                     void const * input_buffer, void * output_buffer,
+                     long nframes)
+{
+  if (stm->emergency_bailout) {
+    return CUBEB_ERROR;
+  }
+  return stm->data_callback(stm, user_ptr, input_buffer, output_buffer,
+                            nframes);
+}
+
+void
+wasapi_state_callback(cubeb_stream * stm, void * user_ptr, cubeb_state state)
+{
+  if (stm->emergency_bailout) {
+    return;
+  }
+  return stm->state_callback(stm, user_ptr, state);
+}
+
 char const *
 intern_device_id(cubeb * ctx, wchar_t const * id)
 {
@@ -873,8 +892,11 @@ refill(cubeb_stream * stm, void * input_buffer, long input_frames_count,
   long out_frames =
       cubeb_resampler_fill(stm->resampler.get(), input_buffer,
                            &input_frames_count, dest, output_frames_needed);
-  /* TODO: Report out_frames < 0 as an error via the API. */
+  if (out_frames < 0) {
-  XASSERT(out_frames >= 0);
+    ALOGV("Callback refill error: %d", out_frames);
+    wasapi_state_callback(stm, stm->user_ptr, CUBEB_STATE_ERROR);
+    return out_frames;
+  }
 
   float volume = 1.0;
   {
@@ -991,7 +1013,7 @@ get_input_buffer(cubeb_stream * stm)
           (stm->input_stream_params.prefs &
            CUBEB_STREAM_PREF_DISABLE_DEVICE_SWITCHING) ||
           !trigger_async_reconfigure(stm)) {
-        stm->state_callback(stm, stm->user_ptr, CUBEB_STATE_ERROR);
+        wasapi_state_callback(stm, stm->user_ptr, CUBEB_STATE_ERROR);
         return false;
       }
       return true;
@@ -1105,7 +1127,7 @@ get_output_buffer(cubeb_stream * stm, void *& buffer, size_t & frame_count)
         (stm->output_stream_params.prefs &
          CUBEB_STREAM_PREF_DISABLE_DEVICE_SWITCHING) ||
         !trigger_async_reconfigure(stm)) {
-      stm->state_callback(stm, stm->user_ptr, CUBEB_STATE_ERROR);
+      wasapi_state_callback(stm, stm->user_ptr, CUBEB_STATE_ERROR);
       return false;
     }
     return true;
@@ -1121,7 +1143,7 @@ get_output_buffer(cubeb_stream * stm, void *& buffer, size_t & frame_count)
   if (stm->draining) {
     if (padding_out == 0) {
       LOG("Draining finished.");
-      stm->state_callback(stm, stm->user_ptr, CUBEB_STATE_DRAINED);
+      wasapi_state_callback(stm, stm->user_ptr, CUBEB_STATE_DRAINED);
       return false;
     }
     LOG("Draining.");
@@ -1190,6 +1212,7 @@ refill_callback_duplex(cubeb_stream * stm)
         static_cast<long>(stm->total_output_frames) - stm->total_input_frames,
         static_cast<float>(stm->total_output_frames) / stm->total_input_frames);
 
+  long got;
   if (stm->has_dummy_output) {
     ALOGV(
         "Duplex callback (dummy output): input frames: %Iu, output frames: %Iu",
@@ -1197,13 +1220,15 @@ refill_callback_duplex(cubeb_stream * stm)
 
     // We don't want to expose the dummy output to the callback so don't pass
     // the output buffer (it will be released later with silence in it)
-    refill(stm, stm->linear_input_buffer->data(), input_frames, nullptr, 0);
+    got =
+        refill(stm, stm->linear_input_buffer->data(), input_frames, nullptr, 0);
+
   } else {
     ALOGV("Duplex callback: input frames: %Iu, output frames: %Iu",
           input_frames, output_frames);
 
-    refill(stm, stm->linear_input_buffer->data(), input_frames, output_buffer,
+    got = refill(stm, stm->linear_input_buffer->data(), input_frames,
-           output_frames);
+                 output_buffer, output_frames);
   }
 
   stm->linear_input_buffer->clear();
@@ -1219,6 +1244,9 @@ refill_callback_duplex(cubeb_stream * stm)
     LOG("failed to release buffer: %lx", hr);
     return false;
   }
+  if (got < 0) {
+    return false;
+  }
   return true;
 }
 
@@ -1245,8 +1273,9 @@ refill_callback_input(cubeb_stream * stm)
 
   long read =
       refill(stm, stm->linear_input_buffer->data(), input_frames, nullptr, 0);
-
+  if (read < 0) {
-  XASSERT(read >= 0);
+    return false;
+  }
 
   stm->linear_input_buffer->clear();
 
@@ -1276,8 +1305,9 @@ refill_callback_output(cubeb_stream * stm)
 
   ALOGV("Output callback: output frames requested: %Iu, got %ld", output_frames,
         got);
-
+  if (got < 0) {
-  XASSERT(got >= 0);
+    return false;
+  }
   XASSERT(size_t(got) == output_frames || stm->draining);
 
   hr = stm->render_client->ReleaseBuffer(got, 0);
@@ -1289,17 +1319,25 @@ refill_callback_output(cubeb_stream * stm)
   return size_t(got) == output_frames || stm->draining;
 }
 
+void
+wasapi_stream_destroy(cubeb_stream * stm);
+
+static void
+handle_emergency_bailout(cubeb_stream * stm)
+{
+  if (stm->emergency_bailout) {
+    CloseHandle(stm->thread);
+    stm->thread = NULL;
+    CloseHandle(stm->shutdown_event);
+    stm->shutdown_event = 0;
+    wasapi_stream_destroy(stm);
+    _endthreadex(0);
+  }
+}
+
 static unsigned int __stdcall wasapi_stream_render_loop(LPVOID stream)
 {
   cubeb_stream * stm = static_cast<cubeb_stream *>(stream);
-  std::atomic<bool> * emergency_bailout = stm->emergency_bailout;
-
-  // Signal wasapi_stream_start that we've copied emergency_bailout.
-  BOOL ok = SetEvent(stm->thread_ready_event);
-  if (!ok) {
-    LOG("thread_ready SetEvent failed: %lx", GetLastError());
-    return 0;
-  }
 
   bool is_playing = true;
   HANDLE wait_array[4] = {stm->shutdown_event, stm->reconfigure_event,
@@ -1332,20 +1370,10 @@ static unsigned int __stdcall wasapi_stream_render_loop(LPVOID stream)
   unsigned timeout_count = 0;
   const unsigned timeout_limit = 3;
   while (is_playing) {
-    // We want to check the emergency bailout variable before a
+    handle_emergency_bailout(stm);
-    // and after the WaitForMultipleObject, because the handles
-    // WaitForMultipleObjects is going to wait on might have been closed
-    // already.
-    if (*emergency_bailout) {
-      delete emergency_bailout;
-      return 0;
-    }
     DWORD waitResult = WaitForMultipleObjects(ARRAY_LENGTH(wait_array),
                                               wait_array, FALSE, 1000);
-    if (*emergency_bailout) {
+    handle_emergency_bailout(stm);
-      delete emergency_bailout;
-      return 0;
-    }
     if (waitResult != WAIT_TIMEOUT) {
       timeout_count = 0;
     }
@@ -1355,7 +1383,7 @@ static unsigned int __stdcall wasapi_stream_render_loop(LPVOID stream)
       /* We don't check if the drain is actually finished here, we just want to
          shutdown. */
       if (stm->draining) {
-        stm->state_callback(stm, stm->user_ptr, CUBEB_STATE_DRAINED);
+        wasapi_state_callback(stm, stm->user_ptr, CUBEB_STATE_DRAINED);
       }
       continue;
     }
@@ -1417,7 +1445,8 @@ static unsigned int __stdcall wasapi_stream_render_loop(LPVOID stream)
     case WAIT_OBJECT_0 + 3: { /* input available */
       HRESULT rv = get_input_buffer(stm);
       if (FAILED(rv)) {
-        return rv;
+        is_playing = false;
+        continue;
       }
 
       if (!has_output(stm)) {
@@ -1436,18 +1465,29 @@ static unsigned int __stdcall wasapi_stream_render_loop(LPVOID stream)
       break;
     default:
       LOG("case %lu not handled in render loop.", waitResult);
-      abort();
+      XASSERT(false);
     }
   }
 
-  if (FAILED(hr)) {
+  // Stop audio clients since this thread will no longer service
-    stm->state_callback(stm, stm->user_ptr, CUBEB_STATE_ERROR);
+  // the events.
+  if (stm->output_client) {
+    stm->output_client->Stop();
+  }
+  if (stm->input_client) {
+    stm->input_client->Stop();
   }
 
   if (mmcss_handle) {
     AvRevertMmThreadCharacteristics(mmcss_handle);
   }
 
+  handle_emergency_bailout(stm);
+
+  if (FAILED(hr)) {
+    wasapi_state_callback(stm, stm->user_ptr, CUBEB_STATE_ERROR);
+  }
+
   return 0;
 }
 
@@ -1696,54 +1736,58 @@ wasapi_init(cubeb ** context, char const * context_name)
 }
 
 namespace {
+enum ShutdownPhase { OnStop, OnDestroy };
+
 bool
-stop_and_join_render_thread(cubeb_stream * stm)
+stop_and_join_render_thread(cubeb_stream * stm, ShutdownPhase phase)
 {
-  bool rv = true;
+  // Only safe to transfer `stm` ownership to the render thread when
-  LOG("Stop and join render thread.");
+  // the stream is being destroyed by the caller.
+  bool bailout = phase == OnDestroy;
+
+  LOG("%p: Stop and join render thread: %p (%d), phase=%d", stm, stm->thread,
+      stm->emergency_bailout.load(), static_cast<int>(phase));
   if (!stm->thread) {
-    LOG("No thread present.");
     return true;
   }
 
-  // If we've already leaked the thread, just return,
+  XASSERT(!stm->emergency_bailout);
-  // there is not much we can do.
-  if (!stm->emergency_bailout.load()) {
-    return false;
-  }
 
