diff --git a/libs/gui/tests/EndToEndNativeInputTest.cpp b/libs/gui/tests/EndToEndNativeInputTest.cpp
index 3344e0b..5f80c16 100644
--- a/libs/gui/tests/EndToEndNativeInputTest.cpp
+++ b/libs/gui/tests/EndToEndNativeInputTest.cpp
@@ -512,6 +512,22 @@
     bgSurface->expectTap(1, 1);
 }
 
+TEST_F(InputSurfacesTest, input_respects_surface_insets_with_replaceTouchableRegionWithCrop) {
+    std::unique_ptr<InputSurface> bgSurface = makeSurface(100, 100);
+    std::unique_ptr<InputSurface> fgSurface = makeSurface(100, 100);
+    bgSurface->showAt(100, 100);
+
+    fgSurface->mInputInfo.surfaceInset = 5;
+    fgSurface->mInputInfo.replaceTouchableRegionWithCrop = true;
+    fgSurface->showAt(100, 100);
+
+    injectTap(106, 106);
+    fgSurface->expectTap(1, 1);
+
+    injectTap(101, 101);
+    bgSurface->expectTap(1, 1);
+}
+
 // Ensure a surface whose insets are cropped, handles the touch offset correctly. ref:b/120413463
 TEST_F(InputSurfacesTest, input_respects_cropped_surface_insets) {
     std::unique_ptr<InputSurface> parentSurface = makeSurface(100, 100);
diff --git a/services/surfaceflinger/FrontEnd/LayerSnapshotBuilder.cpp b/services/surfaceflinger/FrontEnd/LayerSnapshotBuilder.cpp
index a16de1b..2fc9682 100644
--- a/services/surfaceflinger/FrontEnd/LayerSnapshotBuilder.cpp
+++ b/services/surfaceflinger/FrontEnd/LayerSnapshotBuilder.cpp
@@ -21,6 +21,7 @@
 
 #include "LayerSnapshotBuilder.h"
 #include <gui/TraceUtils.h>
+#include <ui/FloatRect.h>
 #include <numeric>
 #include "DisplayHardware/HWC2.h"
 #include "DisplayHardware/Hal.h"
@@ -101,43 +102,52 @@
 }
 
 /**
+ * Returns the bounds used to fill the input frame and the touchable region.
+ *
  * Similar to getInputTransform, we need to update the bounds to include the transform.
  * This is because bounds don't include the buffer transform, where the input assumes
  * that's already included.
  */
-Rect getInputBounds(const LayerSnapshot& snapshot) {
-    if (!snapshot.hasBufferOrSidebandStream()) {
-        return snapshot.croppedBufferSize;
+std::pair<FloatRect, bool> getInputBounds(const LayerSnapshot& snapshot, bool fillParentBounds) {
+    FloatRect inputBounds = snapshot.croppedBufferSize.toFloatRect();
+    if (snapshot.hasBufferOrSidebandStream() && snapshot.croppedBufferSize.isValid() &&
+        snapshot.localTransform.getType() != ui::Transform::IDENTITY) {
+        inputBounds = snapshot.localTransform.transform(inputBounds);
     }
 
-    if (snapshot.localTransform.getType() == ui::Transform::IDENTITY ||
-        !snapshot.croppedBufferSize.isValid()) {
-        return snapshot.croppedBufferSize;
+    bool inputBoundsValid = snapshot.croppedBufferSize.isValid();
+    if (!inputBoundsValid) {
+        /**
+         * Input bounds are based on the layer crop or buffer size. But if we are using
+         * the layer bounds as the input bounds (replaceTouchableRegionWithCrop flag) then
+         * we can use the parent bounds as the input bounds if the layer does not have buffer
+         * or a crop. We want to unify this logic but because of compat reasons we cannot always
+         * use the parent bounds. A layer without a buffer can get input. So when a window is
+         * initially added, its touchable region can fill its parent layer bounds and that can
+         * have negative consequences.
+         */
+        inputBounds = fillParentBounds ? snapshot.geomLayerBounds : FloatRect{};
     }
-    return snapshot.localTransform.transform(snapshot.croppedBufferSize);
+
+    // Clamp surface inset to the input bounds.
+    const float inset = static_cast<float>(snapshot.inputInfo.surfaceInset);
+    const float xSurfaceInset = std::clamp(inset, 0.f, inputBounds.getWidth() / 2.f);
+    const float ySurfaceInset = std::clamp(inset, 0.f, inputBounds.getHeight() / 2.f);
+
+    // Apply the insets to the input bounds.
+    inputBounds.left += xSurfaceInset;
+    inputBounds.top += ySurfaceInset;
+    inputBounds.right -= xSurfaceInset;
+    inputBounds.bottom -= ySurfaceInset;
+    return {inputBounds, inputBoundsValid};
 }
 
