diff --git a/extras/backends/sdl/backend_sdl.c b/extras/backends/sdl/backend_sdl.c
index 1fcfdb41..9471e27d 100644
--- a/extras/backends/sdl/backend_sdl.c
+++ b/extras/backends/sdl/backend_sdl.c
@@ -169,7 +169,7 @@ static ma_device_state_sdl* ma_device_get_backend_state__sdl(ma_device* pDevice)
 }
 
 
-static ma_result ma_device_step__sdl(ma_device* pDevice);
+static ma_result ma_device_step__sdl(ma_device* pDevice, ma_blocking_mode blockingMode);
 
 
 static void ma_backend_info__sdl(ma_device_backend_info* pBackendInfo)
@@ -562,7 +562,7 @@ static ma_result ma_device_start__sdl(ma_device* pDevice)
     ma_device_type deviceType = ma_device_get_type(pDevice);
 
     /* Step the device once to ensure buffers are pre-filled before starting. */
-    ma_device_step__sdl(pDevice);
+    ma_device_step__sdl(pDevice, MA_BLOCKING_MODE_NON_BLOCKING);
 
     if (deviceType == ma_device_type_capture || deviceType == ma_device_type_duplex) {
         pContextStateSDL->SDL_PauseAudioDevice(pDeviceStateSDL->capture.deviceID, 0);
@@ -593,18 +593,39 @@ static ma_result ma_device_stop__sdl(ma_device* pDevice)
 }
 
 
-static ma_result ma_device_wait__sdl(ma_device* pDevice)
+static ma_result ma_device_step__sdl(ma_device* pDevice, ma_blocking_mode blockingMode)
 {
     ma_device_state_sdl* pDeviceStateSDL = ma_device_get_backend_state__sdl(pDevice);
-    ma_device_state_async_wait(&pDeviceStateSDL->async);
 
-    return MA_SUCCESS;
-}
+    for (;;) {
+        ma_result result;
+
+        if (blockingMode == MA_BLOCKING_MODE_BLOCKING) {
+            ma_device_state_async_wait(&pDeviceStateSDL->async);
+        }
+    
+        if (!ma_device_is_started(pDevice)) {
+            return MA_DEVICE_NOT_STARTED;
+        }
+    
+        result = ma_device_state_async_step(&pDeviceStateSDL->async, pDevice);
+        if (result == MA_SUCCESS) {
+            break;
+        }
+
+        if (result != MA_NO_DATA_AVAILABLE) {
+            return result;
+        }
+
+        /* Getting here means no data was processed. In non-blocking mode we don't care, just get out of the loop. */
+        if (blockingMode == MA_BLOCKING_MODE_NON_BLOCKING) {
+            break;
+        }
+
+        /* Getting here means we're in blocking mode and no data was processed. In this case we'd rather keep waiting for data to be available. */
+        continue;
+    }
 
-static ma_result ma_device_step__sdl(ma_device* pDevice)
-{
-    ma_device_state_sdl* pDeviceStateSDL = ma_device_get_backend_state__sdl(pDevice);
-    ma_device_state_async_step(&pDeviceStateSDL->async, pDevice);
 
     return MA_SUCCESS;
 }
@@ -612,23 +633,19 @@ static ma_result ma_device_step__sdl(ma_device* pDevice)
 static void ma_device_loop__sdl(ma_device* pDevice)
 {
     for (;;) {
-        ma_result result = ma_device_wait__sdl(pDevice);
-        if (result != MA_SUCCESS) {
-            break;
-        }
-
-        /* If the wait terminated due to the device being stopped, abort now. */
-        if (!ma_device_is_started(pDevice)) {
-            break;
-        }
-
-        result = ma_device_step__sdl(pDevice);
+        ma_result result = ma_device_step__sdl(pDevice, MA_BLOCKING_MODE_BLOCKING);
         if (result != MA_SUCCESS) {
             break;
         }
     }
 }
 
+static void ma_device_wake__sdl(ma_device* pDevice)
+{
+    ma_device_state_sdl* pDeviceStateSDL = ma_device_get_backend_state__sdl(pDevice);
+    ma_device_state_async_release(&pDeviceStateSDL->async);
+}
+
 
 static ma_device_backend_vtable ma_gDeviceBackendVTable_SDL =
 {
@@ -643,7 +660,7 @@ static ma_device_backend_vtable ma_gDeviceBackendVTable_SDL =
     NULL,   /* onDeviceRead */
     NULL,   /* onDeviceWrite */
     ma_device_loop__sdl,
-    NULL    /* onDeviceWakeup */
+    ma_device_wake__sdl
 };
 