   BOOL ok = SetEvent(stm->shutdown_event);
   if (!ok) {
-    LOG("Destroy SetEvent failed: %lx", GetLastError());
+    LOG("stop_and_join_render_thread: SetEvent failed: %lx", GetLastError());
+    stm->emergency_bailout = bailout;
+    return false;
   }
 
   /* Wait five seconds for the rendering thread to return. It's supposed to
-   * check its event loop very often, five seconds is rather conservative. */
+   * check its event loop very often, five seconds is rather conservative.
-  DWORD r = WaitForSingleObject(stm->thread, 5000);
+   * Note: 5*1s loop to work around timer sleep issues on pre-Windows 8. */
+  DWORD r;
+  for (int i = 0; i < 5; ++i) {
+    r = WaitForSingleObject(stm->thread, 1000);
+    if (r == WAIT_OBJECT_0) {
+      break;
+    }
+  }
   if (r != WAIT_OBJECT_0) {
-    /* Something weird happened, leak the thread and continue the shutdown
+    LOG("stop_and_join_render_thread: WaitForSingleObject on thread failed: "
-     * process. */
+        "%lx, %lx",
-    *(stm->emergency_bailout) = true;
+        r, GetLastError());
-    // We give the ownership to the rendering thread.
+    stm->emergency_bailout = bailout;
-    stm->emergency_bailout = nullptr;
+    return false;
-    LOG("Destroy WaitForSingleObject on thread failed: %lx, %lx", r,
-        GetLastError());
-    rv = false;
   }
 
-  // Only attempts to close and null out the thread and event if the
+  // Only attempt to close and null out the thread and event if the
-  // WaitForSingleObject above succeeded, so that calling this function again
+  // WaitForSingleObject above succeeded.
-  // attemps to clean up the thread and event each time.
+  LOG("stop_and_join_render_thread: Closing thread.");
-  if (rv) {
+  CloseHandle(stm->thread);
-    LOG("Closing thread.");
+  stm->thread = NULL;
-    CloseHandle(stm->thread);
-    stm->thread = NULL;
 
-    CloseHandle(stm->shutdown_event);
+  CloseHandle(stm->shutdown_event);
-    stm->shutdown_event = 0;
+  stm->shutdown_event = 0;
-  }
 
-  return rv;
+  return true;
 }
 
 void
@@ -1881,9 +1925,6 @@ wasapi_get_preferred_sample_rate(cubeb * ctx, uint32_t * rate)
   return CUBEB_OK;
 }
 
-void
-wasapi_stream_destroy(cubeb_stream * stm);
-
 static void
 waveformatex_update_derived_properties(WAVEFORMATEX * format)
 {
@@ -2259,8 +2300,12 @@ setup_wasapi_stream_one_side(cubeb_stream * stm,
     if (wasapi_create_device(stm->context, device_info,
                              stm->device_enumerator.get(), device.get(),
                              &default_devices) == CUBEB_OK) {
+      if (device_info.latency_hi == 0) {
+        LOG("Input: could not query latency_hi to guess safe latency");
+        wasapi_destroy_device(&device_info);
+        return CUBEB_ERROR;
+      }
       // This multiplicator has been found empirically.
-      XASSERT(device_info.latency_hi > 0);
       uint32_t latency_frames = device_info.latency_hi * 8;
       LOG("Input: latency increased to %u frames from a default of %u",
           latency_frames, device_info.latency_hi);
@@ -2362,10 +2407,10 @@ wasapi_find_bt_handsfree_output_device(cubeb_stream * stm)
     return nullptr;
   }
 
-  int rv = wasapi_enumerate_devices(
+  int rv = wasapi_enumerate_devices_internal(
       stm->context,
       (cubeb_device_type)(CUBEB_DEVICE_TYPE_INPUT | CUBEB_DEVICE_TYPE_OUTPUT),
-      &collection);
+      &collection, DEVICE_STATE_ACTIVE);
   if (rv != CUBEB_OK) {
     return nullptr;
   }
@@ -2561,7 +2606,7 @@ setup_wasapi_stream(cubeb_stream * stm)
   stm->resampler.reset(cubeb_resampler_create(
       stm, has_input(stm) ? &input_params : nullptr,
       has_output(stm) && !stm->has_dummy_output ? &output_params : nullptr,
-      target_sample_rate, stm->data_callback, stm->user_ptr,
+      target_sample_rate, wasapi_data_callback, stm->user_ptr,
       stm->voice ? CUBEB_RESAMPLER_QUALITY_VOIP
                  : CUBEB_RESAMPLER_QUALITY_DESKTOP,
       CUBEB_RESAMPLER_RECLOCK_NONE));
@@ -2794,33 +2839,25 @@ wasapi_stream_destroy(cubeb_stream * stm)
   XASSERT(stm);
   LOG("Stream destroy (%p)", stm);
 
-  // Only free stm->emergency_bailout if we could join the thread.
+  if (!stop_and_join_render_thread(stm, OnDestroy)) {
-  // If we could not join the thread, stm->emergency_bailout is true
+    // Emergency bailout: render thread becomes responsible for calling
-  // and is still alive until the thread wakes up and exits cleanly.
+    // wasapi_stream_destroy.
-  if (stop_and_join_render_thread(stm)) {
+    return;
-    delete stm->emergency_bailout.load();
-    stm->emergency_bailout = nullptr;
   }
 
   if (stm->notification_client) {
     unregister_notification_client(stm);
   }
 
-  CloseHandle(stm->reconfigure_event);
-  CloseHandle(stm->refill_event);
-  CloseHandle(stm->input_available_event);
-
-  // The variables intialized in wasapi_stream_init,
-  // must be destroyed in wasapi_stream_destroy.
-  stm->linear_input_buffer.reset();
-
-  stm->device_enumerator = nullptr;
-
   {
     auto_lock lock(stm->stream_reset_lock);
     close_wasapi_stream(stm);
   }
 
+  CloseHandle(stm->reconfigure_event);
+  CloseHandle(stm->refill_event);
+  CloseHandle(stm->input_available_event);
+
   delete stm;
 }
 
@@ -2875,8 +2912,6 @@ wasapi_stream_start(cubeb_stream * stm)
   XASSERT(stm && !stm->thread && !stm->shutdown_event);
   XASSERT(stm->output_client || stm->input_client);
 
-  stm->emergency_bailout = new std::atomic<bool>(false);
-
   if (stm->output_client) {
     int rv = stream_start_one_side(stm, OUTPUT);
     if (rv != CUBEB_OK) {
@@ -2897,30 +2932,18 @@ wasapi_stream_start(cubeb_stream * stm)
     return CUBEB_ERROR;
   }
 
-  stm->thread_ready_event = CreateEvent(NULL, 0, 0, NULL);
-  if (!stm->thread_ready_event) {
-    LOG("Can't create the thread_ready event, error: %lx", GetLastError());
-    return CUBEB_ERROR;
-  }
-
   cubeb_async_log_reset_threads();
   stm->thread =
       (HANDLE)_beginthreadex(NULL, 512 * 1024, wasapi_stream_render_loop, stm,
                              STACK_SIZE_PARAM_IS_A_RESERVATION, NULL);
   if (stm->thread == NULL) {
     LOG("could not create WASAPI render thread.");
+    CloseHandle(stm->shutdown_event);
+    stm->shutdown_event = 0;
     return CUBEB_ERROR;
   }
 
-  // Wait for wasapi_stream_render_loop to signal that emergency_bailout has
+  wasapi_state_callback(stm, stm->user_ptr, CUBEB_STATE_STARTED);
-  // been read, avoiding a bailout situation where we could free `stm`
-  // before wasapi_stream_render_loop had a chance to run.
-  HRESULT hr = WaitForSingleObject(stm->thread_ready_event, INFINITE);
-  XASSERT(hr == WAIT_OBJECT_0);
-  CloseHandle(stm->thread_ready_event);
-  stm->thread_ready_event = 0;
-
-  stm->state_callback(stm, stm->user_ptr, CUBEB_STATE_STARTED);
 
   return CUBEB_OK;
 }
@@ -2950,15 +2973,12 @@ wasapi_stream_stop(cubeb_stream * stm)
       }
     }
 
-    stm->state_callback(stm, stm->user_ptr, CUBEB_STATE_STOPPED);
+    wasapi_state_callback(stm, stm->user_ptr, CUBEB_STATE_STOPPED);
   }
 
-  if (stop_and_join_render_thread(stm)) {
+  if (!stop_and_join_render_thread(stm, OnStop)) {
-    delete stm->emergency_bailout.load();
-    stm->emergency_bailout = nullptr;
-  } else {
     // If we could not join the thread, put the stream in error.
-    stm->state_callback(stm, stm->user_ptr, CUBEB_STATE_ERROR);
+    wasapi_state_callback(stm, stm->user_ptr, CUBEB_STATE_ERROR);
     return CUBEB_ERROR;
   }
 
@@ -3324,8 +3344,9 @@ wasapi_destroy_device(cubeb_device_info * device)
 }
 
 static int
-wasapi_enumerate_devices(cubeb * context, cubeb_device_type type,
+wasapi_enumerate_devices_internal(cubeb * context, cubeb_device_type type,
-                         cubeb_device_collection * out)
+                                  cubeb_device_collection * out,
+                                  DWORD state_mask)
 {
   com_ptr<IMMDeviceEnumerator> enumerator;
   com_ptr<IMMDeviceCollection> collection;
@@ -3353,8 +3374,7 @@ wasapi_enumerate_devices(cubeb * context, cubeb_device_type type,
     return CUBEB_ERROR;
   }
 
-  hr = enumerator->EnumAudioEndpoints(flow, DEVICE_STATEMASK_ALL,
+  hr = enumerator->EnumAudioEndpoints(flow, state_mask, collection.receive());
-                                      collection.receive());
   if (FAILED(hr)) {
     LOG("Could not enumerate audio endpoints: %lx", hr);
     return CUBEB_ERROR;
@@ -3388,6 +3408,15 @@ wasapi_enumerate_devices(cubeb * context, cubeb_device_type type,
   return CUBEB_OK;
 }
 