-void fillInputFrameInfo(gui::WindowInfo& info, const ui::Transform& screenToDisplay,
-                        const LayerSnapshot& snapshot) {
-    Rect tmpBounds = getInputBounds(snapshot);
-    if (!tmpBounds.isValid()) {
-        info.touchableRegion.clear();
-        // A layer could have invalid input bounds and still expect to receive touch input if it has
-        // replaceTouchableRegionWithCrop. For that case, the input transform needs to be calculated
-        // correctly to determine the coordinate space for input events. Use an empty rect so that
-        // the layer will receive input in its own layer space.
-        tmpBounds = Rect::EMPTY_RECT;
-    }
-
+Rect getInputBoundsInDisplaySpace(const LayerSnapshot& snapshot, const FloatRect& insetBounds,
+                                  const ui::Transform& screenToDisplay) {
     // InputDispatcher works in the display device's coordinate space. Here, we calculate the
     // frame and transform used for the layer, which determines the bounds and the coordinate space
     // within which the layer will receive input.
-    //
-    // The coordinate space within which each of the bounds are specified is explicitly documented
-    // in the variable name. For example "inputBoundsInLayer" is specified in layer space. A
-    // Transform converts one coordinate space to another, which is apparent in its naming. For
-    // example, "layerToDisplay" transforms layer space to display space.
-    //
+
     // Coordinate space definitions:
     //   - display: The display device's coordinate space. Correlates to pixels on the display.
     //   - screen: The post-rotation coordinate space for the display, a.k.a. logical display space.
@@ -145,37 +155,34 @@
     //   - input: The coordinate space in which this layer will receive input events. This could be
     //            different than layer space if a surfaceInset is used, which changes the origin
     //            of the input space.
-    const FloatRect inputBoundsInLayer = tmpBounds.toFloatRect();
-
-    // Clamp surface inset to the input bounds.
-    const auto surfaceInset = static_cast<float>(info.surfaceInset);
-    const float xSurfaceInset =
-            std::max(0.f, std::min(surfaceInset, inputBoundsInLayer.getWidth() / 2.f));
-    const float ySurfaceInset =
-            std::max(0.f, std::min(surfaceInset, inputBoundsInLayer.getHeight() / 2.f));
-
-    // Apply the insets to the input bounds.
-    const FloatRect insetBoundsInLayer(inputBoundsInLayer.left + xSurfaceInset,
-                                       inputBoundsInLayer.top + ySurfaceInset,
-                                       inputBoundsInLayer.right - xSurfaceInset,
-                                       inputBoundsInLayer.bottom - ySurfaceInset);
 
     // Crop the input bounds to ensure it is within the parent's bounds.
-    const FloatRect croppedInsetBoundsInLayer =
-            snapshot.geomLayerBounds.intersect(insetBoundsInLayer);
+    const FloatRect croppedInsetBoundsInLayer = snapshot.geomLayerBounds.intersect(insetBounds);
 
     const ui::Transform layerToScreen = getInputTransform(snapshot);
     const ui::Transform layerToDisplay = screenToDisplay * layerToScreen;
 
-    const Rect roundedFrameInDisplay{layerToDisplay.transform(croppedInsetBoundsInLayer)};
+    return Rect{layerToDisplay.transform(croppedInsetBoundsInLayer)};
+}
+
+void fillInputFrameInfo(gui::WindowInfo& info, const ui::Transform& screenToDisplay,
+                        const LayerSnapshot& snapshot) {
+    auto [inputBounds, inputBoundsValid] = getInputBounds(snapshot, /*fillParentBounds=*/false);
+    if (!inputBoundsValid) {
+        info.touchableRegion.clear();
+    }
+
+    const Rect roundedFrameInDisplay =
+            getInputBoundsInDisplaySpace(snapshot, inputBounds, screenToDisplay);
     info.frameLeft = roundedFrameInDisplay.left;
     info.frameTop = roundedFrameInDisplay.top;
     info.frameRight = roundedFrameInDisplay.right;
     info.frameBottom = roundedFrameInDisplay.bottom;
 
     ui::Transform inputToLayer;
-    inputToLayer.set(insetBoundsInLayer.left, insetBoundsInLayer.top);
-    const ui::Transform inputToDisplay = layerToDisplay * inputToLayer;
+    inputToLayer.set(inputBounds.left, inputBounds.top);
+    const ui::Transform layerToScreen = getInputTransform(snapshot);
+    const ui::Transform inputToDisplay = screenToDisplay * layerToScreen * inputToLayer;
 