 ma_device_backend_vtable* ma_device_backend_sdl = &ma_gDeviceBackendVTable_SDL;
diff --git a/miniaudio.h b/miniaudio.h
index 92d9204a..79537d46 100644
--- a/miniaudio.h
+++ b/miniaudio.h
@@ -9522,8 +9522,9 @@ typedef struct ma_device_state_async
 MA_API ma_result ma_device_state_async_init(ma_device_type deviceType, const ma_device_descriptor* pDescriptorPlayback, const ma_device_descriptor* pDescriptorCapture, const ma_allocation_callbacks* pAllocationCallbacks, ma_device_state_async* pAsyncDeviceState);
 MA_API void ma_device_state_async_uninit(ma_device_state_async* pAsyncDeviceState, const ma_allocation_callbacks* pAllocationCallbacks);
 MA_API ma_result ma_device_state_async_resize(ma_device_state_async* pAsyncDeviceState, ma_uint32 sizeInFramesPlayback, ma_uint32 sizeInFramesCapture, const ma_allocation_callbacks* pAllocationCallbacks);
+MA_API void ma_device_state_async_release(ma_device_state_async* pAsyncDeviceState);
 MA_API void ma_device_state_async_wait(ma_device_state_async* pAsyncDeviceState);
-MA_API void ma_device_state_async_step(ma_device_state_async* pAsyncDeviceState, ma_device* pDevice);
+MA_API ma_result ma_device_state_async_step(ma_device_state_async* pAsyncDeviceState, ma_device* pDevice);  /* Returns MA_SUCCESS if some data was processed, MA_NO_DATA_AVAILABLE if no data was processed. */
 MA_API void ma_device_state_async_process(ma_device_state_async* pAsyncDeviceState, ma_device* pDevice, void* pOutput, const void* pInput, ma_uint32 frameCount);
 /* END ma_device_state_async.h */
 
@@ -46235,6 +46236,21 @@ MA_API ma_result ma_device_state_async_resize(ma_device_state_async* pAsyncDevic
     return MA_SUCCESS;
 }
 
+MA_API void ma_device_state_async_release(ma_device_state_async* pAsyncDeviceState)
+{
+    if (pAsyncDeviceState == NULL) {
+        return;
+    }
+
+    if (pAsyncDeviceState->deviceType == ma_device_type_capture || pAsyncDeviceState->deviceType == ma_device_type_duplex) {
+        ma_semaphore_release(&pAsyncDeviceState->capture.semaphore);
+    }
+
+    if (pAsyncDeviceState->deviceType == ma_device_type_playback || pAsyncDeviceState->deviceType == ma_device_type_duplex) {
+        ma_semaphore_release(&pAsyncDeviceState->playback.semaphore);
+    }
+}
+
 MA_API void ma_device_state_async_wait(ma_device_state_async* pAsyncDeviceState)
 {
     if (pAsyncDeviceState == NULL) {
@@ -46250,10 +46266,12 @@ MA_API void ma_device_state_async_wait(ma_device_state_async* pAsyncDeviceState)
     }
 }
 
-MA_API void ma_device_state_async_step(ma_device_state_async* pAsyncDeviceState, ma_device* pDevice)
+MA_API ma_result ma_device_state_async_step(ma_device_state_async* pAsyncDeviceState, ma_device* pDevice)
 {
+    ma_result result = MA_NO_DATA_AVAILABLE;
+
     if (pAsyncDeviceState == NULL || pDevice == NULL) {
-        return;
+        return MA_INVALID_ARGS;
     }
 
     if (pAsyncDeviceState->deviceType == ma_device_type_capture || pAsyncDeviceState->deviceType == ma_device_type_duplex) {
@@ -46262,6 +46280,7 @@ MA_API void ma_device_state_async_step(ma_device_state_async* pAsyncDeviceState,
             if (pAsyncDeviceState->capture.frameCount > 0) {
                 ma_device_handle_backend_data_callback(pDevice, NULL, pAsyncDeviceState->capture.pBuffer, pAsyncDeviceState->capture.frameCount);
                 pAsyncDeviceState->capture.frameCount = 0;
+                result = MA_SUCCESS;
             }
         }
         ma_spinlock_unlock(&pAsyncDeviceState->capture.lock);
@@ -46275,10 +46294,13 @@ MA_API void ma_device_state_async_step(ma_device_state_async* pAsyncDeviceState,
             if (pAsyncDeviceState->playback.frameCount < pAsyncDeviceState->playback.frameCap) {
                 ma_device_handle_backend_data_callback(pDevice, ma_offset_ptr(pAsyncDeviceState->playback.pBuffer, bytesPerFrame * pAsyncDeviceState->playback.frameCount), NULL, (pAsyncDeviceState->playback.frameCap - pAsyncDeviceState->playback.frameCount));
                 pAsyncDeviceState->playback.frameCount = pAsyncDeviceState->playback.frameCap;
+                result = MA_SUCCESS;
             }
         }
         ma_spinlock_unlock(&pAsyncDeviceState->playback.lock);
     }
+
+    return result;
 }
 
 MA_API void ma_device_state_async_process(ma_device_state_async* pAsyncDeviceState, ma_device* pDevice, void* pOutput, const void* pInput, ma_uint32 frameCount)