diff --git a/libs/gui/BufferItem.cpp b/libs/gui/BufferItem.cpp
index 312fb3b..239da20 100644
--- a/libs/gui/BufferItem.cpp
+++ b/libs/gui/BufferItem.cpp
@@ -64,6 +64,8 @@
         c += mFence->getFlattenedSize();
         FlattenableUtils::align<4>(c);
     }
+    c += mSurfaceDamage.getFlattenedSize();
+    FlattenableUtils::align<4>(c);
     return sizeof(int32_t) + c + getPodSize();
 }
 
@@ -105,6 +107,9 @@
         size -= FlattenableUtils::align<4>(buffer);
         flags |= 2;
     }
+    status_t err = mSurfaceDamage.flatten(buffer, size);
+    if (err) return err;
+    size -= FlattenableUtils::align<4>(buffer);
 
     // check we have enough space (in case flattening the fence/graphicbuffer lied to us)
     if (size < getPodSize()) {
@@ -148,6 +153,9 @@
         if (err) return err;
         size -= FlattenableUtils::align<4>(buffer);
     }
+    status_t err = mSurfaceDamage.unflatten(buffer, size);
+    if (err) return err;
+    size -= FlattenableUtils::align<4>(buffer);
 
     // check we have enough space
     if (size < getPodSize()) {
diff --git a/libs/gui/BufferQueueProducer.cpp b/libs/gui/BufferQueueProducer.cpp
index 4c22ba3..6452cdd 100644
--- a/libs/gui/BufferQueueProducer.cpp
+++ b/libs/gui/BufferQueueProducer.cpp
@@ -525,6 +525,7 @@
     sp<Fence> fence;
     input.deflate(&timestamp, &isAutoTimestamp, &dataSpace, &crop, &scalingMode,
             &transform, &async, &fence, &stickyTransform);
+    Region surfaceDamage = input.getSurfaceDamage();
 
     if (fence == NULL) {
         BQ_LOGE("queueBuffer: fence is NULL");
@@ -621,6 +622,7 @@
         item.mSlot = slot;
         item.mFence = fence;
         item.mIsDroppable = mCore->mDequeueBufferCannotBlock || async;
+        item.mSurfaceDamage = surfaceDamage;
 
         mStickyTransform = stickyTransform;
 
diff --git a/libs/gui/IGraphicBufferProducer.cpp b/libs/gui/IGraphicBufferProducer.cpp
index a3e6fb2..b7982a9 100644
--- a/libs/gui/IGraphicBufferProducer.cpp
+++ b/libs/gui/IGraphicBufferProducer.cpp
@@ -446,7 +446,8 @@
          + sizeof(transform)
          + sizeof(stickyTransform)
          + sizeof(async)
-         + fence->getFlattenedSize();
+         + fence->getFlattenedSize()
+         + surfaceDamage.getFlattenedSize();
 }
 
 size_t IGraphicBufferProducer::QueueBufferInput::getFdCount() const {
@@ -467,7 +468,11 @@
     FlattenableUtils::write(buffer, size, transform);
     FlattenableUtils::write(buffer, size, stickyTransform);
     FlattenableUtils::write(buffer, size, async);
-    return fence->flatten(buffer, size, fds, count);
+    status_t result = fence->flatten(buffer, size, fds, count);
+    if (result != NO_ERROR) {
+        return result;
+    }
+    return surfaceDamage.flatten(buffer, size);
 }
 
 status_t IGraphicBufferProducer::QueueBufferInput::unflatten(
@@ -497,7 +502,11 @@
     FlattenableUtils::read(buffer, size, async);
 
     fence = new Fence();
-    return fence->unflatten(buffer, size, fds, count);
+    status_t result = fence->unflatten(buffer, size, fds, count);
+    if (result != NO_ERROR) {
+        return result;
+    }
+    return surfaceDamage.unflatten(buffer, size);
 }
 
 }; // namespace android
diff --git a/libs/gui/Surface.cpp b/libs/gui/Surface.cpp
index b80890f..245f7a3 100644
--- a/libs/gui/Surface.cpp
+++ b/libs/gui/Surface.cpp
@@ -27,6 +27,7 @@
 #include <utils/NativeHandle.h>
 