     // InputDispatcher expects a display-to-input transform.
     info.transform = inputToDisplay.inverse();
@@ -1008,12 +1015,26 @@
 
     auto cropLayerSnapshot = getSnapshot(requested.touchCropId);
     if (snapshot.inputInfo.replaceTouchableRegionWithCrop) {
-        const Rect bounds(cropLayerSnapshot ? cropLayerSnapshot->transformedBounds
-                                            : snapshot.transformedBounds);
-        snapshot.inputInfo.touchableRegion = Region(displayInfo.transform.transform(bounds));
+        Rect inputBoundsInDisplaySpace;
+        if (!cropLayerSnapshot) {
+            FloatRect inputBounds = getInputBounds(snapshot, /*fillParentBounds=*/true).first;
+            inputBoundsInDisplaySpace =
+                    getInputBoundsInDisplaySpace(snapshot, inputBounds, displayInfo.transform);
+        } else {
+            FloatRect inputBounds =
+                    getInputBounds(*cropLayerSnapshot, /*fillParentBounds=*/true).first;
+            inputBoundsInDisplaySpace =
+                    getInputBoundsInDisplaySpace(*cropLayerSnapshot, inputBounds,
+                                                 displayInfo.transform);
+        }
+        snapshot.inputInfo.touchableRegion = Region(inputBoundsInDisplaySpace);
     } else if (cropLayerSnapshot) {
+        FloatRect inputBounds = getInputBounds(*cropLayerSnapshot, /*fillParentBounds=*/true).first;
+        Rect inputBoundsInDisplaySpace =
+                getInputBoundsInDisplaySpace(*cropLayerSnapshot, inputBounds,
+                                             displayInfo.transform);
         snapshot.inputInfo.touchableRegion = snapshot.inputInfo.touchableRegion.intersect(
-                displayInfo.transform.transform(Rect{cropLayerSnapshot->transformedBounds}));
+                displayInfo.transform.transform(inputBoundsInDisplaySpace));
     }
 
     // Inherit the trusted state from the parent hierarchy, but don't clobber the trusted state
diff --git a/services/surfaceflinger/Layer.cpp b/services/surfaceflinger/Layer.cpp
index 0f2af2f..3b41458 100644
--- a/services/surfaceflinger/Layer.cpp
+++ b/services/surfaceflinger/Layer.cpp
@@ -52,8 +52,11 @@
 #include <system/graphics-base-v1.0.h>
 #include <ui/DataspaceUtils.h>
 #include <ui/DebugUtils.h>
+#include <ui/FloatRect.h>
 #include <ui/GraphicBuffer.h>
 #include <ui/PixelFormat.h>
+#include <ui/Rect.h>
+#include <ui/Transform.h>
 #include <utils/Errors.h>
 #include <utils/Log.h>
 #include <utils/NativeHandle.h>
@@ -2297,62 +2300,21 @@
 }
 
 void Layer::fillInputFrameInfo(WindowInfo& info, const ui::Transform& screenToDisplay) {
-    Rect tmpBounds = getInputBounds();
-    if (!tmpBounds.isValid()) {
+    auto [inputBounds, inputBoundsValid] = getInputBounds(/*fillParentBounds=*/false);
+    if (!inputBoundsValid) {
         info.touchableRegion.clear();
-        // A layer could have invalid input bounds and still expect to receive touch input if it has
-        // replaceTouchableRegionWithCrop. For that case, the input transform needs to be calculated
-        // correctly to determine the coordinate space for input events. Use an empty rect so that
-        // the layer will receive input in its own layer space.
-        tmpBounds = Rect::EMPTY_RECT;
     }
 