+static int
+wasapi_enumerate_devices(cubeb * context, cubeb_device_type type,
+                         cubeb_device_collection * out)
+{
+  return wasapi_enumerate_devices_internal(
+      context, type, out,
+      DEVICE_STATE_ACTIVE | DEVICE_STATE_DISABLED | DEVICE_STATE_UNPLUGGED);
+}
+
 static int
 wasapi_device_collection_destroy(cubeb * /*ctx*/,
                                  cubeb_device_collection * collection)

externals/cubeb/test/test_callback_ret.cpp

@@ -34,6 +34,7 @@ enum test_direction {
 struct user_state_callback_ret {
   std::atomic<int> cb_count{ 0 };
   std::atomic<int> expected_cb_count{ 0 };
+  std::atomic<int> error_state{ 0 };
 };
 
 // Data callback that always returns 0
@@ -98,10 +99,30 @@ long data_cb_ret_nframes(cubeb_stream * stream, void * user, const void * inputb
   return nframes;
 }
 
-void state_cb_ret(cubeb_stream * stream, void * /*user*/, cubeb_state state)
+// Data callback that always returns CUBEB_ERROR
+long
+data_cb_ret_error(cubeb_stream * stream, void * user, const void * inputbuffer,
+                 void * outputbuffer, long nframes)
+{
+  user_state_callback_ret * u = (user_state_callback_ret *)user;
+  // If this is the first time the callback has been called set our expected
+  // callback count
+  if (u->cb_count == 0) {
+    u->expected_cb_count = 1;
+  }
+  u->cb_count++;
+  if (nframes < 1) {
+    // This shouldn't happen
+    EXPECT_TRUE(false) << "nframes should not be 0 in data callback!";
+  }
+  return CUBEB_ERROR;
+}
+
+void state_cb_ret(cubeb_stream * stream, void * user, cubeb_state state)
 {
   if (stream == NULL)
     return;
+  user_state_callback_ret * u = (user_state_callback_ret *)user;
 
   switch (state) {
   case CUBEB_STATE_STARTED:
@@ -110,11 +131,13 @@ void state_cb_ret(cubeb_stream * stream, void * /*user*/, cubeb_state state)
     fprintf(stderr, "stream stopped\n"); break;
   case CUBEB_STATE_DRAINED:
     fprintf(stderr, "stream drained\n"); break;
+  case CUBEB_STATE_ERROR:
+    fprintf(stderr, "stream error\n");
+    u->error_state.fetch_add(1);
+    break;
   default:
     fprintf(stderr, "unknown stream state %d\n", state);
   }
-
-  return;
 }
 
 void run_test_callback(test_direction direction,
@@ -183,6 +206,10 @@ void run_test_callback(test_direction direction,
 
   ASSERT_EQ(user_state.expected_cb_count, user_state.cb_count) <<
     "Callback called unexpected number of times for " << test_desc << "!";
+  // TODO: On some test configurations, the data_callback is never called.
+  if (data_cb == data_cb_ret_error && user_state.cb_count != 0) {
+    ASSERT_EQ(user_state.error_state, 1) << "Callback expected error state";
+  }
 }
 
 TEST(cubeb, test_input_callback)
@@ -190,6 +217,8 @@ TEST(cubeb, test_input_callback)
   run_test_callback(INPUT_ONLY, data_cb_ret_zero, "input only, return 0");
   run_test_callback(INPUT_ONLY, data_cb_ret_nframes_minus_one, "input only, return nframes - 1");
   run_test_callback(INPUT_ONLY, data_cb_ret_nframes, "input only, return nframes");
+  run_test_callback(INPUT_ONLY, data_cb_ret_error,
+                    "input only, return CUBEB_ERROR");
 }
 
 TEST(cubeb, test_output_callback)
@@ -197,6 +226,8 @@ TEST(cubeb, test_output_callback)
   run_test_callback(OUTPUT_ONLY, data_cb_ret_zero, "output only, return 0");
   run_test_callback(OUTPUT_ONLY, data_cb_ret_nframes_minus_one, "output only, return nframes - 1");
   run_test_callback(OUTPUT_ONLY, data_cb_ret_nframes, "output only, return nframes");
+  run_test_callback(OUTPUT_ONLY, data_cb_ret_error,
+                    "output only, return CUBEB_ERROR");
 }
 
 TEST(cubeb, test_duplex_callback)
@@ -204,4 +235,5 @@ TEST(cubeb, test_duplex_callback)
   run_test_callback(DUPLEX, data_cb_ret_zero, "duplex, return 0");
   run_test_callback(DUPLEX, data_cb_ret_nframes_minus_one, "duplex, return nframes - 1");
   run_test_callback(DUPLEX, data_cb_ret_nframes, "duplex, return nframes");
+  run_test_callback(DUPLEX, data_cb_ret_error, "duplex, return CUBEB_ERROR");
 }

src/audio_core/audio_core.cpp

@@ -57,4 +57,12 @@ void AudioCore::PauseSinks(const bool pausing) const {
     }
 }
 
+u32 AudioCore::GetStreamQueue() const {
+    return estimated_queue.load();
+}
+
+void AudioCore::SetStreamQueue(u32 size) {
+    estimated_queue.store(size);
+}
+
 } // namespace AudioCore

src/audio_core/audio_core.h

@@ -65,6 +65,20 @@ public:
      */
     void PauseSinks(bool pausing) const;
 
+    /**
+     * Get the size of the current stream queue.
+     *
+     * @return Current stream queue size.
+     */
+    u32 GetStreamQueue() const;
+
+    /**
+     * Get the size of the current stream queue.
+     *
+     * @param size - New stream size.
+     */
+    void SetStreamQueue(u32 size);
+
 private:
     /**
      * Create the sinks on startup.
@@ -79,6 +93,8 @@ private:
     std::unique_ptr<Sink::Sink> input_sink;
     /// The ADSP in the sysmodule
     std::unique_ptr<AudioRenderer::ADSP::ADSP> adsp;
+    /// Current size of the stream queue
+    std::atomic<u32> estimated_queue{0};
 };
 
 } // namespace AudioCore

src/audio_core/audio_in_manager.cpp

@@ -5,6 +5,8 @@
 #include "audio_core/audio_in_manager.h"
 #include "audio_core/audio_manager.h"
 #include "audio_core/in/audio_in.h"
+#include "audio_core/sink/sink_details.h"
+#include "common/settings.h"
 #include "core/core.h"
 #include "core/hle/service/audio/errors.h"
 
@@ -78,9 +80,12 @@ u32 Manager::GetDeviceNames(std::vector<AudioRenderer::AudioDevice::AudioDeviceN
 
     LinkToManager();
 
-    names.push_back(AudioRenderer::AudioDevice::AudioDeviceName("Uac"));
+    auto input_devices{Sink::GetDeviceListForSink(Settings::values.sink_id.GetValue(), true)};
-
+    if (input_devices.size() > 1) {
-    return 1;
+        names.push_back(AudioRenderer::AudioDevice::AudioDeviceName("Uac"));
+        return 1;
+    }
+    return 0;
 }
 
 } // namespace AudioCore::AudioIn

src/audio_core/device/device_session.cpp

@@ -37,13 +37,16 @@ Result DeviceSession::Initialize(std::string_view name_, SampleFormat sample_for
         sink = &system.AudioCore().GetOutputSink();
     }
     stream = sink->AcquireSinkStream(system, channel_count, name, type);
+    initialized = true;
     return ResultSuccess;
 }
 
 void DeviceSession::Finalize() {
-    Stop();
+    if (initialized) {
-    sink->CloseStream(stream);
+        Stop();
-    stream = nullptr;
+        sink->CloseStream(stream);
+        stream = nullptr;
+    }
 }
 
 void DeviceSession::Start() {

src/audio_core/device/device_session.h

@@ -100,7 +100,7 @@ private:
     /// System
     Core::System& system;
     /// Output sink this device will use
-    Sink::Sink* sink;
+    Sink::Sink* sink{};
     /// The backend stream for this device session to send samples to
     Sink::SinkStream* stream{};
     /// Name of this device session
@@ -119,6 +119,8 @@ private:
     u64 applet_resource_user_id{};
     /// Total number of samples played by this device session
     u64 played_sample_count{};
+    /// Is this session initialised?
+    bool initialized{};
 };
 
 } // namespace AudioCore

src/audio_core/renderer/adsp/audio_renderer.cpp

@@ -128,8 +128,8 @@ void AudioRenderer::CreateSinkStreams() {
     u32 channels{sink.GetDeviceChannels()};
     for (u32 i = 0; i < MaxRendererSessions; i++) {
         std::string name{fmt::format("ADSP_RenderStream-{}", i)};
-        streams[i] = sink.AcquireSinkStream(system, channels, name,
+        streams[i] =
-                                            ::AudioCore::Sink::StreamType::Render, &render_event);
+            sink.AcquireSinkStream(system, channels, name, ::AudioCore::Sink::StreamType::Render);
         streams[i]->SetSystemChannels(streams[i]->GetDeviceChannels());
     }
 }
@@ -199,12 +199,9 @@ void AudioRenderer::ThreadFunc() {
                             command_list_processor.Process(index) - start_time;
                     }
 