 #include <ui/Fence.h>
+#include <ui/Region.h>
 
 #include <gui/IProducerListener.h>
 #include <gui/ISurfaceComposer.h>
@@ -320,6 +321,25 @@
     IGraphicBufferProducer::QueueBufferInput input(timestamp, isAutoTimestamp,
             mDataSpace, crop, mScalingMode, mTransform ^ mStickyTransform,
             mSwapIntervalZero, fence, mStickyTransform);
+
+    if (mDirtyRegion.bounds() == Rect::INVALID_RECT) {
+        input.setSurfaceDamage(Region::INVALID_REGION);
+    } else {
+        // The surface damage was specified using the OpenGL ES convention of
+        // the origin being in the bottom-left corner. Here we flip to the
+        // convention that the rest of the system uses (top-left corner) by
+        // subtracting all top/bottom coordinates from the buffer height.
+        Region flippedRegion;
+        for (auto rect : mDirtyRegion) {
+            auto top = buffer->height - rect.bottom;
+            auto bottom = buffer->height - rect.top;
+            Rect flippedRect{rect.left, top, rect.right, bottom};
+            flippedRegion.orSelf(flippedRect);
+        }
+
+        input.setSurfaceDamage(flippedRegion);
+    }
+
     status_t err = mGraphicBufferProducer->queueBuffer(i, input, &output);
     if (err != OK)  {
         ALOGE("queueBuffer: error queuing buffer to SurfaceTexture, %d", err);
@@ -336,6 +356,9 @@
 
     mConsumerRunningBehind = (numPendingBuffers >= 2);
 
+    // Clear surface damage back to full-buffer
+    mDirtyRegion = Region::INVALID_REGION;
+
     return err;
 }
 
@@ -453,6 +476,9 @@
     case NATIVE_WINDOW_SET_BUFFERS_DATASPACE:
         res = dispatchSetBuffersDataSpace(args);
         break;
+    case NATIVE_WINDOW_SET_SURFACE_DAMAGE:
+        res = dispatchSetSurfaceDamage(args);
+        break;
     default:
         res = NAME_NOT_FOUND;
         break;
@@ -556,6 +582,13 @@
     return setBuffersDataSpace(dataspace);
 }
 
+int Surface::dispatchSetSurfaceDamage(va_list args) {
+    android_native_rect_t* rects = va_arg(args, android_native_rect_t*);
+    size_t numRects = va_arg(args, size_t);
+    setSurfaceDamage(rects, numRects);
+    return NO_ERROR;
+}
+
 int Surface::connect(int api) {
     static sp<IProducerListener> listener = new DummyProducerListener();
     return connect(api, listener);
@@ -582,7 +615,13 @@
     }
     if (!err && api == NATIVE_WINDOW_API_CPU) {
         mConnectedToCpu = true;
+        // Clear the dirty region in case we're switching from a non-CPU API
+        mDirtyRegion.clear();
+    } else if (!err) {
+        // Initialize the dirty region for tracking surface damage
+        mDirtyRegion = Region::INVALID_REGION;
     }
+
     return err;
 }
 
@@ -800,6 +839,27 @@
     }
 }
 
+void Surface::setSurfaceDamage(android_native_rect_t* rects, size_t numRects) {
+    ATRACE_CALL();
+    ALOGV("Surface::setSurfaceDamage");
+    Mutex::Autolock lock(mMutex);
+
+    if (numRects == 0) {
+        mDirtyRegion = Region::INVALID_REGION;
+        return;
+    }
+
+    mDirtyRegion.clear();
+    for (size_t r = 0; r < numRects; ++r) {
+        // We intentionally flip top and bottom here, since because they're
+        // specified with a bottom-left origin, top > bottom, which fails
+        // validation in the Region class. We will fix this up when we flip to a
+        // top-left origin in queueBuffer.
+        Rect rect(rects[r].left, rects[r].bottom, rects[r].right, rects[r].top);
+        mDirtyRegion.orSelf(rect);
+    }
+}
+
 // ----------------------------------------------------------------------
 // the lock/unlock APIs must be used from the same thread