-    // InputDispatcher works in the display device's coordinate space. Here, we calculate the
-    // frame and transform used for the layer, which determines the bounds and the coordinate space
-    // within which the layer will receive input.
-    //
-    // The coordinate space within which each of the bounds are specified is explicitly documented
-    // in the variable name. For example "inputBoundsInLayer" is specified in layer space. A
-    // Transform converts one coordinate space to another, which is apparent in its naming. For
-    // example, "layerToDisplay" transforms layer space to display space.
-    //
-    // Coordinate space definitions:
-    //   - display: The display device's coordinate space. Correlates to pixels on the display.
-    //   - screen: The post-rotation coordinate space for the display, a.k.a. logical display space.
-    //   - layer: The coordinate space of this layer.
-    //   - input: The coordinate space in which this layer will receive input events. This could be
-    //            different than layer space if a surfaceInset is used, which changes the origin
-    //            of the input space.
-    const FloatRect inputBoundsInLayer = tmpBounds.toFloatRect();
-
-    // Clamp surface inset to the input bounds.
-    const auto surfaceInset = static_cast<float>(info.surfaceInset);
-    const float xSurfaceInset =
-            std::max(0.f, std::min(surfaceInset, inputBoundsInLayer.getWidth() / 2.f));
-    const float ySurfaceInset =
-            std::max(0.f, std::min(surfaceInset, inputBoundsInLayer.getHeight() / 2.f));
-
-    // Apply the insets to the input bounds.
-    const FloatRect insetBoundsInLayer(inputBoundsInLayer.left + xSurfaceInset,
-                                       inputBoundsInLayer.top + ySurfaceInset,
-                                       inputBoundsInLayer.right - xSurfaceInset,
-                                       inputBoundsInLayer.bottom - ySurfaceInset);
-
-    // Crop the input bounds to ensure it is within the parent's bounds.
-    const FloatRect croppedInsetBoundsInLayer = mBounds.intersect(insetBoundsInLayer);
-
-    const ui::Transform layerToScreen = getInputTransform();
-    const ui::Transform layerToDisplay = screenToDisplay * layerToScreen;
-
-    const Rect roundedFrameInDisplay{layerToDisplay.transform(croppedInsetBoundsInLayer)};
+    const Rect roundedFrameInDisplay = getInputBoundsInDisplaySpace(inputBounds, screenToDisplay);
     info.frameLeft = roundedFrameInDisplay.left;
     info.frameTop = roundedFrameInDisplay.top;
     info.frameRight = roundedFrameInDisplay.right;
     info.frameBottom = roundedFrameInDisplay.bottom;
 
     ui::Transform inputToLayer;
-    inputToLayer.set(insetBoundsInLayer.left, insetBoundsInLayer.top);
-    const ui::Transform inputToDisplay = layerToDisplay * inputToLayer;
+    inputToLayer.set(inputBounds.left, inputBounds.top);
+    const ui::Transform layerToScreen = getInputTransform();
+    const ui::Transform inputToDisplay = screenToDisplay * layerToScreen * inputToLayer;
 
     // InputDispatcher expects a display-to-input transform.
     info.transform = inputToDisplay.inverse();
@@ -2485,13 +2447,23 @@
         info.inputConfig |= WindowInfo::InputConfig::DROP_INPUT;
     }
 
-    auto cropLayer = mDrawingState.touchableRegionCrop.promote();
+    sp<Layer> cropLayer = mDrawingState.touchableRegionCrop.promote();
     if (info.replaceTouchableRegionWithCrop) {
-        const Rect bounds(cropLayer ? cropLayer->mScreenBounds : mScreenBounds);
-        info.touchableRegion = Region(displayTransform.transform(bounds));
+        Rect inputBoundsInDisplaySpace;
+        if (!cropLayer) {
+            FloatRect inputBounds = getInputBounds(/*fillParentBounds=*/true).first;
+            inputBoundsInDisplaySpace = getInputBoundsInDisplaySpace(inputBounds, displayTransform);
+        } else {
+            FloatRect inputBounds = cropLayer->getInputBounds(/*fillParentBounds=*/true).first;
+            inputBoundsInDisplaySpace =
+                    cropLayer->getInputBoundsInDisplaySpace(inputBounds, displayTransform);
+        }
+        info.touchableRegion = Region(inputBoundsInDisplaySpace);
     } else if (cropLayer != nullptr) {
-        info.touchableRegion = info.touchableRegion.intersect(
-                displayTransform.transform(Rect{cropLayer->mScreenBounds}));
+        FloatRect inputBounds = cropLayer->getInputBounds(/*fillParentBounds=*/true).first;
+        Rect inputBoundsInDisplaySpace =
+                cropLayer->getInputBoundsInDisplaySpace(inputBounds, displayTransform);
+        info.touchableRegion = info.touchableRegion.intersect(inputBoundsInDisplaySpace);
     }
 
     // Inherit the trusted state from the parent hierarchy, but don't clobber the trusted state
@@ -2513,6 +2485,27 @@
     return info;
 }
 