-                    // If the stream queue is building up too much, wait for a signal
+                    if (index == 0) {
-                    // from the backend that a buffer was consumed.
+                        auto stream{command_list_processor.GetOutputSinkStream()};
-                    // In practice this will wait longer than 1 buffer due to timing.
+                        system.AudioCore().SetStreamQueue(stream->GetQueueSize());
-                    auto stream{command_list_processor.GetOutputSinkStream()};
-                    if (stream->GetQueueSize() >= 4) {
-                        render_event.WaitFor(std::chrono::milliseconds(5));
                     }
 
                     const auto end_time{system.CoreTiming().GetClockTicks()};

src/audio_core/renderer/adsp/audio_renderer.h

@@ -14,8 +14,11 @@
 #include "common/thread.h"
 
 namespace Core {
-class System;
+namespace Timing {
+struct EventType;
 }
+class System;
+} // namespace Core
 
 namespace AudioCore {
 namespace Sink {
@@ -194,8 +197,6 @@ private:
     Sink::Sink& sink;
     /// The streams which will receive the processed samples
     std::array<Sink::SinkStream*, MaxRendererSessions> streams;
-    /// An event signalled from the backend when a buffer is consumed, used for timing.
-    Common::Event render_event{};
 };
 
 } // namespace AudioRenderer::ADSP

src/audio_core/renderer/command/effect/light_limiter.cpp

@@ -63,7 +63,8 @@ static void ApplyLightLimiterEffect(const LightLimiterInfo::ParameterVersion2& p
 
         for (u32 sample_index = 0; sample_index < sample_count; sample_index++) {
             for (u32 channel = 0; channel < params.channel_count; channel++) {
-                auto sample{Common::FixedPoint<49, 15>(inputs[channel][sample_index]) *
+                auto sample{(Common::FixedPoint<49, 15>(inputs[channel][sample_index]) /
+                             Common::FixedPoint<49, 15>::one) *
                             params.input_gain};
                 auto abs_sample{sample};
                 if (sample < 0.0f) {
@@ -85,15 +86,17 @@ static void ApplyLightLimiterEffect(const LightLimiterInfo::ParameterVersion2& p
                 auto lookahead_sample{
                     state.look_ahead_sample_buffers[channel]
                                                    [state.look_ahead_sample_offsets[channel]]};
+
                 state.look_ahead_sample_buffers[channel][state.look_ahead_sample_offsets[channel]] =
                     sample;
                 state.look_ahead_sample_offsets[channel] =
                     (state.look_ahead_sample_offsets[channel] + 1) % params.look_ahead_samples_min;
 
-                outputs[channel][sample_index] = static_cast<s32>(std::clamp(
+                outputs[channel][sample_index] = static_cast<s32>(
-                    (lookahead_sample * state.compression_gain[channel] * params.output_gain)
+                    std::clamp((lookahead_sample * state.compression_gain[channel] *
-                        .to_long(),
+                                params.output_gain * Common::FixedPoint<49, 15>::one)
-                    min, max));
+                                   .to_long(),
+                               min, max));
 
                 if (statistics) {
                     statistics->channel_max_sample[channel] =

src/audio_core/renderer/system_manager.cpp

@@ -15,12 +15,17 @@ MICROPROFILE_DEFINE(Audio_RenderSystemManager, "Audio", "Render System Manager",
                     MP_RGB(60, 19, 97));
 
 namespace AudioCore::AudioRenderer {
+constexpr std::chrono::nanoseconds BaseRenderTime{5'000'000UL};
+constexpr std::chrono::nanoseconds RenderTimeOffset{400'000UL};
 
 SystemManager::SystemManager(Core::System& core_)
     : core{core_}, adsp{core.AudioCore().GetADSP()}, mailbox{adsp.GetRenderMailbox()},
       thread_event{Core::Timing::CreateEvent(
-          "AudioRendererSystemManager",
+          "AudioRendererSystemManager", [this](std::uintptr_t, s64 time, std::chrono::nanoseconds) {
-          [this](std::uintptr_t userdata, std::chrono::nanoseconds ns_late) { ThreadFunc2(); })} {}
+              return ThreadFunc2(time);
+          })} {
+    core.CoreTiming().RegisterPauseCallback([this](bool paused) { PauseCallback(paused); });
+}
 
 SystemManager::~SystemManager() {
     Stop();
@@ -30,9 +35,9 @@ bool SystemManager::InitializeUnsafe() {
     if (!active) {
         if (adsp.Start()) {
             active = true;
-            core.CoreTiming().ScheduleLoopingEvent(std::chrono::nanoseconds(2'304'000ULL * 2),
-                                                   thread_event);
             thread = std::jthread([this](std::stop_token stop_token) { ThreadFunc(); });
+            core.CoreTiming().ScheduleLoopingEvent(std::chrono::nanoseconds(0),
+                                                   BaseRenderTime - RenderTimeOffset, thread_event);
         }
     }
 
@@ -95,11 +100,13 @@ void SystemManager::ThreadFunc() {
     constexpr char name[]{"yuzu:AudioRenderSystemManager"};
     MicroProfileOnThreadCreate(name);
     Common::SetCurrentThreadName(name);
-    Common::SetCurrentThreadPriority(Common::ThreadPriority::Critical);
+    Common::SetCurrentThreadPriority(Common::ThreadPriority::High);
     while (active) {
         {
             std::scoped_lock l{mutex1};
+
             MICROPROFILE_SCOPE(Audio_RenderSystemManager);
+
             for (auto system : systems) {
                 system->SendCommandToDsp();
             }
@@ -113,9 +120,43 @@ void SystemManager::ThreadFunc() {
     }
 }
 
-void SystemManager::ThreadFunc2() {
+std::optional<std::chrono::nanoseconds> SystemManager::ThreadFunc2(s64 time) {
+    std::optional<std::chrono::nanoseconds> new_schedule_time{std::nullopt};
+    const auto queue_size{core.AudioCore().GetStreamQueue()};
+    switch (state) {
+    case StreamState::Filling:
+        if (queue_size >= 5) {
+            new_schedule_time = BaseRenderTime;
+            state = StreamState::Steady;
+        }
+        break;
+    case StreamState::Steady:
+        if (queue_size <= 2) {
+            new_schedule_time = BaseRenderTime - RenderTimeOffset;
+            state = StreamState::Filling;
+        } else if (queue_size > 5) {
+            new_schedule_time = BaseRenderTime + RenderTimeOffset;
+            state = StreamState::Draining;
+        }
+        break;
+    case StreamState::Draining:
+        if (queue_size <= 5) {
+            new_schedule_time = BaseRenderTime;
+            state = StreamState::Steady;
+        }
+        break;
+    }
+
     update.store(true);
     update.notify_all();
+    return new_schedule_time;
+}
+
+void SystemManager::PauseCallback(bool paused) {
+    if (paused && core.IsPoweredOn() && core.IsShuttingDown()) {
+        update.store(true);
+        update.notify_all();
+    }
 }
 
 } // namespace AudioCore::AudioRenderer

src/audio_core/renderer/system_manager.h

@@ -6,6 +6,7 @@
 #include <list>
 #include <memory>
 #include <mutex>
+#include <optional>
 #include <thread>
 
 #include "audio_core/renderer/system.h"
@@ -70,7 +71,20 @@ private:
     /**
      * Signalling core timing thread to run ThreadFunc.
      */
-    void ThreadFunc2();
+    std::optional<std::chrono::nanoseconds> ThreadFunc2(s64 time);
+
+    /**
+     * Callback from core timing when pausing, used to detect shutdowns and stop ThreadFunc.
+     *
+     * @param paused - Are we pausing or resuming?
+     */
+    void PauseCallback(bool paused);
+
+    enum class StreamState {
+        Filling,
+        Steady,
+        Draining,
+    };
 
     /// Core system
     Core::System& core;
@@ -92,6 +106,8 @@ private:
     std::shared_ptr<Core::Timing::EventType> thread_event;
     /// Atomic for main thread to wait on
     std::atomic<bool> update{};
+    /// Current state of the streams
+    StreamState state{StreamState::Filling};
 };
 
 } // namespace AudioCore::AudioRenderer

src/audio_core/sink/cubeb_sink.cpp

@@ -44,8 +44,8 @@ public:
      */
     CubebSinkStream(cubeb* ctx_, const u32 device_channels_, const u32 system_channels_,
                     cubeb_devid output_device, cubeb_devid input_device, const std::string& name_,
-                    const StreamType type_, Core::System& system_, Common::Event* event)
+                    const StreamType type_, Core::System& system_)
-        : ctx{ctx_}, type{type_}, system{system_}, render_event{event} {
+        : ctx{ctx_}, type{type_}, system{system_} {
 #ifdef _WIN32
         CoInitializeEx(nullptr, COINIT_MULTITHREADED);
 #endif
@@ -306,7 +306,6 @@ private:
             manager.SetEvent(Event::Type::AudioInManager, true);
             break;
         case StreamType::Render:
-            render_event->Set();
             break;
         }
     }
@@ -469,8 +468,6 @@ private:
     ::AudioCore::Sink::SinkBuffer released_buffer{};
     /// The last played (or received) frame of audio, used when the callback underruns
     std::array<s16, MaxChannels> last_frame{};
-    /// Audio render-only event, signalled when a render buffer is consumed
-    Common::Event* render_event;
 };
 
 CubebSink::CubebSink(std::string_view target_device_name) {
@@ -525,11 +522,9 @@ CubebSink::~CubebSink() {
 }
 
 SinkStream* CubebSink::AcquireSinkStream(Core::System& system, const u32 system_channels,
-                                         const std::string& name, const StreamType type,
+                                         const std::string& name, const StreamType type) {
-                                         Common::Event* event) {
+    SinkStreamPtr& stream = sink_streams.emplace_back(std::make_unique<CubebSinkStream>(
-    SinkStreamPtr& stream = sink_streams.emplace_back(
+        ctx, device_channels, system_channels, output_device, input_device, name, type, system));
-        std::make_unique<CubebSinkStream>(ctx, device_channels, system_channels, output_device,
-                                          input_device, name, type, system, event));
 
     return stream.get();
 }

src/audio_core/sink/cubeb_sink.h

@@ -39,8 +39,7 @@ public:
      * @return A pointer to the created SinkStream
      */
     SinkStream* AcquireSinkStream(Core::System& system, u32 system_channels,
-                                  const std::string& name, StreamType type,
+                                  const std::string& name, StreamType type) override;
-                                  Common::Event* event = nullptr) override;
 
