diff --git a/src/protocols/core/Compositor.cpp b/src/protocols/core/Compositor.cpp
index e8503ae8..db6e3258 100644
--- a/src/protocols/core/Compositor.cpp
+++ b/src/protocols/core/Compositor.cpp
@@ -107,6 +107,7 @@ CWLSurfaceResource::CWLSurfaceResource(SP<CWlSurface> resource_) : resource(reso
 
         pending.damage.intersect(CBox{{}, pending.size});
 
+        auto    previousBuffer       = current.buffer;
         CRegion previousBufferDamage = accumulateCurrentBufferDamage();
 
         current = pending;
@@ -119,11 +120,12 @@ CWLSurfaceResource::CWLSurfaceResource(SP<CWlSurface> resource_) : resource(reso
             //    current.damage.copy().scale(current.scale).transform(current.transform, current.size.x, current.size.y).add(current.bufferDamage).add(previousBufferDamage);
             current.buffer->update(CBox{{}, {INT32_MAX, INT32_MAX}}); // FIXME: figure this out to not use this hack. QT apps are wonky without this.
 
-            // release the buffer, glTexImage2D is synchronous (as in, data is consumed after the call returns)
+            // release the buffer if it's synchronous as update() has done everything thats needed
             // so we can let the app know we're done.
-            // for dma buffers, this doesn't matter.
-            current.buffer->sendRelease();
-            bufferReleased = true;
+            if (current.buffer->isSynchronous()) {
+                current.buffer->sendRelease();
+                bufferReleased = true;
+            }
         }
 
         // TODO: we should _accumulate_ and not replace above if sync
@@ -146,6 +148,12 @@ CWLSurfaceResource::CWLSurfaceResource(SP<CWlSurface> resource_) : resource(reso
                 },
                 nullptr);
         }
+
+        // for async buffers, we can only release the buffer once we are unrefing it from current.
+        if (previousBuffer && !previousBuffer->isSynchronous() && !bufferReleased) {
+            previousBuffer->sendRelease();
+            bufferReleased = true;
+        }
     });
 
     resource->setDamage([this](CWlSurface* r, int32_t x, int32_t y, int32_t w, int32_t h) { pending.damage.add(CBox{x, y, w, h}); });
diff --git a/src/protocols/core/Shm.cpp b/src/protocols/core/Shm.cpp
index 0c01cf80..ce7d3a61 100644
--- a/src/protocols/core/Shm.cpp
+++ b/src/protocols/core/Shm.cpp
@@ -49,6 +49,10 @@ eBufferType CWLSHMBuffer::type() {
     return BUFFER_TYPE_SHM;
 }
 
+bool CWLSHMBuffer::isSynchronous() {
+    return true;
+}
+
 SSHMAttrs CWLSHMBuffer::shm() {
     SSHMAttrs attrs;
     attrs.success = true;
diff --git a/src/protocols/core/Shm.hpp b/src/protocols/core/Shm.hpp
index 862ea112..687e2031 100644
--- a/src/protocols/core/Shm.hpp
+++ b/src/protocols/core/Shm.hpp
@@ -37,6 +37,7 @@ class CWLSHMBuffer : public IWLBuffer {
     virtual eBufferCapability                      caps();
     virtual eBufferType                            type();
     virtual void                                   update(const CRegion& damage);
+    virtual bool                                   isSynchronous();
     virtual SSHMAttrs                              shm();
     virtual std::tuple<uint8_t*, uint32_t, size_t> beginDataPtr(uint32_t flags);
     virtual void                                   endDataPtr();
diff --git a/src/protocols/types/Buffer.hpp b/src/protocols/types/Buffer.hpp
index c9902f8d..9999a4e9 100644
--- a/src/protocols/types/Buffer.hpp
+++ b/src/protocols/types/Buffer.hpp
@@ -49,6 +49,7 @@ class IWLBuffer {
     virtual eBufferCapability                      caps()                        = 0;
     virtual eBufferType                            type()                        = 0;
     virtual void                                   update(const CRegion& damage) = 0;
+    virtual bool                                   isSynchronous()               = 0; // whether the updates to this buffer are synchronous, aka happen over cpu
     virtual SDMABUFAttrs                           dmabuf();
     virtual SSHMAttrs                              shm();
     virtual std::tuple<uint8_t*, uint32_t, size_t> beginDataPtr(uint32_t flags);
diff --git a/src/protocols/types/DMABuffer.cpp b/src/protocols/types/DMABuffer.cpp
index 930e71e6..f26328e9 100644
--- a/src/protocols/types/DMABuffer.cpp
+++ b/src/protocols/types/DMABuffer.cpp
@@ -43,6 +43,10 @@ void CDMABuffer::update(const CRegion& damage) {
     ;
 }
 
+bool CDMABuffer::isSynchronous() {
+    return false;
+}
+
 SDMABUFAttrs CDMABuffer::dmabuf() {
     return attrs;
 }
diff --git a/src/protocols/types/DMABuffer.hpp b/src/protocols/types/DMABuffer.hpp
index e06b5791..dac89493 100644
--- a/src/protocols/types/DMABuffer.hpp
+++ b/src/protocols/types/DMABuffer.hpp
@@ -9,6 +9,7 @@ class CDMABuffer : public IWLBuffer {
 
     virtual eBufferCapability                      caps();
     virtual eBufferType                            type();
+    virtual bool                                   isSynchronous();
     virtual void                                   update(const CRegion& damage);
     virtual SDMABUFAttrs                           dmabuf();
     virtual std::tuple<uint8_t*, uint32_t, size_t> beginDataPtr(uint32_t flags);