+Rect Layer::getInputBoundsInDisplaySpace(const FloatRect& inputBounds,
+                                         const ui::Transform& screenToDisplay) {
+    // InputDispatcher works in the display device's coordinate space. Here, we calculate the
+    // frame and transform used for the layer, which determines the bounds and the coordinate space
+    // within which the layer will receive input.
+
+    // Coordinate space definitions:
+    //   - display: The display device's coordinate space. Correlates to pixels on the display.
+    //   - screen: The post-rotation coordinate space for the display, a.k.a. logical display space.
+    //   - layer: The coordinate space of this layer.
+    //   - input: The coordinate space in which this layer will receive input events. This could be
+    //            different than layer space if a surfaceInset is used, which changes the origin
+    //            of the input space.
+
+    // Crop the input bounds to ensure it is within the parent's bounds.
+    const FloatRect croppedInputBounds = mBounds.intersect(inputBounds);
+    const ui::Transform layerToScreen = getInputTransform();
+    const ui::Transform layerToDisplay = screenToDisplay * layerToScreen;
+    return Rect{layerToDisplay.transform(croppedInputBounds)};
+}
+
 sp<Layer> Layer::getClonedRoot() {
     if (mClonedChild != nullptr) {
         return sp<Layer>::fromExisting(this);
@@ -3461,20 +3454,46 @@
 }
 
 /**
+ * Returns the bounds used to fill the input frame and the touchable region.
+ *
  * Similar to getInputTransform, we need to update the bounds to include the transform.
  * This is because bounds don't include the buffer transform, where the input assumes
  * that's already included.
  */
-Rect Layer::getInputBounds() const {
-    if (!hasBufferOrSidebandStream()) {
-        return getCroppedBufferSize(getDrawingState());
+std::pair<FloatRect, bool> Layer::getInputBounds(bool fillParentBounds) const {
+    Rect croppedBufferSize = getCroppedBufferSize(getDrawingState());
+    FloatRect inputBounds = croppedBufferSize.toFloatRect();
+    if (hasBufferOrSidebandStream() && croppedBufferSize.isValid() &&
+        mDrawingState.transform.getType() != ui::Transform::IDENTITY) {
+        inputBounds = mDrawingState.transform.transform(inputBounds);
     }
 
-    Rect bufferBounds = getCroppedBufferSize(getDrawingState());
-    if (mDrawingState.transform.getType() == ui::Transform::IDENTITY || !bufferBounds.isValid()) {
-        return bufferBounds;
+    bool inputBoundsValid = croppedBufferSize.isValid();
+    if (!inputBoundsValid) {
+        /**
+         * Input bounds are based on the layer crop or buffer size. But if we are using
+         * the layer bounds as the input bounds (replaceTouchableRegionWithCrop flag) then
+         * we can use the parent bounds as the input bounds if the layer does not have buffer
+         * or a crop. We want to unify this logic but because of compat reasons we cannot always
+         * use the parent bounds. A layer without a buffer can get input. So when a window is
+         * initially added, its touchable region can fill its parent layer bounds and that can
+         * have negative consequences.
+         */
+        inputBounds = fillParentBounds ? mBounds : FloatRect{};
     }
-    return mDrawingState.transform.transform(bufferBounds);
+
+    // Clamp surface inset to the input bounds.
+    const float inset = static_cast<float>(mDrawingState.inputInfo.surfaceInset);
+    const float xSurfaceInset = std::clamp(inset, 0.f, inputBounds.getWidth() / 2.f);
+    const float ySurfaceInset = std::clamp(inset, 0.f, inputBounds.getHeight() / 2.f);
+
+    // Apply the insets to the input bounds.
+    inputBounds.left += xSurfaceInset;
+    inputBounds.top += ySurfaceInset;
+    inputBounds.right -= xSurfaceInset;
+    inputBounds.bottom -= ySurfaceInset;
+
+    return {inputBounds, inputBoundsValid};
 }
 
 bool Layer::simpleBufferUpdate(const layer_state_t& s) const {
diff --git a/services/surfaceflinger/Layer.h b/services/surfaceflinger/Layer.h
index 2fb122c..9ed5899 100644
--- a/services/surfaceflinger/Layer.h
+++ b/services/surfaceflinger/Layer.h
@@ -570,6 +570,8 @@
 
     FloatRect getBounds(const Region& activeTransparentRegion) const;
     FloatRect getBounds() const;
+    Rect getInputBoundsInDisplaySpace(const FloatRect& insetBounds,
+                                      const ui::Transform& displayTransform);
 
     // Compute bounds for the layer and cache the results.
     void computeBounds(FloatRect parentBounds, ui::Transform parentTransform, float shadowRadius);
@@ -931,7 +933,7 @@
      * "replaceTouchableRegionWithCrop" is specified. In this case, the layer will receive input
      * in this layer's space, regardless of the specified crop layer.
      */
-    Rect getInputBounds() const;
+    std::pair<FloatRect, bool> getInputBounds(bool fillParentBounds) const;
 
     // constant
     sp<SurfaceFlinger> mFlinger;