     /**
      * Close a given stream.

src/audio_core/sink/null_sink.h

@@ -18,8 +18,7 @@ public:
     SinkStream* AcquireSinkStream([[maybe_unused]] Core::System& system,
                                   [[maybe_unused]] u32 system_channels,
                                   [[maybe_unused]] const std::string& name,
-                                  [[maybe_unused]] StreamType type,
+                                  [[maybe_unused]] StreamType type) override {
-                                  [[maybe_unused]] Common::Event* event = nullptr) override {
         return &null_sink_stream;
     }
 

src/audio_core/sink/sdl2_sink.cpp

@@ -46,8 +46,8 @@ public:
      */
     SDLSinkStream(u32 device_channels_, const u32 system_channels_,
                   const std::string& output_device, const std::string& input_device,
-                  const StreamType type_, Core::System& system_, Common::Event* event)
+                  const StreamType type_, Core::System& system_)
-        : type{type_}, system{system_}, render_event{event} {
+        : type{type_}, system{system_} {
         system_channels = system_channels_;
         device_channels = device_channels_;
 
@@ -279,7 +279,6 @@ private:
             manager.SetEvent(Event::Type::AudioInManager, true);
             break;
         case StreamType::Render:
-            render_event->Set();
             break;
         }
     }
@@ -417,8 +416,6 @@ private:
     ::AudioCore::Sink::SinkBuffer released_buffer{};
     /// The last played (or received) frame of audio, used when the callback underruns
     std::array<s16, MaxChannels> last_frame{};
-    /// Audio render-only event, signalled when a render buffer is consumed
-    Common::Event* render_event;
 };
 
 SDLSink::SDLSink(std::string_view target_device_name) {
@@ -441,10 +438,9 @@ SDLSink::SDLSink(std::string_view target_device_name) {
 SDLSink::~SDLSink() = default;
 
 SinkStream* SDLSink::AcquireSinkStream(Core::System& system, const u32 system_channels,
-                                       const std::string&, const StreamType type,
+                                       const std::string&, const StreamType type) {
-                                       Common::Event* event) {
     SinkStreamPtr& stream = sink_streams.emplace_back(std::make_unique<SDLSinkStream>(
-        device_channels, system_channels, output_device, input_device, type, system, event));
+        device_channels, system_channels, output_device, input_device, type, system));
     return stream.get();
 }
 

src/audio_core/sink/sdl2_sink.h

@@ -37,8 +37,7 @@ public:
      * @return A pointer to the created SinkStream
      */
     SinkStream* AcquireSinkStream(Core::System& system, u32 system_channels,
-                                  const std::string& name, StreamType type,
+                                  const std::string& name, StreamType type) override;
-                                  Common::Event* event = nullptr) override;
 
     /**
      * Close a given stream.

src/audio_core/sink/sink.h

@@ -63,8 +63,7 @@ public:
      * @return A pointer to the created SinkStream
      */
     virtual SinkStream* AcquireSinkStream(Core::System& system, u32 system_channels,
-                                          const std::string& name, StreamType type,
+                                          const std::string& name, StreamType type) = 0;
-                                          Common::Event* event = nullptr) = 0;
 
     /**
      * Get the number of channels the hardware device supports.

src/core/core.cpp

@@ -139,7 +139,6 @@ struct System::Impl {
 
         kernel.Suspend(false);
         core_timing.SyncPause(false);
-        cpu_manager.Pause(false);
         is_paused = false;
 
         audio_core->PauseSinks(false);
@@ -155,7 +154,6 @@ struct System::Impl {
 
         core_timing.SyncPause(true);
         kernel.Suspend(true);
-        cpu_manager.Pause(true);
         is_paused = true;
 
         return status;
@@ -170,7 +168,6 @@ struct System::Impl {
         std::unique_lock<std::mutex> lk(suspend_guard);
         kernel.Suspend(true);
         core_timing.SyncPause(true);
-        cpu_manager.Pause(true);
         return lk;
     }
 
@@ -178,7 +175,6 @@ struct System::Impl {
         if (!is_paused) {
             core_timing.SyncPause(false);
             kernel.Suspend(false);
-            cpu_manager.Pause(false);
         }
     }
 
@@ -348,13 +344,14 @@ struct System::Impl {
             gpu_core->NotifyShutdown();
         }
 
+        kernel.ShutdownCores();
+        cpu_manager.Shutdown();
         debugger.reset();
         kernel.CloseServices();
         services.reset();
         service_manager.reset();
         cheat_engine.reset();
         telemetry_session.reset();
-        cpu_manager.Shutdown();
         time_manager.Shutdown();
         core_timing.Shutdown();
         app_loader.reset();

src/core/core_timing.cpp

@@ -22,11 +22,11 @@ std::shared_ptr<EventType> CreateEvent(std::string name, TimedCallback&& callbac
 }
 
 struct CoreTiming::Event {
-    u64 time;
+    s64 time;
     u64 fifo_order;
     std::uintptr_t user_data;
     std::weak_ptr<EventType> type;
-    u64 reschedule_time;
+    s64 reschedule_time;
 
     // Sort by time, unless the times are the same, in which case sort by
     // the order added to the queue
@@ -59,7 +59,8 @@ void CoreTiming::Initialize(std::function<void()>&& on_thread_init_) {
     event_fifo_id = 0;
     shutting_down = false;
     ticks = 0;
-    const auto empty_timed_callback = [](std::uintptr_t, std::chrono::nanoseconds) {};
+    const auto empty_timed_callback = [](std::uintptr_t, u64, std::chrono::nanoseconds)
+        -> std::optional<std::chrono::nanoseconds> { return std::nullopt; };
     ev_lost = CreateEvent("_lost_event", empty_timed_callback);
     if (is_multicore) {
         worker_threads.emplace_back(ThreadEntry, std::ref(*this), 0);
@@ -77,6 +78,7 @@ void CoreTiming::Shutdown() {
         thread.join();
     }
     worker_threads.clear();
+    pause_callbacks.clear();
     ClearPendingEvents();
     has_started = false;
 }
@@ -94,6 +96,14 @@ void CoreTiming::Pause(bool is_paused_) {
         }
     }
     paused_state.store(is_paused_, std::memory_order_relaxed);
+
+    if (!is_paused_) {
+        pause_end_time = GetGlobalTimeNs().count();
+    }
+
+    for (auto& cb : pause_callbacks) {
+        cb(is_paused_);
+    }
 }
 
 void CoreTiming::SyncPause(bool is_paused_) {
@@ -117,6 +127,14 @@ void CoreTiming::SyncPause(bool is_paused_) {
             wait_signal_cv.wait(main_lock, [this] { return pause_count == 0; });
         }
     }
+
+    if (!is_paused_) {
+        pause_end_time = GetGlobalTimeNs().count();
+    }
+
+    for (auto& cb : pause_callbacks) {
+        cb(is_paused_);
+    }
 }
 
 bool CoreTiming::IsRunning() const {
@@ -130,12 +148,12 @@ bool CoreTiming::HasPendingEvents() const {
 
 void CoreTiming::ScheduleEvent(std::chrono::nanoseconds ns_into_future,
                                const std::shared_ptr<EventType>& event_type,
-                               std::uintptr_t user_data) {
+                               std::uintptr_t user_data, bool absolute_time) {
 
     std::unique_lock main_lock(event_mutex);
-    const u64 timeout = static_cast<u64>((GetGlobalTimeNs() + ns_into_future).count());
+    const auto next_time{absolute_time ? ns_into_future : GetGlobalTimeNs() + ns_into_future};
 
-    event_queue.emplace_back(Event{timeout, event_fifo_id++, user_data, event_type, 0});
+    event_queue.emplace_back(Event{next_time.count(), event_fifo_id++, user_data, event_type, 0});
     pending_events.fetch_add(1, std::memory_order_relaxed);
 
     std::push_heap(event_queue.begin(), event_queue.end(), std::greater<>());
@@ -145,30 +163,15 @@ void CoreTiming::ScheduleEvent(std::chrono::nanoseconds ns_into_future,
     }
 }
 
-void CoreTiming::ScheduleEventAt(std::chrono::nanoseconds next_time,
+void CoreTiming::ScheduleLoopingEvent(std::chrono::nanoseconds start_time,
-                                 const std::shared_ptr<EventType>& event_type,
+                                      std::chrono::nanoseconds resched_time,
-                                 std::uintptr_t user_data) {
-    std::unique_lock main_lock(event_mutex);
-    const u64 timeout = static_cast<u64>(next_time.count());
-
-    event_queue.emplace_back(Event{timeout, event_fifo_id++, user_data, event_type, 0});
-    pending_events.fetch_add(1, std::memory_order_relaxed);
-
-    std::push_heap(event_queue.begin(), event_queue.end(), std::greater<>());
-
-    if (is_multicore) {
-        event_cv.notify_one();
-    }
-}
-
-void CoreTiming::ScheduleLoopingEvent(std::chrono::nanoseconds time,
                                       const std::shared_ptr<EventType>& event_type,
-                                      std::uintptr_t user_data) {
+                                      std::uintptr_t user_data, bool absolute_time) {
     std::unique_lock main_lock(event_mutex);
-    const u64 timeout = static_cast<u64>((GetGlobalTimeNs() + time).count());
+    const auto next_time{absolute_time ? start_time : GetGlobalTimeNs() + start_time};
 
     event_queue.emplace_back(
-        Event{timeout, event_fifo_id++, user_data, event_type, static_cast<u64>(time.count())});
+        Event{next_time.count(), event_fifo_id++, user_data, event_type, resched_time.count()});
     pending_events.fetch_add(1, std::memory_order_relaxed);
 
     std::push_heap(event_queue.begin(), event_queue.end(), std::greater<>());
@@ -247,6 +250,11 @@ void CoreTiming::RemoveEvent(const std::shared_ptr<EventType>& event_type) {
     }
 }
 
+void CoreTiming::RegisterPauseCallback(PauseCallback&& callback) {
+    std::unique_lock main_lock(event_mutex);
+    pause_callbacks.emplace_back(std::move(callback));
+}
+
 std::optional<s64> CoreTiming::Advance() {
     global_timer = GetGlobalTimeNs().count();
 
@@ -257,21 +265,31 @@ std::optional<s64> CoreTiming::Advance() {
         event_queue.pop_back();
 
         if (const auto event_type{evt.type.lock()}) {
-
             event_mutex.unlock();
 
-            const s64 delay = static_cast<s64>(GetGlobalTimeNs().count() - evt.time);
+            const auto new_schedule_time{event_type->callback(
-            event_type->callback(evt.user_data, std::chrono::nanoseconds{delay});
+                evt.user_data, evt.time,
+                std::chrono::nanoseconds{GetGlobalTimeNs().count() - evt.time})};
 
             event_mutex.lock();
             pending_events.fetch_sub(1, std::memory_order_relaxed);
-        }
 
-        if (evt.reschedule_time != 0) {
+            if (evt.reschedule_time != 0) {
-            event_queue.emplace_back(
+                // If this event was scheduled into a pause, its time now is going to be way behind.
-                Event{global_timer + evt.reschedule_time - (global_timer - evt.time),
+                // Re-set this event to continue from the end of the pause.
-                      event_fifo_id++, evt.user_data, evt.type, evt.reschedule_time});
+                auto next_time{evt.time + evt.reschedule_time};
-            std::push_heap(event_queue.begin(), event_queue.end(), std::greater<>());
+                if (evt.time < pause_end_time) {
+                    next_time = pause_end_time + evt.reschedule_time;
+                }
+
+                const auto next_schedule_time{new_schedule_time.has_value()
+                                                  ? new_schedule_time.value().count()
+                                                  : evt.reschedule_time};
+                event_queue.emplace_back(
+                    Event{next_time, event_fifo_id++, evt.user_data, evt.type, next_schedule_time});
+                pending_events.fetch_add(1, std::memory_order_relaxed);
+                std::push_heap(event_queue.begin(), event_queue.end(), std::greater<>());
+            }
         }
 
         global_timer = GetGlobalTimeNs().count();

src/core/core_timing.h

@@ -20,8 +20,9 @@
 namespace Core::Timing {
 
 /// A callback that may be scheduled for a particular core timing event.
-using TimedCallback =
+using TimedCallback = std::function<std::optional<std::chrono::nanoseconds>(
-    std::function<void(std::uintptr_t user_data, std::chrono::nanoseconds ns_late)>;
+    std::uintptr_t user_data, s64 time, std::chrono::nanoseconds ns_late)>;
+using PauseCallback = std::function<void(bool paused)>;
 
 /// Contains the characteristics of a particular event.
 struct EventType {
@@ -93,18 +94,15 @@ public:
 
     /// Schedules an event in core timing
     void ScheduleEvent(std::chrono::nanoseconds ns_into_future,
-                       const std::shared_ptr<EventType>& event_type, std::uintptr_t user_data = 0);
+                       const std::shared_ptr<EventType>& event_type, std::uintptr_t user_data = 0,
-
+                       bool absolute_time = false);
-    /// Schedules an event with an absolute time in core timing
-    void ScheduleEventAt(std::chrono::nanoseconds next_time,
-                         const std::shared_ptr<EventType>& event_type,
-                         std::uintptr_t user_data = 0);
 
     /// Schedules an event which will automatically re-schedule itself with the given time, until
     /// unscheduled
-    void ScheduleLoopingEvent(std::chrono::nanoseconds time,
+    void ScheduleLoopingEvent(std::chrono::nanoseconds start_time,
+                              std::chrono::nanoseconds resched_time,
                               const std::shared_ptr<EventType>& event_type,
-                              std::uintptr_t user_data = 0);
+                              std::uintptr_t user_data = 0, bool absolute_time = false);
 
     void UnscheduleEvent(const std::shared_ptr<EventType>& event_type, std::uintptr_t user_data);
 
@@ -136,6 +134,9 @@ public:
     /// Checks for events manually and returns time in nanoseconds for next event, threadsafe.
     std::optional<s64> Advance();
 
+    /// Register a callback function to be called when coretiming pauses.
+    void RegisterPauseCallback(PauseCallback&& callback);
+
 private:
     struct Event;
 
@@ -147,7 +148,7 @@ private:
 
     std::unique_ptr<Common::WallClock> clock;
 
-    u64 global_timer = 0;
+    s64 global_timer = 0;
 
     // The queue is a min-heap using std::make_heap/push_heap/pop_heap.
     // We don't use std::priority_queue because we need to be able to serialize, unserialize and
@@ -173,10 +174,13 @@ private:
     bool shutting_down{};
     bool is_multicore{};
     size_t pause_count{};
+    s64 pause_end_time{};
 
     /// Cycle timing
     u64 ticks{};
     s64 downcount{};
+
+    std::vector<PauseCallback> pause_callbacks{};
 };
 
 /// Creates a core timing event with the given name and callback.

src/core/cpu_manager.cpp

@@ -25,10 +25,8 @@ void CpuManager::ThreadStart(std::stop_token stop_token, CpuManager& cpu_manager
 }
 
 void CpuManager::Initialize() {
-    running_mode = true;
     num_cores = is_multicore ? Core::Hardware::NUM_CPU_CORES : 1;
     gpu_barrier = std::make_unique<Common::Barrier>(num_cores + 1);
-    pause_barrier = std::make_unique<Common::Barrier>(num_cores + 1);
 
     for (std::size_t core = 0; core < num_cores; core++) {
         core_data[core].host_thread = std::jthread(ThreadStart, std::ref(*this), core);
@@ -36,8 +34,11 @@ void CpuManager::Initialize() {
 }
 
 void CpuManager::Shutdown() {
-    running_mode = false;
+    for (std::size_t core = 0; core < num_cores; core++) {
-    Pause(false);
+        if (core_data[core].host_thread.joinable()) {
+            core_data[core].host_thread.join();
+        }
+    }
 }
 
 void CpuManager::GuestThreadFunction() {
@@ -64,6 +65,10 @@ void CpuManager::IdleThreadFunction() {
     }
 }
 
+void CpuManager::ShutdownThreadFunction() {
+    ShutdownThread();
+}
+
 ///////////////////////////////////////////////////////////////////////////////
 ///                             MultiCore                                   ///
 ///////////////////////////////////////////////////////////////////////////////
@@ -176,41 +181,13 @@ void CpuManager::PreemptSingleCore(bool from_running_enviroment) {
     }
 }
 
-void CpuManager::SuspendThread() {
+void CpuManager::ShutdownThread() {
     auto& kernel = system.Kernel();
-    kernel.CurrentScheduler()->OnThreadStart();
+    auto core = is_multicore ? kernel.CurrentPhysicalCoreIndex() : 0;
+    auto* current_thread = kernel.GetCurrentEmuThread();
 
-    while (true) {
+    Common::Fiber::YieldTo(current_thread->GetHostContext(), *core_data[core].host_context);
-        auto core = is_multicore ? kernel.CurrentPhysicalCoreIndex() : 0;
+    UNREACHABLE();
-        auto& scheduler = *kernel.CurrentScheduler();
-        Kernel::KThread* current_thread = scheduler.GetSchedulerCurrentThread();
-        Common::Fiber::YieldTo(current_thread->GetHostContext(), *core_data[core].host_context);
-
-        // This shouldn't be here. This is here because the scheduler needs the current
-        // thread to have dispatch disabled before explicitly rescheduling. Ideally in the
-        // future this will be called by RequestScheduleOnInterrupt and explicitly disabling
-        // dispatch outside the scheduler will not be necessary.
-        current_thread->DisableDispatch();
-
-        scheduler.RescheduleCurrentCore();
-    }
-}
-
-void CpuManager::Pause(bool paused) {
-    std::scoped_lock lk{pause_lock};
-
-    if (pause_state == paused) {
-        return;
-    }
-
-    // Set the new state
-    pause_state.store(paused);
-
-    // Wake up any waiting threads
-    pause_state.notify_all();
-
-    // Wait for all threads to successfully change state before returning
-    pause_barrier->Sync();
 }
 
 void CpuManager::RunThread(std::size_t core) {
@@ -241,27 +218,9 @@ void CpuManager::RunThread(std::size_t core) {
         system.GPU().ObtainContext();
     }
 
-    {
+    auto* current_thread = system.Kernel().CurrentScheduler()->GetIdleThread();
-        // Set the current thread on entry
+    Kernel::SetCurrentThread(system.Kernel(), current_thread);
-        auto* current_thread = system.Kernel().CurrentScheduler()->GetIdleThread();
+    Common::Fiber::YieldTo(data.host_context, *current_thread->GetHostContext());
-        Kernel::SetCurrentThread(system.Kernel(), current_thread);
-    }
-
-    while (running_mode) {
-        if (pause_state.load(std::memory_order_relaxed)) {
-            // Wait for caller to acknowledge pausing
-            pause_barrier->Sync();
-
-            // Wait until unpaused
-            pause_state.wait(true, std::memory_order_relaxed);
-
-            // Wait for caller to acknowledge unpausing
-            pause_barrier->Sync();
-        }
-
-        auto current_thread = system.Kernel().CurrentScheduler()->GetSchedulerCurrentThread();
-        Common::Fiber::YieldTo(data.host_context, *current_thread->GetHostContext());
-    }
 }
 
 } // namespace Core

src/core/cpu_manager.h

@@ -50,16 +50,14 @@ public:
     void Initialize();
     void Shutdown();
 
-    void Pause(bool paused);
-
     std::function<void()> GetGuestThreadStartFunc() {
         return [this] { GuestThreadFunction(); };
     }
     std::function<void()> GetIdleThreadStartFunc() {
         return [this] { IdleThreadFunction(); };
     }
-    std::function<void()> GetSuspendThreadStartFunc() {
+    std::function<void()> GetShutdownThreadStartFunc() {
-        return [this] { SuspendThread(); };
+        return [this] { ShutdownThreadFunction(); };
     }
 
     void PreemptSingleCore(bool from_running_enviroment = true);
@@ -72,6 +70,7 @@ private:
     void GuestThreadFunction();
     void GuestRewindFunction();
     void IdleThreadFunction();
+    void ShutdownThreadFunction();
 
     void MultiCoreRunGuestThread();
     void MultiCoreRunGuestLoop();
@@ -83,7 +82,7 @@ private:
 
     static void ThreadStart(std::stop_token stop_token, CpuManager& cpu_manager, std::size_t core);
 
-    void SuspendThread();
+    void ShutdownThread();
     void RunThread(std::size_t core);
 
     struct CoreData {
@@ -91,12 +90,7 @@ private:
         std::jthread host_thread;
     };
 
-    std::atomic<bool> running_mode{};
-    std::atomic<bool> pause_state{};
-    std::unique_ptr<Common::Barrier> pause_barrier{};
     std::unique_ptr<Common::Barrier> gpu_barrier{};
-    std::mutex pause_lock{};
-
     std::array<CoreData, Core::Hardware::NUM_CPU_CORES> core_data{};
 
     bool is_async_gpu{};

src/core/hle/kernel/k_thread.cpp

@@ -269,7 +269,7 @@ Result KThread::InitializeIdleThread(Core::System& system, KThread* thread, s32
 Result KThread::InitializeHighPriorityThread(Core::System& system, KThread* thread,
                                              KThreadFunction func, uintptr_t arg, s32 virt_core) {
     return InitializeThread(thread, func, arg, {}, {}, virt_core, nullptr, ThreadType::HighPriority,
-                            system.GetCpuManager().GetSuspendThreadStartFunc());
+                            system.GetCpuManager().GetShutdownThreadStartFunc());
 }
 
 Result KThread::InitializeUserThread(Core::System& system, KThread* thread, KThreadFunction func,
@@ -739,6 +739,19 @@ void KThread::Continue() {
     KScheduler::OnThreadStateChanged(kernel, this, old_state);
 }
 
+void KThread::WaitUntilSuspended() {
+    // Make sure we have a suspend requested.
+    ASSERT(IsSuspendRequested());
+
+    // Loop until the thread is not executing on any core.
+    for (std::size_t i = 0; i < static_cast<std::size_t>(Core::Hardware::NUM_CPU_CORES); ++i) {
+        KThread* core_thread{};
+        do {
+            core_thread = kernel.Scheduler(i).GetSchedulerCurrentThread();
+        } while (core_thread == this);
+    }
+}
+
 Result KThread::SetActivity(Svc::ThreadActivity activity) {
     // Lock ourselves.
     KScopedLightLock lk(activity_pause_lock);

src/core/hle/kernel/k_thread.h

@@ -208,6 +208,8 @@ public:
 
     void Continue();
 
+    void WaitUntilSuspended();
+
     constexpr void SetSyncedIndex(s32 index) {
         synced_index = index;
     }

src/core/hle/kernel/kernel.cpp

@@ -76,7 +76,7 @@ struct KernelCore::Impl {
         InitializeMemoryLayout();
         Init::InitializeKPageBufferSlabHeap(system);
         InitializeSchedulers();
-        InitializeSuspendThreads();
+        InitializeShutdownThreads();
         InitializePreemption(kernel);
 
         RegisterHostThread();
@@ -131,9 +131,9 @@ struct KernelCore::Impl {
         CleanupObject(system_resource_limit);
 
         for (u32 core_id = 0; core_id < Core::Hardware::NUM_CPU_CORES; core_id++) {
-            if (suspend_threads[core_id]) {
+            if (shutdown_threads[core_id]) {
-                suspend_threads[core_id]->Close();
+                shutdown_threads[core_id]->Close();
-                suspend_threads[core_id] = nullptr;
+                shutdown_threads[core_id] = nullptr;
             }
 
             schedulers[core_id]->Finalize();
@@ -238,26 +238,27 @@ struct KernelCore::Impl {
 
     void InitializePreemption(KernelCore& kernel) {
         preemption_event = Core::Timing::CreateEvent(
-            "PreemptionCallback", [this, &kernel](std::uintptr_t, std::chrono::nanoseconds) {
+            "PreemptionCallback",
+            [this, &kernel](std::uintptr_t, s64 time,
+                            std::chrono::nanoseconds) -> std::optional<std::chrono::nanoseconds> {
                 {
                     KScopedSchedulerLock lock(kernel);
                     global_scheduler_context->PreemptThreads();
                 }
-                const auto time_interval = std::chrono::nanoseconds{std::chrono::milliseconds(10)};
+                return std::nullopt;
-                system.CoreTiming().ScheduleEvent(time_interval, preemption_event);
             });
 
         const auto time_interval = std::chrono::nanoseconds{std::chrono::milliseconds(10)};
-        system.CoreTiming().ScheduleEvent(time_interval, preemption_event);
+        system.CoreTiming().ScheduleLoopingEvent(time_interval, time_interval, preemption_event);
     }
 
-    void InitializeSuspendThreads() {
+    void InitializeShutdownThreads() {
         for (u32 core_id = 0; core_id < Core::Hardware::NUM_CPU_CORES; core_id++) {
-            suspend_threads[core_id] = KThread::Create(system.Kernel());
+            shutdown_threads[core_id] = KThread::Create(system.Kernel());
-            ASSERT(KThread::InitializeHighPriorityThread(system, suspend_threads[core_id], {}, {},
+            ASSERT(KThread::InitializeHighPriorityThread(system, shutdown_threads[core_id], {}, {},
                                                          core_id)
                        .IsSuccess());
-            suspend_threads[core_id]->SetName(fmt::format("SuspendThread:{}", core_id));
+            shutdown_threads[core_id]->SetName(fmt::format("SuspendThread:{}", core_id));
         }
     }
 
@@ -779,7 +780,7 @@ struct KernelCore::Impl {
     std::weak_ptr<ServiceThread> default_service_thread;
     Common::ThreadWorker service_threads_manager;
 
-    std::array<KThread*, Core::Hardware::NUM_CPU_CORES> suspend_threads;
+    std::array<KThread*, Core::Hardware::NUM_CPU_CORES> shutdown_threads;
     std::array<Core::CPUInterruptHandler, Core::Hardware::NUM_CPU_CORES> interrupts{};
     std::array<std::unique_ptr<Kernel::KScheduler>, Core::Hardware::NUM_CPU_CORES> schedulers{};
 
@@ -1093,16 +1094,27 @@ const Kernel::KSharedMemory& KernelCore::GetHidBusSharedMem() const {
 
 void KernelCore::Suspend(bool suspended) {
     const bool should_suspend{exception_exited || suspended};
-    const auto state{should_suspend ? ThreadState::Runnable : ThreadState::Waiting};
+    const auto activity = should_suspend ? ProcessActivity::Paused : ProcessActivity::Runnable;
-    {
+
-        KScopedSchedulerLock lk{*this};
+    for (auto* process : GetProcessList()) {
-        for (auto* thread : impl->suspend_threads) {
+        process->SetActivity(activity);
-            thread->SetState(state);
+
-            thread->SetWaitReasonForDebugging(ThreadWaitReasonForDebugging::Suspended);
+        if (should_suspend) {
+            // Wait for execution to stop
+            for (auto* thread : process->GetThreadList()) {
+                thread->WaitUntilSuspended();
+            }
         }
     }
 }
 
+void KernelCore::ShutdownCores() {
+    for (auto* thread : impl->shutdown_threads) {
+        void(thread->Run());
+    }
+    InterruptAllPhysicalCores();
+}
+
 bool KernelCore::IsMulticore() const {
     return impl->is_multicore;
 }

src/core/hle/kernel/kernel.h

@@ -283,6 +283,9 @@ public:
     /// Exceptional exit all processes.
     void ExceptionalExit();
 
+    /// Notify emulated CPU cores to shut down.
+    void ShutdownCores();
+
     bool IsMulticore() const;
 
     bool IsShuttingDown() const;

src/core/hle/kernel/time_manager.cpp

@@ -11,15 +11,17 @@
 namespace Kernel {
 
 TimeManager::TimeManager(Core::System& system_) : system{system_} {
-    time_manager_event_type =
+    time_manager_event_type = Core::Timing::CreateEvent(
-        Core::Timing::CreateEvent("Kernel::TimeManagerCallback",
+        "Kernel::TimeManagerCallback",
-                                  [this](std::uintptr_t thread_handle, std::chrono::nanoseconds) {
+        [this](std::uintptr_t thread_handle, s64 time,
-                                      KThread* thread = reinterpret_cast<KThread*>(thread_handle);
+               std::chrono::nanoseconds) -> std::optional<std::chrono::nanoseconds> {
-                                      {
+            KThread* thread = reinterpret_cast<KThread*>(thread_handle);
-                                          KScopedSchedulerLock sl(system.Kernel());
+            {
-                                          thread->OnTimer();
+                KScopedSchedulerLock sl(system.Kernel());
-                                      }
+                thread->OnTimer();
-                                  });
+            }
+            return std::nullopt;
+        });
 }
 
 void TimeManager::ScheduleTimeEvent(KThread* thread, s64 nanoseconds) {

src/core/hle/service/hid/hid.cpp

@@ -74,26 +74,34 @@ IAppletResource::IAppletResource(Core::System& system_,
     // Register update callbacks
     pad_update_event = Core::Timing::CreateEvent(
         "HID::UpdatePadCallback",
-        [this](std::uintptr_t user_data, std::chrono::nanoseconds ns_late) {
+        [this](std::uintptr_t user_data, s64 time,
+               std::chrono::nanoseconds ns_late) -> std::optional<std::chrono::nanoseconds> {
             const auto guard = LockService();
             UpdateControllers(user_data, ns_late);
+            return std::nullopt;
         });
     mouse_keyboard_update_event = Core::Timing::CreateEvent(
         "HID::UpdateMouseKeyboardCallback",
-        [this](std::uintptr_t user_data, std::chrono::nanoseconds ns_late) {
+        [this](std::uintptr_t user_data, s64 time,
+               std::chrono::nanoseconds ns_late) -> std::optional<std::chrono::nanoseconds> {
             const auto guard = LockService();
             UpdateMouseKeyboard(user_data, ns_late);
+            return std::nullopt;
         });
     motion_update_event = Core::Timing::CreateEvent(
         "HID::UpdateMotionCallback",
-        [this](std::uintptr_t user_data, std::chrono::nanoseconds ns_late) {
+        [this](std::uintptr_t user_data, s64 time,
+               std::chrono::nanoseconds ns_late) -> std::optional<std::chrono::nanoseconds> {
             const auto guard = LockService();
             UpdateMotion(user_data, ns_late);
+            return std::nullopt;
         });
 
-    system.CoreTiming().ScheduleEvent(pad_update_ns, pad_update_event);
+    system.CoreTiming().ScheduleLoopingEvent(pad_update_ns, pad_update_ns, pad_update_event);
-    system.CoreTiming().ScheduleEvent(mouse_keyboard_update_ns, mouse_keyboard_update_event);
+    system.CoreTiming().ScheduleLoopingEvent(mouse_keyboard_update_ns, mouse_keyboard_update_ns,
-    system.CoreTiming().ScheduleEvent(motion_update_ns, motion_update_event);
+                                             mouse_keyboard_update_event);
+    system.CoreTiming().ScheduleLoopingEvent(motion_update_ns, motion_update_ns,
+                                             motion_update_event);
 
     system.HIDCore().ReloadInputDevices();
 }
@@ -135,13 +143,6 @@ void IAppletResource::UpdateControllers(std::uintptr_t user_data,
         }
         controller->OnUpdate(core_timing);
     }
-
-    // If ns_late is higher than the update rate ignore the delay
-    if (ns_late > pad_update_ns) {
-        ns_late = {};
-    }
-
-    core_timing.ScheduleEvent(pad_update_ns - ns_late, pad_update_event);
 }
 
 void IAppletResource::UpdateMouseKeyboard(std::uintptr_t user_data,
@@ -150,26 +151,12 @@ void IAppletResource::UpdateMouseKeyboard(std::uintptr_t user_data,
 
     controllers[static_cast<size_t>(HidController::Mouse)]->OnUpdate(core_timing);
     controllers[static_cast<size_t>(HidController::Keyboard)]->OnUpdate(core_timing);
-
-    // If ns_late is higher than the update rate ignore the delay
-    if (ns_late > mouse_keyboard_update_ns) {
-        ns_late = {};
-    }
-
-    core_timing.ScheduleEvent(mouse_keyboard_update_ns - ns_late, mouse_keyboard_update_event);
 }
 
 void IAppletResource::UpdateMotion(std::uintptr_t user_data, std::chrono::nanoseconds ns_late) {
     auto& core_timing = system.CoreTiming();
 
     controllers[static_cast<size_t>(HidController::NPad)]->OnMotionUpdate(core_timing);
-
-    // If ns_late is higher than the update rate ignore the delay
-    if (ns_late > motion_update_ns) {
-        ns_late = {};
-    }
-
-    core_timing.ScheduleEvent(motion_update_ns - ns_late, motion_update_event);
 }
 
 class IActiveVibrationDeviceList final : public ServiceFramework<IActiveVibrationDeviceList> {

src/core/hle/service/hid/hidbus.cpp

@@ -50,12 +50,15 @@ HidBus::HidBus(Core::System& system_)
     // Register update callbacks
     hidbus_update_event = Core::Timing::CreateEvent(
         "Hidbus::UpdateCallback",
-        [this](std::uintptr_t user_data, std::chrono::nanoseconds ns_late) {
+        [this](std::uintptr_t user_data, s64 time,
+               std::chrono::nanoseconds ns_late) -> std::optional<std::chrono::nanoseconds> {
             const auto guard = LockService();
             UpdateHidbus(user_data, ns_late);
+            return std::nullopt;
         });
 
-    system_.CoreTiming().ScheduleEvent(hidbus_update_ns, hidbus_update_event);
+    system_.CoreTiming().ScheduleLoopingEvent(hidbus_update_ns, hidbus_update_ns,
+                                              hidbus_update_event);
 }
 
 HidBus::~HidBus() {
@@ -63,8 +66,6 @@ HidBus::~HidBus() {
 }
 
 void HidBus::UpdateHidbus(std::uintptr_t user_data, std::chrono::nanoseconds ns_late) {
-    auto& core_timing = system.CoreTiming();
-
     if (is_hidbus_enabled) {
         for (std::size_t i = 0; i < devices.size(); ++i) {
             if (!devices[i].is_device_initializated) {
@@ -82,13 +83,6 @@ void HidBus::UpdateHidbus(std::uintptr_t user_data, std::chrono::nanoseconds ns_
                         sizeof(HidbusStatusManagerEntry));
         }
     }
-
-    // If ns_late is higher than the update rate ignore the delay
-    if (ns_late > hidbus_update_ns) {
-        ns_late = {};
-    }
-
-    core_timing.ScheduleEvent(hidbus_update_ns - ns_late, hidbus_update_event);
 }
 
 std::optional<std::size_t> HidBus::GetDeviceIndexFromHandle(BusHandle handle) const {

src/core/hle/service/nvflinger/nvflinger.cpp

@@ -69,21 +69,20 @@ NVFlinger::NVFlinger(Core::System& system_, HosBinderDriverServer& hos_binder_dr
 
     // Schedule the screen composition events
     composition_event = Core::Timing::CreateEvent(
-        "ScreenComposition", [this](std::uintptr_t, std::chrono::nanoseconds ns_late) {
+        "ScreenComposition",
+        [this](std::uintptr_t, s64 time,
+               std::chrono::nanoseconds ns_late) -> std::optional<std::chrono::nanoseconds> {
             const auto lock_guard = Lock();
             Compose();
 
-            const auto ticks = std::chrono::nanoseconds{GetNextTicks()};
+            return std::max(std::chrono::nanoseconds::zero(),
-            const auto ticks_delta = ticks - ns_late;
+                            std::chrono::nanoseconds(GetNextTicks()) - ns_late);
-            const auto future_ns = std::max(std::chrono::nanoseconds::zero(), ticks_delta);
-
-            this->system.CoreTiming().ScheduleEvent(future_ns, composition_event);
         });
 
     if (system.IsMulticore()) {
         vsync_thread = std::jthread([this](std::stop_token token) { SplitVSync(token); });
     } else {
-        system.CoreTiming().ScheduleEvent(frame_ns, composition_event);
+        system.CoreTiming().ScheduleLoopingEvent(frame_ns, frame_ns, composition_event);
     }
 }
 

src/core/memory/cheat_engine.cpp

@@ -184,10 +184,12 @@ CheatEngine::~CheatEngine() {
 void CheatEngine::Initialize() {
     event = Core::Timing::CreateEvent(
         "CheatEngine::FrameCallback::" + Common::HexToString(metadata.main_nso_build_id),
-        [this](std::uintptr_t user_data, std::chrono::nanoseconds ns_late) {
+        [this](std::uintptr_t user_data, s64 time,
+               std::chrono::nanoseconds ns_late) -> std::optional<std::chrono::nanoseconds> {
             FrameCallback(user_data, ns_late);
+            return std::nullopt;
         });
-    core_timing.ScheduleEvent(CHEAT_ENGINE_NS, event);
+    core_timing.ScheduleLoopingEvent(CHEAT_ENGINE_NS, CHEAT_ENGINE_NS, event);
 
     metadata.process_id = system.CurrentProcess()->GetProcessID();
     metadata.title_id = system.GetCurrentProcessProgramID();
@@ -237,8 +239,6 @@ void CheatEngine::FrameCallback(std::uintptr_t, std::chrono::nanoseconds ns_late
     MICROPROFILE_SCOPE(Cheat_Engine);
 
     vm.Execute(metadata);
-
-    core_timing.ScheduleEvent(CHEAT_ENGINE_NS - ns_late, event);
 }
 
 } // namespace Core::Memory

src/core/tools/freezer.cpp

@@ -53,8 +53,10 @@ Freezer::Freezer(Core::Timing::CoreTiming& core_timing_, Core::Memory::Memory& m
     : core_timing{core_timing_}, memory{memory_} {
     event = Core::Timing::CreateEvent(
         "MemoryFreezer::FrameCallback",
-        [this](std::uintptr_t user_data, std::chrono::nanoseconds ns_late) {
+        [this](std::uintptr_t user_data, s64 time,
+               std::chrono::nanoseconds ns_late) -> std::optional<std::chrono::nanoseconds> {
             FrameCallback(user_data, ns_late);
+            return std::nullopt;
         });
     core_timing.ScheduleEvent(memory_freezer_ns, event);
 }

src/tests/core/core_timing.cpp

@@ -9,6 +9,7 @@
 #include <cstdlib>
 #include <memory>
 #include <mutex>
+#include <optional>
 #include <string>
 
 #include "core/core.h"
@@ -25,13 +26,15 @@ u64 expected_callback = 0;
 std::mutex control_mutex;
 
 template <unsigned int IDX>
-void HostCallbackTemplate(std::uintptr_t user_data, std::chrono::nanoseconds ns_late) {
+std::optional<std::chrono::nanoseconds> HostCallbackTemplate(std::uintptr_t user_data, s64 time,
+                                                             std::chrono::nanoseconds ns_late) {
     std::unique_lock<std::mutex> lk(control_mutex);
     static_assert(IDX < CB_IDS.size(), "IDX out of range");
     callbacks_ran_flags.set(IDX);
     REQUIRE(CB_IDS[IDX] == user_data);
     delays[IDX] = ns_late.count();
     ++expected_callback;
+    return std::nullopt;
 }
 
 struct ScopeInit final {